Democracy and Development

Is economic development conducive to political democracy? Does

democracy foster or hinder material welfare? These two questions
are examined by looking at the experiences of 135 countries
between 1950 and 1990. Descriptive information, statistical analy-
ses, and historical narratives are interwoven to gain an under-
standing of the dynamics of political regimes and their impact on
economic development and other aspects of material welfare. The
findings, several of them quite surprising, dispel any notion of a
trade-off between democracy and development. Economic devel-
opment does not tend to generate democracies, but democracies
are much more likely to survive in wealthy societies. The type of
political regime has no impact on the growth of total national
income, and political instability affects growth only in dictator-
ships. Per capita incomes rise more rapidly in democracies
because populations increase faster under dictatorships. In
general, political regimes have greater effects on demography than
on economics.

Adam Przeworski is Carroll and Milton Petrie Professor in the

Department of Politics at New York University.

Michael E. Alvarez is Assistant Professor of Political Science at

DePaul University.

José Antonio Cheibub is Assistant Professor of Political Science at

Yale University.

Fernando Limongi is Associate Professor of Political Science at the

University of São Paulo.
 CAMBRIDGE STUDIES IN THE THEORY OF
DEMOCRACY

General Editor
ADAM PRZEWORSKI New York University

OTHER BOOKS IN THE SERIES

Jon Elster, ed., Deliberative Democaracy
Adam Przeworski, Susan Stokes, and Bernard Manin, eds.,
Democracy, Accountability, and Representation
Democracy and Development
Political Institutions and
Well-Being in the World,
1950–1990

Adam Przeworski
New York University
Michael E. Alvarez
DePaul University
José Antonio Cheibub
Yale University
Fernando Limongi
University of São Paulo
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo

Cambridge University Press

The Edinburgh Building, Cambridge CB2 2RU, UK

Published in the United States of America by Cambridge University Press, New York

www.cambridge.org
Information on this title: www.cambridge.org/9780521790321

© Adam Przeworski, Michael E. Alvarez, José Antonio Cheibub, and Fernando

Limongi 2000

This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without
the written permission of Cambridge University Press.

First published 2000

Reprinted 2002, 2003

A catalogue record for this publication is available from the British Library

ISBN 13 978 0 521 79032 1 hardback

ISBN 10 0 521 79032 8 hardback

ISBN 13 978 0 521 79379 7 paperback

ISBN 10 0 521 79379 3 paperback

Transferred to digital printing 2005

 Contents

List of Tables and Figures page xi

Acknowledgments xvii
Introduction 1
Questions 1
Methods 7
Plan 10

1 Democracies and Dictatorships 13

Introduction 13
Democracy and Dictatorship 14
Operational Rules: Filling Offices by Contested
Elections 18
“Botswana” and the Alternation Rule 23
Summary of Rules 28
Distinguishing among Democracies and Dictatorships 30
What We Did Not Include 33
Stability and Change of Political Regimes 36
Stability and Change of Political Leadership 51
Conclusion 54
Appendix 1.1: Alternative Approaches 55
Appendix 1.2: Classification of Political Regimes,
1950–1990 59
Appendix 1.3: Basic Data about Regime Dynamics 69
Appendix 1.4: The “Short” Data Base 77

2 Economic Development and Political Regimes 78

Introduction 78
Development and Democracy 78
Regime Dynamics 88

vii
 Contents

Level of Economic Development and Regime Dynamics 92

Economic Growth and Regime Dynamics 106
Income Inequality 117
Economic Factors in Context 122
Democratic Institutions and the Sustainability of
Democracy 128
Conclusion 136
Appendix 2.1: Dynamic Probit Model 137
Appendix 2.2: Survival Models 139

3 Political Regimes and Economic Growth 142

Introduction 142
Political Regimes and Economic Growth 146
Poor and Wealthy Countries 158
Miracles and Disasters 176
Conclusion 178
Appendix 3.1: Identification, Specification, and
Robustness 179
Appendix 3.2: Estimating the Contribution of Public
Investment 184

4 Political Instability and Economic Growth 187

Introduction 187
Contemporaneous Political Upheavals 189
Past Instability 193
Expectations 200
Conclusion 211
Appendix 4.1: Estimating the Impact of Expectations
on Economic Performance 213

5 Political Regimes and Population 216

Total Income, Population, and Per Capita Income 216
Political Regimes and Population Growth 218
Explanations 233
Human Capital, Fertility, and Economic Growth 256
Conclusion 264
Appendix 5.1: Calculating Age Structure 265

Conclusion 269
Appendix C.1: Simulation 278

viii
Contents

Appendix I: Selection Model 279

Selection as a Problem of Identification 279
Statistical Models of Selection 281
Statistical Models of Performance 285
Estimating Selection-Corrected Models of
Performance 287
Appendix II: Codebook 291
References 301
Author Index 313
Country Index 316
Subject Index 320

ix
 Tables and Figures

Tables
1.1 Distribution of observations by criteria used
for classification as dictatorships page 30
1.2 Number of countries, old and new, by year 37
1.3 Democracies and dictatorships in old and
new countries 39
1.4 Expected and observed numbers of transitions,
by period 45
1.5 Regime duration: average age in years at
end of regime spell or last year of observation 46
1.6 Regimes and regime transitions by region 49
1.7 Transitions between political regimes:
parliamentarism, mixed, presidentialism,
bureaucracies, and autocracies 51
1.8 Distribution of changes of chief executives
(HEADS) by regime 52
1.9 Average duration (in years) of chief executives’
spells by regime 53
1.10 Leadership turnover rates by regime type 54
2.1 Predictions of probit models of regimes 82
2.2 Incorrect predictions, by country and period,
and the probability that a regime is a
dictatorship during the period 84
2.3 Transitions, by lagged per capita income
(LEVLAG) 93
2.4 Highest levels of per capita income (LEVEL)
under which dictatorships survived in different
countries 95

xi
 Tables and Figures

2.5 Countries that developed over long periods

under dictatorship and reached incomes
above $4,115 96
2.6 Transitions to dictatorship, 1951–1990, by last
full year of democracy, per capita income, and
type of democracy 100
2.7 Regime transitions, by lagged per capita income
and average education 102
2.8 Countries (in our sample) that experienced
democracy before 1950 104
2.9 Democracies that lasted at least 20 years, by per
capita income and by income distribution 107
2.10 Observed rates of transition, by lagged per
capita income (LEVLAG) and lagged rate of
economic growth (GLAG) 110
2.11 Regime transitions, by longer-term dynamics of
per capita income 111
2.12 Dynamic probit analysis of the impact of growth
on regime survival 113
2.13 Incidence of political mobilization, by longer-term
dynamics of per capita income 115
2.14 Growth and mobilization during the year
preceding regime transition 116
2.15 Income inequality and regime dynamics 118
2.16 Regime transitions, by level and labor share 122
2.17 Dynamic probit analysis of regime transitions 124
2.18 Impact of intra-regime instability (turnover of
heads of governments) on regime transition
rates 125
2.19 Observed rates of transition to dictatorship,
by lagged per capita income (LEVLAG) and
lagged rate of economic growth (GLAG), of
parliamentary and presidential democracies and
the hypothetical numbers of transitions if there
had been as many presidential democracies as
parliamentary democracies under all
circumstances 130
2.20 Dynamic probit analysis of regime transitions,
by type of democracy 133
2.21 Hazard rates for parliamentary and presidential
democracies, by political conditions 135

xii
Tables and Figures

3.1A Investment share in GDP (INV), by per capita income

(LEVEL) 147
3.1B Selection-corrected estimators of investment share
(INV) 148
3.2A Rate of growth of capital stock (KSG), by per capita
income (LEVEL) 149
3.2B Selection-corrected estimators of the rate of growth of
capital stock (KSG) 150
3.3A Rate of growth of labor force (LFG), by per capita
income (LEVEL) 151
3.3B Selection-corrected estimators of labor force growth
(LFG) 152
3.4 Barebones model, by low and high per capita incomes,
decomposed by sources of growth, with LFG taken as
exogenous 154
3.5 Selection-corrected estimators of the rate of growth of
income (YG) 157
3.6 Rate of growth of GDP (YG), by per capita income
(LEVEL) 159
3.7 Per capita income at the beginning and end of the
period, by bands of $1,000 163
3.8A Labor exploitation (labor-share sample, N = 2,061) 168
3.8B Accounting for the difference in wages 169
3.9 Miracles and disasters (fastest- and slowest-growing
regimes, over periods of at least ten years) 177
A3.1.1 Robustness 183
A3.1.2 Selection-corrected estimators of the rate of growth of
income (YG), under modified classification of regimes 184
A3.2.1 Coefficients (W) on public capital, by capital per worker 186
4.1 Selection-corrected effects of socio-political unrest on
growth 192
4.2 Growth of income (YG), by the number of past regime
transitions (STTR) and the current regime (REG) 194
4.3A Growth of income (YG), by the number of past
changes of chief executives (HEADS) and the
current regime (REG) 195
4.3B Growth of income (YG), by the frequency of
past changes of chief executives (TURNOVER) and the
current regime (REG) 195
4.4 Effect of regime age (AGER) on the growth of income
(YG); selection-corrected panel estimates 197

xiii
 Tables and Figures

4.5 Effects of expected hazards to regime (EXPHAZR) and

to heads of governments (EXPHAZH) on the growth of
income (YG); selection-corrected two-way (country
and year) fixed-effects estimates 202
4.6 Effects of expected hazards to regime (EXPHAZR) and
to heads of governments (EXPHAZH) on the share of
investment in GDP (INV); selection-corrected, AR1,
one-way (country) fixed-effects estimates 207
A4.1.1 Alternative estimators of the effects of hazards on
growth and investment 214
5.1A Observed rates of population growth (POPG), by per
capita income (LEVEL) 219
5.1B Selection-corrected estimators of population growth
(POPG) 220
5.2 Observed diachronic effects of regimes on population
growth 222
5.3A Observed rates of population growth under
dictatorship and democracy (small sample) 223
5.3B Observed birth and death rates under dictatorship and
democracy 227
5.3C Observed infant mortality (INFMORT) and life
expectancy (LIVES) under dictatorship and democracy 229
5.3D A decomposition of death rates under dictatorship and
democracy 230
5.4 Selection-corrected rates of births and deaths and of
life expectancy 232
5.5 Observed diachronic effects of regimes on births,
fertility, deaths, and life expectancy 233
5.6A Effect of regimes on the logarithm of death rates 238
5.6B Effect of regimes on the logarithm of death rates 240
5.7A Observed patterns of contraception and fertility, by per
capita income and regime 245
5.7B Summary of selection-corrected regressions of fertility,
by regime 250
5.8 Observed enrollment ratios, by gender, under
dictatorship and democracy 252
5.9 Impact of fertility on the growth of per capita
income 262
5.10 Countries in the trap and their long-term growth 263
C.1 Average values of selected indicators observed or
predicted during various years 275

xiv
Tables and Figures

C.2 Distribution of countries in entry year, exit year,

and 2030 276

Figures
1.1 Proportions of democracies in the world, in old
countries, and in new countries 41
1.2 Proportions of democracies in old countries: world,
Latin America, and other regions 43
1.3 Transitions to democracy and to dictatorship by year 44
2.1 Probability that a regime will be democratic, by per
capita income 80
3.1 Labor shares in Singapore and Austria at similar
levels of per capita income from year to year 172
3.2 Labor shares in Mexico and Portugal at similar levels
of per capita income from year to year 174
3.3 Argentina’s labor share by regime 175
5.1 Demographic transitions by per capita income and
regime 225
5.2 Growth of countries in and out of the trap 258
5.3A Countries in the trap 260
5.3B Countries out of the trap 261

xv
 Acknowledgments

Because several of these chapters were circulated independently and

were presented as papers at various conferences, they accumulated
numerous comments. As a result, our intellectual debts are so exten-
sive that we can no longer be sure to whom we owe which inspiration
and which criticism. Hence, with a profound apology to all those who
generously taught us economics, suggested questions to be studied,
offered alternative explanations, or pointed out silly mistakes, we are
unable to list them.
Much of the work on this project was done when we were all at the
University of Chicago.
This project was supported in part by grants from the National
Science Foundation (SES-9022605 and SES-9311793). Fernando
Limongi was supported by a fellowship from the Fundacão de Amparo
à Pesquisa do Estado de São Paulo (FAPESP).

xvii
 Introduction

Questions
Is economic development conducive to political democracy? Does
democracy foster or hinder material welfare? These have been central
questions on intellectual and political agendas over the past fifty years,
ever since the Atlantic Charter, signed by Churchill and Roosevelt in
1941, offered the “assurance that all men in all the lands might live
out their lives in freedom from fear and want.” For the first time in
history, democracy and development, freedom from fear and from
want, were conceived as the future of all the people in the world, not
as the privilege of the “civilized” nations.
In some ways, these questions are badly formulated. Political
regimes, however one thinks about them, are complex. They combine
many institutional features that can have emergent effects and that
may work at cross-purposes. They may, at the same time, encourage
economic rationality but hinder economic initiative, grant govern-
ments the authority necessary to promote development but also allow
them to evade popular control, and foster long-term thinking at the
cost of short-term disasters, or vice versa. Development, in turn, is a
multifaceted process of structural transformations, not only economic,
that becomes manifest in the growth of income, productivity, con-
sumption, investment, education, life expectancy, and employment –
all that makes for a better life. But again, these good things do not
necessarily go together, and they certainly do not go together for all.
Average income and even consumption can grow at the cost of in-
creasing unemployment, of growing inequality, of immiseration for
large segments of the population, of degradation of the environment.
Hence, the question of the relationship between democracy and devel-
opment encompasses more specific issues concerning the impacts of

1
 Introduction

particular features of political regimes on various aspects of economic

performance.
Nevertheless, asking these questions in their general form is
inescapable. Following World War II, various dictatorial regimes
appealed to the masses of the poor by presenting themselves as forces
for progress, as agents of development, as shortcuts to modernity. Their
claims to legitimacy, their appeals to loyalty, were that they were
uniquely capable of mobilizing resources and energies to break the
chains of poverty, to build a better future, to lead their respective coun-
tries to affluence, power, and prestige. Whatever their particular
ideological stripes, such regimes plastered walls and minds with
images that pictured how everything – homes, schools, hospitals,
armies – would grow in the radiant future. They were to eradicate
poverty, generate affluence, enable their countries to assume their
rightful places among the powers of the world, and by the example of
their own success convert others to the righteousness of their dictato-
rial ways. They were always the “tigers”: The most rapidly growing
country in the world in the 1950s, at least if we are to believe its own
statistics, was communist Romania. The economic miracle of the early
1970s was the military-ruled Brazil. The economic tigers of the 1980s
were the dictatorships of Singapore, South Korea, and Taiwan. In the
1990s the leader was China. And those spectacular successes re-
peatedly sowed doubt in the minds of even committed democrats:
whether or not development did indeed require order and discipline,
whether or not rationality would flow from authority, whether or not
democrats could continue to trust that their own ways would be
capable of lifting the masses of the world’s poor from their plight.
Around 1960, when decolonialization was giving birth to many “new
nations,” in an international context in which communist regimes
still appeared to be developing impressively, many scholars and po-
liticians concluded that the perceived economic effectiveness of dicta-
torships was simply a fact of life, one that should be confronted
courageously by admitting that democracy was a luxury that could be
afforded only after the hard task of development had been accom-
plished. To cite just a few typical voices of the time, Galenson (1959:
3) claimed that “the more democratic a government is, . . . the greater
the diversion of resources from investment to consumption.” De
Schweinitz (1959) argued that if the less-developed countries “are
to grow economically, they must limit democratic participation in
political affairs.” La Palombara (1963: 57) thought that “if economic
development is the all-embracing goal, the logic of experience dictates

2
Introduction

that not too much attention can be paid to the trappings of demo-
cracy.” Dictatorships were needed to generate development: As
Huntington and Nelson (1976: 23) put it, “political participation must
be held down, at least temporarily, in order to promote economic
development.”1
The world thus faced, it seemed, a trade-off between democracy and
development. Yet the future was not so bleak for democracy. Whereas
dictatorships, even though despicable, were needed to generate devel-
opment, they would self-destroy as a result of their own success.
According to the dominant canon of the time, democracy would natu-
rally emerge after a society had undergone necessary economic and
social transformations. The basic assumption of modernization theory
was that societies undergo one general process, of which democrati-
zation is but the final facet. Hence, the emergence of democracy would
be an inexorable consequence of development. Because in that view
dictatorships would generate development, and development would
lead to democracy, the best route to democracy was seen as a circuitous
one. The policy prescriptions that resulted from that mode of thinking
rationalized support for dictatorships, at least those that were “capable
of change,” that is, anti-communist ones.
Communism now appears dead, and the idea that it ever portended
the future seems ludicrous, albeit in omniscient retrospect. And democ-
racy has been discovered to have many economic virtues. It is claimed
that democracy encourages investment by safeguarding property
rights, promotes allocational efficiency by allowing a free flow of ideas,
and prompts governments to choose good policies by imposing the
threat of electoral sanctions. Yet doubts remain. For many, Pinochet’s
Chile was the paradigm of successful economic reforms, and the eco-
nomic success of authoritarian China is the model for the new Russia.
Even though democratic ideals are nourishing political forces from
Argentina to Mongolia, the allure of a “strong government,” “insulated
from pressures,” guided by technical rationality, capable of imposing
order and discipline, continues to seduce. Whether in the case of the
Tiananmen Square massacre or the autogolpe of President Alberto

1
Still in 1975, impressed by the growth of the communist countries, Huntington and
Dominguez (1975: 60) observed that “the interest of the voters generally [leads] parties
to give the expansion of personal consumption a higher priority vis-à-vis investment
than it would receive in a non-democratic system. In the Soviet Union, for instance, the
percentage of GDP devoted to consumption was driven down from 65% in 1928 to 52%
in 1937. It is most unlikely that a competitive party system would have sustained a rev-
olution from above like this.”

3
 Introduction

Fujimori in Peru, international financial institutions, as well as gov-

ernments in the developed countries, are still willing to close their eyes
to violations of democratic rights and even human rights in deference
to the purported economic effectiveness of dictatorships. A myopic
public has to be protected from itself; economic reforms must be
carried out before the people have time to react. This is still a popular
belief.
Thus, whether well formulated or not, the question of the relative
merits of political regimes continues to evoke intellectual analysis as
well as political passion. Much of the appeal of dictatorships stems from
the fact that at various moments they have seemed to offer “the best
practice.” The “tigers” have tended to be dictatorships. But are dicta-
torships necessarily tigers? The list of disasters generated by authori-
tarianism is long and tragic. Even the economic collapse of communism
pales in comparison with the destruction caused by dictatorships in
many African countries, the squandering of resources in the Middle
East, or the havoc spawned by military governments in Central
America. For every developmental miracle, there have been several
dictatorships that have engaged in grandiose projects that have
ended in ruin, or else dictatorships that have simply stolen and squan-
dered. In turn, the record of performance among the democracies,
which has featured examples of spectacular growth (notably in
Western Europe until the mid-1970s) as well as of rapid deterioration
(as in Latin America in the 1980s), has not reached the extremes
seen among the dictatorships. Hence, to assess the impacts of
political regimes, we must examine their full record, not just the best
performances.
We want to know the effects of political regimes on the material
well-being of the people who live under them. But well-being, like its
cognates, such as welfare, utility, ophelimity, or simply happiness, is
an elusive concept. It combines objective and subjective elements: the
resources and capabilities that enable people to lead the lives they
choose, the choices they actually make, what Sen (1988) calls “func-
tionings,” as well as their evaluations of their lives. Well-being entails
being able to work and to consume, being sufficiently educated to
know what choices one can make in life, being able to choose the
number of children one wants, being able to live a healthy life. But
whereas the conditions that people face independently of their actions
can be measured objectively, well-being is difficult to assess in inter-
personal terms when these conditions do afford people the opportu-

4
Introduction

nity to make choices. To borrow an example from Sen (1988: 18), one
may starve because one does not have enough to eat or because
one is fasting. Conditions per se are often difficult to distinguish
from the consequences of the choices that people make under those
conditions.
Nevertheless, though well-being is difficult to evaluate, destitution is
manifest (Dasgupta 1993). Even such a rough measure as income can
tell us whether or not people are materially deprived. A life without
adequate food and shelter, threatened by disease, condemned to igno-
rance, is a life of destitution: The most subjectivist economist needs but
a few hours to determine that what a poor village needs is clean water
and a school.
We are willing to make judgments about the degrees of well-being
that people experience under different political regimes because we do
not believe that preferences about those aspects we study are endoge-
nous to regimes. Whether living under dictatorship or democracy,
people do not want to consume less, live shorter lives, get less educa-
tion, or see their children die. Yet even if preferences were endogenous
(e.g., suppose that mothers would happily offer their sons as cannon
fodder for dictators), such preferences could not justify normative
judgments unless they were informed by the actual experience of the
alternatives.
What, then, are the observable indications of well-being? We do not
neglect the intrinsic value of political and civic liberties inherent in
democracy. As Dasgupta (1993: 47) argues, the view that the poor do
not care about the freedoms associated with democracy “is a piece of
insolence that only those who don’t suffer from their lack seem to enter-
tain” (see also Sen 1994a). Yet our question is whether or not these
liberties, those necessary for people to freely choose their rulers, affect
well-being in other realms. This is why we refer to “material” well-
being. We are interested in the material consequences of political
regimes.
Dasgupta (1993: 76) proposes that, besides political liberties, one
can think of well-being along three dimensions: income and all it
affords, health, and education. With all the innumerable caveats,
income is simply the best overall indicator of the choices people enjoy
in their lives. It is in many ways far from perfect: Even if its com-
ponents were weighted at competitive market prices, those prices
would necessarily ignore the capacitating effects of certain forms of
consumption, as well as the very capacity to make choices that income

5
 Introduction

affords.2 Yet, at least when it is accurately computed, more income is

better than less: Although income is only a means to well-being, it
determines what bundles of consumption a person can choose.
One way, therefore, to assess the effects of regimes on well-being is
to examine their impacts on economic growth, something about which
people all around the world care.3 The monetary value of consumption,
as distinct from total income (which also includes savings), is an alter-
native measure. We shall ask under which regime economies develop
faster, which is more likely to generate miracles and which disasters,
which is more likely to ensure that development will be sustained,
which is more apt to exploit advantageous conditions, and which is
more adept at coping with adversities. We shall examine the impacts
of political regimes on the growth of per capita income and per capita
consumption, on productivity, on investment, and on the growth of
employment. We shall also attempt to say something about income dis-
tribution, about poverty, and about the material security that these
regimes generate, but given the paucity and poor quality of the data,
there is little we can report with certainty.
Yet people’s lives under democracy and dictatorship need not be the
same even if those people have the same incomes. As Sen (1993: 40)
put it, if “the economic success of a nation is judged only by income
and by other traditional indicators of opulence and financial sound-
ness, as it so often is, the important goal of well-being is missed.” Birth
rates, death rates, fertility, infant mortality, life expectancy, and school
enrollment rates conjure an image of what one can expect of life under
either regime. Several of these indicators focus specifically on the
conditions of women.
These are, then, the aspects of material well-being we examine. Our
approach is largely inductive: We consider almost all of the countries
that existed at any time between 1950 and 1990 and classify their
political regimes. We then examine why so many rich countries enjoy

2
For a discussion of income as an indicator of well-being, see Dasgupta (1993: chap. 7).
The UNDP Human Development Report provides an index that aggregates several indi-
cators into a single measure of “human development” (HDI). Although that may be a
useful summary measure, we prefer to study its components separately.
3
Average satisfaction with life rises steeply as the per capita incomes of countries increase
from low levels. The same is true of the relationship between income and life satisfac-
tion observed among individuals within a given country. In turn, it appears that the
average satisfaction level does not increase as the average per capita income of a country
increases over time, but this latter observation is based on countries that already have
high incomes. For a recent summary of these findings and an interpretation, see Frank
(1997).

6
Introduction

democracy and so very few poor countries do. Armed with knowledge
of how political regimes rise and fall, we compare their performances
in different realms.
Many of the patterns we have discovered are surprising. Indeed, our
findings contradict so many preconceived notions that at times we have
a feeling that we are digging a cemetery for old theories. Some find-
ings are so puzzling that they leave us speechless. Others lend them-
selves to rival explanations that cannot be evaluated with the available
data. All these findings do call for explanations, and we hope that
others will try to understand them. But our own posture is largely min-
imalist: We simply try to establish what one should reasonably believe
about the experience of the forty years we examine, the “facts.”

Methods
Democracy and dictatorship constitute different ways of organizing
political lives: of selecting rulers, processing conflicts, making and
implementing public decisions. We want to know if these ways make
a difference for people’s lives. And this means that we need to observe
democracies and dictatorships wherever they are, under the full range
of conditions under which they have existed. To speak of “democracy,”
one must reach beyond the experience of the industrialized countries.
Democracies breathe also in countries where it does not snow: Brazil,
Mauritius, Jamaica. Indeed, when we look for democratic regimes that
existed during any time between 1950 and 1990, we find two-thirds
of them in the less-developed countries. So if we want to know about
the impact of “democracy,” rather than of snow, we must study all
democracies we can find, wherever they are.
Yet knowledge of all the democracies that ever existed is not enough
to ascertain the impact of democracy. Because this methodological
point is crucial for all that follows, we need to explain the logic of our
analysis. We are not the first to study the impact of regimes on eco-
nomic performance. Indeed, there have been at least forty empirical
studies of this subject, the first dating as far back as 1966. Yet if we
think that we have something new to say, it is because we believe that
all of those studies employed a faulty methodology and drew invalid
conclusions.
Our aim is to assess the effects of political regimes on material well-
being. How can we go about it? Take the growth of per capita income.
We observe Chile in 1985 and discover that its regime was authori-
tarian and that its per capita income grew at the rate of 2.87 percent.

7
 Introduction

But what we want to know is whether Chile would have grown faster
or slower had its regime been a democracy in 1985. And the infor-
mation we have does not suffice to answer this question. Unless we
know what would have been the growth of Chile in 1985 had it been
democratic, how can we tell if it would have grown faster or slower
than it did under dictatorship?
Had we observed in 1985 a Chile that was simultaneously a dicta-
torship and a democracy, we would have the answer. But that is not
possible. Still, there seems to be a way out: We could look for some
case that was exactly like 1985 Chile in all respects other than its
regime and, perhaps, its rate of growth, and we could match this
country with Chile. We could then compare the growth of dictatorial
Chile in 1985 with the performance of its democratic match and draw
a conclusion.
But if the fact that Chile was a dictatorship in 1985 had some causes
in common with the fact that it grew at the rate observed, then a case
that matches Chile in all respects other than the regimes and the rate
of growth may not be found. Among all the observed cases, there will
be observations without a match, or, stated differently, we will not be
able to observe dictatorships under the same conditions as democra-
cies. And this means that we will not be able to distinguish whether
the observed differences in economic performance are due to regimes
or to the circumstances under which they found themselves. Conclu-
sions based on observed cases, even on all that can be observed, will
be then invalid.
Imagine the following thought (or computer) experiment. Assume
that the expected rate of growth is the same for all countries regard-
less of their regimes and that the growth of each country during each
year differs from the mean only by a random error. Assume further
that democracies are more likely to die, that is, to become dictator-
ships, when economic conditions are bad. Then it will turn out that,
on the average, countries observed as democracies will grow faster
than dictatorships. Regression, a conditional mean, will show the
same: The coefficient on regimes will show a difference in favor of
democracies. And because we have just assumed that the rate of
growth does not vary across regimes, the observed difference will be
due purely to the bias caused by the non-random selection of regimes.
The inference that democracy promotes growth will be thus fallacious:
We will observe this difference because democracies disappeared
whenever they faced economic crises while dictatorships survived
them, not because regimes made a difference for growth. Under these

8
Introduction

conditions, relying exclusively on the observed values will lead us

astray. Our inferences will suffer from “selection bias.”
The lesson is that appearances can be misleading. Comparing means
of the observed cases or performing standard regression on these
observations may lead to invalid conclusions. In the presence of non-
random selection, the observed differences are due at least in part to
the conditions under which the observations were generated and not
to effects of regimes. Whenever observations are not selected randomly,
inferences based on observed cases may yield inconsistent and biased
estimates of the effect of being in particular states (institutions, poli-
cies) on outcomes (performance). This much is now standard statisti-
cal wisdom. Yet the implications of this failure are profound: We can
no longer use the standard model of induction to make valid inferences
from the observed to the unobserved cases.
There are several ways in which such a bias may arise. To take one
example, democracies may be more likely to be found in countries that
are already wealthy, and, in turn, such countries may systematically
develop slower. This is a case of an exogenous selection on observ-
ables: Some observable factor, say per capita income, affects both the
selection of regimes and the economic growth of countries. We have
already provided an example in which selection is observable but
endogenous: Say that democracies are more likely than dictatorships
to fall when they face adverse economic conditions. Yet there may be
also common causes of regime selection and economic development
that we are unable to observe: Imagine that if a country happens to
have an enlightened leadership, it is more likely both to have a demo-
cratic regime and to develop.
The world around us is not generated randomly. The world we
observe is a result of actions of people pursuing their ends. It nurtures
successes and eliminates failures. These processes are systematic, and
so are their effects. The search for randomness is futile; we must
proceed differently. The question is how.
To cope with the non-random nature of our observations, we revert
to counterfactuals. To evaluate institutions, to assess their impact, we
must ask what the outcomes would have been if other institutions had
been observed under the same conditions. And if their selection is not
random, these institutions are not observable under the same condi-
tions. Hence, a recourse to counterfactuals is unavoidable. To cite
Tetlock and Belkin (1996), “counterfactual reasoning is a prerequisite
for any learning from history.”
To correct for non-random selection, we use the observations we

9
 Introduction

have to generate the counterfactuals we lack. To make such inferences,

we study the process by which observations are selected into par-
ticular states, that is, the rise and fall of the particular regimes. Then
we revert to an artifice: a distribution from which the observed and
the unobserved cases were generated. Having assumed the population
distribution and having identified the rule by which regimes are
selected, we can do the rest: use the actual observations to make infer-
ences about their counterfactual pairs, compare the outcomes, and
arrive at a conclusion about the impact of regimes. And this is what
we do.

Plan
This methodology underlies the organization of the book. Even
though our primary question concerns the impact of democracy on
development, more precisely of political regimes on material well-
being, in order to study this question we must first learn how coun-
tries happen to be living under particular regimes – the impact of
development on democracy.
Obviously, our preliminary task is to define “democracy” and “dic-
tatorship” and to classify the observed regimes into these categories.
Although several classifications of regimes, covering different periods
and sets of countries, are now available, they can be improved by (1)
a better grounding in political theory, (2) exclusive reliance on observ-
ables rather than on subjective judgments, (3) an explicit distinction
between systematic and random errors, and (4) more extensive cover-
age. In Chapter 1, we develop a classification of political regimes guided
by these objectives. This may appear to be the crucial step, because
everything that follows hinges upon it. But even if regime classification
has been the subject of some controversies, alternative definitions of
“democracy” give rise to almost identical classifications of the actual
observations. Our own classification, though more extensive than most,
is no exception. It differs little from all others with which we could
compare it. Hence, it turns out that in the end little depends on the
way we classify regimes. If they were available for the same sample,
other classifications would have generated the same results.
Once the regimes have been classified, we need to understand their
dynamic. In particular, we must examine whether or not the rise and
fall of political regimes have something to do with factors that also
affect economic performance. We already know that if the same factors,
whether or not we can observe them, affect both the selection of

10
Introduction

regimes and their performance, we must take this fact into account
when studying the impact of regimes on performance.
We are thus led to reopen the problematic inaugurated by Lipset
(1959, 1960) and subjected to innumerable empirical analyses since
then. In Chapter 2, we ask first whether or not the incidence of demo-
cratic regimes is related to the level of development in those countries.
Then we distinguish whether this relation is due to the fact that democ-
racies are more likely to emerge as a result of economic development
or only are more likely to survive when a country is already affluent.
Having examined the impact of economic factors, we bring into the
analysis a whole host of other considerations, such as the international
political climate, the political histories of the particular countries, and
their social conditions. Our purpose is to predict as well as we can
which regime a country has had at a particular time, that is, to iden-
tify what we earlier called the “selection rule.”
Once we know this rule, we proceed to the analysis of the impact of
political regimes on material well-being. In Chapter 3, we study the
impact of regimes on the growth of total income, total consumption,
per worker output, investment, labor force, and the functional distrib-
ution of income. Armed with knowledge of the selection rule, we isolate
the effect of regimes from the effects of the conditions under which
these regimes find themselves. Because we discover that regimes have
different effects in poorer and in wealthier countries, we then analyze
these groups of countries separately.
Whereas the question whether or not regimes affect economic
growth is controversial, the assertion that political instability thwarts
economic growth seems to be almost consensual. But “instability” turns
out to be a muddled notion; distinctions are needed. In Chapter 4, we
examine the impacts of past, current, and expected instability on
economic growth, separately under democracy and dictatorship.
In Chapter 5, we extend the analysis to demographic aspects of well-
being. This is not a topic we intended to treat when we planned the
book; we were driven to it serendipitously by the logic of analysis.
Many, perhaps most, students of economic growth take as the depen-
dent variable the rate of growth of income per capita. Yet whereas per
capita rather than the aggregate income is what matters for individual
welfare, per capita income is a result of two distinct processes: the
growth of total output and the growth of population. Hence, we rea-
soned, we should examine them separately. To understand population
growth, we study the effects of regimes on birth rates and death rates,
fertility, and life expectancy. Once we assess the impact of regimes on

11
 Introduction

population, we return to study their effect on the growth of per capita

incomes.
Because technical, typically statistical issues are entailed in many of
the analyses, all the chapters include appendixes in which our methods
are explained and additional tests are offered. Two general appendixes
follow the text. Appendix I presents a general model of selection that
organizes the entire analysis. Appendix II is a listing of variables and
the sources from which they were derived.
We will not reveal our conclusions here; an impatient reader can flip
the pages. We will disclose only this: Whenever regimes do make a dif-
ference, lives under dictatorships are miserable. The Churchillian view
may be not enough, but it is accurate. Democracies are far from perfect,
but they are better than all the alternatives.

12
 Chapter One

Democracies and Dictatorships

Introduction
To study systematically the origins and the consequences of politi-
cal regimes, we need first to determine what regime each country has
had during each period of its history. This is a complicated and irre-
deemably controversial undertaking. Although “democracy” and “dic-
tatorship” are concepts of everyday usage, the intuitions we associate
with them are not always sufficient to determine that a particular
country operated according to one or the other at a given time. Thus
we must formalize these concepts in terms of rules that can be deci-
sively and reliably applied to the observable aspects of national histo-
ries. And once we have such rules, we must apply them. Any such
procedure necessitates going back and forth between conceptual analy-
sis and historical observation. Closure is difficult to reach: In some
cases, history simply has not generated the information called for by
the rules, and we must accept the fact that there will be systematic
errors. Other cases are so idiosyncratic that we must accept the fact
that they have to be treated as random errors. Nevertheless, although
no classification can be free of errors, the encouraging lesson is that
independent endeavors by different researchers have led to very
similar results.
Although several classifications of regimes, covering different
periods and sets of countries, are now available, they can, in our
view, be improved by (1) a better grounding in political theory, (2)
exclusive reliance on observables, rather than on subjective judgments,
(3) an explicit distinction between systematic and random errors,
and (4) more extensive coverage. Thus, the purpose of this chapter
is to develop a classification of political regimes guided by these
objectives.

13
 Democracies and Dictatorships

This chapter is organized as follows: The first section covers con-

ceptual issues. The second section spells out the three basic rules we
use to classify regimes. The third section focuses on the treatment of
systematic errors and offers an additional rule that applies to a par-
ticular class of cases. The fourth section summarizes these rules and
shows their effects on regime classification. The fifth section considers
distinctions among democratic regimes and among authoritarian
regimes. The sixth section enumerates some criteria we did not
include. The seventh and eighth sections describe the results of apply-
ing our rules to historical observations of 141 countries between 1950
(or the year of independence) and 1990. Appendix 1.1 compares our
approach with alternative measures. Appendix 1.2 lists the regimes
by country and period. Appendix 1.3 provides some background in-
formation on regime transitions. Finally, Appendix 1.4 describes a
somewhat smaller data set, our “short” data set, which is used in
the remaining part of this book.

Democracy and Dictatorship

Our purpose is to classify the political regime observed in each
country during each year either as a democracy or as a dictatorship,
a term we use interchangeably with “authoritarian regime.” Although
we later distinguish different types of democracies and dictatorships,
our basic classification is dichotomous.
Our general stance is minimalist. Perusing the innumerable defini-
tions, one discovers that “democracy” has become an altar on which
everyone hangs his or her favorite ex voto.1 Almost all normatively
desirable aspects of political life, and sometimes even of social and eco-
nomic life, are credited as definitional features of democracy: repre-
sentation, accountability, equality, participation, dignity, rationality,
security, freedom – the list goes on. Indeed, according to many defini-
tions, the set of true democracies is an empty set. And from an ana-
lytical point of view, lumping all good things together is of little use.
The typical research problem is to examine the relationships among
them. We want to know if holding repeated elections induces govern-

1
To cite Macpherson (1966: 1): “Democracy used to be a bad word. Everybody who was
anybody knew that democracy, in its original sense of rule by the people or government
in accordance with the will of the bulk of the people, would be a bad thing – fatal to
individual freedom and to all graces of civilized living. That was the position taken by
pretty nearly all men of intelligence from the earliest historical times down to about a
hundred years ago. Then, within fifty years, democracy became a good thing.”

14
Democracy and Dictatorship

mental accountability, if participation generates equality, if freedom

imbues political systems with rationality. Hence, we want to define
“democracy” narrowly.
Three major distinctions dominate modern political thought con-
cerning forms of government.2 Montesquieu’s legacy (1995 [1748]) is
the distinction between limited regimes and despotic regimes. Kelsen’s
contribution (1945), going back to Rousseau and Kant, was to distin-
guish between “autonomy” (systems in which norms are determined
by those to whom they apply) and “heteronomy” (systems in which the
legislators are distinct from those who are subject to the laws). Finally,
Schumpeter’s innovation (1942) was to emphasize competition, or, to
borrow Dahl’s term (1971), which we prefer, “contestation,” as the
essential feature of democracy.
We focus on contestation. Our purpose is to distinguish between
(1) regimes that allow some, even if limited, regularized competition
among conflicting visions and interests and (2) regimes in which some
values or interests enjoy a monopoly buttressed by the threat or the
actual use of force. Thus “democracy,” for us, is a regime in which
those who govern are selected through contested elections. This defi-
nition has two parts: “government” and “contestation.”
In no regime are all governmental offices filled by elections. Outside
of classical Greece, generals, who are public officials, have never been
elected. Judges rarely are. What is essential in order to consider a
regime as democratic is that two kinds of offices be filled, directly or
indirectly, by elections: the chief executive office and the seats in the
effective legislative body.
Moreover, in a democracy, the offices that are being filled by con-
tested elections grant their occupants the authority to exercise gover-
nance free of the legal constraint of having to respond to a power not
constituted as a result of the electoral process. Thus, governmental
responsibility either directly to voters or to a parliament elected by
them is a defining feature of democracy. If we were to use only con-
tested elections as the criterion, we would date the beginnings of
democracy in many Western European countries to the period in which
governments still were not autonomous, not independent of the crown,
well before World War I. Yet, whenever that was the last step in the
process of democratization, and in Western Europe most often it was,
we date the advent of democracy to the time of transfer of govern-
mental responsibility from the crown (or a non-elected upper chamber)

2
See Bobbio (1989: 100–25, and chap. 4).

15
 Democracies and Dictatorships

to the parliament. Thus we date democracy in Great Britain to 1911

(Dahl 1990), and in Germany to 1919.
Until the early years of this century, the struggle for democracy con-
cerned primarily suffrage. In most countries, democracy emerged only
gradually, in a sequence of steps. Typically, what happened first was
that legislatures, elected on a nonpartisan basis and under highly
restricted suffrage, divided along partisan lines and began to contest
elections on a partisan basis. Then there followed extensions of polit-
ical rights, which were sometimes very gradual, as in Norway or Chile,
and at times instantaneous, as in Finland in 1906. In contrast, in those
countries that have only recently confronted the eventuality of having
to establish democratic institutions, suffrage is not an issue: It is taken
for granted that it will be “universal.”3 Neither is governmental respon-
sibility an issue.4 In the recent cases, the only focus of conflicts has
been contestation: whether or not divergent political forces would be
able to compete for governmental offices and to assume office if they
won elections.
Contestation occurs when there exists an opposition that has
some chance of winning office as a consequence of elections. We take
Przeworski’s dictum that “democracy is a system in which parties lose
elections” (1991: 10) quite literally: Whenever in doubt, we classify as
democracies only those systems in which incumbent parties actually
did lose elections. Alternation in office constitutes prima facie evidence
of contestation. In most nineteenth-century Latin American countries,
the incumbent presidents, even if they could not be reelected, con-
trolled their succession. We consider such regimes democratic only
when that system no longer functioned, when the opposition won an
election and assumed office.
Contestation entails three features: (1) ex-ante uncertainty, (2) ex-
post irreversibility, and (3) repeatability.
By “ex-ante uncertainty” we mean that there is some positive prob-
ability that at least one member of the incumbent coalition will lose in
a particular round of elections. Uncertainty is not synonymous with

3
The quotation marks are needed because the very notion of “universal” suffrage is a
matter of convention. Suffrage was “universal” in Europe before the voting qualification
was lowered from 21 to 18 years of age, and it is now “universal” at 18, rather than
16, 14, or 12. And “universal” is in turn defined relative to “citizen,” itself a legally
regulated notion. Immigrants often do not have the right to vote in national elections
even if they have lived in a country for a long time, and in several Western European
countries they constitute more than 8% of the adult population.
4
Except in Western Samoa.

16
Democracy and Dictatorship

unpredictability: The probability distribution of electoral chances is

typically known. All that is necessary for outcomes to be uncertain is
that it be possible for some incumbent party to lose.5 The best illus-
tration of such uncertainty is the surprise expressed by an editorial in
the Chilean right-wing newspaper, El Mercurio, in the aftermath of
Salvador Allende’s victory in the first round of the presidential elec-
tions of 1970: “No one expected that a marxist candidate could win
elections through a universal, secret, bourgeois franchise.” The fran-
chise may have been “bourgeois,” the chances skewed, and the victory
of a Marxist candidate may have been known to be unlikely. But it was
possible. The eventual outcome was not certain ex ante.
This feature of democracies has practical consequences. Most people
think that Argentina under President Arturo Illia (1963–66) was demo-
cratic, even though the largest party in the country was prohibited
from competing in the elections of July 1963. In turn, most agree that
Mexico is not democratic, even though no party is legally banned from
contesting elections. The reason is that Illia won narrowly, with 26.2
percent of votes cast, and he could have lost. In contrast, in Mexico it was
certain that the Partido Revolucionario Institucional (PRI) would win.
By “ex-post irreversibility” we mean the assurance that whoever
wins election will be allowed to assume office. The outcomes of elec-
tions must be irreversible under democracy even if the opposition wins.
In 1929, the dictator of El Salvador, General Romero, announced that
his country was about to join the family of civilized nations by cele-
brating the first free and honest election. He issued a decreto-ley that
specified when the elections would take place, who would be qualified
to vote, what the ballots would look like, when the polling places would
be open, and so forth. The last point declared that “Army contingents
will be stationed in the polling places in case the Opposition wins.”
That was not a democratic election.

5
A strong notion of uncertainty would require that there be some chance that the major
member of a coalition might find itself out of office as the result of the next election; a
weak notion would extend to any member of the incumbent coalition. Under some con-
ditions, changes of minor partners may be considered inconsequential: Italy under
Christian Democratic domination may be a case in point. But in other cases, the elec-
toral fates of minor parties can affect the orientation of a government: Israel and
Germany are the relevant cases here. We have opted for the weak version: We consider
any change of the governing coalition as alternation. Moreover, we do not want to imply
that all alternations result directly from elections. In several countries – Poland between
1919 and 1926 is an example – a government has changed between opposing parties
without an election intervening and the new government has remained in office as a
result of the subsequent election.

17
 Democracies and Dictatorships

The practical consequence of this feature is to exclude sham elec-

tions as well as periods of liberalization. Liberalization is typically
intended by dictatorial regimes to be a controlled opening of the
political space. When it fails, that is, when the opposition does win, a
clampdown sometimes follows. Hence, there is no certainty that the
opposition will be able to celebrate its victory.
The final feature of contestation is that elections must be repeated.
Whoever wins the current round of elections cannot use office to make
it impossible for the competing political forces to win next time. Democ-
racy, as Linz (1984) put it, is government pro tempore. All political out-
comes must be temporary: Losers do not forfeit the right to compete
in the future, to negotiate again, to influence legislation, to pressure
the bureaucracy, or to seek recourse to courts. Even constitutional
provisions are not immutable; rules, too, can be changed according
to rules.
The practical consequence of this last feature is that we should
reserve judgment about elections, since an electoral victory may
serve only to establish an authoritarian rule. This has been true in
several African countries following independence. Unless the losers
are given political guarantees that their ability to contest future
elections will be protected, the mere fact that elections have been
held does not suffice to qualify the regime as democratic. Only if the
losers are allowed to compete, win, and assume office is a regime
democratic.
Throughout this discussion, we have focused on democracy. We treat
dictatorship simply as a residual category, perhaps better denominated
as “not democracy.” Our procedure is to establish rules that will dis-
qualify a particular regime as democratic, without worrying about the
nature of the regimes eliminated in this manner. Only then do we intro-
duce some features that distinguish among different non-democratic
regimes.

Operational Rules: Filling Offices by Contested Elections

Following Cardoso (1979: 38), as well as O’Donnell and Schmitter
(1986: 73), by “regime” we mean the system of relations between the
civil society and the state. A regime is a system of rules and practices
that determine who has political rights, how they can be exercised, and
with what effects for the control over the state. Hence, even if dicta-
tors succeed one another, the regime, in our sense of the term, remains
the same as long as it remains a dictatorship. Thus, when General Viola

18
Operational Rules: Offices by Contested Elections

overthrows General Videla, we consider the entire period as one

dictatorship, rather than as “Videla” and “Viola” regimes. And even
in those rare cases when the basic democratic institutions are
transformed without a break in legality, as in France in 1958, we still
consider each of them as a single “democratic regime.”
Democracy is a regime in which government offices are filled by con-
tested elections. The first part of this definition is easy to operational-
ize: it is relatively simple to observe which offices, if any, are filled as
a result of elections. But whether or not these elections are contested,
in the sense defined earlier, is not always apparent. The existence of
more than one independent party is a sine qua non of contestation, but
it may not be sufficient.
We next specify the rules we use to classify regimes: first those that
are applied to assess whether or not the relevant offices were filled via
elections, then those that are used to assess whether or not elections
were contested.6 Our rules are the following:
Rule 1: The chief executive must be elected.

The “chief executive” may be the president, the prime minister, or,
in rare cases, a collegial body. Following Banks (1996), we define as
the “chief executive” the occupant of the office formally designated as
that of the head of government, thus excluding éminences grises:
strongmen who effectively rule their countries but do not occupy a
formal position.7
For a regime to be qualified as democratic, the executive must be
directly or indirectly elected in popular elections and must be respon-
sible only directly to voters or to a legislature elected by them. Indirect
elections qualify as popular only if the electors are themselves elected.
Elections by bodies that are themselves nominated do not qualify as
popular elections.
Rule 2: The legislature must be elected.

The legislature can be a congress, an assembly, or a parliament.

Only the lower house is considered. A constituent assembly that does
not have ordinary legislative powers is not considered a legislature.

6
The information about elections is based on Banks (1996), revised and updated.
7
Such as Deng Tsao Ping, whose only formal position for many years was president of
the Chinese Bridge Association. Banks sometimes considers first secretaries of the Com-
munist Party as chief executives, while at other times he takes presidents and prime
ministers as chief executives in communist regimes. We could not discover what rules
he used, and we took occupants of the formal office as chief executives.

19
 Democracies and Dictatorships

Our rule is that the legislature must be elected for a regime to qualify
as democratic.
Rule 3: There must be more than one party.

In some cases, there were no parties: Either there were no elections

or elections were conducted but all political parties were banned. In
other cases, there was only one party. We consider such regimes
authoritarian.8
By “party” we mean an independent list of candidates presented to
voters in elections. In communist Poland, for example, three parties
and a number of Catholic groups were represented in the Sejm, but
until June 1989 voters were offered only one list: a National Front or
Patriotic Front or whatever front it was called at the moment. Hence,
in cases where the share of seats held by the major party in the legis-
lature was less than 100 percent we checked to see if there was more
than one list in legislative elections. For example, although the ruling
Vanguard of the Malagasy Revolution (Arema) did not control all
the seats in the parliament in Madagascar after 1976, according to
Freedom House (1992: 318), “Until March of 1990, when a High
Constitutional Court decree permitted multi-partyism, political associ-
ations had to operate within the FNDR as the nation’s sole legal polit-
ical entity.” The FNDR (National Front of the Malagasy Revolution) was
thus the only list offered to voters – one party by our definition.
Applying this rule, we classified as dictatorships all regime-years
during which legislatures were elected but parties were banned or
during which a single party held 100 percent of the seats in the legis-
lature (except for Jamaica and Trinidad and Tobago, as discussed later)
or in which only one list was offered to voters in elections.
We also extended this rule to disqualify as democratic those regimes
in which incumbents used an electoral victory to establish (1) non-party
rule or (2) one-party rule or (3) a permanent electoral domination. This
is called the “consolidation” rule.
Consolidation of non-party or one-party rule occurred whenever
incumbents either banned all parties, or all opposition parties, or
forced all parties to merge with the ruling one. If an incumbent party
consolidated during its tenure in office a one-party rule or a non-party
rule, then the regime is considered to have been authoritarian from
8
Note that we do not assume that the existence of political parties is a necessary condi-
tion for contestation; after all, most pre-1900 elections were non-partisan. We exclude
from the democracies only those regimes in which all or all but one political parties
were banned. Bernard Manin made us sensitive to this point.

20
Operational Rules: Offices by Contested Elections

the moment at which the incumbent party assumed office. Note that
we are not examining intentions: If they tried and failed, the regime is
democratic.
We say that consolidation of incumbent rule also occurred whenever
there was more than one party but at some time the incumbents uncon-
stitutionally closed the legislature and rewrote electoral rules to their
own advantage.9 The entire period preceding the closing of the legis-
lature during which the same party was in office is then considered
authoritarian. In some analyses later in this book, we relax this rule
in order to identify cases in which incumbents perpetuated an auto-
golpe. In these instances, regimes are classified as democracies until
the year in which the autogolpe occurred, allowing us to separate the
cases of transition to authoritarianism caused by the incumbent chief
executive.10 These cases are listed in Appendix 1.3.
To understand how the consolidation rule was applied, consider
Malaysia, a country where three elections were held between inde-
pendence in 1957 and 1969. The incumbents won absolute majorities
of votes in the first two elections, but not in the third. They then
declared a state of emergency, closed the congress, and changed the
rules in such a way as to make this unpleasant experience unrepeat-
able. According to Ahmad (1988: 357), “the better showing by the
opposition caused a temporary loss of confidence and even the con-
clusion by some in the ruling party that it had lost its mandate.” The
parliament was dissolved in 1969, a state of emergency was pro-
claimed, and a tough internal-security law, still in effect, was adopted.
The constitution was rewritten to ensure that no more electoral defeats
would occur. Ahmad (1988: 358) comments on this event: “What is
more interesting about the conduct of elections as part of the democ-
ratic process, however, was probably the unstated notion that losing
an election meant virtually total political defeat. Therefore 1969 served
notice to the Alliance leadership that it might have to one day face the
prospect of an electoral defeat. . . . The rules of the game of Malaysian
democracy were therefore set for modification after 1969 because the

9
The mere act of dissolving the legislature is not sufficient to qualify as consolidation. In
many cases legislatures are dissolved or suspended according to extant constitutional
provisions: The 1975 Indian state of emergency was duly approved by the two houses
of the legislature. In some cases, notably Australia in 1975, the constitutionality of the
dissolution has been dubious. Our rule is that if elections were immediately proclaimed
and took place within the immediate future under the same rules, we did not treat such
a dissolution as a breakdown of democracy.
10
Note that what we observe are the outcomes of conflicts, rather than their initiation;
hence, an inference is entailed in interpreting these results.

21
 Democracies and Dictatorships

prospect of a zero-sum electoral result would be unacceptable if Malay

political supremacy was not to be assured.” As a result, “the fear of an
electoral defeat has been diminished under the Barisan Nasional coali-
tion concept. The parties that have not succumbed to the taste of power
by joining the BN cannot pretend to be able to form the national gov-
ernment at any time in the foreseeable future.”
In South Korea, President Chung-Hee Park held elections once and
won enough votes; then he held them again and became dissatisfied
with the result, closed the congress, and assumed dictatorial powers.
Five years later, he reopened the congress under new rules. President
Ferdinand Marcos in the Philippines won election twice and assumed
dictatorial power when he could not amend the constitution to enjoy
more terms.
Because in these cases we have prima facie evidence that incum-
bents were not prepared to yield office as a result of elections (although
one could argue that although the Malays were not willing to do it in
1969, they might have been willing to accept defeat earlier), we clas-
sify these regimes as dictatorships.
Note that this part of the consolidation rule applies only up to the
moment of the unlawful change of electoral rules. If a regime uncon-
stitutionally changed the rules in its favor but subsequently yielded
office under these new rules, then the regime is considered authori-
tarian up to the time of the openly dictatorial interregnum, and demo-
cratic subsequently. This is why we have classified the João Batista
Figueiredo term in Brazil as democratic: Although his predecessor
temporarily closed the congress and made it more difficult for the oppo-
sition to win, Figueiredo’s successor, Tancredo Neves, won election
against the candidate supported by the military under the same rules
as Figueiredo.
Thus the “party” rule is that if (1) there were no parties or (2) there
was only one party or (3) the incumbents’ term in office ended in the
establishment of non-party or one-party rule or (4) the incumbents
unconstitutionally closed the legislature and rewrote the rules in their
favor, then the regime was a dictatorship. As shown later, the absence
of legal opposition is the most frequent reason for classifying regimes
as dictatorships.
These three rules appear to us to be non-controversial, and they are
easy to apply. The first thing we learned from applying them is that the
great majority of cases, 91.8 percent of country-years, are unambigu-
ously classified by these three rules. There is, however, one particular
class of regimes that cannot be classified one way or another.

22
“Botswana” and the Alternation Rule

“Botswana” and the Alternation Rule

Thus far we have classified as democracies those regimes in which
the chief executive and the legislature are elected in multi-party elec-
tions. But we do not know that all regimes that satisfy these criteria
are in fact democracies.
Consider Botswana. Government offices in Botswana are filled by
elections, more than one party competes, there is little repression, and
there are no exceptional allegations of fraud. Hence, by the rules intro-
duced thus far, Botswana should be considered a democracy, and
indeed it is generally considered to be one. Yet the same party has ruled
Botswana since independence, always controlling an overwhelming
majority in the legislature. Thus, the question arises whether or not
elections are held in Botswana only because the ruling party is certain
to win them and whether or not the ruling party would yield office if
it ever lost. These are not moot questions: Looking into the future, a
specialist on this country speculates that “the resulting conflict could
well force the BDP [Botswana Democratic Party] to choose between
losing in parliamentary elections and abandoning elections as a
method of leadership selection. Given the paternalistic attitude of the
BDP from President Masire down, the latter choice would not be sur-
prising” (Holm 1988: 208). Hence, if democracy is a system in which
elections are held even if the opposition has a chance to win and in
which the winners can assume office, then the observable evidence is
not sufficient to classify Botswana one way or another.
Botswana is an ideal type: no constraints on the opposition, little
visible repression, no apparent fraud. But the issue is more general. If
the same party or coalition of parties had won every single election
from some time in the past until it was deposed by force or until now,
we cannot know if it would have held elections when facing the
prospect of losing or if it would have yielded office had it in fact lost.
We must thus decide which way to err: whether we prefer to commit
the error of excluding from the democracies some systems that are in
fact democracies (type I) or the error of including as democracies some
systems that are not in fact democratic (type II). Err we must; the ques-
tion is which way.
In some cases, either antecedent or subsequent events have pro-
vided additional information. In the United Kingdom, we knew even
before Blair’s victory that the Conservatives had lost elections and
allowed their opponents to assume office. In Japan, after a long tenure
in office, the incumbent party finally lost elections and allowed the

23
 Democracies and Dictatorships

opposition to assume office. Because this is the only information we

have, we use it – not without a leap of inference – to conclude that
these regimes are or were democratic. The same is true when we know
that incumbents unconstitutionally prevented the opposition from
winning elections or assuming office. In all these cases, we use this
information retroactively. This clearly is not a very satisfactory solu-
tion: One might easily imagine that even if certain incumbents were
willing to allow a peaceful alternation in office later on, they might not
have been willing to tolerate it earlier; conversely, even if they sup-
pressed the opposition later on, they might not have done so earlier.
But this is the only information we have; we cannot observe what might
have happened. The only alternative would be to attempt to assess the
degree of repression, intimidation, or fraud for each election, but, in
our view, such assessments cannot be made in a reliable way.
Japan is a paradigmatic case of a long tenure in office that ended with
a lawful alternation. The LDP (Liberal Democratic Party) was in office
continually until the 1993 election. Yet when the incumbents finally
lost, they allowed the opposition to assume office. The same was true in
Mauritius, the Bahamas, Barbados, Trinidad and Tobago, the Domi-
nican Republic, Grenada, Jamaica, St. Lucia, and the Solomon Islands.
In each of these countries the same party stayed in power for at least
two terms, either following independence or the birth of democracy; yet
eventually it lost an election and gave up office peacefully. We use this
information retroactively: Whenever a ruling party eventually suffered
an electoral defeat and allowed the opposition to assume office, the
regime is classified as democratic for the entire period this party was in
power under the same rules. Alternation thus overrides the party rule:
In Jamaica, as well as in Trinidad and Tobago, at one point, one party
controlled all the seats in the legislature. Yet it lost the subsequent elec-
tion and relinquished office. We therefore consider these regimes to
have been democratic even during the period of one-party rule.
We have already discussed cases in which incumbents, facing the
prospect of an electoral defeat or having actually been defeated, uncon-
stitutionally closed the legislature, introduced a state of emergency, and
rewrote the rules in their own favor. In such cases, we evoke the rule
about the consolidation of incumbent advantage and classify the regime
as authoritarian during the entire period before the openly dictatorial
interregnum. In South Korea, President Park won the 1963 elections,
and had he not instituted dictatorial rule nine years later, we would
never have known whether or not he would ever have been ready to
relinquish office. A Korean student of military politics (Se Jin Kim, cited

24
“Botswana” and the Alternation Rule

by Han 1988: 275) commented with regard to the 1963 elections that
“Park’s victory was in fact a blessing for the future of democracy in
Korea. Had the military lost, it can be safely assumed that the military
would have ignored the electoral outcome and continued to rule even
though such rule would have meant a total destruction of constitu-
tionalism.” But subsequently Park did close the congress and change
the rules, and we use this information to infer that he would not have
been ready to yield office during the preceding nine years.
Even when incumbents hold elections only because they expect to
win, they sometimes make mistakes: They hold elections and lose.
Then they have to decide whether to accept the popular verdict or over-
ride it. They can revert to post-election fraud: Anastasio Somoza is
purported to have said to his electoral opponent, “You poor s.o.b.,
perhaps you won the voting, but I won the counting,” a recipe appar-
ently applied by the Mexican PRI in 1988. Blatant fraud constitutes
prima facie evidence that the incumbents were not predisposed to
permit a lawful alternation in office. Or they can publish voting results
and still not allow the opposition to assume office.
Yet, to return to Botswana, in some cases history has not been kind
enough to provide even the information that we have for Japan or
Malaysia: All we know is that the incumbents always win. Presumably,
we would want to think that if Botswana is like Japan, it should be
considered democratic, but if it is like Malaysia, it should be consid-
ered authoritarian. But we do not know if Botswana is like Japan or
like Malaysia. Elections may be held in Botswana only because the
ruling party is sure to win, but how are we to know what would happen
if they expected to lose or in fact lost?
To provide more intuition, consider Turkey between 1950 and 1960,
another period generally considered democratic. The Democratic Party
(DP) came to power in 1950, holding 83.8 percent of the seats. It won
in 1954 with 93.0 percent of the seats, and in 1957 with 69.5 percent,
until it was ousted by the military in 1960. After the 1957 elections
“the DP responded to its declining support by resorting to increasingly
authoritarian measures against the opposition. . . . The last straw in the
long chain of authoritarian measures was the establishment by the gov-
ernment party in April 1960 of a parliamentary committee of inquiry
to investigate the ‘subversive’ activities of the RPP [main opposition
party]” (Ozbudun 1988: 200). Would the DP ever have yielded power
peacefully had it not been deposed by force?
We choose to take a cautious stance, that is, to avoid type-II errors.
While examining the histories of particular countries, we were

25
 Democracies and Dictatorships

impressed that the dream of many political elites is to rule perpetually

and to rule with consent: Politicians are just PRIstas by nature. The
Mexican system has been the ideal for many politicians in Latin
America and, until the defeat of the LDP, the Japanese system in Asia.
Attempts at creating a hegemonic system, in which some or even
all opposition would be allowed but the ruling party would not be
threatened with losing office, have been made at various moments
in Botswana, Gambia, Senegal,11 Argentina, Bolivia, Colombia,12
Ecuador, El Salvador, Guatemala, Honduras,13 Guyana, Bangladesh,
Egypt, Malaysia, Pakistan,14 the Philippines,15 South Korea,16

11
Coulon (1988: 154) writes about Senegal: “In 1978, a constitutional reform was adopted
which put into place a system of ‘controlled democracy.’ The number of parties was
limited to three. . . . The legislative and presidential elections of 1978 were a great
success for the Socialist Party [former UPS, the ruling party] (which received 81.7
percent of the ballots cast and 82 of the 100 seats in the Assembly) and for President
Senghor personally (who won 82.5 percent of the vote). . . . It must be emphasized,
however, that the elections were held in a tense climate and organized in a way that
threatened the secrecy of the ballot.”
12
In Colombia, Laureano Gómez was elected president with 83.8% of the vote and took
office in August 1950. He continued tight censorship and increased repression against
labor and violence against liberals and Protestants. He attempted to reform the con-
stitution in order to impose a falangist-corporatist framework, freeing the presidency
from most congressional constraints, centralizing power, and converting the senate into
a corporatist body.
13
In Honduras between 1965 and 1970, “the conservative, authoritarian civil-military
government suppressed popular organizations and rigged the electoral machinery to
assure National party victories in the 1965 and 1968 elections” (McDonald and Ruhl
1989: 113).
14
In Pakistan, according to Rose (1988: 114–15), “Ayub’s intention initially had been to
establish a nonparty system but it quickly became clear that this would be counter-
productive. Ayub then moved to the opposite extreme, legalized virtually all parties that
applied, and formed his own party. . . . What was rather astonishing was [that] the 1962
constitution, Ayub Khan’s rather cleverly disguised authoritarian system, went along
from 1961 to 1969 with no serious political challenges.”
15
In the Philippines, according to Jackson (1988: 246), “the final structure created by
Marcos was the Kilusang Bagong Lipuna (New Society Movement) or KBL. The KBL
initially was not referred to as a political party, but was designed to select and elect
candidates to local, provincial, and national offices. KBL candidates, in an atmosphere
of restricted press and speech, triumphed in the 1978 interim assembly elections as
well as in the 1980 local elections. The degree of limited participation is indicated by
the fact that the 1978 opposition was led from a jail cell by former Senator Benigno
Aquino. The interim assembly contained only fourteen non-KBL members out of 200
members. . . . President Marcos ran for reelection in June 1981 but his logical oppo-
nent, former Senator Aquino, was excluded by the constitution, which required all nom-
inees to be at least fifty years of age (Aquino was forty-eight.) With virtually no
opposition, President Marcos was reelected.”
16
In South Korea, according to Han (1988: 268–9), “the new government was born with
a democratic constitution and with the expectation that it would usher in democratic

26
“Botswana” and the Alternation Rule

Singapore, Turkey, Taiwan, and most likely other countries where the
evidence is not that direct. Indeed, it seems that many dictatorships
and some democracies are just failed attempts at creating a Mexico or
a Japan: Sometimes the ruling party overdoes it, and the result is a
naked dictatorship; sometimes the ruling party is forced to compro-
mise, and the result is democracy.
Suppose that politicians want power but also want to be admired and
adored. Ideally, they would hold office as a result of elections. Yet the
hunger for power overwhelms other motivations: They prefer to remain
in office by force rather than lose power. Incumbents have some notion
of how likely it is that they will win the next election. If they think they
will win, they hold elections. If they think they will lose, they do not. If
these assumptions are correct, then the observed sample of regimes
that hold regular elections is biased in favor of “democracies,” that is,
regimes that look like democracies, in the sense that they permit con-
testation and fill offices by elections, yet are not democracies in the
sense that the opposition has a chance to assume office as a result of
elections. Among the observed democracies, there are some that hold
elections only because the opposition cannot win and some in which the
opposition would not be allowed to assume office if it won. Hence,
holding elections is not sufficient to classify a regime as democratic.
We thus need one more rule: alternation. This rule is applicable only
to cases that qualify under the preceding three rules and in which in
the immediate past the incumbents either held office by virtue of elec-
tions for more than two terms or initially held office without being
elected. If all these conditions are satisfied and if the incumbents sub-
sequently held but never lost elections, we consider such regimes
authoritarian.
In making this decision, we are buttressed by an empirical obser-
vation. Among those cases in which alternation in office via elections
did occur, except for some Caribbean islands, the share of seats of the
incumbents was almost always smaller than two-thirds. Hence, the
conditional probability that the seat share will be larger than two-thirds
given that alternation occurs is very small. Because alternations via
elections are generally less frequent than seat shares in excess of two-
thirds, Bayes’s rule implies that the conditional probability that an

politics for South Korea. But the Rhee government was determined to remain in power
– for life – which required several constitutional changes, election rigging, and repres-
sion of the opposition. Rhee was able to establish his personal dictatorship by making
use of the state power as exemplified by the national police.”

27
 Democracies and Dictatorships

alternation will occur given that seat share is larger than two-thirds is
also very small: 8.8 percent.17 Countries in which one party wins an
overwhelming share of seats are not likely to be democracies; this rule
should classify regimes accurately about seven-eighths of the time.
Although there are some countries where a ruling party had been
winning by very large margins and yet subsequently left office via elec-
tions, the striking finding is that we could have used this ex-post cri-
terion to eliminate almost all the cases in which the same party
continually held office. If we were to decide that a regime in which the
ruling party always won more than two-thirds of seats was not demo-
cratic, Malaysia (where the share of seats after 1971 was always larger
than 68.1) would fail by this criterion, as would Botswana (seats >
77.8), Egypt after 1976 (seats > 75), Gambia (seats > 69.7), Senegal
after 1978 (seats > 83.0), South Africa (seats > 66.6), Mexico (seats >
72.2, or fraud in 1988), Guyana (seats started at 56.6, went to 83.1),
and Singapore (seats = 74.5 in 1965, and all or all but one after 1968).
In South Korea, the share of seats fell from 74.3 in 1967 to 55.4 in
1971, sufficient to prompt President Park to dissolve the congress, and
when the legislature was opened again in 1973, the ruling party con-
trolled 66.7 percent.
The cases distinguished by the alternation rule constitute systematic
error. Those readers who prefer to err in the other direction can reclas-
sify them (they are marked with asterisks in Appendix 1.2). But an
error is unavoidable.

Summary of Rules
For convenience, we restate our rules. A regime is classified as a
dictatorship during a particular year if at least one of these conditions
holds:
Rule 1: “Executive selection.” The chief executive is not elected
(EXSELEC).

Rule 2: “Legislative selection.” The legislature is not elected

(LEGSELEC).

17
Our prior, the probability that during a random country-year the regime will be demo-
cratic, is 0.40. The probability that any election will end with more than two-thirds of
seats going to one party is 0.38. The likelihood, the conditional probability that seats
will be more than two-thirds given that a regime is democratic, is 0.126. Hence, by the
Bayes rule, the posterior, the probability that a country will be democratic given that
seats will be more than two-thirds, is 0.0877.

28
Summary of Rules

Rule 3: “Party.” There is no more than one party. Specifically, this rule
applies if (1) there are no parties or (2) there is only one party or (3)
the current term in office ends in the establishment of non-party or one-
party rule or (4) the incumbents unconstitutionally close the legislature
and rewrite the rules in their favor (PARTY, INCUMB).

Rule 4: “Alternation” (applies only to regimes that have passed the

previous three rules). The incumbents will have or already have held
office continuously by virtue of elections for more than two terms or
have held office without being elected for any duration of their current
tenure in office, and until today or until the time when they were
overthrown they had not lost an election (TYPEII).

At the risk of repetition, we restate the relation between the

“party” and the “alternation” rules. If at any time there are fewer than
two parties, the regime is classified as a dictatorship. If there is
more than one party, the question becomes whether or not there is
a real possibility for the opposition to win and assume office. If
incumbents lose elections at any time in the future, the regime is
considered to have been democratic during their entire tenure in
office. If incumbents repress the opposition at any time in the
future, the regime is classified as a dictatorship during their entire
tenure in office up to this moment. If we observe neither consolidation
nor alternation, we avoid type-II error and classify the regime as a
dictatorship.
Finally, a word is needed about our timing rules. In all cases of
regime transitions, we code the regime that prevailed at the end of the
year, even if it came to power on December 31, as, for example, did
Nigeria in 1983. Transitions to authoritarianism are signaled by a coup
d’état. Transitions to democracy are dated by the time of the inaugu-
ration of the newly elected government, not the time of the election. In
the few cases, like that of the Dominican Republic in 1963, in which a
democratic regime lasted six months (or in Bolivia in 1979, where the
situation changed several times), the information about regimes that
began and ended within the same year is lost.
Our data set covers 141 countries between 1950 (or the year of inde-
pendence) and 1990. Altogether, during this period we found 238
regimes: 105 democracies and 133 dictatorships. They are listed, by
country and period, in Appendix 1.2. Of the total 4,730 years, coun-
tries lived 1,723 (36 percent) years under democracy and 3,007 (64
percent) under dictatorship (REG).
Table 1.1 shows the numbers of cases in the sample that were

29
 Democracies and Dictatorships

Table 1.1. Distribution of observations by

criteria used for classification as dictatorships

Executive selection: non-elected executive 1,513

Legislative selection: 940
No legislature 789
Non-elected legislature 151
Political party 2,250
No political party 651
One political party 1,599
Consolidation of incumbent rule 93
Alternation 389
Executive selection and political party 1,244
Executive selection and legislative selection 89
Legislative selection and political party 767
Executive selection, legislative selection, 731
and political party
Total dictatorships 3,007
Total democracies 1,723
Total regime years 4,730

disqualified as democracies by each of the rules alone and by their

combinations.

Distinguishing among Democracies and Dictatorships

Obviously, neither democracies nor dictatorships are all the same.
Thus, further distinctions are required.
Given the recent popularity of this issue, we classified democracies
as parliamentary, mixed, or presidential (INST). These types are de-
fined as follows. Systems in which the government must enjoy the con-
fidence of the legislature are “parliamentary”; systems in which the
government serves at the pleasure of the elected president are “presi-
dential”; systems in which the government must respond both to a
legislative assembly and to an elected president are “mixed.”18
In a parliamentary system the legislative assembly can dismiss the

18
This criterion coincides almost perfectly with the mode of selection of the government:
by legislatures in parliamentary systems, by voters (directly or indirectly) in presiden-
tial systems. For a review of the differences, see Lijphart (1992).

30
Distinguishing among Democracies and Dictatorships

government, whereas under a presidential system it cannot.19 Some

institutional arrangements, however, do not fit either pure type; they
are “premier-presidential,” “semi-presidential,” or “mixed,” according
to different terminologies. In such systems, the president is elected for
a fixed term and has some executive powers, but the government
serves at the discretion of the parliament. These “mixed” systems are
not homogeneous: Most lean closer to parliamentarism insofar as the
government is responsible to the legislature; others (notably Portugal
between 1976 and 1981) grant the president the power to appoint and
dismiss governments (Shugart and Carey 1992).
Whereas we observed 105 democratic regimes, two countries
changed institutional arrangements without breaking the continuity of
democracy: Brazil twice and France once. Hence, altogether there were
108 democracies, as distinguished by institutional framework. Of these,
55 were parliamentary, 9 were mixed, and 44 were presidential. Of the
total 1,723 years of democracy, countries spent 1,085 (63 percent)
years under parliamentarism, 150 (9 percent) under mixed systems,
and 488 (28 percent) under presidentialism.
To distinguish among dictatorships, we developed three alternative
typologies and examined the relationships among them.
First, some dictatorships are “mobilizing,” whereas others are
“exclusionary.” The former organize permanent political participa-
tion through a single or dominant party and regularly hold acts of
popular mobilization that they call “elections.” They require indi-
viduals to manifest loyalty to the regime by participating. The latter
form, exclusionary dictatorships, may or may not hold elections, but
they do not promote any kind of political participation by the masses.
They only require that individuals not engage in acts oriented against
the regime.
Operationally, we define as “mobilizing” those dictatorships that had
at least one political party, and as “exclusionary” those that did not
have parties (MOBILIZE). Altogether we observed 147 mobilizing and
127 exclusionary dictatorships. Of the 3,007 dictatorial years, 1,991
(66 percent) were spent under mobilizing dictatorships, and 1,016 (34
percent) under exclusionary ones.

19
The Chilean 1891–1925 democracy does not fit this classification. Although it was pop-
ularly called “parliamentary,” that was a misnomer. The Chilean lower house frequently
censured individual ministers, but could not and did not remove the government or the
chief executive, the president. In parliamentary systems, except for some early rare
cases, the responsibility of the government is collective.

31
 Democracies and Dictatorships

Second, we distinguish authoritarian regimes according to the

number of formal powers: executives, legislatures, and parties
(DIVIDED). Our intuition is derived from Machiavelli: Whenever
decision-making is collective, there must exist some rules organizing
the functioning of the government (Bobbio 1989). Hence, even if the
legislature is a rubber stamp or the chief executive obeys dictates of
the single party, the mere existence of such bodies means that there
must exist some formal rules allocating functions and specifying pro-
cedures. We are not claiming, as Kavka (1986) would, that divided gov-
ernments are necessarily limited: Under dictatorship, some of these
bodies may have no autonomous power and do not provide checks and
balances. But the existence of rules distinguishes such regimes from
those dictatorships in which the operation of government need not be
organized by any formal rules.
The “powers” are always the chief executive and, if they exist, leg-
islatures or parties. Hence, a “divided” dictatorship is one that in addi-
tion to the chief executive has a legislature or a party. “Monolithic”
dictatorships have no legislatures and no parties. We observed 167
divided and 91 monolithic dictatorships, with 2,407 (80 percent) years
spent under the former, and 600 (20 percent) under the latter.
Finally, another distinction, in the spirit of Montesquieu, is whether
the dictatorship codifies and announces the rules it intends to apply to
its subjects or governs without such rules (AUT). In the first case, rule
is exercised “by fixed and established laws,” whereas in the second
case “a single person directs everything by his own will and caprice.”
“Bureaucracies” are dictatorships that have some internal rules for
operating the government, at least rules regulating the competence of
the chief executive vis-à-vis the legislature, and some external rules,
namely, laws. To put it differently, bureaucracies are institutionalized
dictatorships. Operationally, bureaucracies are dictatorships that have
legislatures: Because all regimes have chief executives, the existence
of a legislature implies that they must have some rules for regulating
relations among different organs of government. It also indicates that
rule is exercised by law, that is, that people can know the rules that dic-
tators at least intend to enforce at a particular moment and, moreover,
that these rules are universalistic in intent. In turn, “autocracies” are
despotic or, in the language of Linz (1975), “sultanistic” regimes, which
have neither internal rules of operation nor publicly announced univer-
salistic intentions. Operationally, autocracies are systems in which there
is a chief executive, and perhaps a single party, but no legislature. Yet
some of the autocracies and bureaucracies that result from the applica-

32
What We Did Not Include

tion of these rules are, in fact, transitional regimes. We corrected for

these regimes in the list presented in Appendix 1.2. Altogether, we
observed 146 bureaucracies, which lasted 2,117 (70 percent) years, and
116 autocracies, during 890 (30 percent) years.

What We Did Not Include

This conception of democracy in terms of contested elections for ex-
ecutive and legislative offices is clearly minimalist. Hence, it may be
useful to make explicit at least some of the features that we did not
consider when classifying regimes as democracies or dictatorships.
First, we do not include in our conception of democracy any social
or economic aspects of a society. Many scholars (Weffort 1992) and, as
surveys from many countries demonstrate, most citizens perceive
social or economic equality as an essential feature of democracy. Yet
the questions whether or not, on the one hand, contested elections tend
to generate equality in the social or economic realm (Jackman 1974;
Muller 1988) and, on the other hand, whether or not economic equal-
ity makes democracy more durable (Muller 1988) are just too inter-
esting to be resolved by a definitional fiat. We prefer to define
democracy narrowly and to study its causes and consequences.
Second, we do not think that “accountability,” “responsibility,”
“responsiveness,” or “representation” should be treated as definitional
features of democracy. When Dahl (1971: 1) says that “a key charac-
teristic of a democracy is the continued responsiveness of the govern-
ment to the preferences of its citizens,” or when Riker (1965: 31)
asserts that “democracy is a form of government in which the rulers
are fully responsible to the ruled,” they mean either (1) that when and
only when the government is responsive, the regime is democratic
regardless of anything else, or (2) that if a system is democratic by
some other criteria, then the government will behave responsibly. The
standard way of thinking follows Dahl, who lists several conditions that
are necessary and sufficient for governments to be responsive. And it
is the presence of these conditions, not responsiveness, that defines a
regime as democratic: The statement that “if these conditions hold,
then governments will be responsive” is a theorem, not a definition.
Moreover, this theorem is most likely false unless additional conditions
are specified: First, the very notion of “responsiveness” or “account-
ability” is muddled, and second, probably only some otherwise demo-
cratic governments are “accountable” in any intuitive sense of this term
(Przeworski, Manin, and Stokes 1999). Hence, the question whether or

33
 Democracies and Dictatorships

not regimes characterized by freedom of opinion, widespread partici-

pation, and repeated elections are in fact responsive is best left open
for investigation, rather than resolved by definition.
Third, whereas some degree of political freedom is a sine qua non
condition for contestation, democracy cannot be sufficiently defined in
terms of “liberties” or “freedom,” or human rights, which underlie the
Gastil (1980, 1990) or the Freedom House scales. The American con-
ception of “freedom” perceives it as a condition, not as a predicate of
actions: People are free, even if they never exercise their freedom. Thus
U.S. citizens are free to form political parties; yet they almost never
form them. They are certainly free to vote, yet about half do not. From
our point of view, to paraphrase Rosa Luxemburg, the point is not to
be free but to act freely. And acting freely in the political realm entails
enabling conditions, institutional as well as social. Whereas democracy
is a system of political rights – these are definitional – it is not a system
that necessarily furnishes the conditions for effective exercise of these
rights.20 Thus, assessing “freedom” or “liberty” without determining
the conditions that enable its exercise can easily lead to ideologically
motivated labels that measure only similarities to the United States,
rather than the actual exercise of political rights.
Fourth, we do not include participation as a definitional feature of
democracy. In Dahl’s conception of “polyarchy,” both contestation and
participation are necessary to classify a regime as democratic. Indeed,
Dahl sets the participation threshold so high that by his criterion
the United States would not qualify as a democracy until the 1950s.
Vanhanen (1992) sets it lower, but still disqualifies as democracies
those regimes in which elections are contested but participation is
very limited. Yet we want to distinguish regimes in which at least some,
but not necessarily all, conflicting interests contest elections. And
empirical evidence from Western Europe (Przeworski 1975) as well as
from Latin America (Coppedge 1992) indicates that the distribution of
votes among parties changes only slowly after each extension of
suffrage, implying that even when suffrage is highly restricted, diver-
gent interests are being represented. Moreover, we want to be able to
test theories about the effects of participation on the performance
and the durability of democracy (Huntington 1968; O’Donnell 1973;

20
Mueller’s libertarian view, “political equality is something that evolves without much
further ado when people are free” (1992: 988), should thus be contrasted with J. S.
Mill’s insistence that “high wages and universal reading are the two elements of democ-
racy” (quoted in Burns 1969: 290).

34
What We Did Not Include

Huntington and Nelson 1976). Using any threshold would produce a

censored sample and a bias we prefer to avoid.21
Fifth, as long the chief executive and the legislature were elected in
contested elections, we did not delve further into civil–military rela-
tions. Several distinctions could be made here. In some regimes that
we classified as democracies, civilian institutions do not control the mil-
itary, who in turn do not intervene in politics. In other democracies,
civilian politicians use the threat of military intervention in strategic
interactions among themselves (“praetorian politics”). Finally, in some
democracies (Honduras and Thailand are prototypes), civilian rule is
but a thin veneer over military power, exercised by defrocked gener-
als. Yet as long as officeholders are elected in elections that someone
else has some chance of winning, and as long as they do not use the
incumbency to eliminate the opposition, the fact that the chief execu-
tive is a general or a lackey of generals does not add any relevant infor-
mation. Most generals who get elected only because they are generals
are eliminated from consideration on the basis of other rules. Some
probably sneak through; there is no measurement without error.
Finally, several countries have been ravaged by civil wars: El Sal-
vador, Guatemala, Uganda, and Sudan are obvious cases. If a regime
is a set of institutions that regulate the relationship between the civil
society and the state, then there can be no regime where there is no
state. To varying degrees, the very question whether a regime is or is
not democratic turns out to be irrelevant. Voting typically occurs in only
some parts of such countries, and legislatures frequently turn out to be
ineffective.22 We considered excluding such periods altogether. Yet the
degree to which a civil war disrupts the normal functioning of the polit-
ical system is difficult to assess.
Thus, to repeat, our approach is minimalist. We want to be able to
examine empirically, rather than decide by definition, whether or not
the repeated holding of contested elections is associated with economic
performance. Democracy, to repeat, is a regime in which some gov-
ernmental offices are filled by contested elections. Whether or not

21
Note that if we were to use Dahl’s participation threshold of 50% of adults to qualify
countries as democratic, we would date Western European democracies quite late: in
the case of Belgium or France, after World War II. The proportion of the population
age 20 years or older who could vote in the 1946 election, the last one before women
got the right to vote, in Belgium was 45.5%, of whom 90.3% voted, which yields a par-
ticipation rate of 41.1%. In France in 1936, 40.1% of those age 20 and over could vote,
and the turnout was 84.4%, implying a participation rate of 33.7%.
22
Most of Banks’s coding for “ineffective legislature” was due to civil wars.

35
 Democracies and Dictatorships

regimes in which the rulers are elected tend to generate social and eco-
nomic equality, control by citizens over politicians, effective exercise of
political rights, widespread participation, or freedom from arbitrary
violence, as well as economic growth, high employment, low inflation,
and public services, should be studied empirically rather than decided
definitionally.

Stability and Change of Political Regimes

Of the 141 countries that we observed for at least some time between
1950 and 1990,23 73 were already independent in 1950, and 68 gained
sovereignty after 1950 (Table 1.2).
Independence, as we know, came in waves. From 1950 through
1959, eight new countries entered the world of sovereign states: Oman,
Laos, Morocco, Sudan, Tunisia, Ghana, Malaysia, and Guinea. The
largest wave occurred between 1960 and 1968, when forty-one new
countries appeared in the world (with the single largest expansion
having occurred in 1960, when seventeen African countries gained
independence). The remaining nineteen countries became independent
between 1969 and 1981, with the largest additions occurring in 1971,
when Bangladesh and some of the Persian Gulf states were created
(Bahrain, Qatar, and the United Arab Emirates), and in 1975, with the
independence of the last Portuguese colonies in Africa (Angola, Cape
Verde Islands, and Mozambique) and of the Comoro Islands, Suriname,
and Papua New Guinea. The last country to become independent prior
to 1990 was Belize, in 1981.
Because our observations begin with 1950, and about half of the
countries became independent only later, the aggregate distribution of
regimes over the years depends on two factors: changes of regimes in
the already existing countries and the entrance of new countries into
the world (Table 1.3).
Given that the number of countries changed over time, one should

23
Because our observations end with 1990, the data set does not include any of the coun-
tries that resulted from the breakup of the Soviet Union (Armenia, Azerbaijan, Belarus,
Estonia, Georgia, Kazakhstan, Kyrgyzstan, Latvia, Lithuania, Moldova, Tajikstan, Turk-
menistan, Ukraine, and Uzbekistan), Yugoslavia (Bosnia and Herzegovina, Croatia,
Macedonia, Slovenia), Czechoslovakia (Czech Republic and Slovakia), Ethiopia (Eritrea),
and Somalia (Somaliland). The Soviet Union, Yugoslavia, Czechoslovakia, Ethiopia, and
Somalia, however, are part of our data set. Moreover, because no economic data were
available, the data set does not include Albania, Antigua and Barbuda, Bhutan, Brunei,
Cuba, Cyprus, the Democratic People’s Republic of Korea (North Korea), Dominica, Kiri-
bati, Lebanon, Libya, Maldives, Namibia, St. Kitts and Nevis, St. Lucia, St. Vincent and
the Grenadines, and São Tomé and Príncipe.

36
 Table 1.2. Number of countries, old and new, by year

Large data base: 141 countries Short data base: 135 countries

Year Total Old New Entering Total Old Entering/ exiting New Entering/exiting

1950 73 73 0 0 0 0 —
1951 74 73 1 138a 50 50 0
1952 74 73 1 52 52 85, 98 0
1953 74 73 1 52 52 0
1954 75 73 2 88 55 55 83, 84, 87 0
1955 75 73 2 56 56 86 0
1956 78 73 5 30, 40, 44 58 57 82 1 30
1957 80 73 7 18, 89 60 57 3 18, 89
1958 81 73 8 19 60 57 3
1959 81 73 8 60 57 3
1960 98 73 25 3, 5, 7, 9, 10, 12, 16, 65 57 8 3, 5, 19, 33, 46
21, 25, 27, 28, 32, 33,
35, 38, 43, 46
1961 101 73 28 37, 42, 137 90 67 24, 58, 79, 81, 92, 97, 23 7, 9, 10, 12, 16, 21,
104, 121, 127, 128 25, 27, 28, 32, 35,
38, 42, 43, 44
1962 108 73 35 1, 6, 34, 45, 60, 64, 135 97 67 30 1, 6, 34, 37, 45, 60, 64
1963 109 73 36 22 98 67 31 22
1964 112 73 39 26, 47, 116 101 67 34 26, 47, 116
1965 115 73 42 17, 48, 95 104 67 37 17, 48, 95
1966 119 73 46 4, 23, 50, 72 108 67 41 4, 23, 50, 72
1967 120 73 47 100 108 67 41
1968 122 73 49 29, 41 110 67 43 29, 41

(continued)
e
 Table 1.2 (continued)

Large data base: 141 countries Short data base: 135 countries

Year Total Old New Entering Total Old Entering/ exiting New Entering/exiting

1969 122 73 49 110 67 43

1970 123 73 50 130 112 67 45 100, 130
1971 127 73 54 78, 136, 139, 141 117 70 109, 111, 119 47 40, 78
1972 127 73 54 117 70 47
1973 128 73 55 49 117 70 47
1974 130 73 57 20, 56 118 70 48 20
1975 136 73 63 2, 8, 11, 31, 75, 132 124 70 54 2, 8, 11, 31, 75, 132
1976 137 73 64 36 125 70 55 36
1977 138 73 65 13 126 70 56 13
1978 139 73 66 133 127 70 57 49
1979 139 73 66 127 70 57
1980 140 73 67 134 128 70 58 135
1981 141 73 68 51 131 71 103 60 51, 133
1982 141 73 68 131 71 60
1983 141 73 68 131 71 60
1984 141 73 68 132 71 61 134
1985 141 73 68 135 72 90 63 56, 88
1986 141 73 68 135 72 63
1987 141 73 68 132 69 15, 24, 92 63
1988 141 73 68 129 68 83 61 13, 49
1989 141 73 68 126 67 109 59 42, 133
1990 141 73 68 112 63 58, 91, 121, 127 49 2, 4, 32, 38, 41, 46,
50, 75, 100, 116

a
See Appendix 1.2 for country codes.
e
Table 1.3. Democracies and dictatorships in old and new countries

All Old countries New countries Entering sample

Year Total Total Dem Dic Total Dem Dic Total Dem Dic

1950 73 73 35 38 0 0 0 0 0 0
1951 74 73 34 39 1 0 1 1 0 1
1952 74 73 35 38 1 0 1 0 0 0
1953 74 73 35 38 1 0 1 0 0 0
1954 75 73 34 39 2 1 1 1 1 0
1955 75 73 34 39 2 1 1 0 0 0
1956 78 73 34 39 5 2 3 3 1 2
1957 80 73 34 39 7 2 5 2 0 2
1958 81 73 36 37 8 1 7 1 0 1
1959 81 73 37 36 8 0 8 0 0 0
1960 98 73 39 34 25 3 22 17 3 14
1961 101 73 39 34 28 4 24 3 1 2
1962 108 73 36 37 35 6 29 7 2 5
1963 109 73 35 38 36 5 31 1 0 1
1964 112 73 34 39 39 6 33 3 1 2
1965 115 73 33 40 42 7 35 3 0 3
1966 119 73 34 39 46 7 39 4 1 3
1967 120 73 33 40 47 6 41 1 0 1
1968 122 73 31 42 49 7 42 2 1 1
1969 122 73 31 42 49 5 44 0 0 0
1970 123 73 31 42 50 6 44 1 0 1
1971 127 73 32 41 54 6 48 4 0 4
1972 127 73 32 41 54 5 49 0 0 0
1973 128 73 31 42 55 6 49 1 1 0
1974 130 73 32 41 57 7 50 2 1 1
1975 136 73 33 40 63 9 54 6 2 4
1976 137 73 32 41 64 9 55 1 0 1
1977 138 73 31 42 65 9 56 1 0 1
1978 139 73 31 42 66 10 56 1 1 0
1979 139 73 34 39 66 11 55 0 0 0
1980 140 73 33 40 67 12 55 1 1 0
1981 141 73 33 40 68 12 56 1 1 0
1982 141 73 34 39 68 12 56 0 0 0
1983 141 73 37 36 68 11 57 0 0 0
1984 141 73 39 34 68 12 56 0 0 0
1985 141 73 40 33 68 11 57 0 0 0
1986 141 73 42 31 68 13 55 0 0 0
1987 141 73 42 31 68 13 55 0 0 0
1988 141 73 44 29 68 14 54 0 0 0
1989 141 73 45 28 68 13 55 0 0 0
1990 141 73 48 25 68 12 56 0 0 0

Note: See Appendix 1.2 for country codes.

39
 Democracies and Dictatorships

expect that the number of regime transitions would vary as well. As

a null hypothesis, suppose that the probability of any democracy
or dictatorship dying during a particular year is constant over time
and is equal to the average rate over the entire period: 0.0261 for
democracies and 0.0173 for dictatorships. The expected number of
transitions from a given type of regime is then the product of the
number of such regimes and their probability of dying. Comparison
of the expected and the observed numbers of transitions suggests
that until 1961 there were more transitions to democracy and fewer
transitions to dictatorship than one would expect to happen by chance.
The period from 1958 to 1973 was hostile to democracies; more of
them died and fewer were born than one would expect. Finally, the
post-1974 period was again favorable to democracies and hostile to
dictatorships.
This periodization is due to Huntington (1991), according to whom
(1) the “second wave” of democratization began in 1943 and ended in
1962, (2) the “second reverse wave” started in 1958 and ended in
1975, and (3) the “third wave” of democratization began in 1974. Note
that Huntington refers to these patterns as “waves.” If all he means is
that the frequency of regime transitions was not the same during the
entire period, then his observation stands. But were these “waves” in
the sense that each transition made it more likely that another transi-
tion in the same direction would follow?
Consider first the seventy-three countries in our sample that were
independent in 1950, when thirty-five of them (48 percent) were demo-
cratic. By 1960 the number of democracies among these countries
increased to thirty-nine, only to fall to thirty-one by 1968. It was still
thirty-one in 1978, after which it climbed back to thirty-nine in 1984
and to forty-eight in 1990. Hence, with regard to the “old” countries,
those countries that were independent in 1950, our count roughly
agrees with Huntington’s analysis. But even among “old” countries, the
waves depicted by Huntington are less general than they first appear.
Fluctuations in the distribution of democracies among the countries
that were independent in 1950 can be observed only in Latin America.
In 1950, eight of the eighteen Latin American countries were democ-
ratic; that number went down to six in 1955, up to twelve in 1959,
where it remained until 1961, and down again to five in 1976. The
number of democracies went up again after that, to reach fourteen in
1986, where it remained until 1990. Outside Latin America, the pro-
portion of democracies remained relatively constant, around 48
percent, from 1950 through 1985 (Figure 1.1). Only after that year, as

40
 Figure 1.1. Proportions of democracies in the world, in old countries, and in new
countries
e
 Democracies and Dictatorships

a result of the democratization of Bangladesh, Pakistan, the Philip-

pines, South Korea, Poland, Bulgaria, Czechoslovakia, and Hungary,
did the number of democracies in the “old” countries outside Latin
America start to increase: to 51 percent in 1985, to 55 percent in 1986,
and to 62 percent in 1990 (Figure 1.2).
The story for countries that became independent after 1950 is
entirely different. Three out of twenty-five (12 percent) newly inde-
pendent countries were democratic in 1960; subsequently the numbers
were seven out of forty-nine (14.3 percent) in 1968, ten out of sixty-
six (15.1 percent) in 1978, and twelve out of sixty-eight (17.6 percent)
in 1990. Hence, the proportion of democracies among these “new”
countries grew slightly but steadily over the period. In turn, the decline
of the aggregate proportion of democracies in the world during the
1960s was largely due to the emergence of new countries rather than
to transformations of old ones. The “reverse wave” covering the 1960s
was mostly due to the fact the number of countries increased dramat-
ically in the 1960s, and a large proportion of the new countries entered
the world as dictatorships. Thirty-two of the forty-one countries that
became independent between 1960 and 1968 did so under authori-
tarian regimes.24 With a few exceptions, most of them remained
authoritarian for the rest of the period. Hence, the waves of democra-
tization and authoritarianism are at most limited to “old” countries,
and particularly to Latin America.
Another way to examine whether or not regimes come and go in
waves is simply to observe transitions by year. If transitions come in
waves, we would expect that in each successive year more countries
would transit in one direction, cresting at some peak, and then initi-
ating the same pattern going in the other direction.
The annual frequency of transitions presented in Figure 1.3 does
not show such a pattern. Between 1950 and 1961 (the tail end of Hunt-
ington’s second wave of democratization) there were eleven transitions
to democracy, but also nine to dictatorship (Table 1.4). Between 1958
and 1973 (during the “reverse wave”) there were twenty-five transi-
tions to dictatorship, but also sixteen to democracy. Finally, between

24
The countries that emerged as dictatorships after independence are Algeria, Benin,
Botswana, Burkina Faso, Burundi, Cameroon, Central African Republic, Chad, Gabon,
Gambia, Ivory Coast, Kenya, Lesotho, Madagascar, Malawi, Mali, Mauritania, Niger,
Rwanda, Senegal, Swaziland, Tanzania, Togo, Uganda, Zaire, Zambia, Zimbabwe,
Guyana, Singapore, Yemen Arab Republic, Western Samoa, and Kuwait. The countries
that emerged as democracies are Congo, Mauritius, Nigeria, Sierra Leone, Somalia,
Barbados, Jamaica, Trinidad and Tobago, and Malta.

42
 Figure 1.2. Proportions of democracies in old countries: world, Latin America, and
other regions
e
 Figure 1.3. Transitions to democracy and to dictatorship by year
e
Stability and Change of Political Regimes

Table 1.4. Expected and observed numbers of transitions, by period

Transitions to
Transitions to democracy dictatorship

Periods Expected Observed Expected Observed

1950–1961 8.68 11 11.43 9

1958–1973 19.98 16 16.29 25
1974–1990 26.75 32 21.40 15

1974 and 1990 there were many more transitions to democracy, thrity-
two, but still fifteen in the other direction. A simple statistical test is to
correlate the annual numbers of transitions to democracy and to dic-
tatorship. If regimes come in waves, then this correlation should be
highly negative: During the years when there are many transitions in
one direction, there should be few in the other, and vice versa. Yet the
correlation is almost zero: -0.01. Moreover, the numbers of transitions
in the two directions are not even autocorrelated: The annual auto-
correlation of the number of transitions to democracy is -0.011, and
to dictatorship -0.010.25 Hence, however the waves roll, there is always
an undertow pushing in the reverse direction. Huntington’s oceanic
metaphor is only that.
Dictatorships, on average, lasted longer than democracies. But
because some regimes already existed in 1950 and some lasted beyond
1990, these averages could reflect the timing of their emergence as
much as their ability to survive (Table 1.5).
Democracy is a phenomenon of the twentieth century. For that
reason, in 1950 the democracies tended to be younger than the dicta-
torships. The average age of the thirty-four democratic regimes that
existed then was 25.8 years. Only six of them – the United States,

25
A more complicated way of testing whether or not regime transitions come in waves
is to analyze the autoregressive structure of the series using ARIMA identification
methods. If the series are indeed cyclical, they should be at least of order 2, with a pair
of imaginary roots. Because the series are short, these tests are not highly reliable.
Nevertheless, for all the three series – transitions to democracy, transitions to dicta-
torship, and the difference between them – the tests indicate that they are of order 1.
The seasonal component is negligible, and all the series pass the runs test. Hence, there
is no evidence for cycles.

45
 Democracies and Dictatorships

Table 1.5. Regime duration: average age in years at end of regime spell or
last year of observation

All regimes All Democracy Dictatorship

Spells completed between 1950

and 1990 18.6 (97)a 8.5 (45) 27.4 (52)
Spells in course by 1990 38.2 (141) 36.6 (60) 39.4 (81)
Age of regime as of 1950 34.7 (70) 26.2 (34) 42.7 (36)
Spells initiated in or after 1950 15.7 (168) 9.4 (71) 20.3 (97)
and completed by 1990 7.3 (67) 5.1 (33) 9.4 (34)
and in course by 1990 21.3 (101) 13.0 (38) 26.2 (63)

Democracies Parliamentary Mixed Presidential

Spells completed between 1950 11.2 (19) 4.5 (4) 9.4 (25)
and 1990
Spells in course by 1990 41.9 (36) 28.8 (5) 24.4 (19)
Age of regime as of 1950 31.2 (21) 7.0 (2) 20.2 (11)
Spells initiated in or after 1950 10.7 (34) 9.7 (7) 8.2 (33)
and completed by 1990 4.5 (14) 4.5 (4) 5.3 (17)
and in course by 1990 15.0 (20) 16.7 (3) 11.2 (16)

Dictatorships Bureaucracy Autocracy

Spells completed between 1950 20.9 (59) 9.3 (22)

and 1990
Spells in course in 1990 26.0 (87) 13.9 (94)
Age of regime as of 1950 42.1 (28) 37.7 (8)
Spells initiated in or after 1950 12.7 (118) 7.3 (108)
and completed by 1990 7.7 (67) 6.3 (87)
and in course by 1990 19.3 (51) 11.7 (21)

a
Number of spells in parentheses.

France, Luxembourg, the Netherlands, Norway, and Switzerland26 –

had been established before 1900. In turn, the thirty-six dictatorships
that existed in 1950 had an average age of 43.8 years. Seventeen of
them had been established prior to or in 1870: Ethiopia, Liberia, the

26
We extended the age of a regime back as far as 1870. All regimes, democratic or author-
itarian, established before that date were recorded as having been established in 1870.
In Chile, democracy was first established in 1891, but there was a reversal in 1925.

46
Stability and Change of Political Regimes

Dominican Republic, El Salvador, Haiti, Honduras, Mexico, Nicaragua,

Bolivia, Paraguay, China, Iran, Nepal, Thailand, Romania, Turkey, and
the USSR (Russia).
However, democracies also lasted for shorter periods than dic-
tatorships among the countries that did not exist prior to 1950. Of
the 168 regimes that were established in or after 1950, 67 had died
by 1990. Thirty-three were democratic, and they lasted, on average,
5.1 years; 34 were authoritarian and lasted 9.4 years. Of the 101
remaining regimes, that is, the regimes that lasted beyond 1990,
38 were democracies, and 63 were dictatorships. The former, by 1990,
had lasted 13 years, and the latter 26.2 years. Dictatorships, thus,
tended to last longer than democracies, regardless of when they were
observed.
During the 1950–1990 period, most countries each lived under a
single regime. Of the 141 countries we observed, only 41 experienced
transitions between dictatorship and democracy. The remaining 100
countries never experienced regime transitions, and thus each ended
the period with the same regime with which it was first observed
(among these, 67 were dictatorships, and 33 were democracies).
Seventeen countries had just one transition each, of which twelve were
to democracy. The five countries where democracy gave way to dicta-
torships that lasted past 1990 are Laos, where democracy fell in 1959,
Congo in 1963, Sierra Leone in 1967, Somalia in 1969, and Sri Lanka
in 1977. Countries that started the period of observation as dictator-
ships and then established democracies that lasted beyond 1990 are
Colombia in 1958, Venezuela in 1959, the Dominican Republic in 1966,
Portugal in 1976, Spain in 1977, El Salvador and Nicaragua in 1984,
Bangladesh in 1986, Poland in 1989, and Bulgaria, Czechoslovakia,
and Hungary in 1990.27
Two countries began the period of observation as dictatorships,
experienced brief democratic interludes, and became dictatorships
again: Uganda (where democracy lasted for five years, from 1980 to
1984) and Indonesia (where democracy lasted for only two years, from
1955 to 1956). More typically, seven countries were democracies when
we first observed them, went through often long periods of dictator-
ship, and returned to democracy. These are Grenada, Brazil, Chile,
Ecuador, Uruguay, the Philippines, and Greece. Eight countries ex-
perienced three regime transitions: Nigeria, Panama, Suriname, and

27
We do not count East Germany, which we treat as having dropped from the sample in
1990.

47
 Democracies and Dictatorships

Myanmar started the period as democracies, became dictatorships,

returned to democracy, and ended as dictatorships, whereas Bolivia,
South Korea, Thailand, and Turkey began as dictatorships, experi-
mented with democracy, returned to dictatorships, and became democ-
racies again. Two countries had four transitions: Ghana and Pakistan.
Sudan and Honduras had five, Guatemala and Peru had six, and
Argentina, by far the record holder, had eight transitions between
democracy and dictatorship.28
Thus, the regime histories of particular countries are highly hetero-
geneous. Most regimes, as we saw, lasted for a long time, with a major-
ity of countries not experiencing any transition between democracy and
dictatorship during the 1950–1990 period. Some countries alternated
between dictatorship and democracy every few years: There would
be a coup d’état, and a dictatorship would be established; then, often
following another coup d’état, an election would be held and the de-
mocratically elected government would assume office, only to be over-
turned by yet another coup. In some countries this entire cycle occurred
once during the period; in others it occurred twice, and in Argentina
three times.
Moreover, systematic regional differences can be seen: Western
Europe was predominantly democratic, and Eastern Europe was com-
munist; in Africa, only Mauritius was democratic during its entire
history; except for Israel, Middle Eastern countries were dictatorships;
most of the Far Eastern countries, except Japan, were dictatorships;
South Asian countries experienced some transitions; and many, but
not all, Latin American regimes were highly unstable. Indeed, of the
97 transitions that occurred in the 141 countries between 1950 and
1990, 44 were in Latin America, which comprises eighteen countries29
(Table 1.6).
The fact that most countries each lived under the same regime
for most of the time between 1950 and 1990 does not mean that
their rulers or their political orientations or even their institu-
tional frameworks remained the same. The democratic regimes

28
Appendix 1.3 lists countries by the number of transitions they have experienced
between democracy and dictatorship.
29
The rate of transitions per country was highest in Latin America: 2.4. Latin America
was followed by Southeast Asia, where the rate was 1.57 transitions per country, and
South Asia, where the rate was 1.2 transitions per country. In all other regions (includ-
ing the countries of the Organization for Economic Cooperation and Development, the
OECD), the rate of transitions per country was well below 1.

48
 Stability and Change of Political Regimes

Table 1.6. Regimes and regime transitions by region

Years of Transitions to Years of Transitions to

Region democracy democracy dictatorship dictatorship

Sub-Saharan Africa 69 6 1,170 11

South Asia 87 3 97 3
East Asia 4 2 160 1
Southeast Asia 46 5 215 6
Pacific Islands 40 0 50 0
Middle East and North Africa 68 2 513 2
Latin America 366 25 372 19
Caribbean 130 2 80 3
Eastern Europe and 5 4 290 0
Soviet Union
Industrialized countries 908 3 60 1
Total 1,723 52 3,007 45

might be parliamentary, mixed, or presidential. Dictatorships, in

turn, might be “bureaucracies,” institutionalized regimes that pro-
mulgated laws, or “autocracies,” regimes without any proclaimed
rules. Using these distinctions, we observed 55 parliamentary, 9
mixed, and 44 presidential democracies, 146 bureaucracies, and 116
autocracies.
The staying power of democratic institutions was seen to be strong.
During the entire period studied, democratic institutional frameworks
were altered in only three instances: France in 1958, when the par-
liamentary system of the Fourth Republic gave way to the mixed system
of the Fifth; Brazil in 1961, when presidentialism was replaced by a
mixed system; and Brazil again in 1963, when presidentialism was
restored after its overwhelming victory in a plebiscite held in January
of that year. A few countries did change the institutional framework in
their democracies after an authoritarian interregnum: Ghana, Nigeria,
and South Korea replaced the parliamentary systems that had existed
prior to their periods of dictatorship with presidential systems once
democracy was restored. In Suriname the change was from parlia-
mentarism to a mixed system. Pakistan was the only country that went
back and forth: from parliamentarism in 1950–1955 to a mixed system
in 1972–1976 and back to parliamentarism again after 1988. All of the
other seventeen countries that experienced at least one authoritarian

49
 Democracies and Dictatorships

interlude went back to the type of democratic institutions that had

existed before the authoritarian regime.30
Authoritarian institutions, on the other hand, proved highly unsta-
ble. When we classify dictatorships according to the presence or
absence of legislatures, we count 262 regimes, as opposed to 133 when
we do not make any distinctions among them. There are, thus, 129
instances of openings and closings of legislatures (65 cases of closing,
and 64 cases of opening). Again, a few countries account for a large
proportion of the transitions from one type of authoritarianism to the
other. Of the sixty-seven countries that remained under authoritarian
regimes from 1950 through 1990, thirty-two experienced only one type
of dictatorship: twenty-seven as bureaucracies and five as autocracies.
The remaining thirty-five countries experienced seventy-six transitions,
an average of 2.2 changes per country, from one type of authoritari-
anism to the other.31
Autocracies often emerge when democracy is overthrown and the
legislature is temporarily or permanently closed: Of the forty-five cases
of democratic breakdown, thirty-one resulted in this type of dictator-
ship. But autocracies can also emerge as a result of abortive attempts
to liberalize bureaucratic dictatorships. Indeed, the cases in which an
autocracy followed a bureaucracy were most frequent, suggesting that
attempts at liberalization often fail: Of the eighty-seven instances in
which bureaucratic regimes died, twenty-two ended in democracy, but
sixty-five in autocracy.
Autocracy is not an easily sustainable form of authoritarianism. Only
the four Persian Gulf monarchies (Oman, Qatar, Saudi Arabia, and the
United Arab Emirates) were autocracies during the entire period. In
general, autocracies do not last very long: As Table 1.5 indicates, the
average duration of autocratic spells completed by 1990 was 9.3 years,
and of those still in course in 1990 the duration was 13.9 years, com-
pared with 20.9 and 26 years, respectively, for bureaucracies. More-
over, of all the regimes, democratic and authoritarian, autocracies are
the ones at highest risk: During any year, an autocracy has a 10.56
percent chance of experiencing a transition to a different regime, which
compares with 5.12 percent for presidential democracies, 4.11 percent

30
These countries are Sudan, Grenada, Guatemala, Honduras, Panama, Argentina,
Bolivia, Brazil, Chile, Ecuador, Peru, Uruguay, Myanmar, the Philippines, Thailand,
Greece, and Turkey.
31
Benin, Burkina Faso, Morocco, Kuwait, and Jordan were the most unstable authori-
tarian regimes according to this measure: They changed between bureaucracy and
autocracy four, five, five, six, and six times, respectively.

50
 Stability and Change of Political Leadership

Table 1.7. Transitions between political regimes: parliamentarism, mixed,

presidentialism, bureaucracies, and autocracies

Transition to:
Transition
from: Parl Mix Pres Bur Aut Total Number of years Probability

Parl — 1 0 6 12 19 1,085 0.0175

Mix 0 — 1 1 2 4 150 0.0267
Pres 0 1 — 7 17 25 488 0.0513
Bur 8 4 10 — 65 87 2,117 0.0411
Aut 9 0 21 64 — 94 890 0.1056
Total 17 6 32 78 96 223 4,730 0.0471

for bureaucracies, 2.66 percent for mixed democracies, and 1.75

percent for parliamentary democracies (Table 1.7).

Stability and Change of Political Leadership

Rulers changed within each regime. By “rulers” we mean the chief
executives, to whom we refer as “heads” of government, or simply
“heads.” These are presidents in presidential democracies, prime min-
isters in the parliamentary and mixed democracies, and whoever is the
effective ruler in dictatorships. The latter sometimes can be designated
explicitly as dictators, or they may opt for a variety of other titles: heads
of military juntas, presidents, leaders of their ruling parties, executors
of the state of emergency, or kings.
No changes of heads occurred during 3,927 years, one change
occurred in 615 years, two changes in 101 years, three in 14, four in
3, and five in 2 years.32 Thus, altogether there were 881 changes of
heads during the period we observed, once every 5.29 years (Table
1.8). Changes were more frequent in democracies than in dictatorships.
Chief executives in democratic regimes were changed once every 3.48
years, with no significant difference between prime ministers (3.41
years when we combine parliamentary and mixed regimes)33 and

32
This adds to 4,662 years. The difference from the total of 4,730 is due to the exclusion
of Switzerland, Uruguay up to 1966, and Yugoslavia after 1980, each of which had a
collective executive.
33
Separately, the average is 3.77 for prime ministers in parliamentary regimes and 2.03
for prime ministers in mixed regimes.

51
 Democracies and Dictatorships

Table 1.8. Distribution of changes of chief executives (HEADS) by regimea

Number of Dem Parl Mix Pres Dict Bur Aut Total

changes of
heads by year

0 1,254 838 89 327 2,673 1,935 738 3,927

1 354 212 50 92 261 146 115 615
2 49 31 9 9 52 23 29 101
3 6 2 2 2 8 1 7 14
4 2 2 0 0 1 0 1 3
5 0 0 0 0 2 2 0 2
Total 1,665 1,085 150 430 2,997 2,107 890 4,662
No. of changes 478 288 74 116 403 205 198 881
Average duration 3.48 3.77 2.03 3.71 7.44 10.28 4.49 5.29

a
Excludes Switzerland, Uruguay until 1966, and Yugoslavia after 1981 because they had col-
lective executives.

presidents (3.71 years). The difference, however, was large across

types of dictatorships: Whereas in autocracies we observed one change
of chief executive every 4.49 years, in bureaucracies we observed one
change every 10.28 years, with an average for all dictatorships equal
to 7.44 years.
Some incumbents experienced changes in their political regimes
while in office. This happened during the tenure of sixty-eight chief
executives; fifty-four of them survived one change of regime, nine sur-
vived two changes, three survived three changes, and one each sur-
vived five and six changes. Most of these changes were between
different types of dictatorships and were due to the opening and closing
of legislatures.34 A few, however, were from democracy to dictatorship
or vice versa: for example, from presidentialism to autocracy in
Uruguay under Juan Bordaberry in 1973, and from autocracy to pres-

34
The most extreme case is Jordan, where the legislature was closed for one year in
1966, as well as during 1974–1984 and again during 1985–1989, representing six
regime changes from bureaucracy to autocracy. Other cases of frequent changes of
regimes without a change of chief executive are as follows: Morocco, where the legis-
lature was closed during 1963–1965, 1970–1972, and in 1978; Burkina Faso, where
Sangoulé Lamizana allowed an elective legislature to convene in 1970 only to close it
in 1974 and reopen it in 1978; Laos, where Souvanna Phouma closed the legislature
two times, in 1966 and 1974; and Nepal, where King Mahendra experimented with
legislative bodies in 1959 and 1963.

52
Stability and Change of Political Leadership

Table 1.9. Average duration (in years) of chief executives’ spellsa by

regime

Type of spell Average Maximum N

All 5.8 44 858

b
Censored 8.1 39 138
Not censored 5.4 44 720
Regime changec 13.3 39 68
No regime change 5.2 44 790
No regime change and not censored 4.9 44 671
Democracies 3.7 23 395
Parliamentary 3.9 23 242
Mixed 2.3 7 57
Parliamentary and mixed 3.6 23 295
Presidential 4.2 12 96
Dictatorships 6.6 44 276
Bureaucracies 8.0 36 167
Autocracies 4.5 44 109
Regime change and not censored 11.8 38 49
Regime change and censored 17.2 39 19
No regime change and censored 6.6 31 119

a
Continuous years of ruling by the same person.
b
Spells in course by 1990.
c
A spell with regime change is one during which the incumbent changed the type of
political regime.

identialism in Nicaragua under Daniel Ortega Saavedra in 1984. Once

we take those two facts into consideration, we find that prime minis-
ters in both parliamentary and mixed regimes had the shortest average
tenure (3.6 years), that the durations of democratic presidents and
autocratic rulers were about the same (4.2 and 4.5 years, respectively),
and that the chief executives in bureaucracies were the ones who lasted
the longest (8 years) (Table 1.9).
A similar conclusion follows when we examine the rates of leader-
ship turnover, defined as the annual number of changes in chief exec-
utive accumulated over the life span of a regime. As a benchmark,
consider that the average turnover rate for all the countries we

53
 Democracies and Dictatorships

Table 1.10. Leadership turnover

rates by regime typea

Regime Turnover rate

Democracy 0.164
Parliamentary 0.140
Mixed 0.271
Presidential 0.186
Dictatorship 0.073
Bureaucracy 0.047
Autocracy 0.190
All 0.103

a
Excluding spells in course in 1990.

observed was 0.17, somewhat less frequently than once every five
years. Because, on the average, we observed each country for about
thirty-three years, this turnover rate corresponds to an average of
about six changes of chief executive per country.35 When we consider
leadership turnover rates across political regimes, we find a similar
rate only in democracies, where it is 0.16 (Table 1.10). The turnover
rate is higher in autocracies and in mixed presidential democracies.
These rates imply that an average democratic spell experiences about
eighteen changes of leadership, whereas an average authoritarian spell
experiences nine changes. Among democracies, the number of changes
of heads is thirteen in the average parliamentary regime, and six in
the average presidential regime.

Conclusion
These are, then, the basic facts about political regimes in the world
between 1950 and 1990. Democracy is a system in which incumbents
lose elections and leave office when the rules so dictate. Dictatorships
are a residual category: If a political regime is not democratic, we con-
sider it to be a dictatorship of one stripe or another. Moreover, we do
not distinguish between dictatorships that succeed one another.
35
In this case, changes in chief executives were accumulated over the entire period during
which we observed each country, regardless of political regime.

54
Appendix 1.1: Alternative Approaches

In most cases it is simple to apply this conception of democracy to

classify the regimes that existed in the particular countries at the par-
ticular moments. All one needs to do is to observe whether or not the
chief executive was elected, whether or not the legislature was elected,
and whether or not there was political opposition. In some cases,
however, history did not provide the necessary evidence: There was an
opposition, officials were elected, but the same party always won. All
one can do in such instances is to decide which error to avoid.
The resulting classification of regimes is not idiosyncratic. Whereas
we were concerned to justify our approach theoretically and to ground
the classification on observations, rather than judgments, our classifi-
cation is almost identical with those produced by several alternative
scales of democracy (see Appendix 1.1). Indeed, it seems that in spite
of all their conceptual and observational differences, the various
approaches yield highly similar classifications of regimes. Hence, there
is no reason to think that the results that follow depend on the partic-
ular way regimes were classified.
In the chapters that follow, we first explain some of the patterns
described here and then explore their consequences for economic per-
formance and material well-being.

Appendix 1.1: Alternative Approaches

Conceptually, our scale is close to that of Bollen (1980), as well as
that of Coppedge and Reinicke (1990). Bollen used four indicators: (1)
whether or not elections were fair, (2) whether or not the chief execu-
tive was elected, (3) whether or not the legislature was elected, and
(4) whether or not the legislature was effective. Coppedge and Reinicke
coded answers to three questions: (1) whether or not elections pre-
sented voters with a meaningful choice, (2) whether or not the out-
come was affected by significant fraud, and (3) whether all or some or
no political organizations were banned. We used Bollen’s second and
third dimensions and Coppedge and Reinicke’s third dimension. We
did experiment with Banks’s measure of legislative effectiveness, but
found his assessments too unreliable. It is clear that allegations of
fraud are even more frequent than its actual occurrence, and by all
indications some fraud is a ubiquitous phenomenon in democracies.
Screaming “Fraud!” is just part of the standard repertoire of de-
mocratic competition. Indeed, there are cases in which the opposition
has withdrawn from the competition, claiming that the elections would

55
 Democracies and Dictatorships

not be conducted fairly. We conclude that there is no way to assess the

validity of such allegations in a standardized way. For example, the
opposition decided not to contest the 1984 Nicaraguan elections, but
some of its leaders later expressed regret about 1984 once they dis-
covered that they had won the subsequent elections in 1990. Hence,
although our approach is theoretically akin to those of Bollen and
Coppedge and Reinicke, we have tried to the extent possible to avoid
subjective judgments by relying only on observables. The Gurr (1990)
measure in Polity II is conceptually somewhat different, because it con-
siders the limited character of the government by coding “constraints
on the chief executive.” His assessments, however, are not easy to
reproduce.
Although we have been careful to specify our understanding of
democracy and to distinguish it from some rival conceptions, it appears
that from a practical point of view alternative measures of democracy
generate highly similar results. The dimensions used to assess whether
or not and to what extent a particular regime is democratic seem to
make little difference.36 To cite Inkeles (1990: 5–6), “the indicators most
commonly selected to measure democratic systems generally form a
notably coherent syndrome, achieving high reliability as measurement
scales. . . . A testimonial to the robustness of the underlying common
form and structure of the democratic systems is found in the high
degree of agreement produced by the classification of nations as de-
mocratic or not, even when democracy is measured in somewhat
different ways by different analysts. . . . Thus Coppedge and Reinicke,
following a quite independent theoretical model, end up with a scale
of polyarchy which correlates .94 with Gastil’s civil liberties measure
for some 170 countries in 1985. Gurr’s measure performs similarly
in relation to Bollen’s [and] his ratings of 118 countries circa 1965
correlate .83 with Bollen’s measure and .89 with a score combining
Gastil’s separate measures of political and civil liberties for 113 coun-
tries in 1985.”
Our measure is no exception. The Coppedge-Reinicke scale for 1978
predicts 92 percent of our dichotomous regimes, the Bollen 1965 scale
predicts 85 percent, and the Gurr scales of Autocracy and Democracy
for 1950–1986 jointly predict 91 percent. The Gastil scale of political
liberties, covering the period from 1972 to 1990, predicts 93.2 percent

36
Note, however, that different measures appear to be biased in somewhat different direc-
tions. See Bollen (1993).

56
Appendix 1.1: Alternative Approaches

of our classification; his scale of civil liberties predicts 91.5 percent;

and the two scales jointly predict 94.2 percent of our regimes.37 Hence,
our classification is by no means idiosyncratic. Different views of
democracy, including those that entail highly subjective judgments,
yield a robust classification.
The main difference between our approach and the alternatives is
that we use a nominal classification, rather than a ratio scale. We
believe that although some regimes are more democratic than others,
unless the offices are contested, they should not be considered
democratic. The analogy with the proverbial pregnancy is thus that
whereas democracy can be more or less advanced, one cannot be half-
democratic: There is a natural zero point. Note that Bollen and Jackman
(1989) are confused: It is one thing to argue that some democracies are
more democratic than others, but it is another to argue that democracy
is a continuous feature over all regimes, that is, that one can distinguish
the degrees of “democracy” for any pair of regimes.38
Bollen and Jackman (1989: 612) argue that difficulties in classifying
some cases speak in favor of using continuous scales: “Dichotomizing
democracy,” in their view, “blurs distinctions between borderline
cases.” Yet why are there “borderline cases”? Suppose that we have
defined democracy and not-democracy, established operational rules,
and found that some cases cannot be unambiguously classified by these
rules. Does this mean that there are borderline cases and that democ-
racy is thus “inherently continuous”? And should we stick the cases
that cannot be unambiguously classified, given our rules, into an “inter-
mediate” category, halfway between democracy and dictatorship? That
view strikes us as ludicrous. If we cannot classify some cases given our
rules, all this means is that we either have unclear rules or have insuf-
ficient information to apply them.
We have already seen that some “borderline cases” constitute sys-

37
Because other scales are ordinal (and pretend to be cardinal), whereas ours is nominal,
we use probit maximum likelihood to predict our classification on the basis of these
scales.
38
They also argue by assertion, referring to “the inherently continuous nature of the
concept of political democracy” (1989: 612), claiming that “since democracy is con-
ceptually continuous, it is best measured in continuous terms” (p. 612), and that
“democracy is always a matter of degree” (p. 618). Hence, in their view, the “degrees
of democracy” in Mexico, in Salazar’s Portugal, and in Franco’s Spain were different.
How they decide that “democracy is conceptually continuous,” whatever that means,
remains mysterious, but we are admonished that “it is important that the measure-
ment history of this construct not repeat itself” (p. 612).

57
 Democracies and Dictatorships

tematic error, whereas others bring random error. Systematic errors

can be treated by explicit rules, such as our “alternation” rule, and
their consequences can be examined statistically. There are some
regimes that cannot be unambiguously classified on the basis of all
the evidence produced by history. Because history produces a biased
sample of democracies – sampling is endogenous (Pudney 1989) – we
must revert to counterfactual judgments. In such cases we must decide
which error we prefer to avoid: classifying as democracies regimes that
may not be democracies, or rejecting as democracies regimes that may
in fact qualify. Yet, once this decision is made, the classification is
unambiguous. Mexico is not a regime intermediate between democracy
and dictatorship, not a “borderline case.” It is a regime in which the
ruling party allows some contestation but always wins: either a democ-
racy or a dictatorship, depending in which direction one wants to err
systematically.
In turn, some errors that are random with regard to the rules will
remain, and we will have to live with them. But errors are errors, not
“intermediate” categories. And there are no grounds to think that
a finer classification would be more precise. A finer scale would gen-
erate smaller errors, but more of them, and a rougher scale would
generate larger errors, but fewer of them. And if errors of larger mag-
nitude are less likely, the dichotomous scale will have a lower expected
error.
Suppose that the true nature of democracy lies on a J-point scale,
j = 1, . . . , J, but its measurement is subject to error. Let the unobserved
true score be DT and the assigned value D, and let the probability of a
j-point error be P( j) = Pr{|D - DT| = j} = aj. The reliability of the scale
is then Pr{|D - DT| = 0} = 1 - Sj=1 Pr( j). Assume that the distribution of
the true observations is uniform. Then the expected value of the error
will be

E ( D - DT ) = Â j =1 Pr( j ) * j * 2( J - j ),

where the first factor is the probability of an error of a given mag-

nitude, the second factor is the magnitude, and the third is the number
of such errors. Assume, as an illustration, that the probability of
making an error of magnitude 1 is a = 0.2, so that Pr( j = 0) ~ - 0.75.
Suppose that this is a Gastil scale, with seven points. Then the expected
error for seven observations will be about 3.5.
Now dichotomize this seven-point scale in such a way that if D £ 4,

58
Appendix 1.2: Classification of Regimes

then the assigned score is D = 2.5 (which is the midpoint value for one
regime), and if D > 4, then the assigned score is D = 5.5 (midpoint for
the other regime), so that each error costs three points on the seven-
point scale. Let the probabilities of errors and the distributions of the
true scores on the seven-point scale be the same. Then the expected
value of the error is

E ( D - DT ) = Â j =1 Pr( j ) * 3 * 2[d * j + (1 - d )( J - j )],

where the last factor in each expression is the number of relevant

errors (e.g., the only relevant one-point error is between 4 and 5, and
there are two of them, misclassifying 4 as 5 or 5 as 4), and d = 1 if
j £ 4, and d = 0 otherwise. At a = 0.2, the expected error for seven
observations of a dichotomous scale will be about 2.
Hence, there is less measurement error when a dichotomous scale
is used. If the distribution of true observations is unimodal and close
to symmetric, a more refined classification will have a smaller error,
but in fact observations on all the polychotomous scales tend to be U-
shaped, which advantages a dichotomous classification even more than
our example with the uniform distribution.
In sum, we think that our classification has some advantages. First,
it is grounded in theory. Second, it is based exclusively on observed
facts. Third, it separates cases subject to systematic error. Fourth,
it contains less random error than polychotomous scales. Finally, it
covers every year for 141 countries during forty-one years.

Appendix 1.2: Classification of Political Regimes, 1950–1990

At least some of the years for regimes marked with asterisks have
been classified as bureaucracies on the basis of our “alternation”
rule.
Country Regime Entry Exit

1. Algeria Bureaucracy 1962 1964

Autocracy 1965 1976
Bureaucracy 1977 1990
2. Angola Autocracy 1975 1979
Bureaucracy 1980 1990
3. Benin Bureaucracy 1960 1964

(continued)

59
 Democracies and Dictatorships

Country Regime Entry Exit

Autocracy 1965 1978

Bureaucracy 1979 1989
Autocracy 1990 1990
4. Botswana Bureaucracy* 1966 1990
5. Burkina Faso Bureaucracy 1960 1965
Autocracy 1966 1969
Bureaucracy 1970 1973
Autocracy 1974 1977
Bureaucracy* 1978 1979
Autocracy 1980 1990
6. Burundi Bureaucracy 1962 1965
Autocracy 1966 1981
Bureaucracy 1982 1986
Autocracy 1987 1990
7. Cameroon Bureaucracy 1960 1970
Autocracy 1971 1972
Bureaucracy 1973 1990
8. Cape Verde Bureaucracy 1975 1990
9. Central African Republic Bureaucracy 1960 1965
Autocracy 1966 1986
Bureaucracy 1987 1990
10. Chad Bureaucracy 1960 1974
Autocracy 1975 1990
11. Comoros Autocracy 1975 1977
Bureaucracy 1978 1990
12. Congo Presidentialism 1960 1962
Bureaucracy 1963 1976
Autocracy 1977 1978
Bureaucracy 1979 1990
13. Djibouti Bureaucracy 1977 1990
14. Egypt Bureaucracy* 1950 1990
15. Ethiopia Autocracy 1950 1956
Bureaucracy 1957 1973
Autocracy 1974 1986
Bureaucracy 1987 1990
16. Gabon Bureaucracy* 1960 1990
17. Gambia Bureaucracy* 1965 1990
18. Ghana Bureaucracy 1957 1964
Autocracy 1965 1969

(continued)

60
Appendix 1.2: Classification of Regimes

Country Regime Entry Exit

Parliamentarism 1970 1971

Autocracy 1972 1978
Presidentialism 1979 1980
Autocracy 1981 1990
19. Guinea Bureaucracy 1958 1983
Autocracy 1984 1990
20. Guinea-Bissau Bureaucracy 1974 1990
21. Ivory Coast Bureaucracy* 1960 1990
22. Kenya Bureaucracy 1963 1990
23. Lesotho Bureaucracy 1966 1969
Autocracy 1970 1990
24. Liberia Bureaucracy 1950 1979
Autocracy 1980 1984
Bureaucracy 1985 1989
Autocracy 1990 1990
25. Madagascar Bureaucracy* 1960 1971
Autocracy 1972 1976
Bureaucracy 1977 1990
26. Malawi Bureaucracy 1964 1990
27. Mali Bureaucracy 1960 1967
Autocracy 1968 1981
Bureaucracy 1982 1990
28. Mauritania Bureaucracy 1960 1977
Autocracy 1978 1990
29. Mauritius Parliamentarism 1968 1990
30. Morocco Autocracy 1956 1962
Bureaucracy 1963 1964
Autocracy 1965 1969
Bureaucracy 1970 1971
Autocracy 1972 1976
Bureaucracy 1977 1990
31. Mozambique Bureaucracy 1975 1990
32. Niger Bureaucracy 1960 1973
Autocracy 1974 1990
33. Nigeria Parliamentarism 1960 1965
Autocracy 1966 1978
Presidentialism 1979 1982
Autocracy 1983 1990
34. Rwanda Bureaucracy 1962 1972

(continued)

61
 Democracies and Dictatorships

Country Regime Entry Exit

Autocracy 1973 1980

Bureaucracy 1981 1990
35. Senegal Bureaucracy* 1960 1990
36. Seychelles Bureaucracy 1976 1990
37. Sierra Leone Parliamentarism 1961 1966
Autocracy 1967 1967
Bureaucracy 1968 1990
38. Somalia Mixed 1960 1968
Autocracy 1969 1978
Bureaucracy 1979 1990
39. South Africa Bureaucracy* 1950 1990
40. Sudan Parliamentarism 1956 1957
Autocracy 1958 1964
Parliamentarism 1965 1968
Bureaucracy 1969 1984
Autocracy 1985 1985
Parliamentarism 1986 1988
Autocracy 1989 1990
41. Swaziland Bureaucracy 1968 1972
Autocracy 1973 1977
Bureaucracy 1978 1990
42. Tanzania Bureaucracy 1961 1990
43. Togo Bureaucracy 1960 1966
Autocracy 1967 1978
Bureaucracy 1979 1990
44. Tunisia Bureaucracy 1956 1990
45. Uganda Bureaucracy 1962 1970
Autocracy 1971 1979
Presidentialism 1980 1984
Autocracy 1985 1990
46. Zaire Autocracy 1960 1960
Bureaucracy 1961 1962
Autocracy 1963 1969
Bureaucracy 1970 1990
47. Zambia Bureaucracy 1964 1990
48. Zimbabwe Bureaucracy 1965 1990
49. Bahamas Parliamentarism 1973 1990
50. Barbados Parliamentarism 1966 1990
51. Belize Parliamentarism 1981 1990

(continued)

62
Appendix 1.2: Classification of Regimes

Country Regime Entry Exit

52. Canada Parliamentarism 1950 1990

53. Costa Rica Presidentialism 1950 1990
54. Dominican Republic Bureaucracy 1950 1961
Autocracy 1962 1965
Presidentialism 1966 1990
55. El Salvador Bureaucracy 1950 1959
Autocracy 1960 1960
Bureaucracy* 1961 1983
Presidentialism 1984 1990
56. Grenada Parliamentarism 1974 1978
Autocracy 1979 1983
Parliamentarism 1984 1990
57. Guatemala Presidentialism 1950 1953
Bureaucracy 1954 1957
Presidentialism 1958 1962
Autocracy 1963 1965
Presidentialism 1966 1981
Bureaucracy 1982 1985
Presidentialism 1986 1990
58. Haiti Bureaucracy 1950 1985
Autocracy 1986 1989
Bureaucracy 1990 1990
59. Honduras Bureaucracy 1950 1955
Autocracy 1956 1956
Presidentialism 1957 1962
Autocracy 1963 1964
Bureaucracy* 1965 1970
Presidentialism 1971 1971
Autocracy 1972 1981
Presidentialism 1982 1990
60. Jamaica Parliamentarism 1962 1990
61. Mexico Bureaucracy* 1950 1990
62. Nicaragua Bureaucracy* 1950 1970
Autocracy 1971 1971
Bureaucracy* 1972 1978
Autocracy 1979 1983
Presidentialism 1984 1990
63. Panama Presidentialism 1950 1950
Bureaucracy 1951 1951

(continued)

63
 Democracies and Dictatorships

Country Regime Entry Exit

Presidentialism 1952 1967

Autocracy 1968 1977
Bureaucracy 1978 1990
64. Trinidad & Tobago Parliamentarism 1962 1990
65. United States Presidentialism 1950 1990
66. Argentina Presidentialism 1950 1954
Autocracy 1955 1957
Presidentialism 1958 1961
Autocracy 1962 1962
Presidentialism 1963 1965
Autocracy 1966 1972
Presidentialism 1973 1975
Autocracy 1976 1982
Presidentialism 1983 1990
67. Bolivia Bureaucracy 1950 1950
Autocracy 1951 1955
Bureaucracy* 1956 1963
Autocracy 1964 1978
Presidentialism 1979 1979
Autocracy 1980 1981
Presidentialism 1982 1990
68. Brazil Presidentialism 1950 1960
Mixed 1961 1962
Presidentialism 1963 1963
Bureaucracy 1964 1967
Autocracy 1968 1969
Bureaucracy 1970 1978
Presidentialism 1979 1990
69. Chile Presidentialism 1950 1972
Autocracy 1973 1989
Presidentialism 1990 1990
70. Colombia Bureaucracy 1950 1953
Autocracy 1954 1957
Presidentialism 1958 1990
71. Ecuador Presidentialism 1950 1962
Autocracy 1963 1967
Bureaucracy 1968 1969
Autocracy 1970 1978
Presidentialism 1979 1990

(continued)

64
Appendix 1.2: Classification of Regimes

Country Regime Entry Exit

72. Guyana Bureaucracy* 1966 1990

73. Paraguay Bureaucracy* 1950 1990
74. Peru Bureaucracy 1950 1955
Presidentialism 1956 1961
Autocracy 1962 1962
Presidentialism 1963 1967
Autocracy 1968 1979
Presidentialism 1980 1989
Bureaucracy 1990 1990
75. Suriname Parliamentarism 1975 1979
Autocracy 1980 1986
Bureaucracy 1987 1987
Mixed 1988 1989
Bureaucracy 1990 1990
76. Uruguay Presidentialism 1950 1972
Autocracy 1973 1984
Presidentialism 1985 1990
77. Venezuela Autocracy 1950 1951
Bureaucracy 1952 1958
Presidentialism 1959 1990
78. Bangladesh Autocracy 1971 1971
Bureaucracy 1972 1974
Autocracy 1975 1978
Bureaucracy* 1979 1981
Autocracy 1982 1985
Presidentialism 1986 1990
79. China, People’s Republic (PR) Autocracy 1950 1953
Bureaucracy 1954 1990
80. India Parliamentarism 1950 1990
81. Indonesia Autocracy 1950 1954
Parliamentarism 1955 1956
Bureaucracy 1957 1959
Autocracy 1960 1970
Bureaucracy 1971 1990
82. Iran Bureaucracy 1950 1960
Autocracy 1961 1962
Bureaucracy 1963 1983
Autocracy 1984 1990
83. Iraq Autocracy 1950 1950

(continued)

65
 Democracies and Dictatorships

Country Regime Entry Exit

Bureaucracy 1951 1957

Autocracy 1958 1979
Bureaucracy 1980 1990
84. Israel Parliamentarism 1950 1990
85. Japan Parliamentarism 1952 1990
86. Jordan Bureaucracy 1950 1965
Autocracy 1966 1966
Bureaucracy 1967 1973
Autocracy 1974 1983
Bureaucracy 1984 1984
Autocracy 1985 1988
Bureaucracy 1989 1990
87. South Korea Bureaucracy* 1950 1959
Parliamentarism 1960 1960
Bureaucracy 1961 1971
Autocracy 1972 1972
Bureaucracy 1973 1987
Presidentialism 1988 1990
88. Laos Parliamentarism 1954 1958
Bureaucracy 1959 1965
Autocracy 1966 1966
Bureaucracy 1967 1973
Autocracy 1974 1990
89. Malaysia Bureaucracy 1957 1968
Autocracy 1969 1970
Bureaucracy 1971 1990
90. Mongolia Bureaucracy* 1950 1990
91. Myanmar Parliamentarism 1950 1957
Autocracy 1958 1959
Parliamentarism 1960 1961
Autocracy 1962 1973
Bureaucracy 1974 1987
Autocracy 1988 1989
Bureaucracy 1990 1990
92. Nepal Autocracy 1950 1958
Bureaucracy 1959 1959
Autocracy 1960 1962
Bureaucracy 1963 1990
93. Pakistan Parliamentarism 1950 1955

(continued)

66
Appendix 1.2: Classification of Regimes

Country Regime Entry Exit

Bureaucracy 1956 1957

Autocracy 1958 1961
Bureaucracy* 1962 1971
Mixed 1972 1976
Autocracy 1977 1984
Bureaucracy 1985 1987
Parliamentarism 1988 1990
94. Philippines Presidentialism 1950 1964
Bureaucracy 1965 1971
Autocracy 1972 1977
Bureaucracy 1978 1985
Presidentialism 1986 1990
95. Singapore Bureaucracy* 1965 1990
96. Sri Lanka Parliamentarism 1950 1976
Bureaucracy* 1977 1990
97. Syria Bureaucracy 1950 1960
Autocracy 1961 1969
Bureaucracy 1970 1990
98. Taiwan Bureaucracy* 1950 1990
99. Thailand Bureaucracy 1950 1956
Autocracy 1957 1968
Bureaucracy 1969 1970
Autocracy 1971 1974
Parliamentarism 1975 1975
Autocracy 1976 1976
Bureaucracy 1977 1982
Parliamentarism 1983 1990
100. Yemen Arab Republic Autocracy 1967 1977
Bureaucracy 1978 1990
101. Austria Parliamentarism 1950 1990
102. Belgium Parliamentarism 1950 1990
103. Bulgaria Bureaucracy 1950 1989
Parliamentarism 1990 1990
104. Czechoslovakia Bureaucracy 1950 1989
Parliamentarism 1990 1990
105. Denmark Parliamentarism 1950 1990
106. Finland Mixed 1950 1990
107. France Parliamentarism 1950 1957
Mixed 1958 1990

(continued)

67
 Democracies and Dictatorships

Country Regime Entry Exit

108. East Germany Bureaucracy 1950 1990

109. West Germany Parliamentarism 1950 1990
110. Greece Parliamentarism 1950 1966
Autocracy 1967 1970
Bureaucracy 1971 1973
Parliamentarism 1974 1990
111. Hungary Bureaucracy 1950 1989
Parliamentarism 1990 1990
112. Iceland Mixed 1950 1990
113. Ireland Parliamentarism 1950 1990
114. Italy Parliamentarism 1950 1990
115. Luxembourg Parliamentarism 1950 1990
116. Malta Parliamentarism 1964 1990
117. Netherlands Parliamentarism 1950 1990
118. Norway Parliamentarism 1950 1990
119. Poland Bureaucracy 1950 1988
Mixed 1989 1990
120. Portugal Bureaucracy 1950 1975
Mixed 1976 1990
121. Romania Bureaucracy 1950 1990
122. Spain Autocracy 1950 1976
Parliamentarism 1977 1990
123. Sweden Parliamentarism 1950 1990
124. Switzerland Presidentialism 1950 1990
125. Turkey Bureaucracy* 1950 1960
Parliamentarism 1961 1979
Autocracy 1980 1982
Parliamentarism 1983 1990
126. United Kingdom Parliamentarism 1950 1990
127. Soviet Union Bureaucracy 1950 1990
128. Yugoslavia Bureaucracy 1950 1990
129. Australia Parliamentarism 1950 1990
130. Fiji Bureaucracy 1970 1986
Autocracy 1987 1990
131. New Zealand Parliamentarism 1950 1990
132. Papua New Guinea Parliamentarism 1975 1990
133. Solomon Islands Parliamentarism 1978 1990
134. Vanuatu Parliamentarism 1980 1990

(continued)

68
Appendix 1.3: Basic Data about Regime Dynamics

Country Regime Entry Exit

135. Western Samoa Autocracy 1962 1978

Bureaucracy* 1979 1990
136. Bahrain Autocracy 1971 1972
Bureaucracy 1973 1974
Autocracy 1975 1990
137. Kuwait Autocracy 1961 1962
Bureaucracy 1963 1975
Autocracy 1976 1980
Bureaucracy 1981 1985
Autocracy 1986 1990
138. Oman Autocracy 1951 1990
139. Qatar Autocracy 1971 1990
140. Saudi Arabia Autocracy 1950 1990
141. United Arab Emirates Autocracy 1971 1990

Appendix 1.3: Basic Data about Regime Dynamics

Asterisks indicate cases classified as regime transitions according to
the regular coding rules.

(A) Transitions to Dictatorships by Incumbents

Country Year

Cameroon 1963
Central African Republic 1962
Chad 1962
Djibouti 1982
Gabon 1967
Ghana 1972*
Kenya 1969
Lesotho 1970
Malawi 1966
Rwanda 1965
Sierra Leone 1967*
Uganda 1970
Zambia 1973

(continued)

69
 Democracies and Dictatorships

Country Year

Zimbabwe 1980
Ecuador 1970
Uruguay 1973*
Bangladesh 1975
South Korea 1972
Malaysia 1969
Pakistan 1956*
Philippines 1972
Turkey 1980*

(B) Countries by Regime Type and the Number of Transitions to

Authoritarianism Experienced by 1950
Democracies in 1950 that had experienced no transition to authori-
tarianism by then:

Canada
Guatemala
Panama
United States
Brazil
Ecuador
Uruguay
India
Israel
Myanmar
Pakistan
Philippines
Sri Lanka
Belgium
Denmark
France
Iceland
Ireland
Luxembourg
Netherlands
Norway
Sweden
Switzerland
United Kingdom

70
Appendix 1.3: Basic Data about Regime Dynamics

Australia
New Zealand

Dictatorships in 1950 that had experienced no transition to authori-

tarianism by then:

Egypt
Ethiopia
Liberia
South Africa
Dominican Republic
El Salvador
Haiti
Honduras
Mexico
Nicaragua
Bolivia
Paraguay
China (PR)
Indonesia
Iran
Iraq
Jordan
South Korea
Mongolia
Nepal
Syria
Taiwan
Thailand
Hungary
Romania
Turkey
Soviet Union
Saudi Arabia

Democracies in 1950 that had experienced at least one transition to

authoritarianism by then:

Costa Rica
Argentina
Chile
Austria
Finland

71
 Democracies and Dictatorships

West Germany
Greece
Italy

Dictatorships in 1950 that had experienced at least one transition to

authoritarianism by then:
Colombia
Peru
Venezuela
Bulgaria
Czechoslovakia
East Germany
Poland
Portugal
Spain
Yugoslavia

(C) Regime Transitions, by Country

NO TRANSITIONS (100 cases)
Dictatorships (67 cases)

Algeria 1962–1990
Angola 1975–1990
Benin 1960–1990
Botswana 1966–1990
Burkina Faso 1960–1990
Burundi 1962–1990
Cameroon 1960–1990
Cape Verde 1975–1990
Central African Republic 1960–1990
Chad 1960–1990
Comoros 1975–1990
Djibouti 1977–1990
Egypt 1950–1990
Ethiopia 1950–1990
Gabon 1960–1990
Gambia 1965–1990
Guinea 1958–1990
Guinea-Bissau 1974–1990

(continued)

72
Appendix 1.3: Basic Data about Regime Dynamics

Dictatorships (67 cases, cont.)

Ivory Coast 1960–1990

Kenya 1963–1990
Lesotho 1966–1990
Liberia 1950–1990
Madagascar 1950–1990
Malawi 1964–1990
Mali 1960–1990
Mauritania 1960–1990
Morocco 1956–1990
Mozambique 1975–1990
Niger 1960–1990

Democracies (33 cases)

Rwanda 1962–1990
Senegal 1960–1990
Seychelles 1976–1990
South Africa 1950–1990
Swaziland 1968–1990
Tanzania 1961–1990
Togo 1960–1990
Tunisia 1956–1990
Zaire 1960–1990
Zambia 1964–1990
Zimbabwe 1965–1990
Haiti 1950–1990
Mexico 1950–1990
Guyana 1966–1990
Paraguay 1950–1990
Bahrain 1971–1990
China (PR) 1950–1990
Iran 1950–1990
Iraq 1950–1990
Jordan 1950–1990
Kuwait 1961–1990
Malaysia 1957–1990
Mongolia 1950–1990
Nepal 1950–1990
Oman 1951–1990

(continued)

73
 Democracies and Dictatorships

Democracies (33 cases, cont.)

Qatar 1971–1990
Saudi Arabia 1950–1990
Singapore 1965–1990
Syria 1950–1990
Taiwan 1950–1990
United Arab Emirates 1971–1990
Yemen 1967–1990
East Germany 1970–1988
Romania 1961–1989
Soviet Union 1950–1990
Yugoslavia 1950–1990
Fiji 1970–1990
Western Samoa 1962–1990
Mauritius 1968–1990
Bahamas 1973–1990
Barbados 1966–1990
Belize 1980–1990
Canada 1950–1990
Costa Rica 1950–1990
Jamaica 1962–1990
Trinidad & Tobago 1962–1990
United States 1950–1990
India 1950–1990
Israel 1950–1990
Japan 1952–1990
Austria 1950–1990
Belgium 1950–1990
Denmark 1950–1990
Finland 1950–1990
France 1950–1990
West Germany 1950–1990
Iceland 1950–1990
Ireland 1950–1990
Italy 1950–1990
Luxembourg 1950–1990
Malta 1964–1990
Netherlands 1950–1990
Norway 1950–1990

(continued)

74
Appendix 1.3: Basic Data about Regime Dynamics

Democracies (33 cases, cont.)

Sweden 1950–1990
Switzerland 1950–1990
United Kingdom 1950–1990
Australia 1950–1990
New Zealand 1950–1990
Papua New Guinea 1975–1990
Solomon Islands 1978–1990
Vanuatu 1980–1990

ONE TRANSITION (17 cases)

To Dictatorship (5 cases)

Congo 1963
Sierra Leone 1967
Somalia 1968
Sri Lanka 1977

To Democracy (12 cases)

Laos 1958
Dominican Republic 1966
El Salvador 1984
Nicaragua 1984
Colombia 1958
Bangladesh 1985
Venezuela 1959
Bulgaria 1989
Czechoslovakia 1989
Hungary 1989
Poland 1989
Portugal 1975
Spain 1976

TWO TRANSITIONS (9 cases)

Dic Æ Dem Æ Dic (2 cases)

Uganda
Indonesia

75
 Democracies and Dictatorships

Dem Æ Dic Æ Dem (7 cases)

Grenada
Brazil
Chile
Ecuador
Uruguay
Philippines
Greece

THREE TRANSITIONS (8 cases)

Dem Æ Dic Æ Dem Æ Dic (5 cases)

Nigeria
Panama
Suriname
Myanmar
Thailand

Dic Æ Dem Æ Dic Æ Dem (3 cases)

Bolivia
South Korea
Turkey

FOUR TRANSITIONS (2 cases)

Ghana
Pakistan

FIVE TRANSITIONS (2 cases)

Sudan
Honduras

SIX TRANSITIONS (2 cases)

Guatemala
Peru

EIGHT TRANSITIONS (1 case)

Argentina

76
Appendix 1.4: The “Short” Data Base

Appendix 1.4: The “Short” Data Base

Because the economic data are not available for all the countries
and years described earlier, we shall be working with a somewhat
smaller data set. As the earliest year for which we have data on per
capita income is 1950, our observations on the rate of economic growth
begin in 1951. Moreover, because the patterns of economic develop-
ment for countries that rely for most of their income on oil are sui
generis, we excluded six countries in which the ratio of fuel exports to
total exports in 1984–1986 exceeded 50 percent.39 These limitations
delineate what we call our “short” data base.
The basic patterns that have been described remain unchanged as
we move to the smaller data set. Overall, we lose 604 observations,
171 in the six excluded oil-producing countries, and the rest where the
economic data are not available. The most significant losses are con-
centrated in East Asia (31.7 percent), the Pacific islands (30 percent),
and Eastern Europe (36.9 percent). Because of data unavailability, we
lose 78 years of democracy (4.5 percent) and 355 years of dictatorship
(11.8 percent).
In the end, thus, the data set with which we work in the rest of
this book contains observations for 1,645 years of democracy (1,022
of parliamentary democracies, 147 of mixed democracies, and 476 of
presidential democracies) and 2,481 years of dictatorship (1,812 of
bureaucracies and 669 of autocracies), for a total of 4,126 observa-
tions. They compose 99 spells of democracies (or 50 of parliamen-
tarism, 9 of mixed systems, and 43 of presidentialism) and 123 spells
of dictatorships (or 133 of bureaucracy and 98 of autocracy). This
yields thirty-nine transitions from democracy to dictatorship, and forty-
nine from dictatorship, to democracy.

39
These countries are Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, and the United Arab
Emirates.

77
 Chapter Two

Economic Development and

Political Regimes

Introduction
Any casual glance at the world will show that poor countries tend
to have authoritarian regimes, and wealthy countries democratic ones.
The question is why. What are the conditions that determine whether
democracy or dictatorship prevails? What causes political regimes to
rise, endure, and fall? Can their transformations be explained gener-
ally, or are they caused by circumstances idiosyncratic to each country
or period? Are they driven by economic development or by other
factors, such as the preceding political history, cultural traditions, polit-
ical institutions, or the international political climate?
We begin with the observation that the incidence of democracy is
undoubtedly related to the level of economic development. Having
established the central importance of development, we distinguish
two causal mechanisms that may generate this relationship, asking
whether democracies are more likely to emerge as countries develop
economically under dictatorships or, having emerged for reasons other
than economic development, are only more likely to survive in coun-
tries that are already developed. This analysis is extended first to the
impact of economic performance and then to a panoply of social and
political factors. A separate section is devoted to the impacts of differ-
ent types of democratic institutions. A summary closes the chapter. The
statistical models on which the analysis is based are presented in the
appendixes.

Development and Democracy

First advanced in 1959, S. M. Lipset’s observation that democracy
is related to economic development has generated the largest body of

78
Development and Democracy

research on any topic in comparative politics. It has been supported

and contested, revised and extended, buried and resuscitated. And yet,
though several articles in the Festschrift honoring Lipset (Marks and
Diamond 1992) proclaim conclusions, neither the theory nor the facts
are clear.
Aggregate patterns, such as that in Figure 2.1, show that the rela-
tionship between the level of economic development and the incidence
of democratic regimes is strong and tight. Indeed, one can correctly
predict 77.5 percent of the 4,126 annual observations of regimes just
by looking at per capita income.1 What remains controversial, however,
is the relative importance of the level of development as compared with
other factors, such as the political legacy of a country, its past history,
its social structure, its cultural traditions, the specific institutional
framework, and, last but not least, the international political climate.
To compare the impacts of different factors, we proceed as follows:
(1) We estimate the probability that the regime in a given country
during a particular year will have been a dictatorship or a democracy,
conditional on the value of each independent variable and their various
combinations. (2) We take as the predicted regime the one for which
this probability is higher, more than 0.50, and (3) We compare these
predictions with the observations. Such predictions obviously are quite
rough, because they do not distinguish whether the probability that a
regime is, say, a dictatorship is 0.99 or 0.51. But because all we need
is a yardstick with which to compare the influences of different factors,
this simple procedure is sufficient. Hence, our criterion in compar-
ing the effects of different variables is simply the number of correct
predictions. Given the controversies about comparing the fit of non-
linear models, we also provide, however, an alternative measure, the
Zavoina-McKelvey (1975) pseudo-R2.

1
These predictions are derived from probit, a form of non-linear regression, in which
the probability that a country i will have had a dictatorial (as opposed to democratic)
regime at time t is modeled as Pr(REGit = Dictatorship) = F(Xitb), where F(·) is the
cumulative distribution function (c.d.f.) of the normal distribution. A fair amount of ink
has been spilled over the issue whether or not the relationship between development
and democracy is linear (Jackman 1973; Arat 1988). We now know better. Democracy,
however it is measured, is a qualitative or limited variable (it assumes the value of 0 or
1 under our measurement, it ranges from 2 to 14 on the Freedom House scale, from 0
to 100 on the Bollen scale, and so on). Hence, no predicted index of democracy
can become negative as the level of development tends to zero, and no predicted index
of democracy can exceed whatever is the maximum value of a particular scale as the
level gets very large. Only a non-linear function, such as the normal or logistic (as
suggested by Dahl 1971), can satisfy these constraints. This is why we use probit
throughout.

79
 Figure 2.1. Probability that a regime will be democratic, by per capita income
e
Development and Democracy

Following the lead of extensive work reported by others, the factors

we consider include the following:

1. The level of development (LEVEL), as measured by per capita

income.2
2. The political legacies of a country, as summarized by two dummy
variables that indicate whether or not the country became indepen-
dent after 1945 (NEWC) and whether or not it was a British colony
in 1919 (BRITCOL).
3. The political history of the country, as indicated by the number of
past transitions to authoritarianism (STRA).3
4. The religious structure of the country, as indicated by proportions of
Catholics (CATH), Protestants (PROT), and Moslems (MOSLEM) in the
population.
5. The ethnolinguistic (ELF60) and religious (RELDIF) fractionalization
of the country, measured as the probability that two randomly
chosen individuals will not belong to the same group.
6. The international political environment, as measured by the pro-
portion of other democracies in the world (ODWP) during the par-
ticular year. By “other” we mean in countries other than the one
under consideration.

Table 2.1 shows the predictions based on different combinations of

these variables. How one evaluates these results depends on the null
model. If one knows absolutely nothing and is forced to think that a ran-
domly chosen country has a fifty-fifty chance of having either regime,
then correctly predicting more than half, 2,063 cases, is already an
improvement. Because we do know, however, that there were more dic-
tatorial years than democratic years, 60 versus 40 percent, one could
correctly predict 2,481 country-years by guessing that all countries were
dictatorships at all times. Except for ODWP, we list in Table 2.1 only those
variables that considered alone or in thematic combinations at least
match the random guess conditioned on the relative frequencies.

2
All income figures are expressed in 1985 constant purchasing power parity (PPP) dollars,
from the Penn World Tables (PWT 5.6). One might question that per capita income
is an adequate indicator of the level of development. In some countries it clearly is
not. That is why we excluded from our analysis six countries that derive most of their
income from oil revenues. Even with this exclusion, income may not correspond per-
fectly to whatever one means by “development.” Yet in our view it is the best indicator
– better than energy consumption, literacy, industrialization, and other alternative
measures.
3
But to discount somewhat the distant past, we assign the value of 1 for all the transi-
tions that occurred before 1950.

81
 Table 2.1. Predictions of probit models of regimes

Number of correct
predictions of

Democracies Dictatorships Total

N = 1,645 N = 2,481 N = 4,126 Proportion
Variables (40.0%) (60.0%) (100.0%) correct ZMa

LEVEL 973 2,226 3,199 77.5 0.63

Legacies: NEWC + BRITCOL 1,264 1,501 2,765 67.0 0.47
Religion: CATH + PROT + MOSLEM 1,074 1,697 2,771 67.2 0.49
ELF60 0 2,481 2,481 60.0 0.40
RELDIF 0 2,481 2,481 60.0 0.39
History: STRA 19 2,459 2,478 60.0 0.39
ODWP 68 2,376 2,444 59.2 0.39
LEVEL + legacies 1,003 2,201 3,204 77.7 0.63
LEVEL + religion 1,028 2,216 3,244 78.6 0.65
LEVEL + ELF60 1,020 2,279 3,299 80.0 0.66
LEVEL + RELDIF 969 2,225 3,194 77.4 0.63
LEVEL + history 974 2,227 3,201 77.6 0.63
LEVEL + ODWP 1,005 2,233 3,238 78.5 0.64
Everything 1,136 2,241 3,377 81.4 0.70

a
ZM stands for the Zavoina and McKelvey (1975) pseudo-R2.
e
Development and Democracy

Per capita income (LEVEL) clearly outperforms a random guess.

“Legacies,” the colonial experience, can improve on a random guess
by 7 percent, as can the distribution of religions. Ethnic and religious
heterogeneity, as well as the number of past transitions to dictatorship,
can correctly predict only as many cases as one would guess by
knowing the marginal distribution. The proportion of other democra-
cies in the world (ODWP) predicts poorly when considered alone.4
When considered along with LEVEL, no other variables, alone or in
their thematic combinations, can contribute much to the prediction.
Adding to LEVEL all the other variables can improve the guess by only
4 percent.5 Hence, all the evidence indicates that it is the level of devel-
opment that can best predict the incidences of various political regimes.
Whereas these variables accurately predicted 81.4 percent of the
regimes that countries had at any given time, there were 749 cases in
which the regime expected was not the observed one. Can we learn
anything from these mistakes? Table 2.2 lists the incorrect predictions
by country and period and also shows the predicted probability that
the regime would be a dictatorship. Several features of this list merit
comment:
1. In many countries, democracy was a colonial legacy, left to fend for
itself against all odds. That was true for the Congo, Nigeria, Sierra
Leone, Somalia, Suriname, Burma (Myanmar), Pakistan, the Philip-
pines, and Ceylon (Sri Lanka), where democracy did not survive. Yet
democracies that emerged from decolonialization survived in Mau-
ritius (which by our prediction should have become a democracy
only 14 years after independence, in 1982), in India (which was

4
We also examined another variable not reported in the table, namely, the proportion of
other democracies in the region (ODRP). This variable generates the best overall predic-
tion, which would seem to indicate that international factors operate through a process
of geographically based contagion, rather than via a worldwide political climate. We rush
to emphasize, however, that this results should be interpreted with skepticism, because
we estimated the model as if all the observations were independent or at least as if the
number of countries in each region were large, and in some regions it was not. Although
international determinants of regimes have been subjects of some speculation (White-
head 1991; Schmitter 1991), as well as of statistical analysis (Starr 1991), a model that
can distinguish different mechanisms through which international influences operate has
yet to be constructed and tested. Our findings suggest, at best, that the topic merits further
study, and we decided not to include the regional variable in our final analysis.
5
One could reasonably ask why we take LEVEL as the benchmark, asking by how much
other variables improve the prediction with regard to LEVEL, rather than vice versa.
The reason is the asymmetry: When “legacies” are added to LEVEL, the prediction
improves by 0.2 percent, but when LEVEL is added to “legacies”, the prediction
improves by 10.7 percent. Similarly for other variables.

83
 Table 2.2. Incorrect predictions, by country and period, and the probability that a
regime is a dictatorship during the period

Democracies predicted as dictatorships Dictatorships predicted as democracies

Pr(REG = Dic) Pr(REG = Dic)

Country Period highest Democracy Country Period lowest Dictatorship

Congo 1961–62 0.89 Fell 1963 Botswona 1986–89 0.42 Survived

Ghana 1970–71 0.86 Fell 1972 Gabon 1976–77 0.37 Survived
1979–80 0.82 Fell 1981 1990 0.49 Survived
Mauritius 1968–82 0.73 Survived Seychelles 1976–90 0.31 Fell 1993
Nigeria 1960–65 0.91 Fell 1966 South Africa 1950–61 0.40 Fell 1995
1979–82 0.87 Fell 1983 1974–76 0.48 Fell 1995
Sierra Leone 1962–66 0.86 Fell 1967 1979–90 0.31 Fell 1995
Somalia 1961–68 0.98 Fell 1969 El Salvador 1953 0.50 Fell 1984
Sudan 1986–88 0.87 Fell 1989 Mexico 1950–90 0.11 Survived
Belize 1981–90 0.75 Survived Panama 1980–90 0.34 Fell 1994
Costa Rica 1950–90 0.65 Survived Argentina 1955–57 0.19 Fell 1958
Dominican R. 1966–87 0.76 Survived 1962 0.26 Fell 1963
El Salvador 1984–89 0.61 Survived 1966–72 0.24 Fell 1973
Grenada 1985–90 0.80 Survived 1976–82 0.14 Fell 1983
Guatemala 1951–53 0.66 Fell 1954 Bolivia 1952 0.48 Fell 1979
1958–62 0.71 Fell 1963 Chile 1981 0.48 Fell 1990
1966–81 0.78 Fell 1982 1985–89 0.32 Fell 1990
1986–90 0.63 Survived Colombia 1950–57 0.44 Fell 1958

84
e
 Honduras 1957–62 0.75 Fell 1963 Paraguay 1952 0.50 Fell 1991
1971 0.82 Fell 1972 1955 0.50 Fell 1991
1982–90 0.75 Survived 1988–90 0.43 Fell 1991
Jamaica 1962–83 0.58 Survived Peru 1950–55 0.45 Fell 1956
Nicaragua 1984–90 0.65 Survived 1990 0.41 Survived
Panama 1952–67 0.65 Fell 1968 Uruguay 1979–81 0.45 Fell 1985
Bolivia 1979 0.63 Fell 1964 Venezuela 1951–58 0.05 Fell 1959
1982–89 0.63 Survived Iraq 1979–80 0.28 Survived
Brazil 1950–63 0.67 Fell 1964 South Korea 1986–87 0.37 Fell 1988
Chile 1951 0.52 Fell 1973 Malaysia 1988–90 0.31 Survived
1960 0.50 Fell 1973 Singapore 1973–90 0.02 Survived
1962–72 0.63 Fell 1973 Taiwan 1984–90 0.11 Fell 1995
Colombia 1960–82 0.69 Survived Bulgaria 1988–89 0.48 Fell 1990
Ecuador 1950–62 0.64 Fell 1963 East Germany 1979–88 0.10 Collapsed
1979–83 0.51 Survived Greece 1967–73 0.29 Fell 1974
Peru 1960 0.51 Fell 1962 Poland 1988 0.49 Fell 1989
1963–67 0.58 Fell 1968 Portugal 1952–75 0.37 Fell 1976
1980–83 0.53 Fell 1990 Spain 1951–76 0.11 Fell 1977
Suriname 1975–89 0.99 Fell 1990 Soviet Union 1988–89 0.45 Collapsed
Uruguay 1962–72 0.69 Fell 1973 Fiji 1990 0.47 Survived
1985 0.54 Survived
Bangladesh 1986–90 0.99 Survived
India 1950–83 0.83 Survived
Japan 1952–65 0.66 Survived
South Korea 1960 0.81 Fell 1961
Myanmar 1950–57 0.65 Fell 1958
1960–61 0.66 Fell 1962
Pakistan 1950–55 0.79 Fell 1956

(continued)
85
e
 Table 2.2 (continued)

Democracies predicted as dictatorships Dictatorships predicted as democracies

Pr(REG = Dic) Pr(REG = Dic)

Country Period highest Democracy Country Period lowest Dictatorship

1972–76 0.91 Fell 1977

1988–90 0.74 Survived
Philippines 1950–64 0.84 Fell 1965
1986–90 0.75 Survived
Sri Lanka 1950–76 0.94 Fell 1977
Thailand 1975 0.86 Fell 1976
1983–89 0.75 Survived
Bulgaria 1990 0.52 Survived
Czechoslovakia 1990 0.75 Survived
West Germany 1951 0.52 Survived
Greece 1963 0.50 Fell 1967
Hungary 1990 0.57 Survived
Poland 1989–90 0.54 Survived
Turkey 1961–79 0.95 Fell 1980
1983–90 0.87 Survived
Papua New Guinea 1975–85 0.64 Survived
Solomon Islands 1981–88 0.78 Survived
Vanuatu 1984–90 0.96 Survived

86
e
Development and Democracy

predicted as a dictatorship during the entire period), in Belize, in

Jamaica (which was predicted as a dictatorship until 1983), in Papua
New Guinea (until 1985), in the Solomon Islands (until 1988), and
in Vanuatu (not by 1990). The odds against democracy in India were
extremely high.
2. Some countries succeeded in living under democracy when all the
observable conditions conspired against it. These attempts suc-
ceeded in Bangladesh after 1986, Thailand after 1983, and Turkey
after 1983. They failed in Ghana (twice), in Sudan, and in South
Korea in 1960.
3. In some countries democracy was imposed by the Allies in the after-
math of World War II. These countries include West Germany, which
by our prediction should have become a democracy only by 1952,
and Japan, which should have become democratic only by 1965.
4. Some countries experienced dictatorships that should not have been
warranted by their observed conditions. That was true notably of
the four military dictatorships in Argentina, the Fujimori autogolpe
in Peru, the Perez Jimenez dictatorship in Venezuela between 1948
and 1958, and the military rule in Greece between 1967 and 1973.
5. Several countries have waited much longer to make the transit
to democracy than their conditions would predict. They include
Seychelles, South Africa, Taiwan, Chile, Portugal, and Spain. Other
countries have remained in the grip of authoritarianism even though
all the observable factors indicate that they should have had demo-
cratic regimes: Mexico, which we would have predicted as a democ-
racy already by 1951 and which had a probability of 0.11 of being
a dictatorship in 1990; Singapore, which had a 0.02 probability of
being a dictatorship in 1990, and Malaysia. Communist countries
are a topic apart, and we shall return to it.
6. Finally, these results cast a new light on the instability of political
regimes in many Latin American countries. It is often observed that
democracy is particularly unstable in Latin America. Yet that may be
a wrong conclusion: What is unstable in Latin America is dictator-
ship. If we take all countries in the world that fall in the Latin Amer-
ican income range per capita, from $971 to $8,233, we discover that
democracies are more likely in Latin America than in other regions:
Being in Latin America makes democracy 12 percent more likely
(t value of the dummy is 3.470) than elsewhere. It thus appears that
several Latin American countries experimented with democracy in
the face of adverse conditions under which countries elsewhere in the
world tended to remain in the grip of dictatorships, and under those
conditions, democracies had little chance for survival.

87
 Economic Development and Political Regimes

To summarize, the level of economic development, as measured by

per capita income, is by far the best predictor of political regimes. Yet
there are countries in which dictatorships persist when all the observ-
able conditions indicate they should not; there are others in which
democracies flourish in spite of all the odds. Thus some factors influ-
encing the incidence of the different kinds of regimes are not identi-
fied by this analysis.

Regime Dynamics
There are two distinct reasons that the incidence of democracy may
be related to the level of economic development: Democracies may be
more likely to emerge as countries develop economically, or, having
been established for whatever reasons, democracies may be more
likely to survive in developed countries. We call the first explanation
“endogenous” and the second “exogenous.”
Because we are dealing with only two kinds of regimes, democra-
cies emerge whenever dictatorships die.6 Hence, to assert that democ-
racies emerge as a result of economic development is the same as
saying that dictatorships die as countries ruled by them become eco-
nomically developed. Democracy thus is said to be secreted out of
dictatorships by economic development. A story told about country
after country is that as a country develops, its social structure becomes
complex, new groups emerge and organize, labor processes require the
active cooperation of employees, and, as a result, the system can no
longer be effectively run by command: The society is too complex, tech-
nological change endows the direct producers with autonomy and
private information, civil society emerges, and dictatorial forms of
control lose their effectiveness. Various groups, whether the bour-
geoisie, workers, or just the amorphous “civil society,” rise against the
dictatorial regime, and it falls.
The endogenous explanation is a “modernization” theory. The basic
assumption of this theory is that there is one general process, of which
democratization is but the final facet. Modernization consists of a
gradual differentiation and specialization of social structures culmi-
nating in a separation of the political from other structures, and making
democracy possible. The specific causal chains consist in sequences of

6
This is not quite true of our data set, because different countries enter and exit the
sample at different moments. For the moment, we consider the population of countries
as fixed.

88
Regime Dynamics

industrialization, urbanization, education, communication, mobiliza-

tion, political incorporation, and innumerable other “-ations”: a pro-
gressive accumulation of social changes that make a society ready to
proceed to the final one, democratization.
Modernization may be one reason that the incidence of democracy
is related to economic development, and this is the reading imputed
to Lipset by most commentators (Diamond 1992: 125; Huber,
Rueschemeyer, and Stephens 1993: 711). His most influential critic,
O’Donnell (1973: 3), paraphrases Lipset’s thesis as saying that “if other
countries become as rich as the economically advanced nations, it is
highly probable that they will become political democracies.” Democ-
racy is endogenous, because it results from economic development
under authoritarianism. The hypothesis is that if authoritarian coun-
tries develop, they become democratic. The sequence of events we
would thus expect to observe is one of poor authoritarian countries
developing and becoming democratic once they reach some level of
development, a “threshold.”
Yet suppose, just suppose, that dictatorships are equally likely to die,
and democracies to emerge, at any level of development. They may die
for so many different reasons that development, with all its moderniz-
ing consequences, plays no privileged role. After all, as Therborn
(1977) emphasized, many European countries became democratized
because of wars, not because of “modernization,” a story repeated by
the Argentine defeat in the Malvinas and elsewhere. Some dictator-
ships have fallen in the aftermath of the death of the founding
dictator, such as a Franco, uniquely capable of maintaining the author-
itarian order. Some have collapsed because of economic crises, some
because of foreign pressures, and perhaps some for purely idiosyn-
cratic reasons.
If dictatorships die and democracies emerge randomly with regard
to economic development, is it still possible that there should be more
democracies among wealthy countries than among poor countries? If
one is to judge Lipset (1959: 56) by his own words – “The more well-
to-do a nation, the greater the chances it will sustain democracy” –
then even if the emergence of a democracy is independent of the level
of development, the chance that this regime will survive will be greater
if it is established in an affluent country. We would thus expect democ-
racies to appear randomly with regard to levels of development, and
then to die in the poorer countries and to survive in the wealthier coun-
tries. And because every time a dictatorship happened to die in an
affluent country democracy would be there to stay, history should grad-

89
 Economic Development and Political Regimes

ually accumulate wealthy democracies. This is no longer a modern-

ization theory, because the emergence of democracy is not brought
about by development. Democracy appears exogenously, deus ex
machina. It tends to survive if a country is “modern,” but it is not a
product of “modernization.”
Some algebra may help elucidate what is entailed. Let the pro-
bability that a country, i = 1, . . . , N, will have an authoritarian regime
during a particular year, t = 1, . . . , T, be pA(it), where the subscript
“A” stands for “authoritarian,” and let the probability that it will have
a democratic regime be pD(it) = 1 - pA(it). Let the probability that a
dictatorship will die from one year to another be pAD(it), so that the
probability that it will survive is pAA(it) = 1 - pAD(it). Similarly, let
the probability that a democracy will die be pDA(it) = 1 - pDD(it). If
we assume for the time being that these “transition probabilities,”
pjk, j = A, D, k = A, D, are constant each year and are the same for all
countries, then we can describe the evolution of regimes by
pD (t ) = pDD pD (t - 1) + pAD pA (t - 1)
pA (t ) = pDA pD (t - 1) + pAA pA (t - 1).

Therefore the proportion of regimes that will be democracies next year

will depend on the proportion of democracies that survive from the
current year, pDD, and the proportion of dictatorships that will die, that
is, become democracies, pAD. Similarly for dictatorships. In matrix
form,

È pD (t )˘ È pDD pAD ˘ È pD (t - 1)˘

ÍÎ pA (t )˙˚ = ÍÎ pDA pAA ˙˚ ÍÎ pA (t - 1)˙˚
.

Given the transition rates, there exists a distribution of regimes that,

if reached, will remain stable in the absence of exogenous distur-
bances. These equilibrium probabilities are

pAD pDA
pD* = and pA* = .
pDA + pAD pDA + pAD

This long-run distribution of regimes depends only on the relative rates

at which they die each year, not on their initial distribution. If pAD >
pDA, then in the long run there will be more democracies than dicta-
torships in the world. Moreover, whatever the initial distribution of
regimes, their proportions will over time tend to these equilibrium
values. And because the probabilities that regimes will die during any
particular year are likely to be low – in fact they are low – this con-

90
Regime Dynamics

vergence will be monotonic, that is, the proportion of one regime will
continue to increase, and that of the other to decline.7 If at the begin-
ning the proportion of democracies was lower than pD*, this propor-
tion will continually increase over time, and the proportion of
dictatorships will decline.
Suppose now that whereas dictatorships die at some constant
annual rate, democracies never die, so that pDA = 0. Then in the
long run all countries will be democracies. Every time a dictator-
ship dies, a democracy is established, and, once it is established, it
survives forever. The speed of this process will depend on the rate at
which dictatorships die, but the accumulation of democracies is
inexorable.
Now, to return to the issue at stake, imagine that these transition
probabilities are not constant, but depend on the level of development.
To keep matters simple, suppose that there are only two levels: low (L)
and high (H). At the low level, both regimes have some probability of
dying that is more than zero and less than one. Now consider two pos-
sibilities: One is that while pAD(L) < 1, once dictatorships pass the
threshold that defines the high level, they are certain to die, so that
pAD(H) = 1, whereas democracies die at the same rate, pDA = pDA(L) =
pDA(H), at either level. The transition probabilities are thus

Level = low Level = high

È p DD p AD ˘ È p DD 1.00˘
ÍÎp DA p AA ˙˚ ÍÎp DA 0.00˙˚

Whereas the long-run proportion of democracies at the low level

will be pD*(L) = pAD/(pAD + pDA), at the high level it will be pD*(H) =
1/(1 + pDA), so that the equilibrium proportion of democracies will
be higher at the higher level, pD*(L) < pD*(H). Thus, the proportion
of democracies will be higher at the high level of development
because democracies are more likely to emerge as a result of devel-
opment. This is the “endogenous” (modernization) version of the
explanation.
But suppose, alternatively, that authoritarian regimes die at exactly
the same rate whether in poor countries or developed ones, so that
pAD(L) = pAD(H) = pAD, and in turn democratic regimes never die once
they are established in affluent countries, so that pDA(H) = 0.00. The
transition probabilities are then
7
Convergence will be monotonic if pAD + pDA < 1; otherwise, the proportions of regimes
will oscillate around the equilibrium.

91
 Economic Development and Political Regimes

Level = low Level = high

È p DD p AD ˘ È1.00 p AD ˘
ÍÎp DA p AA ˙˚ ÍÎ0.00 p AA ˙˚

and we already know that whereas the long-run proportion of democ-

racies at the low level will be pD*(L) < 1, at the high level all countries
will become democracies in the long run. Hence, we will observe an
aggregate relationship between the level of development and the inci-
dence of democracies even though democracies are equally likely to
emerge at any level, that is, even if development under authoritarian-
ism does not increase the probability that a country will become demo-
cratic. This is then the “exogenous” version.
Thus, to decide which mechanism generates the relationship
between development and democracy, we need to determine how the
respective transition probabilities change with the level of develop-
ment. Appendix 2.1 describes how we do it.

Level of Economic Development and Regime Dynamics

Examine first some descriptive patterns, presented in column 5 of
Table 2.3. If the theory according to which the emergence of democracy
is a result of economic development is true, transitions to democracy
should be more likely when authoritarian regimes reach higher levels
of development. In fact, dictatorships survive almost invariably in the
very poor countries, those whose per capita incomes are under $1,000,
or at least they succeed one another and the regime remains the same.8
They are less stable in countries with incomes between $1,001 and
$4,000, and even less so between $4,001 and $7,000. But if income
reaches the level of $7,000, the trend reverses and they become more
likely to survive. As the lower panel of Table 2.3 shows, transitions to
democracy are less likely in poor countries and in rich ones, but they
are more likely at the intermediate income levels. If we take all the dic-
tatorships, their probability of dying during any year is 0.0198; for those
with incomes over $1,000, this probability is 0.0280, over $5,000 it is
0.0526, over $6,000 it is 0.0441, and over $7,000 it is 0.0286; the two
very wealthy dictatorships with incomes above $8,000 still survived in
1990. Hence, it appears that Huntington was correct, albeit only with
regard to authoritarian regimes, when he argued that one should expect
to observe “a bell-shaped pattern of instability” (1968: 43). Economic
development seems to destabilize dictatorships in countries at interme-
8
Remember that we treat dictatorships that succeed one another as a single spell.

92
 Table 2.3. Transitions, by lagged per capita income (LEVLAG)

Low–high PJK TJK TOT PAD TAD TA PDA TDA TD

0–1,000 0.0147 15 1,019 0.0063 6 945 0.1216 9 74

1,001–2,000 0.0321 32 997 0.0242 18 745 0.0556 14 252
2,001–3,000 0.0325 16 493 0.0261 8 306 0.0428 8 187
3,001–4,000 0.0201 7 349 0.0146 3 205 0.0278 4 144
4,001–5,000 0.0339 8 236 0.0469 6 128 0.0185 2 108
5,001–6,000 0.0308 6 195 0.0595 5 84 0.0090 1 111
6,001–7,000 0.0190 3 158 0.0606 2 33 0.0080 1 125
7,001– 0.0015 1 679 0.0286 1 35 0.0000 0 644
Total 0.0213 88 4,126 0.0198 49 2,481 0.0237 39 1,645

Above PJK TJK TOT PAD TAD TA PDA TDA TD

0 0.0213 88 4,126 0.0198 49 2,481 0.0237 39 1,645

1,000 0.0235 73 3,105 0.0280 43 1,535 0.0191 30 1,570
2,000 0.0194 41 2,110 0.0316 25 791 0.0121 16 1,319
3,000 0.0155 25 1,616 0.0351 17 484 0.0071 8 1,132
4,000 0.0142 18 1,268 0.0500 14 280 0.0040 4 988
5,000 0.0097 10 1,032 0.0526 8 152 0.0023 2 880
6,000 0.0048 4 837 0.0441 3 68 0.0013 1 769
7,000 0.0015 1 679 0.0286 1 35 0.0000 0 644

Notes: PJK stands for the probability of any regime transition; TJK is their number. TOT is the total
number of annual observations. PAD stands for the probability of transition from authoritarianism to
democracy; TAD is their number. TA is the total number of annual observations of authoritarianism.
PDA stands for the probability of transition from democracy to authoritarianism; TDA is their number.
TD is the total number of annual observations of democracy.
e
 Economic Development and Political Regimes

diate levels of income, but not in those that are poor nor in those that
are wealthy.
Indeed, dictatorships survived for years in countries that were
wealthy by comparative standards. Whatever the threshold at which
development is supposed to dig the grave for an authoritarian regime,
it is clear that many dictatorships must have passed it in good health.
Note that we have already excluded six wealthy countries that derive
large proportions of their revenues from oil. Yet dictatorships flour-
ished also in Singapore, East Germany, Taiwan, the Soviet Union,
Spain, and Mexico for many years after those countries rose to incomes
well above $5,000, an income that Austria, Belgium, France, West
Germany, Iceland, Ireland, Italy, the Netherlands, and Norway did not
have by 1951. Table 2.4 lists the dictatorships that survived even
though the probability that they should be democracies, as predicted
by the level of development alone, was above 0.50, which corresponds
to per capita income of $4,115.
Conversely, many dictatorships fell in countries with low income
levels. Six fell in countries with incomes below $1,000, and eighteen
in countries between $1,000 and $2,000, and altogether thirty-six col-
lapsed when the probability that the regime should be democratic, as
predicted by per capita income alone, was less than 0.50. Hence, with
twenty-five dictatorships surviving in wealthy countries and thirty-six
falling in poor ones, the causal power of development in generating
democracies cannot be very strong. The distribution of levels at which
transitions to democracy occur is highly scattered.
Yet this may not be a fair test of modernization theory. After all, this
theory supposes that countries develop over a longer period, so that
all the modernizing consequences have time to accumulate. Let us
therefore examine more closely those countries that did develop under
authoritarian regimes and that at some time became “modern,” which
we will take somewhat arbitrarily to mean that they had a per capita
income of $4,115 (Table 2.5).
Twenty dictatorships (to remind, out of 123) did develop over longer
periods of time and reached “modernity.” Gabon, Mexico, Syria, and
Yugoslavia developed continuously for at least a decade, reached the
level at which democracy would be expected to be the more likely
regime, and, having remained under dictatorships, experienced a series
of economic crises. Singapore and Malaysia developed over a long
period, became wealthy, and remained dictatorships. In East Germany,
Taiwan, the Soviet Union, Spain, Bulgaria, and Hungary competitive
elections eventually took place, but at very different levels of income.

94
Level of Development and Regime Dynamics

Table 2.4. Highest levels of per capita income

(LEVEL) under which dictatorships survived in
different countries

Highest Pr(REG
Country Year level = Dem)a

Singapore 1990 11,698 0.992

East Germany 1988 10,433 0.977
Iraq 1979 8,598 0.923
Taiwan 1990 8,067 0.895
Soviet Union 1989 7,744 0.875
Spain 1976 7,390 0.851
Gabon 1976 6,969 0.818
Venezuela 1957 6,939 0.815
Bulgaria 1988 6,866 0.809
Argentina 1980 6,505 0.776
Mexico 1981 6,463 0.772
Iran 1976 6,434 0.769
Argentina 1972 5,815 0.705
Yugoslavia 1979 5,674 0.690
Hungary 1987 5,650 0.687
Greece 1973 5,218 0.637
Uruguay 1981 5,162 0.630
Malaysia 1990 5,117 0.625
Poland 1978 5,102 0.623
South Korea 1987 5,080 0.620
Syria 1981 4,668 0.569
Portugal 1974 4,657 0.568
Argentina 1962 4,541 0.553
Argentina 1957 4,355 0.530
Suriname 1981 4,220 0.513

a
Pr(REG = Dem) is the probability that a regime will be
democratic given the level of income. It is calculated as
1 - F(a + b * LEVEL), where F(·) is the c.d.f. of the normal
distribution.

Given its 1974 income level, Uruguay should have never been a dicta-
torship. The economic history of the Chilean dictatorship is convoluted:
Its income in 1974 was $3,561; it climbed with downs and ups to $4,130
by 1981, collapsed to $3,199 by 1983, recovered to surpass the 1974
level only by 1986, and passed the threshold of $4,115 in 1989, exactly

95
 Economic Development and Political Regimes

Table 2.5. Countries that developed over long periods under dictatorships
and reached incomes above $4,115

Entry Peak
Passes Transition
Country Year LEVEL Pr = 0.50 Year LEVEL Pr year at Pr

Gabon 1961 1,969 1973 1976 6,969 0.82 Never

Mexico 1951 2,317 1971 1981 6,463 0.77 Never
Brazil 1965 1,864 1980 1978 3,881 0.47 1978 0.47
Chile 1974 3,561 1981 1981 4,130 0.50 No
1989 1989 4,355 0.53 1989 0.53
Uruguay 1974 4,148 1974 1981 5,162 0.63 1985 0.48
South Korea 1961 911 1985 1988 5,606 0.68 1988 0.68
Malaysia 1957 1,282 1982 1990 5,117 0.63 Never
Singapore 1965 1,845 1972 1990 11,698 0.99 Never
Syria 1961 1,607 1978 1981 4,668 0.57 Never
Taiwan 1952 968 1979 1990 8,067 0.90 Post-1990
Bulgaria 1981 4,216 ???? 1989 6,739 0.80 1990 0.80
Czechoslovakia 1961 1,709 1989 1990 4,094 0.49 1990 0.49
East Germany 1971 4,995 ???? 1988 10,433 0.98 1990 ????
Greece 1967 3,308 1970 1974 4,966 0.61 1974 0.61
Hungary 1971 3,657 1974 1987 5,650 0.69 1989 0.68
Poland 1971 3,109 1974 1978 5,102 0.62 No
1985 1988 4,529 0.55 1989 0.55
Portugal 1951 1,314 1973 1974 4,657 0.57 1975 0.52
Spain 1951 2,205 1964 1976 7,390 0.85 1976 0.85
Soviet Union 1961 2,536 1971 1989 7,744 0.88 Collapsed
Yugoslavia 1961 2,073 1974 1979 5,674 0.69 Collapsed

Note: This table lists countries that grew over a period of at least seven years and at some time reached
a per capita income of $4,115. “Entry” is the first year of the dictatorship or 1951 or the year after the
country became independent or the year after economic data became available. “Passes Pr = 0.50” is
the year when the country reached a per capita income of $4,115. “Peak” gives the time when the
country reached the highest income level under the particular dictatorship and the probability, as pre-
dicted by per capita income, that it would be a democracy. Finally, “Transition” gives the year the dic-
tatorship fell, if ever, and the probability of democracy at that time.

the year of transition. The history of Poland is similar: By our criteria,

it reached the threshold of democracy in 1974; it experienced an eco-
nomic crisis in 1979 and a mass movement for democracy in 1980,
passed the threshold again in 1985, and became a democracy in 1989.
In turn, Brazil, Czechoslovakia, Portugal, and perhaps even South Korea
and Greece are the dream cases for a modernization theorist. Those
countries developed under dictatorships, became wealthy, and threw off

96
Level of Development and Regime Dynamics

their dictatorships more or less when their levels of development would

have predicted. But they are few.
This is not to say that democracy did not emerge in some countries
when they became modern. Indeed, perhaps in those countries that did
develop over a long period, the very thought of democracy appeared on
the political agenda because they were too modern – not only in those
countries that became democratic just when our model predicts but also
those that waited much longer: Taiwan, the Soviet Union, Spain, and
Bulgaria. Modernization may create the “prerequisites” for political
conflict over the form of regime. But the manner in which these conflicts
will develop remains unpredictable. When conflicts over regimes are
examined at a micro level, by looking at the political actors involved,
their motives and their beliefs, it becomes apparent that these are
situations laden with uncertainty (O’Donnell and Schmitter 1986;
Przeworski 1991). Game-theoretic analyses of transitions to democracy
make it apparent that the actors involved often do not know each other’s
preferences, the relationships of physical forces, or the outcomes of
eventual conflicts (Wantchekon 1996; Zielinski 1997). And under such
conditions, various equilibria can prevail: Whereas transition to democ-
racy is one feasible outcome, so is the perpetuation of the dictatorial
status quo, or even a solidification of dictatorship. Hence, even if mod-
ernization may generate conflicts over democracy, the outcomes of such
conflicts are open-ended.
But if modernization theory is to have any predictive power, there
must be some level of income at which one can be relatively sure that
the country will throw off its dictatorship. And one is hard put to find
this level: Even among the countries that satisfy the premise of the
modernization theory, those listed in Table 2.5, the range of incomes
at which dictatorships survived is very wide. Few authoritarian
regimes have developed over a long period, and even if most of them
should eventually become democracies, no level of income can predict
when that should occur.
Moreover, even if to predict is not the same as to explain, “explain-
ing” can easily entail an ex-post fallacy. Take Taiwan, which in 1952
had a per capita income of $968. It developed rapidly, passing by 1979
our threshold of $4,115; it had a probability of 0.10 of being a dicta-
torship in 1990, and in 1995, for the first time, elected its president in
contested elections. Suppose that during all that time the Taiwanese
dictatorship had faced each year a probability of 0.02 of dying, for
reasons not related to development. It thus would have had a cumula-
tive chance of about 50 percent of not being around by 1995 even if it

97
 Economic Development and Political Regimes

had not developed at all. Thus we might erroneously attribute to devel-

opment what may have been just a cumulation of random hazards.9
And, indeed, the Taiwanese dictatorship most likely democratized to
mobilize international support against the threat from China: for
geopolitical reasons, not for economic reasons.
In sum, the causal power of economic development in bringing down
dictatorships appears paltry. The level of development, at least as mea-
sured by per capita income, gives little information about the chances
of transition to democracy.
On the other hand, per capita income has a strong impact on the sur-
vival of democracies. As column 8 of Table 2.3 shows, in countries with
per capita incomes under $1,000, the probability that a democracy
would die during a particular year was 0.1216, which implies that their
expected life was about eight years. Between $1,001 and $2,000, this
probability was 0.0556, for an expected duration of about eighteen
years, and the probability that a democracy would die in a country with
an income above $4,000 was almost zero. Indeed, no democracy has
ever been subverted, not during the period we studied nor ever before
nor after, regardless of everything else, in a country with a per capita
income higher than that of Argentina in 1975: $6,055. There is no doubt
that democracy is stable in affluent countries: The probability of it col-
lapsing is almost zero; the coefficient on LEVEL in statistical analyses of
survival (see Appendix 2.2) is positive and significant, and the stark fact
is that up until 1990 thirty-one democracies had lived 742 years with
incomes above that of Argentina in 1975, and not one had ever fallen.
A question that has been extensively debated is whether or not the
stability of democracy is monotonic with regard to the level of devel-
opment. Although there are important theoretical differences and even
sharper political differences between Huntington (1968; Huntington
and Nelson 1976) and O’Donnell (1973), both have claimed that there
is a level beyond which further development decreases the probability
that democracy will survive. Huntington has argued that both regime
types become unstable when a country undergoes modernization,
which occurs at some intermediate level of development. O’Donnell,
in turn, has claimed that democracies tend to die when a country
exhausts “the easy stage of import substitution,” again at some inter-
mediate level of economic development.

9
An analogy may be useful. Suppose that a woman runs a risk of 0.01 of dying from acci-
dental causes during each year of her life, and then at the age of 78 she gets hit by a falling
brick. To attribute her death to development would be to conclude that she died of old age.

98
Level of Development and Regime Dynamics

Huntington (1968: 1) was concerned with stability and did not care
whether regimes were democratic or authoritarian. “The most impor-
tant political distinction among countries,” he thought, “concerns not
their form of government but their degree of government.” Hence, the
United States, the United Kingdom, and the Soviet Union were all
systems in which “the government governs.” Whether it was the Polit-
buro, the cabinet, or the president mattered little. “The problem,” he
insisted, “was not to hold elections but to create organizations.” Indeed,
we were told, “the primary problem is not liberty but the creation of a
legitimate public order” (1968: 7). Though never explicitly referring to
Lipset, Huntington (1968: 35–6) observed that “in actuality, only some
of the tendencies encompassed in the concept of ‘political moderniza-
tion’ characterized the ‘modernizing’ areas. Instead of a trend toward
competitiveness and democracy, there was an ‘erosion of democracy’
and a tendency to autocratic military regimes and one-party regimes.
Instead of stability, there were repeated coups and revolts.” We should
expect “a bell-shaped pattern of political instability” (p. 43) among
democratic as well as authoritarian regimes.
O’Donnell dragged Lipset over the coals for various methodological
transgressions. Reflecting on his criticisms in retrospect, he observed
that “Chapter I is now an archeological remnant – testimony of a debate
that in 1971 had recently begun and today is finished: it is no longer nec-
essary to lead the reader through tedious series of data to demonstrate
that ‘socio-economic development’ does not foster ‘democracy and/or
political stability’ ” (1979: 204). What the data show, O’Donnell asserted,
is that “in contemporary South America, the higher and the lower levels
of modernization are associated with non-democratic political systems,
while political democracies are found at intermediate levels of modern-
ization.” Hence, at least within the range observed by O’Donnell, we
should observe that democracies fall as economies develop.
Is there some level of development beyond which democracies are
more likely to die than they were earlier? We have already seen in Table
2.3 that the probability of a democracy dying declines monotonically
with per capita income. Although O’Donnell did cite a countercase
against Lipset, his account of the rise of bureaucratic authoritarianism
does not undermine Lipset’s theory.10 O’Donnell studied a country that
turns out to be a distant outlier: As Table 2.6 shows, three of the four
10
O’Donnell was careful not to make general claims: His purpose was to explain the down-
fall of democracies in the Southern Cone. But his theory of “bureaucratic authoritari-
anism” captured the imaginations of scholars all around the world, who treated it as
applicable almost everywhere.

99
Table 2.6. Transitions to dictatorship, 1951–
1990, by last full year of democracy, per capita
income, and type of democracy

Country Year LEVEL Type

Argentina 1975 6,055 Presidential

Argentina 1965 5,011 Presidential
Argentina 1965 4,790 Presidential
Uruguay 1972 4,034 Presidential
Argentina 1954 3,989 Presidential
Suriname 1979 3,923 Parliamentary
Chile 1972 3,857 Presidential
Greece 1966 3,176 Parliamentary
Turkey 1979 2,957 Parliamentary
Peru 1967 2,694 Presidential
Guatemala 1981 2,534 Presidential
Suriname 1989 2,491 Mixed
Peru 1989 2,247 Presidential
Panama 1967 2,227 Presidential
Peru 1961 2,148 Presidential
Bolivia 1979 2,037 Presidential
Brazil 1963 1,889 Presidential
Guatemala 1962 1,693 Presidential
Thailand 1975 1,686 Parliamentary
Guatemala 1953 1,509 Presidential
Ecuador 1962 1,451 Presidential
Nigeria 1982 1,419 Presidential
Sri Lanka 1976 1,336 Parliamentary
Honduras 1971 1,236 Presidential
Philippines 1964 1,217 Presidential
Congo 1962 1,120 Presidential
Sierra Leone 1966 1,097 Parliamentary
Ghana 1971 1,042 Parliamentary
Honduras 1962 1,042 Presidential
Somalia 1968 1,015 Mixed
Ghana 1980 978 Presidential
Pakistan 1976 943 Presidential
South Korea 1960 898 Parliamentary
Sudan 1988 765 Parliamentary
Nigeria 1965 621 Parliamentary
Pakistan 1955 577 Parliamentary
Uganda 1984 576 Presidential
Myanmar 1961 312 Parliamentary
Myanmar 1957 267 Parliamentary

100
Level of Development and Regime Dynamics

transitions to authoritarianism at per capita incomes above $4,000

occurred in Argentina, and the fourth in Uruguay. Thus, Lipset was right
in thinking that the richer the country, the more likely it is to sustain
democracy.
Clearly, this fact cries for an explanation. One possible account for
the durability of democracies in wealthy countries, proposed already
by Lipset, is that, through various sociological mechanisms, wealth
lowers the intensity of distributional conflicts. An alternative explana-
tion is that income is just a proxy for education, and more highly edu-
cated people are more likely to embrace democratic values. Education,
specifically accumulated years of education for an average member of
the labor force, does increase the probability of survival of democracy
at each level of income.11 The probability that a democracy will die in
a country where the average member of the labor force has fewer than
three years of formal education is 0.1154; it is 0.0620 when the level
of education is between three and six years, 0.0080 when it is six to
nine years, and zero when the average worker has more than nine
years of education. The highest level of education under which a
country experienced a transition to dictatorship was 8.36 years in Sri
Lanka in 1977, but that was an outlier. The next highest level of edu-
cation when democracy fell was 6.85 years in Uruguay.
But income is not a proxy for education. Even though these two vari-
ables are highly correlated (0.78), their effects are to a large measure
independent. As Table 2.7 shows, whereas at each income level the
probability of democracy falling decreases with increasing education,
the converse is also true: At each level of education, the probability of
democracy dying decreases with income. Hence, for reasons that are
not easy to identify, wealth does make democracies more stable, inde-
pendently of education.
Finally, we find no evidence of “consolidation.” Democracies become
“consolidated” if the conditional probability that a democratic regime
will die during a particular year given that it has survived thus far (the
“hazard rate”) declines with its age, so that, as Dahl (1990) has argued,
democracies are more likely to survive if they have lasted for some
time. Examining the ages at which democracies die indicates that this

11
We have data only for 2,900 country-years of education. The mean is 4.85 years, and
the standard deviation is 3.12, with a minimum of 0.03 (Guinea in 1966) and a maxi-
mum of 12.81 (United States in 1985); 27.6% of the sample had educational levels lower
than three years, 64.4% lower than six years, and 90.8% lower than nine years. Only
13.0% of the sample had education levels higher than Sri Lanka in 1977, and 26.1%
higher than Uruguay in 1973.

101
 Table 2.7. Regime transitions, by lagged per capita income and average education

Education (in years)

Income 0–3 3–6 6–9 9– Total

0–4,000
All 0.0208 20 961 0.0393 27 687 0.0198 5 252 0.0000 0 35 0.0269 52 1,936
Dic 0.0098 15 511 0.0294 15 511 0.0194 3 155 0.0000 0 28 0.0167 26 1,557
Dem 0.1212 12 99 0.0681 12 176 0.0206 2 97 0.0000 0 7 0.0686 26 379
4,001–8,000
All 0.0000 0 36 0.0458 6 131 0.0172 4 233 0.0000 0 68 0.0214 10 468
Dic 0.0000 0 31 0.0434 3 69 0.0390 3 77 0.0000 0 7 0.0326 6 184
Dem 0.0000 0 5 0.0484 3 62 0.0064 1 156 0.0000 0 61 0.0141 4 284
8,001–
All 0.0000 0 1 0.0000 0 3 0.0000 0 129 0.0000 0 251 0.0000 0 384
Dic 0.0000 0 1 0.0000 0 0 0.0000 0 4 0.0000 0 0 0.0000 0 5
Dem 0.0000 0 0 0.0000 0 3 0.0000 0 125 0.0000 0 251 0.0000 0 379
Total
All 0.0200 20 998 0.0401 33 822 0.0147 9 614 0.0000 0 353 0.0222 62 2,788
Dic 0.0089 8 894 0.0310 18 580 0.0253 6 237 0.0000 0 34 0.0183 32 1,746
Dem 0.1154 12 104 0.0620 15 242 0.0080 3 377 0.0000 0 319 0.0288 30 1,042

Note: The first number under each level of education is the probability of transition away from a given
regime; the second number is the number of such transitions, and the third is the total number of
annual observations of this regime at that level of education.
e
Level of Development and Regime Dynamics

is true, but once the level of development is taken into account, the
hazard rates become independent of age, meaning that for a given level
of development, democracies are about equally likely to die at any age
(see Appendix 2.2 for details). These findings indicate that the hazard
rates uncorrected for the level of development decline because coun-
tries develop, not because a democracy that has long been around is
more likely to continue.
The conclusion reached thus far is that whereas economic develop-
ment under dictatorship has at most a non-linear relationship to the
emergence of democracies, once they are established, democracies are
much more likely to endure in more highly developed countries. Yet
because our systematic observations begin in 1950, the question arises
whether or not these patterns also characterize the earlier period.
Studies in the Lipset tradition have assumed that they do: They have
inferred the historical process from cross-sectional observations. Yet
the validity of such inferences is contested by followers of Moore
(1966), who claimed that the Western European route to democracy
was unique, not to be repeated. Note that when Rustow (1970) pointed
out that the levels of development at which different countries perma-
nently established democratic institutions varied widely, Lipset’s rejoin-
der (1981) was that the thresholds at which democracy was established
were lower for the early democracies. Is that true?
Although economic data for the pre-war period are not comparable
to those at our disposal after 1950, Maddison (1995) reconstructed per
capita income series for several countries going back to the nineteenth
century. Table 2.8 portrays the pre-1950 experiences with democracy.
The levels at which democracies were established before 1950 vary
as widely as they do for the later period; indeed, they cover almost
the entire range of incomes observed. The poorest countries in which
democracy was experimented with before 1950 were Pakistan, which
became independent in 1947 when it had per capita income of 631 (1990
G-K dollars), and India, which in 1947 had income of 641. Yugoslavia
had income of 1,064 in 1921; Bulgaria had income of 1,169 in 1926; Por-
tugal had income around 1,354 in 1910 (1913 figure); Brazil had income
of 1,460 in 1946. The United States in 1830 (interpolated from 1820 to
1850) and Norway in 1884 must have had about the same income as
Brazil in 1946. In turn, New Zealand had an income of 5,367 when it
became independent in 1907, Venezuela had 5,102 when it first exper-
imented with democracy in 1945, and the United Kingdom had 5,052 in
1911. Hence, the levels at which transitions to democracy occurred
before 1950 were highly dispersed. Again, there was no clear threshold.

103
Table 2.8. Countries (in our sample) that experienced democracy before 1950

Last pre-1950 Situation

First democracy Pre-1950 reversal democacy in 1950

Country Date LEVEL Date LEVEL Date LEVEL LEVEL Regime

Canada (1920) 1920 3,659 None 1920 7,047 Dem

Costa Rica 1919 1948 1948 Dem
Guatemala 1945 None 1945 Dem
United States 1830 1,464a None 1830 9,573 Dem
Argentina 1912 3,904 1930 4,080 1946 4,665 4,987 Dem
Brazil 1946 1,460 1946 1,460 1,673 Dem
Chile 1891 1,949(1900) 1925 2,876 1932 2,274 3,827 Dem
Colombia 1910 1,236(1913) 1949 2,107 2,089 Dic
Ecuador 1947 None 1947 Dem
Peru 1939 1,884 1948 2,094 2,263 Dic
Uruguay 1942 1942 Dem
Venezuela 1945 5,102 1948 7,394 7,424 Dic
Israel (1948) 1948 None 1948 Dem
India (1947) 1947 641 None 1947 597 Dem
Myanmar (1948) 1948 None 1948 393 Dem
Pakistan (1947) 1947 631 None 1947 650 Dem
Philippines (1946) 1946 None 1946 1,293 Dem
Sri Lanka (1948) 1948 None 1948 Dem
Austria 1918 2,572 1934 2,871 1945 3,731 Dem
Belgium 1919 3,318 None 1919 5,346 Dem
Bulgaria (1908) 1926 1,169 1934 1,309 1,651 Dic
Czechoslovakia (1918) 1920 1,933 1948 3,088 3,501 Dic
Denmark 1901 2,986 None 1901 6,683 Dem
Finland (1917) 1919 1,610 1930 2,589 1944 4,131 Dem
Franceb 1875 2,198 None(?) 1875 5,221 Dem
Germany 1919 2,763 1933 3,591 1949 4,281 Dem
Greece 1926 2,368(1929) 1936 2,501 1946 1,412 1,951 Dem
Iceland (1918) 1920 None 1920 Dem
Ireland (1921) 1923 2,625(1926) None 1923 3,518 Dem
Italy 1919 2,783 1922 2,574 1946 3,425 Dem
Luxembourg 1868 None 1,868 Dem
Netherlands 1868 2,640(1870) None 1868 5,850 Dem
Norway (1905) 1884 1,466 None 1884 4,969 Dem
Poland (1918) 1919 1926 2,117(1929) 2,447 Dic
Portugal 1910 1,354(1913) 1926 1,536(1929) 2,132 Dic
Spain 1931 2,713 1936 2,304 2,397 Dic
Sweden 1918 2,533 None 1918 6,738 Dem
Switzerland 1870 2,172 None 1870 8,939 Dem
United Kingdom 1911 5,052 None 1911 6,847 Dem
Yugoslavia (1918) 1921 1,064 1929 1,376 1,546 Dic
Australia (1901) 1901 4,112 None 1901 7,218 Dem
New Zealnad (1907) 1907 5,367 None 1907 8,495 Dem

Notes: Income figures are not the same as in the rest of the text. They are expressed in 1990 Geary-
Khamis dollars, as reported by Maddison (1995). Blanks indicate that income data are not available.
Democratization is dated by (1) the occurrence of contested elections organized on a partisan basis and
(2) legislative sovereignty of the lower house elected by broadest suffrage (rather than responsibility to
the crown or a non-elected upper chamber), whichever came later, but not by the extent of franchise
or participation, and (3) when relevant, the first victory of the opposition candidate in presidential elec-
tions. For countries that became independent after 1871, dates in parentheses are for the year of inde-
pendence.
a
Interpolated, based on 1820 and 1850.
b
The question mark refers to the Vichy regime.
 Economic Development and Political Regimes

Reversals occurred in four out of ten countries. And, again, they

were more likely to occur in countries where democracy was estab-
lished when they were poor. Among the countries for which income
data are available, eight democracies subsequently fell and four sur-
vived until 1950 (but in Pakistan democracy did fall soon after) where
democracy was established with incomes under $2,000. In turn, six
fell and twelve survived until 1950 (indeed, until today) in countries
that had incomes above $2,000 when democracy was first established.
The collapse of democracy in Chile, Colombia, Peru, Austria, Bulgaria,
Germany, Greece, Finland, Italy, Poland, Portugal, Spain, and
Yugoslavia occurred when these countries had incomes below 3,000
1990 G-K dollars, which means well below 3,000 1985 dollars, which
we have been using. The highest level at which democracy collapsed
was in Venezuela in 1948, but note that the 7,394 1990 G-K dollars
corresponds to roughly 4,880 1985 PPP dollars, which we have been
using throughout. Hence, the Argentine 1975 income of 6,055 still
stands as the highest at which democracy was ever subverted.
To conclude, there is no doubt that democracies are more likely to
be found in the more highly developed countries. Yet the reason is not
that democracies are more likely to emerge when countries develop
under authoritarianism, but that, however they do emerge, they are
more likely to survive in countries that are already developed.

Economic Growth and Regime Dynamics

The conditions that countries inherit are not sufficient to explain
why regimes survive or die. Dictatorships lasted in many countries that
not only were wealthy but also enjoyed other conditions that should
have predisposed them toward democracy. And, conversely, some
democracies were established in countries that were poor and yet
endured the passage of time.
Table 2.9 presents a list of democracies that lasted at least twenty
years, organized by ascending levels of per capita income at the time
when they were established or, when we could not determine their
initial income, when data were first available. The income figures for
the pre-1950 period are based on extrapolations and are at best
approximative.12 But the range of incomes at which lasting democra-
cies were established is so large that all the inaccuracies do not change

12
To compare the pre- and post-1950 incomes, in Table 2.9 we are extrapolating incomes
expressed in 1985 PPP USD, rather than the 1990 G-K dollars used in Table 2.8.

106
 Table 2.9. Democracies that lasted at least 20 years, by per capita income and by
income distribution

First observed Income distribution

Established Lasted
Country (year) Year LEVEL until Year INEQ GINI Year INEQ GINI

India 1947 1947 556 Now 1951 6.14 35.56 1990 4.30 29.69
Philippines 1946 1946 697 1965 1957 7.42 46.14 1965 16.00 51.32
Brazil 1946 1946 917 1964 1960 18.72 53.00 1970 19.28 57.61
Netherlands 1868 1868 1,050 Now 1975 4.43 28.60 1989 5.11 29.60
Austria 1945 1945 1,093 Now
Sri Lanka 1948 1951 1,107 1977 1953 10.35 47.80 1979 8.70 43.50
United States 1830 1830 1,119 Now 1947 8.20 34.28 1990 9.60 37.80
Norway 1884 1884 1,228 Now 1962 8.08 37.52 1991 7.69 33.31
Malta 1964 1964 1,377 Now
Dominican R. 1966 1966 1,413 Now 1976 45.00 1989 13.26 50.46
Costa Rica 1948 1951 1,449 Now 1961 8.87 50.00 1989 12.67 46.07
Colombia 1958 1958 1,613 Now 1970 8.63 52.02 1988 15.11 51.20
Chile 1932 1932 1650 1973 1968 11.42 45.64 1971 12.16 46.00
Italy 1946 1946 1,708 Now 1974 41.00 1989 4.56 32.74
France 1875 1875 1,748 Now 1956 49.00 1984 6.38 34.91
Japan 1952 1952 1,768 Now 1962 7.71 37.20 1990 35.00
Jamaica 1962 1962 1,802 Now 1958 47.71 1990 8.09 41.79
Papua NG 1975 1975 1,870 Now
Sweden 1918 1918 1,919 Now 1967 8.92 33.41 1990 5.16 32.52
Denmark 1901 1901 2,213 Now 1976 5.44 31.00 1992 6.90 33.20

(continued)
e
 Table 2.9 (continued)

First observed Income distribution

Established Lasted
Country (year) Year LEVEL until Year INEQ GINI Year INEQ GINI

Switzerland 1870 1870 2,226 Now

West Germany 1949 1949 2,567 Now 1963 3.58 28.13 1984 5.90 32.19
Israel 1947 1954 2,585 Now
Finland 1944 1944 2,636 Now 1966 5.08 31.80 1991 4.34 26.11
Ireland 1922 1951 2,816 Now 1973 8.79 38.69 1987 9.05 34.60
Belgium 1919 1919 2,960 Now 1979 4.57 28.25 1992 4.31 26.92
United Kingdom 1911 1911 3,016 Now 1961 3.95 25.30 1990 5.27 32.30
Mauritius 1968 1968 3,074 Now 1980 39.10 1991 6.48 36.69
Barbados 1966 1966 3,353 Now 1951 12.45 45.49 1979 22.67 48.86
Uruguay 1937 1937 3,492 1973
Iceland 1922 1951 3,675 Now
Canada 1920 1920 3,838 Now 1951 5.72 32.56 1990 4.49 27.56
Portugal 1976 1976 4,471 Now 1973 4.61 30.63 1990 7.44 36.76
Greece 1974 1974 4,966 Now 1974 6.52 35.11 1988 6.65 35.19
Luxembourg 1868 1951 5,964 Now 1985 4.11 27.13
Trinidad & Tobago 1962 1962 6,006 Now 1958 14.29 46.02 1981 13.08 41.72
New Zealand 1907 1951 6,264 Now 1973 5.43 30.05 1990 9.77 40.21
Venezuela 1959 1959 6,718 Now 1971 14.44 47.65 1990 16.18 53.84
Spain 1977 1977 7,446 Now 1973 30.51 1989 4.20 25.91
Bahamas 1972 1978 8,740 Now

Note: All the pre-1951-level figures are based on extrapolations and are expressed in 1985 PPP USD.
INEQ is the ratio of incomes of the top quintile to the bottom quintile, GINI is the Gini index.
e
Economic Growth and Regime Dynamics

the picture. The fact is that some democracies survived for long periods
even in very poor countries, including notably the United States, which
in 1830 must have had the same per capita income as many contem-
porary African nations, about the same as today’s Nigeria.
Hence, though we already know that in affluent countries democ-
racy is impregnable, wealth is not necessary for democracies to
survive. Some democracies, like some dictatorships, appear to survive
even when they face adverse conditions. The hypothesis to investigate
is that the survival of regimes is due to their economic performance,
that is, that they are subject to endogenous attrition.
Let us again examine some descriptive patterns. In Table 2.10, the
hazard rates (the probability that a regime will die in a particular year)
are calculated separately for different bands of the rates of growth of
per capita income, lagged one year, at each income level. When we
look at the entire sample, it is apparent that growth matters for regime
survival: When per capita income has declined during the preceding
year, the probability that either type of regime will die is 0.0324, but
when income has grown, that probability is 0.0164, one-half. And if
this difference appears small, think in terms of frequencies: One in
thirty-one regimes will die when the economies are shrinking, and one
in sixty-one when they are expanding.
Democracies appear to be more sensitive to growth performance.
When they face a decline in income, they die at the rate of 0.0512, so
that about one in twenty of them dies, but when incomes are growing,
they die at the rate of 0.0152, one in sixty-six. Moreover, democracies
that grow slowly, at rates of less than 5 percent per annum, die at the
rate of 0.0173, whereas those that grow at rates faster than 5 percent
die at the rate of 0.0132. Thus, Olson (1963) and Huntington (1968)
could not have been more wrong when they thought that rapid growth
destabilizes democracies.
Dictatorships are less sensitive to economic crises: Their respective
probabilities of dying are 0.0240, one in forty-two, when the economy
decays, and 0.0174, one in fifty-seven, when it grows.
The difference between the two regimes becomes even more pro-
nounced when we examine the longer-term dynamics of growth. With
some exceptions, the longer the economic crisis, the more likely it is that
democracy will fall: The chance that a democracy will die is 1 in 135
when incomes grow during any three or more consecutive years, and 1
in 13 when incomes fall during any two consecutive years. Conversely,
most deaths of democracy are accompanied by some economic crisis:
In twenty-eight out of thirty-nine instances, deaths of democracies were

109
Table 2.10. Observed rates of transition, by lagged per capita income
(LEVLAG) and lagged rate of economic growth (GLAG)

All Dictatorships Democracies

LEVLAG
GLAG PJK TJK TOT PAD TAD TA PDA TDA TD

0–1,000 0.0147 15 1,019 0.0063 6 945 0.1216 9 74

G£0 0.0193 9 467 0.0091 4 440 0.1852 5 27
G>0 0.0109 6 552 0.0040 2 505 0.0851 4 47
1,001–2,000 0.0321 32 997 0.0242 18 745 0.0556 14 252
G£0 0.0447 14 313 0.0313 7 224 0.0787 7 89
G>0 0.0263 18 684 0.0211 11 521 0.0429 7 163
2,001–3,000 0.0325 16 493 0.0261 8 306 0.0428 8 187
G£0 0.0522 7 134 0.0341 3 88 0.0870 4 46
G>0 0.0251 9 359 0.0229 5 218 0.0284 4 141
3,001–4,000 0.0201 7 349 0.0146 3 205 0.0278 4 144
G£0 0.0303 3 99 0.0172 1 58 0.0488 2 41
G>0 0.0160 4 250 0.0136 2 147 0.0194 2 103
4,001–5,000 0.0339 8 236 0.0469 6 128 0.0185 2 108
G£0 0.0500 3 60 0.0588 2 34 0.0385 1 26
G>0 0.0284 5 176 0.0426 4 94 0.0122 1 82
5,001–6,000 0.0308 6 195 0.0595 5 84 0.0090 1 111
G£0 0.0541 2 37 0.0952 2 21 0.0000 0 16
G>0 0.0253 4 158 0.0476 3 63 0.0105 1 95
6,001–7,000 0.0190 3 158 0.0606 2 33 0.0080 1 125
G£0 0.0857 3 35 0.3333 2 6 0.0345 1 29
G>0 0.0000 0 123 0.0000 0 27 0.0000 0 96
7,001– 0.0015 1 679 0.0286 1 35 0.0000 0 644
G£0 0.0000 0 120 0.0000 0 3 0.0000 0 117
G>0 0.0018 1 559 0.0313 1 32 0.0000 0 527
Total 0.0213 88 4,126 0.0198 49 2,481 0.0237 39 1,645
G£0 0.0324 41 1,265 0.0240 21 874 0.0512 20 391
G>0 0.0164 47 2,861 0.0174 28 1,607 0.0152 19 1,254
Moving average (2)
Total 0.0218 87 3,991 0.0200 48 2,396 0.0245 39 1,595
G£0 0.0293 31 1,059 0.0241 18 748 0.0418 13 311
G>0 0.0191 56 2,932 0.0182 30 1,648 0.0202 26 1,284
Moving average (3)
Total 0.0223 86 3,856 0.0208 48 2,312 0.0246 38 1,544
G£0 0.0286 26 910 0.0226 15 665 0.0449 11 245
G>0 0.0204 60 2,946 0.0200 33 1,647 0.0208 27 1,299

Notes: G £ 0 means that per capita income declined or remained the same; G > 0 means that
per capita income increased. PJK stands for the probability of any regime transition; TJK is their
number. TOT is the total number of annual observations. PAD stands for the probability of tran-
sition from authoritarianism to democracy; TAD is their number. TA is the total number of annual
observations of authoritarianism. PDA stands for the probability of transition from democracy
to authoritarianism; TDA is their number. TD is the total number of annual observations of
democracy.

110
Economic Growth and Regime Dynamics

Table 2.11. Regime transitions, by longer-term dynamics of per capita

income

Incomes
increased
or decreased
during these Dictatorships Democracies
consecutive
years PAD TAD TA PDA TDA TD

Increased
3 or more years 0.0170 14 824 0.0074 6 807
2 years 0.0139 4 288 0.0265 5 189
1 year 0.0205 10 488 0.0315 8 254
Decreased
1 year 0.0220 10 455 0.0549 13 237
2 years 0.0181 4 221 0.0778 7 90
3 years 0.0455 4 88 0.0000 0 31
4 or more years 0.0230 2 87 0.0000 0 27
Total 48 2,471 39 1,635

accompanied by a fall in income during at least one of the two preced-

ing years. Dictatorships, in turn, die under all kinds of economic condi-
tions. The probabilities of dictatorships falling are almost the same
regardless whether the economy grew or declined during any number
of consecutive years. And, conversely, whereas fourteen out of forty-
eight transitions to democracy followed at least three years of consecu-
tive growth, ten dictatorships fell after their countries experienced at
least two consecutive years of economic decline, six after declines of
three years, and two after declines of four or more years. Thus, albeit
not without exceptions, deaths of democracies follow a clear pattern:
They are more likely when a country experiences an economic crisis,
and in most cases they are accompanied by one. But dictatorships die
under the entire range of economic conditions (Table 2.11).
We know already that democracies never die in wealthy countries.
Yet it is still striking how fragile poor democracies are. In countries
with incomes under $2,000, of the 116 years during which declines in
incomes occurred, twelve democracies fell the following year: about
one in ten. Even among countries with incomes between $2,001 and

111
 Economic Development and Political Regimes

$5,000, declining incomes resulted in the fall of seven democracies in

113 years during which this happened: one in sixteen. And then, above
$6,055 the miracle occurs. Longer-term patterns are the same: Among
democracies with per capita incomes of less than $3,000, one in thirty-
five died when incomes had grown during two or more consecutive
years, but one in ten when incomes had declined during one or more
consecutive years. Poor dictatorships, in turn, are impervious to eco-
nomic crises, but wealthy dictatorships are more vulnerable when
incomes decline. One in forty-two of them died when incomes
increased during at least two consecutive years, but one in sixteen
when incomes declined for at least one year.
These descriptive patterns are confirmed by statistical analyses. In
Table 2.12 we show the effects of growth rates on the probabilities
of regime transitions during the single preceding year, and of moving
averages over two and three preceding years. Democracies turn out to
be most sensitive to their growth performance during one single year;
the average over two years matters less, and over three years not at
all. Dictatorships as well are most sensitive to last year’s growth, but
longer crises still matter.
Is it then true that, to put it in the words of Diamond and Linz (1989:
17), “economic crisis represents one of the most common threats to
democratic stability”? Do economic crises threaten regime stability, or
does regime instability cause economic crises?
The question of what causes what is difficult, perhaps impossible,
to answer. One can think of rival stories. Take democracies: It may
happen that for some exogenous reason, say a jump in energy prices
or international interest rates, an economy suffers a shock. No longer
able to satisfy popular demands, democracy becomes vulnerable to
political forces that put the blame on the “anarchy” of democratic
competition; such forces promise to establish “order” and to sanitize
the economy. But it is equally plausible that the shock could be politi-
cal, say an electoral victory by a left-wing party: That party pursues
a redistributive policy, investors are thrown into a panic, and the
economy grinds to a halt. Similar stories can be told about dictator-
ships: Facing an economic shock, the regime can no longer legitimize
itself by its economic performance, and forced to liberalize, it unleashes
the forces for transition. But again, the shock might be political,
say the imminent death of a founding dictator or a power struggle
within the authoritarian bloc. Facing the imminent death of the regime,
investors run away, and workers openly press their demands, causing
economic decline. In either case, one has to be able to tell a rather

112
Table 2.12. Dynamic probit analysis of the impact of growth on regime
survival

Lagged rate of growth

Log-likelihood -372.9192
Restricted (slope = 0) log-likelihood -2,685.421
Chi-squared (7) 4,625.003
Significance level 0.0000000
Transitions to dictatorship Transitions to democracy

Variable Coefficient t ratio Pr |t| ≥ x Coefficient t ratio Pr |t| ≥ x

Constant -1.1444 -5.288 0.00000 -2.5238 -13.704 0.00000

LEVLAG -0.20098E-03 -1.399 0.16195 0.32883E-03 0.699 0.48446
LEVSQR -0.29429E-05 -0.158 0.87411 -0.29161E-04 -1.309 0.19050
GLAG -0.42345E-01 -2.988 0.00281 -0.21067E-01 -3.823 0.00013

Lagged moving average of growth over two years

Log-likelihood -368.7513
Restricted (slope = 0) log-likelihood -2,595.780
Chi-squared (7) 4,454.058
Significance level 0.0000000

Transitions to dictatorship Transitions to democracy

Variable Coefficient t ratio Pr |t| ≥ x Coefficient t ratio Pr |t| ≥ x

Constant -1.0673 -4.993 0.00000 -2.4135 -13.442 0.00000

LEVLAG -0.22871E-03 -1.633 0.10249 0.33118E-03 0.569 0.56911
LEVSQR -0.39442E-06 -0.022 0.98215 -0.28839E-04 -1.231 0.21821
MA2GL -0.31652E-01 -1.675 0.09402 -0.26569E-01 -2.642 0.00823

Lagged moving average of growth over three years

Log-likelihood -364.1207
Restricted (slope = 0) log-likelihood -2,506.529
Chi-squared (7) 4,284.818
Significance level 0.0000000
Transitions to dictatorship Transitions to democracy

Variable Coefficient t ratio Pr |t| ≥ x Coefficient t ratio Pr |t| ≥ x

Constant -1.0418 -4.800 0.00000 -2.4894 -13.117 0.00000

LEVLAG -0.24784E-03 -1.773 0.07627 0.33155E-03 0.466 0.64152
LEVSQR 0.10868E-05 0.063 0.95015 -0.28881E-04 -1.182 0.23730
MA3GL -0.14568E-01 -0.648 0.51677 -0.30060E-01 -1.697 0.08977

Notes: LEVLAG, lagged per capita income; LEVSQR, lagged squared per capita income;
GLAG, lagged rate of growth; MA2GL, lagged moving average of growth over two years;
MA3GL, lagged moving average of growth over three years. The t ratios under “Transi-
tions to democracy” refer to shift coefficients (a in Appendix I), not to the total coefficients
(a + b).

113
 Economic Development and Political Regimes

complex story in which political and economic events feed on each

other in turn.
Ideally, we would like to be able to observe something like “pres-
sures toward transition” and relate them to economic dynamics. But
we cannot observe them. All we can see are visible manifestations of
political mobilization: strikes (STRIKES), anti-government demonstra-
tions (AGDEMONS), or riots (RIOTS). To draw inferences from these
observations, we can reason as follows: If regimes fall when incomes
decline and mobilization is not any higher than when they do not fall
(“low”), then we can attribute the causal effect to economic crises. If
regimes fall when incomes grow but mobilization is high, we can
suspect that it is political mobilization that brings them down. If they
fall when incomes decline and mobilization is high, we cannot tell
which is the cause. Finally, if they fall when incomes are growing and
mobilization is low, neither is the cause.
Before we follow this reasoning, note in Table 2.13 that the inci-
dence of mobilization – the sum of the preceding three events (MOBI-
LIZATION) – is independent of economic growth in democracies, but
in dictatorships it is higher during economic crises.13 Moreover, the
incidence of mobilization is more or less the same whether or not a
democracy is to fall next year, 2.47 as opposed to 1.87 during other
years, but it is much higher in the years preceding deaths of dictator-
ships, 4.37 versus 1.01 (Table 2.14).
Given that in dictatorships political mobilization and economic crises
coincide, there were only three dictatorships that fell following a year
during which incomes declined but mobilization was not any higher
than during other years. In turn, in eleven instances incomes were
growing but mobilization was high. Hence, following our reasoning, we
can attribute three deaths of dictatorships to economic crises alone and
eleven to political mobilization alone. But in eighteen cases economic
crises and political mobilization coincided, so that we cannot identify
the cause. And in seventeen instances, dictatorships fell even though
the economy was growing and political mobilization was low, and we
have no way of telling what caused those dictatorships to die.
The number of cases in which democracies died while incomes fell
and mobilization remained low is larger: twelve out of thirty-eight.
In turn, eight democracies died while their economies were growing

13
In addition to the analysis in Table 2.13, we conducted statistical analyses, which
showed that the coefficient of growth on mobilization was negative and significant
under dictatorships and almost zero under democracies.

114
Economic Growth and Regime Dynamics

Table 2.13. Incidence of political mobilization, by longer-term dynamics

of per capita income

Incomes
increased
or decreased
during these Dictatorships Democracies
consecutive
years Surviving Dying Surviving Dying

Increased
3 or more years 1.05 5.79 1.97 2.00
2 years 0.94 1.25 1.83 2.60
1 year 0.71 2.30 1.65 1.13
Decreased
1 year 1.09 5.60 1.79 2.00
2 years 1.34 5.50 1.91 5.14
3 years 1.56 5.50 1.77 —
4 or more years 0.77 1.00 1.78 —

Actions
experienced

STRIKES 0.08 0.53 0.24 0.34

AGDEMONS 0.46 2.18 0.77 0.71
RIOTS 0.46 1.65 0.86 1.42
MOBILIZATION 1.01 4.36 1.87 2.47
Total 2,354 49 1,548 38

Note: The cell entries measure average incidences of the respective variables.

yet mobilization was high. Yet again, seven deaths of democracies

occurred when incomes declined while mobilization was high, and
eleven when incomes grew while mobilization was low.
Because we can think of no other way to tell what causes what, all
we can do is to conclude from what we can see. It is obvious that regime
transitions, events that are quite rare, occur under a wide variety of
circumstances. Economic crises are not sufficient to bring regimes
down. Neither are waves of political mobilization. Indeed, eleven
democracies and seventeen dictatorships fell when incomes were
growing and mobilization was low. There are obviously other causes:

115
 Economic Development and Political Regimes

Table 2.14. Growth and mobilization during the

year preceding regime transition

Regime and Mobilization

income
trend Low High Total

Dictatorships
Income growth
Negative 3 18 21
Positive 17 11 28
Total 20 29 49

Democracies
Income growth
Negative 12 7 19
Positive 11 8 19
Total 23 15 38

Note: “Low” mobilization is an incidence below the average

for years other than that preceding a regime transition;
“high” mobilization is above that average.

deaths of the founding dictator, institutional stalemates, external wars,

foreign pressures – the list goes on.
Just consider the fall of communism. Most communist countries
experienced a sharp economic downturn between 1978 and 1980.
Czechoslovakia, Hungary, and Yugoslavia never recuperated from that
crisis. Poland, Romania, and the Soviet Union recovered from that
downturn by the early 1980s, but their growth slowed down again after
1985. Given that dictatorships are more vulnerable at income levels
at which those countries found themselves, and given their economic
slowdown, the chances that those regimes would die increased. But
they were never large: We calculated them (using the model presented
in Table 2.12) to be about one in twenty. It took an economic crisis,
a wave of popular mobilization in Poland, the Soviet defeat in
Afghanistan, a massive United States armament, and probably several
strategic miscalculations (Kaminski 1997; Przeworski 1997) for one of
them to fall. And then Henry Kissinger’s dominoes tumbled, although
in the direction opposite to the one he had so eloquently predicted.

116
Income Inequality

We are thus not claiming that the economic factors are sufficient
to account for the different fates of regimes. The chances of regimes
surviving appear to depend somewhat on these conditions, but the
chances are only the odds according to which dice are thrown. Obvi-
ously, other conditions matter, as do the actions of the people living
under these conditions and, perhaps, sheer luck. Thus, regime selec-
tion is in part due to observable factors, such as level of income and
its growth, but in part to factors that we cannot or at least did not sys-
tematically observe.
Yet there are some things we do know. Foremost among them is that
democracies never die in wealthy countries. But all the evidence we
have examined also indicates that democracies in poorer countries are
more likely to die when they experience economic crises than when
their economies grow. In turn, dictatorships die under much more
varied economic circumstances: Indeed, it appears that economic
circumstances have little to do with the deaths of dictatorships.

Income Inequality
The entire analysis thus far has concerned only average or total indi-
cators for each country. But there are good reasons to think that a
factor that matters for regime stability is also the distribution of income
among different groups. Under dictatorships, high income inequality
may stimulate movements attracted by the egalitarian promise of
democracy. Under democracy, dominant social groups may seek
recourse to authoritarianism when the exercise of political rights by
the poor – whether in the form of suffrage or freedom of association –
results in egalitarian pressures.
Unfortunately, these hypotheses are almost impossible to test. The
best available data set on income distribution (Deininger and Squire
1996) is still far from complete and combines numbers collected by dif-
ferent methods. For many countries, information is not available at all,
and for many others it is available only for irregularly scattered years.
In the end, we thus have only 542 annual observations that match our
period. Moreover, some of the numbers are on an income basis, and
others on an expenditure basis; some are pre- and others post-fisc.
Hence, there is little to go on.
An additional complication is that the data show somewhat of a
Kuznets effect, that is, the degree of inequality, as measured by the Gini
index, increases and then falls with per capita income. The top panel
of Table 2.15 shows Gini indices by per capita income and by regime.

117
Table 2.15. Income inequality and regime dynamics

A: Inequality (Gini indices) by per capita income

Average Gini indices

LEVEL All Dictatorships Democracies

0–1,000 34.2806 (47) 35.7348 (23) 32.8871 (24)

1,001–2,000 38.6957 (74) 37.6443 (53) 41.3490 (21)
2,001–3,000 45.5331 (52) 43.1266 (29) 48.5674 (23)
3,001–4,000 39.0088 (49) 34.7779 (29) 45.1435 (20)
4,001–5,000 36.1776 (45) 32.4200 (33) 46.5108 (12)
5,001–6,000 35.0812 (24) 33.7861 (18) 38.9667 (6)
6001– 33.0105 (251) 32.2900 (18) 33.0662 (233)
Total 35.9952 (542) 36.1355 (203) 35.9112 (339)

B1: Regime transitions by Gini indices

PJK TJK TOT PAD TAD TA PDA TDA TD

All 0.0188 25 1,327 0.0322 19 590 0.0081 6 737

<35 0.0167 10 597 0.0376 9 239 0.0028 1 358
≥35 0.0205 15 730 0.0285 10 351 0.0132 5 379

B2: Regime transitions by ratios of top 20% to bottom 20%

PJK TJK TOT PAD TAD TA PDA TDA TD

All 0.0192 22 1,133 0.0364 17 467 0.0075 5 666

<9 0.0146 10 685 0.0347 10 288 0.0000 0 397
≥9 0.0268 12 448 0.0391 7 179 0.0187 5 269

C1: Regime transitions by changes of Gini indices

PJK TJK TOT PAD TAD TA PDA TDA TD

All 0.0240 29 1,209 0.0371 19 512 0.0143 10 697

<0 0.0272 17 626 0.0542 13 240 0.0104 4 386
>0 0.0206 12 583 0.0221 6 272 0.0193 6 311

C2: Regime transitions by changes in ratios of top 20% to bottom 20%

PJK TJK TOT PAD TAD TA PDA TDA TD

All 0.0224 23 1,027 0.0361 15 416 0.0131 8 611

<0 0.0208 11 528 0.0323 7 217 0.0129 4 311
>0 0.0240 12 499 0.0402 8 199 0.0133 4 300

(continued)
118
 Income Inequality

Table 2.15 (continued)

C3: Regime transitions by changes in the share of the bottom 40%

PJK TJK TOT PAD TAD TA PDA TDA TD

All 0.0222 23 1,036 0.0356 15 421 0.0130 8 615

<0 0.0246 14 568 0.0352 8 227 0.0176 6 341
>0 0.0192 9 468 0.0361 7 194 0.0073 2 274

C4: Regime transitions by changes in the share of the top 20%

PJK TJK TOT PAD TAD TA PDA TDA TD

All 0.0222 23 1,036 0.0356 15 421 0.0130 8 615

<0 0.0163 9 551 0.0287 6 209 0.0088 3 342
>0 0.0289 14 485 0.0424 9 212 0.0183 5 273

D1: Regime transitions by changes in the income of the bottom 40%

PJK TJK TOT PAD TAD TA PDA TDA TD

All 0.0224 23 1,027 0.0361 15 416 0.0131 8 611

<0 0.0367 11 300 0.0636 7 110 0.0211 4 190
>0 0.0165 12 727 0.0261 8 306 0.0095 4 421

D2: Regime transitions by changes in the income of the top 20%

PJK TJK TOT PAD TAD TA PDA TDA TD

All 0.0224 23 1,027 0.0361 15 416 0.0131 8 611

<0 0.0412 10 243 0.0761 7 92 0.0199 3 151
>0 0.0166 13 784 0.0247 8 324 0.0109 5 460

E1: Regime transitions by changes in average welfare

PJK TJK TOT PAD TAD TA PDA TDA TD

All 0.0224 23 1,027 0.0361 15 416 0.0131 8 611

<0 0.0494 12 243 0.0745 7 94 0.0336 5 149
>0 0.0140 11 784 0.0248 8 322 0.0065 3 462

Notes: PJK stands for the probability of any regime transition; TJK is their number. TOT is the
total number of annual observations. PAD stands for the probability of transition from authori-
tarianism to democracy; TAD is their number. TA is the total number of annual observations of
authoritarianism. PDA stands for the probability of transition from democracy to authoritarian-
ism; TDA is their number. TD is the total number of annual observations of democracy.

119
 Economic Development and Political Regimes

As we see, the average index increases from 34.28 for all countries
with incomes under $1,000 to 45.53 for all countries between $2,001
and $3,000 and then falls to 33.01 for all countries with incomes above
$6,000. Dictatorships and democracies show similar patterns and have
the same average income inequality, as measured by the Gini index.
Yet with all these caveats, some suggestive patterns do emerge. To
test whether or not the level of income inequality destabilizes regimes,
we considered five years (two before and two after, unless another
observation was available during these years or the regime changed)
around each observation of income distribution, thus obtaining 1,327
years for Gini indices and 1,133 for ratios of incomes of top and bottom
quintiles. This data set still contains only six (for Gini) and five (for the
ratio) transitions to dictatorship; hence the caveats should be kept in
mind. Nevertheless, dichotomizing the Gini index approximately at the
observed mean, that is, 35 (the actual mean is 38.3), shows that only
one democracy fell during the 358 years when the Gini index was
below this threshold, whereas five fell during the 379 years when
inequality was higher (panel B1 of Table 2.15). The effect of inequal-
ity on dictatorships, for which we observe 19 transitions, was less pro-
nounced. The ratio of top to bottom incomes shows the same: When
this ratio was less than nine, no democracy fell during 397 years, but
when it was higher, five fell during 269 years (panel B2). And, again,
the effect on dictatorships was negligible. Hence, whatever these
numbers are worth, they do suggest that democracy is more stable in
more egalitarian societies, and the durability of dictatorships is unaf-
fected by income distribution.
To test whether or not increasing or decreasing inequality has an
effect on the stability of regimes, we took all cases in which more than
one observation was available for a country, observed how income dis-
tribution changed, and calculated the rates of transition during these
periods.14 The numbers are again small, and they tend to be dispro-
portionately drawn from the more developed countries. Not much can
be learned from them. Changes in absolute income levels (see panels
D of Table 2.15) matter more for the stability of both regimes than do

14
Note that some of these data were presented in Table 2.9, which lists democracies that
lasted at least twenty years. If we consider only poorer countries, we see that democ-
racy survived in the Dominican Republic even though income inequality increased, but
it fell in the three other poor countries where inequality increased: the Philippines,
Brazil, and Chile. In turn, among countries where income inequality was reduced,
democracy survived in India, Japan, Finland, and Mauritius, and fell only in Sri
Lanka.

120
Income Inequality

changes in the overall distribution or the income shares of particular

groups (panels C). Both regimes are less stable when absolute incomes,
whether of the top 20 percent or the bottom 40 percent of earners,
decline.
In turn, changes in the overall distribution, whether measured by
the Gini index or by the ratio, have no clear effect on the stability of
regimes. Yet both regimes are slightly less stable when the share of the
top 20 percent increases. Democracies, and only democracies, are
somewhat less stable when the income share of the bottom 40 percent
declines. Thus, with all the caveats, it appears that both regimes are
threatened when the rich get relatively richer, but only democracy is
vulnerable when the poor get relatively poorer.
We also calculated changes in average welfare, attaching lower
weights to increases of incomes that were already higher (Atkinson
1970; logarithmic utility function). This transformation allows us to dis-
tinguish the effect of growth of the average income from the effect of
changes in its distribution: There are some cases in which the average
income grows but inequality increases to such an extent that the
average welfare declines. Both regimes appear to be more likely to die
when the average welfare falls. Moreover, comparison with Table 2.10
shows that both regimes are more sensitive to changes of welfare than
of income alone, implying that when the average income declines and
inequality increases – income declines at the cost of the poor – both
regimes are much more vulnerable than when the average income
increases and inequality is reduced.
The functional distribution of income, for which data are more
extensive, shows a much stronger impact of inequality for both
regimes. The data for labor share of value added in manufacturing (LS)
include 2,061 annual observations, as always drawn disproportion-
ately from poor dictatorships and rich democracies. The overall effect
of the functional distribution of income is very strong: Both regimes
are several times more likely to fall when labor receives less than 25
percent of value added in manufacturing. Dictatorships in countries
with incomes below $4,000 are particularly vulnerable when inequal-
ity is high, but dictatorships in wealthier countries survive at some-
what higher rates when the labor share is smaller. Democracies in
poor countries are less likely to survive when labor gets a lower
share, whereas in wealthier countries labor almost always gets more
than 25 percent, and they almost always survive (Table 2.16).
Thus, at the very least, there is no evidence that egalitarian pres-
sures threaten the survival of democracy. Indeed, it appears that

121
 Economic Development and Political Regimes

Table 2.16. Regime transitions, by level and labor share

LEVLAG
LSLAG PJK TJK TOT PAD TAD TA PDA TDA TD

0–1,000 0.0162 5 308 0.0072 2 277 0.0967 3 31

LS £ 25 0.0470 4 85 0.0259 2 77 0.2500 2 8
LS > 25 0.0044 1 223 0.0000 0 200 0.0434 1 23

1,001–2,000 0.0373 16 429 0.0271 9 332 0.0721 7 97

LS £ 25 0.0847 10 118 0.0752 7 93 0.1200 3 25
LS > 25 0.0192 6 311 0.0083 2 239 0.0555 4 72

2,001–3,000 0.0390 11 282 0.0379 6 158 0.0403 5 124

LS £ 25 0.0792 8 101 0.1154 6 52 0.0408 2 49
LS > 25 0.0165 3 181 0.0000 0 106 0.0400 3 75

3,001–4,000 0.0229 5 218 0.0217 3 138 0.0250 2 80

LS £ 25 0.0468 3 64 0.0882 3 34 0.0000 0 30
LS > 25 0.0129 2 154 0.0000 0 104 0.0400 2 50

4,001–5,000 0.0229 3 131 0.0370 3 81 0.0000 0 50

LS £ 25 0.0555 2 36 0.0909 2 22 0.0000 0 14
LS > 25 0.0105 1 95 0.0169 1 59 0.0000 0 36

5,001– 0.0101 7 693 0.0779 6 77 0.0016 1 616

LS £ 25 0.0385 1 26 0.0588 1 17 0.0000 0 9
LS > 25 0.0090 6 667 0.0833 5 60 0.0016 1 609

Total 0.0228 47 2,061 0.0272 29 1,063 0.0180 18 998

LS £ 25 0.0651 28 430 0.0711 21 295 0.0518 7 135
LS > 25 0.0116 19 1,631 0.0104 8 768 0.0127 11 863

Notes: LEVLAG is lagged per capita income; LSLAG is lagged labor share of output in
manufacturing.

democracies are less stable in societies that are more unequal to begin
with, in societies in which household income inequality increases, and
in societies in which labor receives a lower share of value added in
manufacturing. Dictatorships, in turn, particularly in poorer countries,
are much more vulnerable when the functional distribution of income
is more unequal.

Economic Factors in Context

Two questions still remain open. Are the economic factors still
important when one takes into account the cultural, social, and politi-

122
Economic Factors in Context

cal context in which they operate? And even if they do remain impor-
tant, do cultural, social, and political conditions affect the rise and
decline of political regimes independently of the economy?
To move beyond economic factors, we examine the impact of an entire
panoply of variables considered simultaneously. How do the political,
social, and cultural conditions in which regimes find themselves affect
the probability of transitions from democracy to dictatorship and from
dictatorship to democracy? The data show the following (Table 2.17):

1. The impact of per capita income (lagged, LEVLAG), treated linearly,

is apparent for both regimes, but it is orders of magnitude larger for
democracies.
2. The rate of economic growth (lagged, GLAG) matters for the stabil-
ity of both regimes, but less so for democracies. Both are less likely
to die when growth is faster.
3. Intra-regime instability, accumulated turnover of chief executives
(lagged, TLAG), has an important impact on the stability of both
regimes when it is introduced linearly into the analysis. Both regimes
are more likely to die when they experience frequent changes of
heads of government.

There is, however, a fundamental difference between democracies

and dictatorships. A moderate frequency of alternation in office
increases the stability of democracies, but any changes of the heads of
dictatorships shorten their lives. Table 2.18 shows the observed prob-
abilities of both regimes dying in a year that follows a particular fre-
quency of accumulated turnover. Democracies that change their chief
executives less frequently than once in five years die at the rate of
0.0249, with an expected life of forty years. But democracies that
change their heads of governments once between five and four years,
or even more frequently, down to once in two years, have a lower prob-
ability of dying. Only very frequent changes, more frequent that once
in two years, constitute “instability” in the sense of increasing the
chances of democracies dying. Dictatorships, in turn, are most durable
when changes are very infrequent: Their expected life when changes
occur less frequently than once in every five years is seventy-four years.
But when changes occur only somewhat more frequently, once between
five and four years, their expected duration is reduced to thirty years.
As turnover becomes higher, the expected life of dictatorship dwindles,
all the way down to eight years.
Hence, a moderate frequency of alternation in office increases the
stability of democratic regimes. After all, this is the essence of democ-

123
 Economic Development and Political Regimes

Table 2.17. Dynamic probit analysis of regime transitions

Number of observations 3,942

Log-likelihood function -291.0939
Restricted log-likelihood -2,644.349
Chi-squared 4,706.511
Degrees of freedom 23
Significance level 0.0000000

Transitions to dictatorship Transitions to democracy

Variable Coefficient t ratio Pr |t| ≥ x Coefficient t ratio Pr |t| ≥ x

Constant 0.12535 0.139 0.88954 -3.407 -3.119 0.00182

LEVLAG -0.54617E-03 -4.467 0.00001 0.3356E-04 3.953 0.00008
GLAG -0.23844E-01 -1.410 0.15858 -0.1997E-01 -2.215 0.02678
TLAG 0.97218 3.459 0.00054 0.5415 4.437 0.00001
RELDIF 2.5568 2.581 0.00984 0.1096 2.442 0.01459
CATH -1.1295 -2.089 0.03672 0.1535 1.624 0.10427
PROT -2.4463 -1.506 0.13213 -0.2615 -1.565 0.11765
MOSLEM 0.22024E-01 0.044 0.96497 -0.7248E-01 -0.085 0.93196
NEWC -0.16120E-01 -0.036 0.97151 -0.4348 -0.906 0.36489
BRITCOL -0.83942 -1.974 0.04837 0.1645 1.430 0.15260
STRA 0.89495 7.403 0.00000 0.3612 8.992 0.00000
ODWP -3.7446 -1.992 0.04633 3.033 -0.319 0.74958

Frequencies of actual and predicted outcomes; predicted outcome has maximum

probability

Predicted

Actual 0 1 Total

0 1,503 53 1,556
1 32 2,354 2,386
Total 1,535 2,407 3,942

Note: LEVLAG, lagged per capita income; GLAG, lagged rate of growth; TLAG, lagged
cumulative rate of leadership turnover; RELDIF, religious fractionalization; CATH, pro-
portion of Catholics in the population; PROT, proportion of Protestants in the population;
MOSLEM, proportion of Moslems in the population; NEWC, dummy variable indicating
that the country did not exist in 1945; BRITCOL, dummy variable indicating a former
British colony; STRA, cumulative number of past transitions to authoritarianism; ODWP,
proportion of other democracies in the world. The standard errors and the t ratios (includ-
ing their signs) under “Transitions to democracy” refer to shift coefficients (a in Appen-
dix I), not to the total coefficients (a + b).

124
 Economic Factors in Context

Table 2.18. Impact of intra-regime instability (turnover of heads of

governments) on regime transition rates

Frequency:
one change
in: PJK TJK TOT PAD TAD TA PDA TDA TD

>5 years 0.0162 45 2,777 0.0129 26 2,014 0.0249 19 763

5–4 years 0.0222 10 450 0.0292 5 171 0.0179 5 279
4–2 years 0.0351 21 597 0.0619 13 210 0.0206 8 387
2–1 years 0.0442 10 226 0.0588 4 68 0.0379 6 158
<1 year 0.0769 2 26 0.1250 1 8 0.0555 1 18

Total 0.0215 88 4,076 0.0198 49 2,471 0.0243 39 1,605

Notes: PJK stands for the probability of any regime transition; TJK is their number. TOT is the
total number of annual observations. PAD stands for the probability of transition from authori-
tarianism to democracy; TAD is their number. TA is the total number of annual observations of
authoritarianism. PDA stands for the probability of transition from democracy to authoritarian-
ism; TDA is their number. TD is the total number of annual observations of democracy.

racy: alternating in office as a result of application of rules. Dictator-

ship, however, is a system whose stability depends on personal rule.
Any change in the head of government is a threat to its survival.

4. As is frequently claimed, democracies are less stable in countries

that are religiously heterogeneous (RELDIF). We have also experi-
mented with another variable, “ethnolinguistic fractionalization,”
which measures ethnic, rather than religious, heterogeneity. This
variable is statistically significant when considered alone, but plays
no role once religious fractionalization is introduced.

What is striking, however, is that dictatorships are also less stable

in religiously or ethnolinguistically heterogeneous societies. Note that
the standard explanation for the instability of democracies in hetero-
geneous societies, one that goes back to J. S. Mill (1991 [1861]), is that
for democracy to endure there must be some shared values, a “con-
sensus.” Yet because dictatorships are also less likely to endure in
heterogeneous societies, the claim that common values are needed to
support democracy reduces to the observation that regime transitions
are more frequent in heterogeneous countries. Religious or ethnolin-
guistic heterogeneity simply makes all political regimes less stable.

125
 Economic Development and Political Regimes

5. Since Montesquieu (1995 [1748]), much has been said about the
importance of culture for the emergence and durability of democra-
cies. Recent discussions of this topic have revolved mainly around
cultures identified by dominant religions. Even if Weber (1958
[1904]) himself said almost nothing about political institutions (Prze-
worski, Cheibub, and Limongi 1997), the idea that he saw in Protes-
tantism the wellspring of modern democracy is widespread among
contemporary political scientists. Lipset (1959: 165) claimed that “it
has been argued by Max Weber among others that the factors
making for democracy in this area [north-west Europe and their
English-speaking offsprings in America and Australasia] are a his-
torically unique concatenation of elements, part of the complex
which also produced capitalism in this area,” because “the empha-
sis within Protestantism on individual responsibility furthered the
emergence of democratic values.”15 In turn, Catholicism, in Lipset’s
view (1960: 72–3), was antithetical to democracy in pre–World War
II Europe and Latin America. Yet even Catholicism is not the worst
enemy of democracy: Islam and Confucianism hold the palm (Eisen-
stadt 1968: 25–7). Thus, Huntington (1993: 15) reported that “no
scholarly disagreement exists regarding the proposition that tradi-
tional Confucianism was either undemocratic or antidemocratic.”
Similar views about Islam abound (Gellner 1991: 506; Lewis 1993:
96–8).

According to our analysis, none of those assertions can withstand

scrutiny. Indeed, the only effect of religions that emerges from the sta-
tistical examination is that democracies are more likely to survive in
countries in which there are more Catholics. Neither Protestantism nor
Islam seems to have an effect on the emergence or the durability of
democracy.
6. A colonial legacy has little effect on regime stability once all the other
factors are considered. Democracies are somewhat more likely to
survive in countries that were British colonies (BRITCOL), but having
been a colony at all (NEWC) has no effect. Note that the rates of
democratic failure observed between 1950 and 1990 were much
higher among democracies that were established after 1950 (0.0620,
with an expected life of 16.8 years) than in the entire set of democ-
racies (0.0237, with an expected life of 42.2). Yet this effect vanishes

15
Lipset does not point to any specific text of Weber. Neither do Almond and Verba (1963:
10), who assert that “the development of Protestantism, and in particular the noncon-
formist sects, have been considered vital to the development of stable political institu-
tions in Britain, the Old Commonwealth, and the United States” (italics added).

126
Economic Factors in Context

in the statistical analysis, indicating that the observed difference was

due to the low incomes of these countries, not the timing of
independence.
7. The political history of regimes, their past instability, has an impor-
tant role in affecting the chances that the current regime will survive
or die. In countries that experienced more transitions to dictatorship
(STRA) in the past, both types of regimes are less stable, and the
effect is about three times larger for democracies.

This last finding suggests that political learning is a complex matter.

It is frequently argued – Russia is a good example – that the absence
of democratic traditions is an impediment to consolidating new demo-
cratic institutions and, conversely, that democracy is more stable in the
countries that have enjoyed it in the past – here Chile is the paradig-
matic case. Yet what that argument misses is that if a country has had
a democratic regime in its past, it also has experienced subversion of
democracy. Political learning thus cuts both ways. Perhaps democrats
find consolidation easier when they can rely on past traditions, but
anti-democratic forces also have an experience from which they can
draw lessons: People know that an overthrow of democracy is possi-
ble, how it happens, and how to bring it about. If the Russian coup of
1991 was more of a coup de theatre than a coup d’état, it was perhaps
because the Russian golpistas simply did not know how to do it and
were justly ridiculed by their experienced Latin American soulmates.

8. Finally, whereas the international political climate (ODWP) has an

impact on the stability of democracies, it has no effect whatever on
transitions to democracy. The probability that once established a
democracy will die is lower when many countries in the world are
democratic. But the probability that a democracy will be established
does not depend on the proportion of countries in the world that
have democratic regimes. This finding casts doubts on the notion
that democratization comes in waves, a topic we discussed earlier.

To put these findings in perspective, it is necessary to remember that

regime transitions are relatively rare. Although we have seen that some
countries experience a fair amount of regime instability, in most coun-
tries a single regime will continue for long periods. This heterogeneity
appears to be due to the factors that we analyzed earlier, particularly
to past history: Once these factors are considered, the sample is no
longer heterogeneous in the statistical sense. But inertia remains high:
In each year, most countries have the same regimes they had in the
preceding year.

127
 Economic Development and Political Regimes

Yet regimes do die, and the transitions can be explained. When the
impacts of different factors are taken one by one, it again becomes clear
that the level of economic development is crucial for the stability of
democracies, whereas both regimes are affected by their economic per-
formance and by the frequency of turnover of heads of government. Of
all the other variables, past regime instability plays the greatest role.
Religious or ethnolinguistic heterogeneity also affects the stability of
both regimes. In turn, the colonial heritage and cultural patterns, at least
as indicated by the frequency of major religions, play almost no role.
The impact of the level of development, of leadership turnover, and
of past regime instability is robust, in the (Leamer) sense that these
variables are highly significant regardless which other factors are intro-
duced into the analysis. Economic performance variables always
matter, but their statistical significance is somewhat affected by the
presence of other variables. The impact of all the other variables is not
robust: In some specifications, they appear as statistically significant,
but in others they do not. When per capita income is not considered,
Protestantism becomes somewhat important, and ethnolinguistic
fractionalization dominates religious heterogeneity. In some other spec-
ifications, having been a colony appears to have an impact independent
of the British colonial heritage. One should not be surprised, therefore,
that the determinants of regimes have been the subjects of such exten-
sive debates. Yet we think that the basic patterns are clear and robust:
The level of economic development, economic performance, past
regime instability, and leadership turnover tell almost all of the story.

Democratic Institutions and the Sustainability of Democracy

Democracies are not all the same. Systems of representation, divi-
sions of powers, legal doctrines, and the bundles of rights and obliga-
tions associated with citizenship all differ significantly among regimes
that are generally recognized as democratic. And these institutional
arrangements may have an impact on the sustainability of democracy.
The durability of democracies may not be simply a matter of economic,
social, or cultural conditions, because their institutional frameworks
may differ in their capacity to process conflicts, particularly when these
conditions become so adverse that democratic performance is consid-
ered to be inadequate.16 Democracy is sustainable when its institutional

16
Seminal studies along this line are those by Zimmerman (1987, 1988), who found that
the timing and the depth of the recession of 1929–1932 could not predict whether or
not a democracy would survive that crisis.

128
Institutions and the Sustainability of Democracy

framework promotes normatively desirable and politically desired

objectives, but also when these institutions are adept at handling crises
that occur when such objectives are not being fulfilled.
The institutional distinction that appears particularly important
is that between presidential and parliamentary systems. We discussed
this distinction in Chapter 1. Now we must examine whether or not it
makes a difference for the durability of democracies.
A glance at the descriptive patterns shows immediately that Linz
(1990a,b) was right about the durability of these respective institutional
systems. During the period under consideration there were sixteen (28
percent of 57 spells) democracies that died under a parliamentary
system and twenty-three (54 percent of 42 spells) that died under a
presidential one.17 Among those democracies that died, parliamentary
systems lasted on the average two years less than presidential ones:
7.6 years as opposed to 9.6. But the parliamentary democracies that
were still around as of 1990 were much older: on the average about
forty-one years, as compared with twenty-four for presidential regimes.
The probability that a presidential democracy will die during any par-
ticular year is 0.0477, and the probability that a parliamentary democ-
racy will die is 0.0138. Thus, the expected life of democracy under
presidentialism is approximately twenty-one years, whereas under
parliamentarism it is seventy-three years. The oldest democracy in the
world is presidential, but this is a distant exception. Moreover, to dispel
the view that instability of presidential democracies is due to the fact
that they exist primarily in Latin America, it is worth noting that the
completed spells of presidentialism were on the average longer on that
continent (10.6 years) than elsewhere (6.5 years), that is, in the Congo
(1960–1962), Ghana (1979–1980), Nigeria (1979–1982), Uganda
(1980–1984), and the Philippines (1950–1964) (Table 2.19).
We have seen that democracies are more likely to survive in wealthy
countries, and parliamentary democracies are found disproportion-
ately often under such conditions: Whereas two-thirds of presidential
systems were observed in countries with incomes under $4,000, only
30 percent of parliamentary democracies were in countries below that
threshold. Hence, the immediate question is whether or not the dura-
bility of the parliamentary systems is due simply to wealth.

17
Note that we lump mixed systems together with parliamentary ones (see Chapter 1).
Mainwaring (1993) counted democratic breakdowns since 1945. He found nineteen
under parliamentarism, twenty-seven under presidentialism, and four under other
types.

129
 Economic Development and Political Regimes

Table 2.19. Observed rates of transition to dictatorship, by lagged per capita

income (LEVLAG) and lagged rate of economic growth (GLAG), of
parliamentary and presidential democracies and the hypothetical numbers of
transitions if there had been as many presidential democracies as
parliamentary democracies under all circumstances

LEVLAG
GLAG PLA TLA TL PPA TPA TP PPA HTPA TL

0–1,000 0.1094 7 64 0.2000 2 10 0.2000 12.8 64

G£0 0.1667 4 24 0.3333 1 3 0.3333 8 24
G>0 0.0750 3 40 0.1429 1 7 0.1429 5.7 40
1,001–2,000 0.0467 5 107 0.0621 9 145 0.0621 6.6 107
G£0 0.0833 3 36 0.0755 4 53 0.0755 2.7 36
G>0 0.0282 2 71 0.0543 5 92 0.0543 3.9 71
2,001–3,000 0.0233 2 86 0.0594 6 101 0.0594 5.1 86
G£0 0.1176 2 17 0.0690 2 29 0.0690 1.2 17
G>0 0 0 69 0.0556 4 72 0.0556 3.8 69
3,001–4,000 0.0250 2 80 0.0313 2 64 0.0313 2.5 80
G£0 0.0526 1 19 0.0455 1 22 0.0455 0.9 19
G>0 0.0164 1 61 0.0238 1 42 0.0238 1.5 61
4,001–5,000 0 0 72 0.O556 2 36 0.O556 4 72
G£0 0 0 13 0.O769 1 13 0.0769 1 13
G>0 0 0 59 0.0435 1 23 0.0435 2.6 59
5,001–6,000 0 0 100 0.0909 1 11 0.0909 9.1 100
G£0 0 0 13 0 0 3 0 0 13
G>0 0 0 87 0.125 1 8 0.1250 10.9 87
6,001–7,000 0 0 108 0.0588 1 17 0.0588 6.4 108
G£0 0 0 23 0.1667 1 6 0.1667 3.8 23
G>0 0 0 85 0 0 11 0 0 85
7,001– 0 0 546 0 0 98 0 0 546
G£0 0 0 94 0 0 23 0 0 94
G>0 0 0 452 0 0 75 0 0 452
Total 0.0138 16 1,163 0.0477 23 482 0.0400 46.5 1,163
G£0 0.0418 10 239 0.0658 10 152 0.0736 17.6 239
G>0 0.0065 9 924 0.0394 13 330 0.0306 28.3 924

Notes: G £ 0 means that per capita income declined or remained the same; G > 0 means that
per capita income increased. PLA stands for the probability of transition from parliamentarism
to authoritarianism; TLA is their number. TL is the total number of annual observations of par-
liamentarism. PPA stands for the probability of transition from presidentialism to authoritari-
anism; TPA is their number. TP is the total number of annual observations of presidentialism.
HTPA stands for the expected number of transitions from presidentialism if there were exactly
as many presidential as parliamentary democracies.

130
Institutions and the Sustainability of Democracy

Presidential democracies are more likely to die than parliamentary

democracies at any level of development. Indeed, in countries with
incomes above $4,000, no parliamentary democracy died during 826
years they enjoyed such incomes, whereas four presidential democra-
cies died during 162 years (one in forty years) under the same condi-
tions. The richest parliamentary democracy that died was Suriname,
which had a per capita income of $3,923 in 1980 (see Table 2.6). The
second wealthiest was Greece, with income of $3,176 in 1967. In con-
trast, six transitions from presidential democracies to dictatorships
occurred in countries that were wealthier, all the way up to Argentina
in 1976, with an income of $6,055. The difference among countries
with incomes under $4,000 is much smaller, but still in favor of par-
liamentarism: Sixteen parliamentary democracies died under such cir-
cumstances during 337 years (one in twenty-one years), compared with
nineteen presidential democracies during 320 years (one in seventeen
years).
To isolate how much of the difference in the durability of democracy
under these two systems is due to wealth, we calculated what would
have been the rate of transition from presidential democracies to
authoritarianism if there had been exactly as many of them as there
were parliamentary democracies in each band of $1,000 of per capita
income. The answer is that presidential democracies would have died
at an only slightly lower rate than under the observed conditions:
0.0400 (one in twenty-five) instead of 0.0477 (one in twenty-one).
Hence, the difference in longevity between parliamentary and presi-
dential democracies was not due to the wealth of the countries in which
these democracies existed.
How adept are the alternative institutional arrangements in coping
with economic crises? One in twenty-four parliamentary democracies
dies (at the rate of 0.0418) when the economy experiences negative
rates of growth. When the economy grows, only one in 154 parlia-
mentary democracies collapses (0.0065). Hence, parliamentary democ-
racies are sensitive to economic crises. Presidential systems are less
sensitive to economic performance. One in fifteen presidential democ-
racies collapses when incomes decline (0.0658), and one in twenty-five
(0.0394) when they grow. But note that parliamentary systems are
almost as likely to survive when the economy shrinks as presidential
systems are likely to survive when it expands.
We already know that parliamentary democracies endure in coun-
tries with incomes above $4,000 regardless of their economic perfor-
mance. Such democracies survived 143 years during which per capita

131
 Economic Development and Political Regimes

incomes declined, and 683 years when they grew. Presidential democ-
racies, however, died in four instances in such wealthy countries: two
in forty-five years when incomes declined (one in 22.5), and two in 117
years when incomes increased (one in 58.5). Moreover, parliamentary
democracies survive with incomes below $4,000 if the economy grows
– only one in 40 (0.0248) of them died under these conditions – much
better than presidential ones, which died at the rate of one in 19.4
(0.0511). Yet the enthusiasm for parliamentary systems should be
somewhat mitigated, for it seems that their capacity to cope with eco-
nomic crises is limited to countries that are wealthy. When they are
poor, they are vulnerable to economic crises: One in 9.6 of them
(0.1042) dies when incomes decline. This rate is higher than that for
presidential systems – one of which in 13.4 (0.0748) dies under such
conditions – but this difference is due to the fact that the poor parlia-
mentary democracies were poorer than the poor presidential ones. Had
poor presidential democracies occurred at the same income levels as
the poor parliamentary systems, they would have died at the rate of
0.1333, one in 7.5. Thus, presidential democracies are just unstable,
in wealthy as well as in poor countries, when the economy declines
and when it expands. Parliamentary democracies are stable in wealthy
countries and survive at high rates when the economy grows in poor
countries. But they are vulnerable in poor countries that face economic
crises.18
Statistical analyses highlight the different impacts of the level of
development on the survival of parliamentary and presidential democ-
racies. Whereas parliamentary democracies are much less likely to die
in countries in which the per capita income is high, the chances of
survival for presidential democracies are independent of per capita
income. Moreover, once other factors are considered, presidential
democracies appear somewhat more sensitive to their economic per-
formance (Table 2.20).
These two economic factors play their same roles whether or not
they are placed in a broader context. But non-economic variables do
play a role in determining the chances of survival for the two types of

18
Note that in our standard classification of regimes we count as authoritarian the periods
before elected governments overthrow democracy from above (“consolidation rule” of
Chapter 1). But if such instances (listed in Appendix 1.3) should prove to be more fre-
quent under parliamentarism, then our results would be an artifact of classification. To
test that possibility, we reproduced Table 2.20 reclassifying such periods as democra-
tic and treating such autogolpes as transitions. None of the qualitative conclusions was
affected.

132
 Institutions and the Sustainability of Democracy

Table 2.20. Dynamic probit analysis of regime transitions, by type of

democracy

Binomial probit model

Maximum-likelihood estimates
Dependent variable REG
Number of observations 3,927
Iterations completed 10
Log-likelihood function -337.9800
Restricted log-likelihood -2,630.362
Chi-squared 4,584.764
Degrees of freedom 17
Significance level 0.0000000

Transitions to dictatorship

Transitions to democracy From parliamentarism From presidentialism

Variable Coefficient Pr[|Z| ≥ z] Coefficient Pr[|Z| ≥ z] Coefficient Pr[|Z| ≥ z]

Constant -2.4358 0.00000 -1.413 0.00000 -1.867 0.00000

LEVLAG 0.7674E-04 0.04158 -0.4268E-03 0.00401 -0.1261E-03 0.46306
GLAG -0.1627E-01 0.08018 -0.2897E-01 0.08296 -0.3513E-01 0.01543
TLAG 0.6591 0.00038 0.7195 0.00104 0.5302 0.00144
RELDIF -0.3177 0.30042 0.6762 0.62051 1.660 0.07734
STRA 0.3759 0.00000 0.3995 0.00363 0.1244 0.00008

SEATS -0.5318E-02 0.00136 0.3597E-02 0.83106 -0.2993E-02 0.26070

MAJORITY -0.4399 0.00228 0.1734 0.40417 -0.3708 0.00245

DEADLOCK 0.8453 0.01376 -0.6142E-01 0.08420 0.4827 0.00129

MAJORITY 0.3979 0.00706 0.3592 0.94205 -0.1209 0.12600

DEADLOCK 0.6274 0.07104 0.2415 0.17335 0.3989 0.02739

Notes: The first panel presents results for the five variables considered together. Next, the results for
each variable are added to the first five (the values of coefficients and probabilities are almost unaf-
fected by these additions). Finally, the bottom panel shows the results of DEADLOCK and MAJORITY
being added simultaneously to the basic five variables. LEVLAG, lagged per capita income; GLAG, lagged
rate of growth; TLAG, lagged cumulative rate of leadership turnover; RELDIF, religious fractionaliza-
tion; STRA, cumulative number of past transitions to authoritarianism; SEATS, share of seats of the
largest party; DEADLOCK, dummy variable that equals 1 if 0.33 < SEATS < 0.50; MAJORITY, dummy
variable that equals 1 if SEATS > 0.50.

democracies. Religious heterogeneity appears to affect only presiden-

tial systems, which are less stable in heterogeneous societies. In turn,
past regime instability lowers the chances of survival for democracy
under both systems, more so under parliamentarism.
There are good reasons to believe, however, that the functioning of

133
 Economic Development and Political Regimes

these institutions depends not only on economic or social factors

but also on the relationships of the political forces within them. In par-
ticular, a claim has been made repeatedly (Mainwaring 1993; Stepan
and Skach 1993; Jones 1995; Carey 1997; Mainwaring and Shugart
1997) that presidential systems are particularly unstable when they
function under a highly fractionalized party system. One way to
examine this hypothesis is to study the effect of the share of the largest
party in the lower house of the legislature under the two systems
(SEATS). This variable has no impact on the survival of either type of
democracy. The absence of a majority party (MAJORITY), however,
does have a strong impact on the stability of presidential democracies,
which are unstable when no party controls a majority of seats in the
lower house. The chances of survival for parliamentary democracies,
in turn, are independent of the existence of a majority party (Table
2.21).
The story, however, does not end there. Presidential democracies
appear particularly vulnerable in situations in which the largest leg-
islative party controls more than one-third but less than one-half of
seats, a situation that we term “DEADLOCK.” Note that both the
MAJORITY variable and the DEADLOCK variable assume values of zero
in situations of extreme fractionalization, when no single party con-
trols more than one-third of seats. Hence, what distinguishes them
are situations in which the largest party has more than one-third
but less than one-half of seats. And presidential democracies are
much less likely to survive under the conditions of such moderate
fractionalization.19
The reason we refer to these situations as deadlocks is based on the
following intuition. Suppose that the plurality party is the same as the
president’s party, but the opposition parties can muster a majority if
they unite against the president. Then the opposition can pass legisla-
tion. The president can veto it, and, under typical procedures for
overruling the veto, the president will have enough support to make it
stand. Hence, a deadlock ensues. Yet we may be barking up the wrong
tree, for if legislative deadlocks are what underlie such situations, then
such deadlocks should also ensue when a party different from that of
the president controls a majority of seats. Hence, we examined the

19
“Deadlocks” occurred in 30.9% of presidential and 45.0% of parliamentary years.
Majority parties existed in exactly one-half of presidential and 43.1% of parliamentary
years. Finally, situations of extreme fractionalization are not infrequent: In 19.1% of
years under presidentialism and 11.9% under parliamentarism, no party controlled
more than one-third of seats.

134
 Institutions and the Sustainability of Democracy

Table 2.21. Hazard rates for parliamentary and presidential democracies, by

political conditions

Total Parliamentary Presidential

Variable PDA TDA TD PLA TLA TL PPA TPA TP

SEATS
0.0–33.3 0.0220 5 227 0.0147 2 136 0.0330 3 91
33.4–50.0 0.0224 15 668 0.0077 4 519 0.0738 11 149
50.1–66.6 0.0204 11 540 0.0146 5 343 0.0304 6 197
66.7–100 0.0402 8 199 0.0325 5 154 0.0667 3 45
MINORITY 0.0223 20 895 0.0091 6 656 0.0581 14 241
MAJORITY 0.0256 19 741 0.0201 10 498 0.0373 9 241
EFFPARTY
0–2 0.0261 9 345 0.0240 6 250 0.0316 3 95
2–3 0.0154 9 585 0.0051 2 394 0.0366 7 191
3–4 0.0281 9 320 0.0041 1 246 0.1081 8 74
4–5 0.0272 5 184 0.0149 2 134 0.0600 3 50
5– 0.0126 2 158 0.0097 1 103 0.0182 1 55

Notes: Totals differ because some data for seats and for the number of effective parties are not
available. SEATS, share of seats of the largest party; MAJORITY, dummy variable that equals 1
if SEATS > 0.50; EFFPARTY, index of the number of effective parties. PDA stands for the prob-
ability of transition from democracy to authoritarianism. TDA is their number. TD is the total
number of annual observations of democracy. PLA stands for the probability of transition from
parliamentarism to authoritarianism. TLA is their number. TL is the total number of annual
observations of parliamentarism. PPA stands for the probability of transition from presidential-
ism to authoritarianism. TPA is their number. TP is the total number of annual observations of
presidentialism.

effect of a version of the DEADLOCK variable in which we added all

the instances of divided government (which occurred in seven coun-
tries, during forty-nine years) to the previous version. Yet the effect of
this new variable on the stability of presidential democracies is much
weaker. Hence, it may be not deadlock that weakens presidential
democracies, but something about the difficulty of forming legislative
coalitions when there are a few parties with similar strengths. And note
that the descriptive patterns presented in Table 2.21 show that presi-
dential systems are particularly brittle when the number of effective
parties in the legislature is between three and four. The process of
forming legislative coalitions in presidential systems has been little

135
 Economic Development and Political Regimes

studied (but see Figueiredo and Limongi 2000), and we do not know
how it works.
Before concluding, however, that instability of democracy is an
inherent effect of presidentialism, we need to examine one more rival
hypothesis. If presidentialism is a military legacy, then perhaps presi-
dential democracies last for shorter periods simply because they
emerge in countries where the military is politically relevant. We thus
need to compare separately the hazard rates for parliamentary and
presidential democracies distinguished by their origins. It is apparent
that a military legacy shortens the life of democracy regardless of its
institutional framework. Democracies that emerged from civilian dic-
tatorships died at the rate of 0.0158, with an expected life of 63.4 years;
those that succeeded military dictatorships died at the rate of 0.0573,
with an expected life of 17.5 years. Parliamentary democracies,
however, are still more stable regardless of their origins. Given civil-
ian origins, parliamentary democracies died at the rate of 0.0119 and
had an expected life of 83.7 years, and presidential democracies died
at the rate of 0.0329, with an expected life of 30.4 years. Given mili-
tary origins, parliamentary systems died at the rate of 0.0400, with an
expected life of 25 years, and presidential systems died at the rate
0.0628 and had an expected life of 16 years. Thus, again, the stability
of democracies seems to be an effect of their institutional frameworks,
not only of their origins.
Thus, although we remain uncertain about the reasons, it is clear
that presidential democracies are less durable than parliamentary
ones. This difference is not due to the wealth of the countries in which
these institutions were observed, nor to their economic performance.
Neither is it due to any of the political conditions under which they
functioned. Presidential democracies are simply more brittle under all
economic and political conditions.

Conclusion
We began this chapter with the observation that democracies are
much more frequent in developed countries, and dictatorships in poor
ones. Yet this observation is not very illuminating, and neither are the
innumerable cross-sectional analyses of this pattern. The regimes we
observe in particular countries at any moment depend on the condi-
tions under which these regimes were born and on the conditions they
encountered and produced as time passed. And because our system-
atic observations begin only in 1950 or when countries first became

136
Appendix 2.1: Dynamic Probit Model

independent (in some cases when data became available), we took the
regimes under which the countries entered the sample as given and
studied their subsequent dynamics.
The most important lesson we have learned is that wealthy coun-
tries tend to be democratic not because democracies emerge as a con-
sequence of economic development under dictatorships but because,
however they emerge, democracies are much more likely to survive in
affluent societies. We find it difficult to explain why dictatorships die
and democracies emerge. Although we are willing to believe that eco-
nomic development may open the possibility for transition to democ-
racy, even when the conditions for democracy are ripe, the outcomes
of political conflicts are indeterminate. Hence, we failed to detect any
thresholds of development that would make the emergence of democ-
racy predictable. In sum, modernization theory appears to have little,
if any, explanatory power.
In turn, we found that the survival of democracies is quite easily pre-
dictable. Although some other factors play roles, per capita income is
by far the best predictor of the survival of democracies. Democracies
survive in affluent societies whatever may be happening to them. They
are brittle in poor countries. But they are not always sentenced to
die: Education helps them to survive independently of income, and a
balance among the political forces makes them more stable. Institu-
tions also matter: Presidential democracies are less likely to survive
under all circumstances we could observe than are parliamentary
ones.
Yet one should not forget that we are dealing only with chances,
probabilities, not certainties. And if all we can predict are chances,
there must be other factors that matter, some we did not identify. These
may be idiosyncratic, impossible to catch in a statistical analysis; but
on the other hand, they might be systematic, and we may simply have
failed to find them. Whichever they are – this is not a book on the phi-
losophy of history – they remain hidden from our scrutiny. And this is
a fact with consequences for what follows.

Appendix 2.1: Dynamic Probit Model

To decide which mechanism generates the relationship between
development and democracy, we need to determine how the respec-
tive transition probabilities change with the level of development or
other exogenous variables. To estimate the impacts of these variables
on transition probabilities, we rely on Amemiya (1985, chap. 11).

137
 Economic Development and Political Regimes

Our data obey a first-order Markov process,20 that is, the present
regime depends only on the regimes during the preceding year, but not
beyond. Such processes are defined by
E (Rt = A Rt -1, Rt -2 , . . .) = P(t)Rt -1,

where R = A, D stands for regimes, R = A for dictatorship and R = D

for democracy, and P(t) is the matrix of transition probabilities, with
elements pjk(t), j = D, A, k = D, A. Hence, in regression terms,

Rt = P(t ) Rt -1 + ut .

Taking expectations of both sides yields

Èp(Rt = A )˘ = Èp AA p DA ˘ Èp(Rt -1 = A )˘,

ÎÍ p(Rt = D)˚˙ ÍÎ p ADp DD ˚˙ ÎÍ p(Rt -1 = D)˚˙

where the sum of columns of the transition matrix Sjpjk = 1. Hence

p(Rt = A Rt -1 ) = p AA (t)p(Rt -1 = A) + p DA (t)p(Rt -1 = D)

= p DA (t) + [ p AA (t) - p DA (t)]p(Rt -1 = A ).

Now let X be the vector of the exogenous variables. Assume that

pDA (t ) = F ( Xt -1b),

pAA (t ) = F [ X t -1 (a + b)],

where F(·) is the c.d.f. of the normal distribution, with derivatives

dpDA dpAD
= f ( X t -1b)b and = - f [ X t -1 (a + b )](a + b ).
dX dX

Note that pjk(t) is the probability of transition from being in state j

at time t - 1 to being in state k at time t. Given that whenever a tran-
sition occurs we code the regime as the one that became installed
during this year, the probability of transition between t - 1 and t
depends on the conditions at t - 1. Hence, we lag the X’s.
Letting R = 1 when the regime is authoritarian and R = 0 when it is
democratic, so that p(R = A) = pR,

20
If the process is of order 1, then all the information about the lagged values of
the exogenous variables is summarized by the lagged value of the dependent variable.
If the process is of a higher order, then lagged values of the exogenous variables
affect the current value of the dependent variable. Our tests indicate that only the con-
temporaneous values of the exogenous variables matter, that is, the process is of
order 1.

138
Appendix 2.2: Survival Models

p( Rt = A Rt -1 ) = pDA (t ) + [ pAA (t ) - pDA (t )]pR (t - 1)

= F ( X t -1b) + {F [ X t -1 (a + b)] - F ( X t -1b)} pR (t - 1)
= F ( X t -1b) + F ( X t -1a ) pR (t - 1) = F ( X t -1b + X t -1 Rt -1a ).

Hence, to estimate a and b, from which one can calculate pDA

and pAA, and thus pDD = 1 - pDA and pAD = 1 - pAA, all we need is to
maximize
R( t ) 1- R( t )
L = Pit F ( X t -1b + X t -1aRt -1 ) [1 - F ( X t -1b + X t -1aRt -1 )] .

This is the model we used to generate the results presented in

several tables in the main body of the text, with R(0) as observed.
To study the effect of the type of democracy on the transition prob-
abilities, we estimated the equation

p(Rt = A Rt -1 ) = p AA (t)p(Rt -1 = A ) + p LA (t)p(Rt -1 = L) + p PA (t)p(Rt -1 = P),

where L stands for parliamentarism and P for presidentialism, using

the observed fact that pLP ⬇ pPL ⬇ 0 to identify pLA = 1 - pLL and pPA =
1 - pPP, and letting

pAA (t ) = F ( X t -1b),

pLA (t ) = F [ X t -1 (a + b)],

pPA (t ) = F [ X t -1 (a + g )].

Appendix 2.2: Survival Models

Survival models allow a relatively simple test of the hypothesis that
the probability of survival is independent of age. To test whether or not
democracies become consolidated as a result of the passage of time,
that is, whether or not they are more likely to be around the longer
they have been around, we estimate, first, models without any right-
hand-side variables and then control for per capita income.
The hazard function, that is, the probability that a regime will die
at a particular time, t = 1, . . . , given that it has lasted until then, for
the Weibull distribution is
p-1
h(t ) = lp(lt ) .
Assuming that observations are gamma-heterogeneous, the hazard
function for the Weibull distribution is

h(t ) = lp(lt )
p-1
[1 + Q(lt ) ],
p

139
 Economic Development and Political Regimes

where Q is the parameter of the gamma distribution of survival pro-

babilities across observations, with a high Q indicating that observa-
tions are heterogeneous. Finally, for the logistic distribution, the hazard
function is
h(t ) = lp(lt )
p-1
[1 + (lt ) ].
p

The time derivatives of the hazard function for the Weibull distrib-
ution are
dh (t ) dt < 0 if p < 1,
dh (t ) dt = 0 if p = 1,
dh (t ) dt > 0 if p > 1.

For the logistic these derivatives are

p
dh (t ) dt < 0 if p < 1 + (lt ) ,
p
dh (t ) dt = 0 if p = 1 + (lt ) ,
p
dh (t ) dt > 0 if p > 1 + (lt ) .
Fitting the models to data yields the coefficients s = 1/p and an inter-
val estimate of p (t statistics in parentheses).
The hypothesis that the Weibull distribution is homogeneous is not
rejected [Q = 10.73, t statistic = 1.402, Pr(t) = 0.16], so there is no
reason to assume heterogeneity.
Model s min < p < max

Weibull 1.6194 0.32 < 0.61 < 0.91

(4.091)
Logistic 1.2661 0.47 < 0.79 < 1.11
(4.860)

The estimates of p are thus safely smaller than 1.00, which indicates
that the hazard rates are declining with time, in other words, that
democracies become consolidated.
These results are, however, misleading, because as time goes on,
countries develop, and as they develop, the probability that democracy
will fall declines. Hence, to isolate the effect of time, we need to dis-
tinguish it from the effect of development. To do that, we make the
lambdas at each time functions of X = 1, LEVEL, so that
l( X t ) = exp( -bX t ),

and substitute these lambdas into the hazard functions.

140
Appendix 2.2: Survival Models

The distribution is again homogeneous, and the results are

Model s min < p < max LEVEL

Weibull 0.9854 0.70 < 1.01 < 1.33 0.6729

(6.282) (5.178)
Logistic 0.7718 0.94 < 1.29 < 1.65 0.6576
(7.075) (5.386)
Exponential — — 0.6779
(6.265)
Because the Weibull p is almost exactly 1.00, the Weibull hazard rate
is a constant function of time: Given their level of development, democ-
racies are equally likely to die at any age. Hence, what may appear as
“consolidation” is in fact a consequence of development, not merely of
the passage of time. The logistic p = 1.29 implies that with the average
hazard rate of 0.025, this hazard rate declines during the first fifteen
to sixteen years and then increases.
The results for dictatorships, however, show that their hazard rates
tend to decrease somewhat over time (the Weibull distribution is not
heterogeneous).
min < p
Model s < max Q LEVEL LEVSQR

Exponential — — — -0.7780 0.0552

(-2.728) (1.372)
With het — — 0.48 -0.9358 0.0637
(1.757) (-2.344) (1.008)
Weibull 1.3005 0.56 < 0.77 — -1.0372 0.0717
< 0.98
(7.074) (-2.601) (1.340)
Logistic 1.0369 0.71 < 0.97 — -1.0892 0.0776
< 1.23
(7.303) (-2.067) (0.895)

The Weibull estimates of s are higher than 1.00, indicating that p is

somewhat less than 1.00, and hazard rates tend to fall, even if slowly,
over the life of a dictatorship.

141
 Chapter Three

Political Regimes and Economic Growth

Introduction
With the birth of new nations in Asia and Africa, the fear that democ-
racy would undermine economic growth began to be voiced in the United
States. The first statements to that effect were perhaps those by Walter
Galenson and by Karl de Schweinitz, who argued, both in 1959, that
in poor countries democracy unleashes pressures for immediate con-
sumption, which occurs at the cost of investment, hence of growth.
Galenson mentioned both the role of unions and that of governments. He
thought that unions “must ordinarily appeal to the worker on an all-out
consumptionist platform. No matter how much ‘responsibility’ the union
leader exhibits in his understanding of the limited consumption pos-
sibilities existing at the outset of industrialization, he cannot afford to
moderate his demands.” As for governments, he observed that “the
more democratic a government is, . . . the greater the diversion of
resources from investment to consumption.” According to de Schweinitz
(1959: 388), if trade unions and labor parties “are successful in securing
a larger share of the national income and limiting the freedom for action
of entrepreneurs, they may have the effect of restricting investment
surplus so much that the rate of economic growth is inhibited.” That
argument enjoyed widespread acceptance under the influence of Hunt-
ington, who claimed that “the interest of the voters generally leads
parties to give the expansion of personal consumption a higher priority
via-à-vis investment than it would receive in a non-democratic system”
(Huntington and Domiguez 1975: 60; Huntington 1968).
Democracy was thus seen as inimical to economic development.
Moreover, via a rather dubious inference, proponents of that view con-
cluded that dictatorships were therefore better able to force savings
and launch economic growth. To cite a more recent statement: “Eco-

142
Introduction

nomic development is a process for which huge investments in per-

sonnel and material are required. Such investment programs imply
cuts in current consumption that would be painful at the low levels of
living that exist in almost all developing societies. Governments must
resort to strong measures and they enforce them with an iron hand in
order to marshal the surpluses needed for investment. If such mea-
sures were put to a popular vote, they would surely be defeated. No
political party can hope to win a democratic election on a platform of
current sacrifices for a bright future” (Rao 1984: 75).1
The reasoning bears reconstruction. First, that argument assumes
that poor people have a higher propensity to consume.2 This is why
democracy may be compatible with growth at high but not at low levels
of income. Second, the underlying model of growth attributes it to the
increase in the stock of physical capital. Finally, democracy is always
responsive to pressures for immediate consumption. The chain of rea-
soning is thus the following: (1) Poor people want to consume imme-
diately. (2) When workers are able to organize, they drive wages up,
reduce profits, and reduce investment (by lowering either the rate of
return or the volume of profit or both). (3) When people are allowed to
vote, governments tend to distribute income away from investment
(either they tax and transfer or they undertake less public investment).
(4) Lowering investment slows down growth. Note, as well, that this
reasoning implies that the impact of mean-preserving inequality on
growth is ambivalent: In the Kaldor-Pasinetti models, inequality pro-
motes growth, as it increases the incomes of those who save more, but
in the median-voter models it slows down growth to the extent to which
the political system responds to demands for redistribution.
Arguments in favor of democracy are not equally sharp, but they all
focus in one form or another on allocative efficiency: Democracies can
better allocate the available resources to productive uses. One view is
that because authoritarian rulers are not accountable to electorates,

1
At least Huntington and his collaborators wrote during a period when many dictator-
ships, “authoritarian” and “totalitarian,” did grow rapidly. But Rao’s assertion was made
in 1984, after the failure of several Latin American authoritarian regimes and Eastern
European communist regimes was already apparent.
2
Pasinetti (1961) claimed that the propensity to consume is higher for workers than for
capitalists, and Kaldor (1956) believed that it is higher for wages than for profits,
whereas the scholars discussed here seem to assume that in general the marginal
propensity to consume declines with income. Barro and Sala-i-Martin (1995: 77–9) show
that in the optimal growth model the savings rate decreases as a result of the substitu-
tion effect and increases in income as a consequence of the income effect, the net effect
being ambivalent.

143
 Political Regimes and Economic Growth

they have no incentive to maximize total output, but only their own
rents. As a result, democracies better protect property rights, thus
allowing a longer-term perspective to investors. There is also a vague
sense that by permitting a free flow of information, democracies
somehow improve the quality of economic decisions.
According to the first view, the state is always ready to prey on the
society (North 1990), and only democratic institutions can constrain it
to act in a more general interest. Hence, dictatorships, of any stripe,
are sources of inefficiency. Barro (1990), Findlay (1990), Olson (1991),
and Przeworski (1990) have constructed models that differ in detail but
generate the same conclusion. These models assume that some level
of government intervention in the economy is optimal for growth. Then
they all show that, depending on the details of each model, dictator-
ships of various stripes can be expected to undersupply or oversupply
government activities. One interesting variant of this approach is by
Robinson (1995), who thinks that dictators are afraid, at least under
some conditions, that development would give rise to political forces
that would overturn them, and thus they deliberately abstain from
developmentalist policies.
Perhaps the best-known informational argument is based on the
Drèze and Sen (1989) observation that no democracy ever experienced
a famine, which they attribute to the alarm role of the press and the
opposition. Thus, Sen (1994a: 34) observes that “a free press and an
active political opposition constitute the best ‘early warning system’ that
a country threatened by famine can possess.” He also cites an unlikely
source, Mao, reflecting on the great Chinese famine of 1962, to the effect
that “without democracy, you have no understanding of what is hap-
pening down below.” Yet it is not apparent whether this is an argument
strictly about avoiding disasters or about average performance.3
This summary makes no pretense to being exhaustive. All we want
to highlight is that the arguments in favor of dictatorship and those in
favor of democracy are not necessarily incompatible. The arguments
against democracy claim that it hinders growth by reducing invest-
ment; the arguments in its favor maintain that it fosters growth by
promoting allocative efficiency. Both may be true: The rate at which

3
Sah and Stiglitz (1988) compared the quality of the decisions whether or not to under-
take a series of economic projects made under different decision rules. Their conclu-
sions are ambivalent: Although majority rule is conducive to good decisions under many
conditions, decisions by smaller groups are better when the costs of information are
high, whereas decisions by larger groups are superior when the chances of adopting a
bad project are high.

144
Introduction

productive factors grow may be higher under dictatorship, but the use
of resources may be more efficient under democracy. And because
these mechanisms work in opposite directions, the net effect may be
that there is no difference between the two regimes in the average rates
of growth they generate. The patterns of growth may differ, but the
average rates of growth may still be the same.
Rates of growth may thus differ between regimes either because the
productive inputs increase at different rates or because they are used
with different efficiencies. To assess the impact of regimes on growth,
we proceed by the following steps:
First, we examine whether or not it is valid to assume that the pro-
ductive inputs grow at the same rates under the two regimes. This is
a test of the validity of the hypothesis that regimes do not affect the
rates of growth of productive resources.
Second, we estimate separately for each regime a production-
function model, Y = AF(K, L) = AKaLb, in growth form (suppressing the
i, t subscripts):4

Y˙ Y = A˙ A + a K˙ K + b L˙ L + Ql + e,

where
. l is the instrument for the regime variable. The coefficients5
A/A, a, and b provide selection-unbiased estimates of, respectively,
technical progress and the efficiency with which capital stock and labor
force contribute to growth under each regime. This is then a test of the
hypothesis that the efficiency with which resources are used is the
same under the two regimes.
To calculate the effects of regimes independent of the exogenous
conditions under which they were observed, we need to generate the
growth rates that would be expected had each country been observed
each year under each regime. If the hypothesis that the growth of pro-
ductive inputs is exogenous with regard to regimes is not rejected, the
expected values of the growth rates under the two regimes are obtained
by multiplying the coefficients characterizing each regime by the rates
of growth of the productive inputs observed for each country during
each year. If the growth of productive inputs is endogenous with regard
to regimes, the expected growth rates are obtained by multiplying the
coefficients characterizing each regime by the average rate of growth

4
A notational explanation for non-economists:
. The
. dots over the variables indicate time
5
. that for any X, X = dX/dt, and X/X is the rate of growth.
derivatives, so
We refer to A /A as a coefficient because we treat it as a constant, either for regimes or
for countries or for years, depending on estimation methods.

145
 Political Regimes and Economic Growth

of productive inputs observed under each regime. In either case this

procedure generates predicted values of growth for each country
during each year under each regime. These values are then averaged
over all observations, and the regime averages are compared. The
regime effect is the difference between these averages.
Finally, we study whether or not the production function and the
rate of growth of productive inputs are stable for each of the regimes.
Because the two regimes are observed at different income levels, and
because both the production functions and the growth of inputs may
depend on the level of development, we need to test whether the
observed differences are indeed due to regimes or merely to wealth.

Political Regimes and Economic Growth

.
The observed rate of growth of total income (YG = Y/Y ) is higher
under dictatorships, 4.42, than under democracies, 3.95, implying that
income doubles in 15.8 years under dictatorship and in 17.7 years
under democracy. Yet we already know that one should not draw infer-
ences from the observed values.
To identify the effect of regimes, we need to distinguish the effects
of the conditions under which these regimes were observed from the
effect of regimes. We present first the results concerning the effect of
regimes on the growth of productive inputs, then those concerning the
efficiency with which these inputs are used under each regime, and,
finally, the conclusions about the overall effect of regimes on the growth
of total income.
Do productive inputs grow at the same rate under the two regimes?
The claim that democracy undermines investment, whether in general
or only in poor countries, finds no support in the evidence. The
observed average share of investment in gross domestic product (GDP),
INV = I/Y, was in fact much higher in democracies, 20.90 percent,
than in dictatorships, 14.25 percent. But because investment shares
increase with per capita income, and because, as we already know,
dictatorships have generally existed in poorer countries, this could be
just an effect of income. Indeed, controlling for income, as well as for
a number of other variables in a selection model, shows that regimes
have no overall effect on investment.6 Table 3.1 presents the results of

6
The variables in the investment equation include lagged investment share (given that
investment decisions made one year take time to realize, INVLAG), lagged per capita
income (an instrument for expected domestic demand, LEVLAG), the average rate of

146
Political Regimes and Economic Growth

Table 3.1A. Investment share in GDP (INV), by per capita income

(LEVEL)

Investment share
Proportion
LEVEL dictatorships All Dictatorships Democracies

0–1,000 0.9273 8.89 8.79 10.14

1,001–2,000 0.7472 13.53 13.66 13.17
2,001–3,000 0.6207 17.87 17.83 17.93
3,001–4,000 0.5874 20.04 20.47 19.42
4,001–5,000 0.5424 22.76 23.75 21.57
5,001–6,000 0.4308 24.35 25.48 23.50
6,001– 0.0812 25.40 29.92 25.00
Total (N = 4,126) 16.90 14.25 20.90

Note: All cell entries are based on at least 68 observations.

statistical analyses that used different estimation methods.7 Regardless

of the method, the regime difference between the selection-corrected
expected investment shares is never larger than 1 percent. Moreover,
among poorer countries, those with per capita incomes lower than
$3,000, both the observed and the corrected investment shares are
slightly higher in democracies. Hence, there is no reason to think that
regimes affect investment even in poor countries: Had these regimes

growth in the world (a measure of world demand, WORLD), and the relative price of
investment goods (PINV). Alternatively, we estimate a 2SLS model with the predicted
growth rate as an instrument for the current demand. For a review of econometric
models of investment, see Rama (1993).
7
Because we present several tables with the same structure, here is a short guide to
reading them, first by row, then by column. “Biased” stands for the observed values or
ordinary least squares (OLS) regressions in which regime (REG) is a dummy variable.
Then follow various selection-corrected average predicted values of the dependent vari-
able under the two regimes. The correction for selection is always based on the dynamic
probit described in Appendix 2.1. The performance models depend on the structure
of the pooled time series, specifically autocorrelation, homogeneity, and simultaneity.
Finally, the last two rows refer to an alternative manner of estimating selection models,
due to Heckman (1988): “Unobservable” assumes that selection entails unobserved vari-
ables, and “Observable” that it entails only observed ones. The columns “Regime means”
give the average values of the dependent variable under the two regimes. “Constant
regime effect” assumes that the regime effect is an intercept dummy, and “Individual
regime effect” assumes that it is a sum of differences in slopes multiplied by the values
of the independent variables. For details, see Appendix I.

147
 Political Regimes and Economic Growth

Table 3.1B. Selection-corrected estimators of investment share (INV)

Regime means Regime effect

Estimator Dictatorships Democracies Constant Individual

(selection/INV) (N = 2,396) (N = 1,595) (N = 3,991)

Biased 14.28 20.98 0.0860 0.0362

(9.40) (8.02) (0.1041) (0.3340)
MLE 17.19 16.97
(9.26) (9.09)
1F-HATANAKA 21.61 21.25
(6.97) (6.50)
1F-AR1 16.69 17.73
(7.60) (7.69)
2SLS-1F-AR1 16.84 17.69
(7.85) (7.81)
Unobservable 0.0806 0.0348
(0.1092) (0.3437)
Observable 0.0797 0.0042
(0.1019) (0.3118)

By per capita income (LEVEL)

LEVEL < 3,000 LEVEL ≥ $3,000

Dictatorships Democracies Dictatorships Democracies

(N = 1,875) (N = 469) (N = 478) (N = 1,104)

Biased 11.95 14.38 23.41 23.87

(7.80) (6.08) (9.11) (6.99)
1F-AR1 12.17 13.23 23.68 23.61
(2.14) (1.70) (1.95) (0.86)
1F-HATANAKA 14.90 12.91 24.07 24.15
(4.76) (5.30) (4.43) (5.54)

Notes: Probit equations include lagged values of LEVEL, STRA, and RELDIF. The INV equations
include INVLAG, LEVEL, PINV, and WORLD. The 2SLS estimation uses fitted values of YG as
instrument. Standard errors in parentheses.

148
Political Regimes and Economic Growth

Table 3.2A. Rate of growth of capital stock (KSG), by per capita income
(LEVEL)

KSG
Proportion
LEVEL dictatorships All Dictatorships Democracies

0–1,000 0.9273 5.624 5.440 7.947

1,001–2,000 0.7472 7.627 8.112 6.189
2,001–3,000 0.6207 8.925 9.860 7.390
3,001–4,000 0.5874 7.494 7.621 7.314
4,001–5,000 0.5424 8.074 6.806 9.576
5,001–6,000 0.4308 8.038 9.297 7.085
6,001– 0.0812 6.080 10.650 5.676
Total (N = 4,126) 7.007 7.310 6.547

Note: All cell entries are based on at least 68 observations.

existed under the same conditions, specifically in countries with the

same per capita incomes, facing the same relative prices of investment
goods and world demand, and had they been matched for unobserv-
able factors that affect regime selection, their shares of investment in
GDP would have been the same.
Hence, even if pressures for immediate consumption are higher
in poor countries, democracies do not transform them into lower
rates of investment than do dictatorships. Poor countries simply invest
little regardless of their regime: Cape Verde between 1977 and 1983,
Zambia during 1966–68 and 1970–75, and Romania between 1967 and
1981 are the only countries with per capita incomes under $1,000
that maintained investment rates above 30 percent, whereas many
wealthier countries invested at such rates. The share of investment
is the same under the two regimes in wealthier countries, those with
incomes over $3,000.
Another way to think about investment is in terms of the growth . of
capital stock.8 The observed rate of growth of capital stock (KSG = K/K)
is somewhat higher under dictatorships, 7.31, than under democra-
cies, 6.56 (Table 3.2). The observed difference for countries with

. the investment share is INV = I/Y, and the rate of growth of capital stock is
8
Note that
KSG = K /K = (I - dK)/K. The depreciation rate implied by the Penn World Tables (PWT)
data is quite high.

149
 Political Regimes and Economic Growth

Table 3.2B. Selection-corrected estimators of the rate of growth of capital

stock (KSG)

Regime means Regime effect

Estimator Dictatorships Democracies Constant Individual

(selection/INV) (N = 2,396) (N = 1,595) (N = 3,991)

Biased 7.24 6.49 0.5420 -0.3571

(14.00) (10.78) (0.5094) (1.4630)
MLE 7.50 6.99
(2.57) (2.26)
2SLS-MLE 7.01 7.06
(9.99) (9.57)
Unobservable 0.5856 -0.3601
(0.5350) (1.4642)
Observable 0.6202 0.0056
(0.5127) (5.6502)

By per capita income (LEVEL)

LEVEL < 3,000 LEVEL ≥ 3,000

Dictatorships Democracies Dictatorships Democracies

(N = 1,917) (N = 487) (N = 479) (N = 1,108)

Biased 7.03 6.71 8.05 6.39

(14.47) (13.29) (11.91) (8.29)
2SLS-MLE 6.88 7.09 7.94 6.94
(11.46) (12.25) (6.85) (6.61)

Notes: Probit equations include lagged values of LEVEL, STRA, and RELDIF. The KSG equations
include LEVLAG, WORLD, PINV, PRIME, and COMEX. The 2SLS estimation adds YG, with
YGLAG as intrument. In 1F PANEL: LEVLAG, WORLD, PINV, and PRIME. Standard errors in
parentheses.

incomes below $3,000 is negligible. Above $3,000, capital stock grew

faster under dictatorship, at the rate of 8.05, than under democracy,
where it increased at the rate of 6.39. Correcting for selection, with the
same exogenous variables we used for the investment share, shows
again, however, that regimes do not affect the growth of capital stock
in the sample as a whole and not among countries with incomes under

150
Political Regimes and Economic Growth

Table 3.3A. Rate of growth of labor force (LFG); by per capita income
(LEVEL)

Rate of labor-force growth

Proportion
LEVEL dictatorships All Dictatorships Democracies

0–1,000 0.9273 2.314 2.338 2.019

1,001–2,000 0.7472 2.320 2.312 2.342
2,001–3,000 0.6207 2.295 2.365 2.181
3,001–4,000 0.5874 2.013 2.254 1.670
4,001–5,000 0.5424 1.561 1.979 1.066
5,001–6,000 0.4308 1.205 1.711 0.822
6,001– 0.0812 1.395 1.711 1.367
Total (N = 4,128) 2.006 2.270 1.608

Note: All cell entries are based on at least 68 observations.

$3,000. Above $3,000, the selection-corrected values are somewhat

higher under dictatorship, 7.94, as opposed to 6.94 under democracy.
Hence, different regimes affect neither the share of investment in
GDP nor the rate of growth of capital stock. Moreover, contrary to all
the arguments, poor countries invest little regardless of the regime, and
it is in wealthier countries that investment is slightly higher under
dictatorships. .
In turn, the rate of growth of the labor force (LFG = L /L) is higher
under dictatorships (Table 3.3). The observed values are 2.27 percent
per annum under dictatorships and 1.61 under democracies. Note
again the observed patterns. The rates of growth of the labor force are
about the same under the two regimes in countries with per capita
incomes of less than $3,000. But in countries with higher incomes, the
difference between them becomes dramatic. Indeed, the labor force
grows faster in dictatorships with incomes above $6,000 than in
democracies with incomes between $3,000 and $4,000.
Because the labor force grows at a slower rate in wealthier coun-
tries, one might suspect again that this difference is due to the distri-
bution of regimes by income. Some of it is, but not enough to eradicate
the effect of regimes. Even when the regimes are matched for their
income, their colonial heritage, and the frequencies of Catholics,

151
Table 3.3B. Selection-corrected estimators of labor force growth (LFG)

Regime means Regime effect

Estimator Dictatorships Democracies Constant Individual

(selection/LFG) (N = 2,396) (N = 1,595) (N = 3,991)

Biased 2.28 1.59 0.2565 0.0820

(1.94) (1.20) (0.0695) (0.3354)
2SLS 2.17 1.72
(0.37) (0.28)
AR1 2.17 1.72
(0.37) (0.28)
HATANAKA 2.16 1.75
(0.53) (0.68)
2F PANEL 2.00 1.81
(0.79) (0.53)
Unobservable 0.2795 0.0870
(0.0728) (0.3353)
Observable 0.2863 0.0893
(0.0702) (0.3462)

LFPW sample N = 2,076 N = 1,254

Biased 2.39 1.65

(1.93) (1.19)
OLS 2.27 1.81
(0.41) (0.66)

By per capita income (LEVEL)

LEVEL < 3,000 LEVEL ≥ 3,000

Dictatorships Democracies Dictatorships Democracies

Estimator (N = 1,917) (N = 487) (N = 479) (N = 1,108)

Biased 2.35 2.22 2.03 1.32

(2.03) (1.02) (1.54) (1.17)
OLS-AR1 2.33 2.35 1.92 1.26
(0.27) (0.44) (0.50) (0.44)

Notes: Probit equations include lagged values of LEVEL, STRA, and RELDIF. The LFG equations
include LEVEL, NEWC, BRITCOL, CATH, PROT, and MOSLEM, plus LFPW in the LFPW sample.
The 2SLS estimation is based on the same variables plus YG, with YGLAG as instrument.
Standard errors in parentheses.

152
Political Regimes and Economic Growth

Protestants, and Moslems, even if they are matched for demand (2SLS),
or for the lagged rate of population growth (HATANAKA estimator), or
for country-specific effects (PANEL estimator), as well as for the factors
affecting selection, and, in a smaller sample, for the labor-force par-
ticipation of women (LFPW), the labor force grows faster under dicta-
torships. The selection-corrected difference is small in countries with
incomes under $3,000, but quite large in those over $3,000.9
Hence these findings generate a surprise. Contrary to all the argu-
ments, the kind of regime does not affect the rate of investment and
the growth of capital stock. And for reasons to be investigated later, it
appears that the regime type does affect the growth of the labor force.
To examine the effect of regimes on the efficiency with which
resources are used in production, we need to compare the coefficients
of the respective production functions. The constant measures total
factor productivity, and the coefficients on capital and labor represent
the elasticities of output with regard to these factors.10
As shown in the first two columns of Table 3.4, which provide the
results for the entire sample, total factor productivity is somewhat
higher in democracies. The elasticity with regard to capital is slightly
higher in dictatorships, but elasticity with regard to labor is higher in
democracies. Hence, one is led to conclude that democracies benefit
more from technical progress and use labor more effectively, but
dictatorships more efficiently employ the physical capital stock. Yet
because there are reasons to suspect that production functions depend
on the composition of each regime sample, we shall return to this topic
later.
What, then, is the overall effect of regimes on growth? Assuming that
the two regimes exist under the same conditions reduces the difference
between the expected growth rates almost to zero. Whether one esti-
mates the model by unconstrained OLS or constrains the coefficients
of the production function11 or uses the appropriate panel methods or

9
The results for the split sample should be treated with some skepticism. A labor-force
growth series is hard to estimate: It has some simultaneity, it is autocorrelated, and
it probably has a cyclical component; given that we need to use country-invariant
variables, we cannot use fixed effects, and the random-effects model does not fit well.
To see if it makes a difference how the model is estimated, we provide in Table 3.3B
several alternative estimators. Fortunately, they all lead to the same conclusion. When
the sample is split, however, statistical problems become overwhelming. The best
estimator seems to be OLS-AR1, which is what we report.
10
For comments on estimation methods, see Appendix 3.1.
11
When selection is estimated by the dynamic probit, the OLS results, constrained or not,
are almost identical with the maximum-likelihood estimates.

153
Table 3.4. Barebones model, by low and high per capita incomes, decomposed by
sources of growth, with LFG taken as exogenous

$3,000 £ Income
Entire Sample Income £ $3,000 < $8,000 Income > $3,000

Estimator Dic Dem Dic Dem Dic Dem Dic Dem

N 2,396 1,595 1,917 487 594 462 479 1,108

Y/P 1,909 6,210 1,227 1,761 4,469 5,490 4,642 8,165
Biased 4.43 3.92 4.30 4.11 4.83 4.30 4.91 3.83
(6.94)a (4.82) (6.97) (6.20) (6.75) (4.51) (6.81) (4.07)

Best estimate:
Method OLS 2F OLS OLS OLS 2F OLS 2F

Constant 0.8966 1.0430 0.8680 0.9226 0.9353 1.7785 0.9230 1.3439

(0.1433) (0.2078) (0.1551) (0.3957) (0.3902) (0.3058) (0.3733) (0.2058)
a 0.4004 0.3447 0.3937 0.3929 0.4409 0.2643 0.4383 0.2751
(0.0060) (0.0086) (0.0064) (0.0470) (0.0178) (0.0160) (0.0166) (0.0119)
b 0.3138 0.3974 0.3173 0.2524 0.2926 0.3838 0.2960 0.5530
(0.0435) (0.1132) (0.0460) (0.1532) (0.1332) (0.1891) (0.1295) (0.1286)
Q -1.1099 -0.0180 -0.1804 -0.0193 -1.2962 -0.0353 -1.1945 0.0719
(0.3355) (0.2461) (0.3941) (0.2748) (0.7035) (0.4791) (0.6903) (0.4288)
YG-HAT 4.30 4.23 4.35 4.23 4.81 4.44 4.40 4.09
(5.03) (4.34) (5.57) (5.56) (4.61) (2.80) (4.21) (2.73)
Capital 2.7776 2.3914 2.7430 2.7376 3.3985 2.0373 3.0201 1.8958
Labor 0.6303 0.7983 0.7370 0.5737 0.4784 0.6274 0.4738 0.8479
TFP 0.8966 1.0430 0.8680 0.9226 0.9353 1.7785 0.9230 1.3439

Constrained OLS (a + b = 1)
F for constraint 41.26 17.19 37.56 5.38 3.90 5.49 4.13 10.06
Probability (0.0000) (0.0001) (0.0000) (0.0207) (0.0489) (0.0185) (0.0427) (0.0017)

Constant 0.1939 0.5796 0.1447 0.1036 0.3323 1.0229 0.3333 0.9053

(0.0934) (0.0934) (0.1016) (0.1798) (0.2438) (0.1713) (0.2355) (0.1110)
a 0.4094 0.3619 0.4033 0.3984 0.4466 0.3179 0.4430 0.3199
(0.0059) (0.0079) (0.0063) (0.0115) (0.0176) (0.0143) (0.0165) (0.0111)
b 0.5906 0.6381 0.5967 0.6016 0.5534 0.6821 0.5570 0.6801
(0.0059) (0.0079) (0.0063) (0.0115) (0.0176) (0.0143) (0.0165) (0.0111)
Q -1.0397 0.2609 -1.1933 -0.0029 -1.0751 0.4813 -0.9831 0.3965
(0.3381) (0.2133) (0.3978) (0.2870) (0.6967) (0.4106) (0.6847) (0.3665)
YGHAT 4.22 4.37 4.34 4.28 4.68 4.59 4.24 4.15
(5.17) (4.60) (5.73) (5.66) (4.71) (3.43) (4.30) (3.19)
Capital 2.8398 2.5104 2.8100 2.7762 3.4424 2.4507 3.0527 2.2045
Labor 1.1864 1.2818 1.3859 1.3972 0.9047 1.1150 0.8540 1.0427
TFP 0.1939 0.5796 0.1447 0.1036 0.3323 1.0229 0.3333 0.9053

a
Standard errors in parentheses.
 Political Regimes and Economic Growth

applies the Heckman (1988) instrumental-variable approach, the differ-

ences between selection-unbiased expected rates of growth are minimal.
Introducing human capital, specifically, the rate of growth in the years
. education attained by the average member of the labor force (EDTG =
of
H /H), reduces the size of the sample, but the results are again the same.
Because the labor force grows faster under dictatorship even when
the two regimes are matched for exogenous conditions, lines with “LFG
endogenous” in Table 3.5 take the rate of growth of the labor force at
the mean values observed for each regime. These results are some-
what more favorable to dictatorships. But the difference between
regimes is still minuscule. Hence, there is no reason to think that the
regime type affects the rate of growth of total income.
Another way to test the effect of regimes is to focus on the countries
that experienced regime changes. Here again, however, one should
proceed prudently. Countries in which regimes are unstable may be
different from those that have had single regimes persisting during the
entire period. Yet the observed average rate of growth was the same in
those countries that did not experience any regime transitions and in
those that underwent one or more regime changes: The rate of growth
for the former was 4.23 percent (N = 2,813), and for the latter 4.25 (N =
1,313). Stable dictatorships grew at the rate of 4.38 percent (N = 1,709),
whereas dictatorships in the countries that also experienced democracy
grew at the rate of 4.51 (N = 772). Stable democracies grew at the rate
of 3.98 percent (N = 1,104), and democracies that rose from or gave way
to dictatorships grew at 3.88 percent (N = 541). Hence, there is no reason
to think that growth in the countries where regimes were stable was
different from that in countries where regimes changed.12
With this reassurance, we can compare the rates of growth of democ-
racies preceding dictatorships with those of democracies following
dictatorchips, and vice versa. The average rate of growth during all
the years of democracies preceding dictatorships was 4.49 percent
(N = 290), and for dictatorships following democracies, 4.37 (N = 425).
Hence, transitions from democracy to dictatorship did not affect the rate
of growth. Growth during all the years of dictatorships preceding
democracies was higher, at the average of 4.74 percent (N = 607), than
the average of 3.64 percent (N = 371) during all the democratic years
that followed. But because recovery from the crises accompanying
transitions to democracy is slow, and the observations are right-hand

12
For a more extensive discussion of the effect of regime stability on growth, see
Chapter 5.

156
 Political Regimes and Economic Growth

Table 3.5. Selection-corrected estimators of the rate of growth of income

(YG)

Regime means Regime effect

Dictatorships Democracies Constant Individual

Estimator (N = 2,396) (N = 1,595) (N = 3,991)

Biased 4.43 3.92 -0.0227 0.0006

(6.94) (4.82) (0.1252) (0.4908)
Unconstrained OLS 4.30 4.24
(5.02) (4.48)
Constrained OLS 4.22 4.37
(LFG exogenous) (5.17) (4.60)
Constrained OLS 4.38 4.11
(LFG endogenous) (5.15) (4.55)
PANEL 4.22 4.23
(LFG exogenous) (5.17) (4.34)
PANEL 4.38 4.07
(LFG endogenous) (5.15) (4.34)
Unobservable 0.0741 0.0461
(0.1312) (0.4871)
Observable -0.0926 0.0490
(0.1537) (0.5061)

EDTG sample N = 1,745 N = 1,042

Biased 4.57 3.91

(6.91) (4.71)
Unconstrained OLS 4.45 4.34
(LFG exogenous) (4.71) (3.88)
Constrained OLS 4.36 4.94
(LFG exogenous) (4.87) (4.01)
2F PANEL 4.45 4.33
(LFG exogenous) (4.71) (3.74)

Notes: Probit equations include lagged values of LEVEL, STRA, and RELDIF. The “barebones”
model includes KSG, LFG, and, in a smaller sample, EDTG. “Constrained” means that the coef-
ficients on the arguments of the production function were constrained to 1.00. PANEL is OLS for
dictatorships and 2F for democracies. “LFG exogenous” is based on the assumption that LFG is
exogenous; “LFG endogenous” takes LFG at the observed mean of each regime.

157
 Political Regimes and Economic Growth

censored, the conclusion that transitions to democracy slow down

growth would be erroneous. Just note that many of the democratic
observations followed transitions from either bureaucratic-authoritar-
ian or communist dictatorships during the 1980s, including the very end
of the decade. Because those democracies did not have time to recover
by 1990, they weigh down the democratic average.
In sum, neither the selection-corrected values for the entire sample
nor the paths of growth associated with regime transitions give any
support to the claim that regime types affect the rate of growth of total
income. Selection-corrected average rates of growth are the same for
the two regimes. And there is no reason to think that steady-state rates
of growth would be different under the two regimes when countries
experience regime transitions.
The first conclusion, therefore, must be that political regimes have
no impact on the rate of growth of total income. The arguments about
the superiority of dictatorships in mobilizing savings for investment
find no support in the evidence. Indeed, the input that grows faster
under dictatorships is not capital but labor. The differences in the
efficiency with which productive inputs are utilized are small. And,
as a consequence, the selection-corrected average expected values of
growth are almost identical. Yet conclusions are premature, for there
are still methodological problems to confront.

Poor and Wealthy Countries

We still face the following difficulty: Dictatorships existed predomi-
nantly in poor countries: 38.5 percent of annual observations (946 out
of 2,481) of dictatorships were in countries with incomes under $1,000,
but only 4.5 percent of democracies (75 out of 1,645) were that poor.
Democracies flourished in wealthy countries: 46.8 percent of them
(769) were observed in countries with incomes above $6,000, whereas
only 2.8 percent (68) of dictatorships existed at such income levels.
Hence, nearly all our observations of countries with incomes below
$1,000, 92.6 percent (946 out of 1,021), are of dictatorships, and nearly
all our observations of countries with incomes above $6,000, 91.9
percent (769 out of 837), are of democracies.
Now, examining the rates of growth in countries classified by inter-
vals of $500 of per capita income (Table 3.6) shows that very poor
countries (under $1,000) grow slowly, at about 3.5 percent. Growth
accelerates in wealthier economies, reaching a peak of 5.1 percent
between $2,000 and $3,000. Then it slows down again to about 3.8

158
Table 3.6. Rate of growth of GDP (YG), by per capita income (LEVEL)

Rate of growth of GDPa

Proportion
LEVEL dictatorships All Dictatorships Democracies

0–1,000 0.9273 3.519 3.464 4.220

1,001–2,000 0.7472 4.636 4.809 4.123
2,001–3,000 0.6207 5.142 5.633 4.335
3,001–4,000 0.5874 4.740 4.915 4.492
4,001–5,000 0.5424 4.552 4.507 4.606
5,001–6,000 0.4308 4.312 4.772 3.963
6,001– 0.0812 3.770 6.054 3.568
Total (N = 4,128) 4.233 4.424 3.945

a
All cell entries are based on at least 68 observations.

Moving averages of rates of growth by bands of $500

LEVEL All Dictatorships Democracies

250–750 3.071 3.107 2.380

500–1,000 3.689 3.647 4.164
750–1,250 4.140 4.050 4.724
1,000–1,500 4.505 4.682 3.848
1,250–1,750 4.969 5.381 3.932
1,500–2,000 4.827 5.021 4.396
1,750–2,250 4.972 5.092 4.744
2,000–2,500 5.444 5.664 5.055
2,250–2,750 5.793 6.989 3.993
2,500–3,000 4.827 5.599 3.653
2,750–3,250 4.808 4.955 4.613
3,000–3,500 5.130 5.238 4.977
3,250–3,750 4.594 4.183 5.445
3,500–4,000 4.317 4.565 3.965
3,750–4,250 4.382 5.065 3.491
4,000–4,500 4.984 4.908 5.086
4,250–4,750 4.742 4.771 4.714
4,500–5,000 3.965 3.881 4.050
4,750–5,250 4.558 4.580 4.535
5,000–5,500 4.217 4.360 4.088
5,250–5,750 4.116 4.003 4.194
5,500–6,000 4.418 5.350 3.845
5,750–6,250 3.363 4.479 2.878
6,000– 3.770 6.054 3.568

159
 Political Regimes and Economic Growth

percent when countries reach incomes above $6,000. Hence, in

accordance with Quah (1996), incomes diverge among poor countries,
until about $2,500, and they converge among wealthy countries.
If very poor and very rich economies both grow slowly regardless of
the regime they have, then this pattern does not present a problem. But
if poor countries grow slowly because they are ruled by dictatorships, or
rich ones because they are democratic, then we cannot make such an
inference, for perhaps if the poor countries had been democratic they
would have grown faster. In fact, the 75 democratic years at incomes
under $1,000 witnessed growth at the rate of 4.22 percent, but dicta-
torships grew at the rate of 3.46 percent in equally poor countries. Con-
versely, if the rich countries had been authoritarian, perhaps they would
have grown faster. Again, the 68 authoritarian years at incomes above
$6,000 enjoyed growth at the rate of 6.05 percent, whereas democracies
had a rate of growth of 3.57 percent at those incomes. Our counterfac-
tual procedure matches the regimes for the conditions under which they
existed, specifically for their productive inputs and a variety of other con-
ditions. But to find out how a country observed, say, as a dictatorship
would have grown had it been a democracy under the same conditions,
we use the information about the way these productive inputs are
transformed into outputs under each regime. And this information,
about production functions, is derived from the actual observations,
which means disproportionately from poor dictatorships and rich
democracies. Hence these production functions may be different not
because of the impact of regimes but because of the effect of wealth.
Thus, to test whether or not the results depend on the samples, we
need to estimate production functions separately for different levels of
development, as always measured by per capita income (Table 3.4).
First we consider only countries with incomes under $3,000, which we
shall call “poor.” Their production functions are almost identical, and
regimes make no difference for the average growth rates. Then we take
countries with incomes above $3,000, “wealthy,” where the difference
between the observed growth rates is particularly high, 4.91 percent
for dictatorships, and 3.83 for democracies. The difference between
the average values almost vanishes when corrected for selection, but
the production functions are quite different. Finally, given that there
are very few dictatorships with incomes above $8,000,13 we need to

13
The wealthiest dictatorship we observed was Singapore, with an income of $11,698,
and the wealthiest democracy in our sample, the United States, had an income of
$18,095. There were 200 democratic years with incomes above that of Singapore.

160
Poor and Wealthy Countries

know if the difference between wealthy dictatorships and wealthy

democracies is still due to the composition of the respective samples,
so we analyze separately countries within the $3,000–$8,000 income
band.
These tests suggest that per capita income of $3,000 is the natural
breaking point.14 The production functions are almost identical in
countries with incomes below $3,000, but they differ between regimes
in wealthier countries. In particular, the difference between the
two regimes becomes visible if we consider only countries within the
$3,000–$8,000 income band. Hence, this difference is not due to
diminishing returns in wealthy democracies.

Poor Countries
In poor countries, the two regimes are almost identical, with
observed rates of growth of 4.34 percent under dictatorship and 4.28
under democracy. The two regimes generate productive inputs at the
same rate and use them in identical ways. They invest about 12.5
percent of GDP and increase capital stock at the rate of about 6 percent,
and labor force at the rate of about 2.2 percent. An increase of 1
percent in the capital stock raises output by about 0.40 percent under
both regimes, and an increase in the labor force by 1 percent augments
output by about 0.60 percent.15 Neither regime benefits much from
technical progress, about 0.1 percent per annum; both get 2.8 percent
in growth from an increase in capital stock, and 1.4 percent from an
increase in labor force. With identical supplies of factors and their iden-
tical utilization, they grow at the same rate under the two regimes: The
selection-corrected average growth rates are the same.
The idea that democracies in poor countries process pressures for
immediate consumption, resulting in lower investment and slower
growth, seemed persuasive at the time it was advanced, and it was not
implausible. There appear to be good reasons to think that people in
poor countries want to consume more immediately: They cannot afford
to make intertemporal trade-offs if they cannot expect to live to benefit
from their short-term sacrifices. It is also plausible that unions, par-
ticularly if they are decentralized, and political parties, competing for

14
We have investigated several more income bands, beginning with $0–$3,000 and
moving the lower and upper cutoffs by $1,000, until $8,000–$11,000.
15
This is a constrained estimate. Constrained estimates are cited in the rest of this
paragraph.

161
 Political Regimes and Economic Growth

votes, would push forward demands for immediate consumption. Yet,

as likely as that view may seem, it simply is not true. Perhaps this only
means that democracy is not very effective at processing what people
want; perhaps developmental goals are not any more attractive to
people under dictatorship than under democracy; perhaps poverty
is so constraining that even dictators cannot squeeze savings out of
indigent people.
The last explanation is most plausible. One piece of evidence is that
very few countries that were very poor when we first observed them
ever developed. Of the forty-eight countries that entered our purview
with incomes below $1,000, only three made it to above $3,000 by
1990. The two miracles were Taiwan, which had an income of $968
in 1950 and $8,067 in 1990, and South Korea, which went from $814
in 1950 to $6,665 in 1990. Thailand had an income of $815 in 1950
and $3,570 in 1990. Four more countries that began under $1,000
made it to more than $2,000, and eleven more to at least $1,000. But
at the end of the period, thirty – out of forty-eight – very poor coun-
tries remained within the income band in which they had begun. The
experiences of countries that were first observed with incomes between
$1,000 and $2,000 were more heterogeneous, but, again, of the forty-
five first observed at that level, only five experienced sustained growth:
Japan, which went from $1,768 in 1950 to $14,317 in 1990, Singa-
pore from $1,845 in 1965 to 11,698 in 1990, Portugal from $1,314 in
1950 to 7,487 in 1990, Greece from $1,480 in 1950 to $6,768 in 1990,
and Malta from $1,377 in 1964 to $6,627 in 1990. Four countries
descended to below $1,000, and twelve still had incomes between
$1,000 and $2,000 in 1990. In turn, none of the forty-two countries
that were first observed with per capita incomes above $2,000 fell
below their starting range, and all but seven of them at least doubled
their incomes by the end of the period. Because the observation periods
were not the same for all countries and typically were shorter for the
poorer ones, many of which became independent around 1960, these
data are somewhat biased against poor countries. Nevertheless, most
countries that we first observed below $2,000 had about thirty years
to grow, and yet most remained poor: evidence of a “low-level trap”
(Table 3.7).
Thus, poverty constrains. Whatever the regime, the society is too
poor to finance an effective state. Collecting total revenues of $127 per
capita, as governments do on the average in countries with incomes
under $1,000, can pay for little else than collecting these revenues.
Government expenditures add up to $167 per person in these

162
 Poor and Wealthy Countries

Table 3.7. Per capita income at the beginning and end of the period, by
bands of $1,000

Exited

Entered 0–1 1–2 2–3 3–4 4–5 5–6 6–7 7–8 8–9 9–10 10– Total

0–1 30 11 4 1 0 0 1 0 1 0 0 48
1–2 4 12 15 6 2 1 2 1 0 0 2 45
2–3 0 0 1 3 3 1 0 1 0 3 1 13
3–4 0 0 0 2 0 2 0 1 0 0 4 9
4–5 0 0 0 0 2 0 1 0 0 0 5 8
5– 0 0 0 0 0 0 1 1 0 0 10 12
Total 34 23 20 12 7 4 5 4 1 3 22 135

Notes: “Entered” stands for 1950 or the year of independence or the first year data were available, and “Exited”
for 1990 or in some cases the last year data were available. Per capita incomes are given in thousands (1985
PPP USD). Cell entries are numbers of countries.

countries, so they run deficits higher than 7 percent of GDP. In coun-

tries with incomes between $1,000 and $2,000, governments collect
$372 per capita and spend $450, still running deficits over 7 percent.
And, like the investment share, government revenue (particularly tax
revenue) as a share of GDP increases monotonically in per capita
income.16 Thus already in countries with incomes between $2,000 and
$3,000, revenues of the central government add to $668 per capita.
Between $3,000 and $4,000 they are $904, and above $6,000 they are
$2,608. To put it differently, per capita public expenditures in coun-
tries with incomes between $3,000 and $4,000 are larger than total
per capita incomes in countries with incomes under $1,000; per capita
public expenditures in countries with incomes above $6,000 are about
the same as total per capita incomes in countries between $3,000 and
$4,000.
Poor countries cannot afford a strong state, and when the state is
weak, the kind of regime matters little for everyday life. In a village
located three days’ travel away from the capital, often the only presence
of the state is a teacher and occasionally roving uniformed bandits.17 Just
calculate: If a mile of road costs about a million dollars, in a country with

16
Cheibub (1998) shows that selection-corrected tax revenues are the same for the two
regimes.
17
The best portrayal of life under a weak state is by Alvaro Mutis (1996).

163
 Political Regimes and Economic Growth

per capita income under $1,000 it would take the total government
expenditures per 600,000 persons to build 100 miles of road. There is
little room for regimes to make a difference when the state is that poor.
Note that we are not arguing that a fiscally large or otherwise large
state is necessarily good for development, but only that if the state is
to be able to make a difference for better conditions, it must have
resources. The role of the state in economic development is a notori-
ously controversial issue. Most of the statistical research on this topic
has been mindless: Studies that discover that the state is bad for growth
simply stick government-consumption expenditures into the equation
for growth and discover that its sign is negative. Needless to say, the
same would happen if one did that with private-consumption expendi-
tures: We did it and know it to be so. If one thinks that government-
consumption expenditures affect growth, the term introduced into the
production-function equation should be the change in government
consumption, not the share of government consumption in GDP.18 But
the real test of the impact of government is to consider separately the
effects of private and public capital stocks. The idea, due to Barro
(1990), is that private capital and public capital play different roles in
development and that they are complementary, so that even if the pro-
duction function exhibits diminishing returns in each stock, the joint
returns will be constant or even increasing. Moreover, because the ideal
combination of private and public capital stocks is one in which their
marginal products are equal (with an appropriate correction if public
investment is financed by distortionary taxes), for each level of private
capital stock (and investment) there is an optimal level of government
capital stock (and investment).19 The state, as measured by the size of
the government capital stock, can thus be too small, just right, or too
big.
This is a plausible view, but it hurls itself against a practical problem,
namely, that standard national accounts do not measure the public
capital stock, only current capital expenditures by the central govern-

18
Ram (1986) has shown that the model that introduces the level of government-
consumption expenditures into the production-function equation is misspecified. He
has developed a specification that allows an assessment of the impact of government
on growth without having explicit information about the public capital stock. We
(Cheibub and Przeworski 1997) applied Ram’s specification to our data set and dis-
covered that the contribution of the state is positive, but for various reasons we have
second thoughts about this approach.
19
It is a different matter whether or not a government, even a benevolent one, would
implement it. On the time inconsistency of optimal taxation, see Benhabib and Velasco
(1996).

164
Poor and Wealthy Countries

ment. And without such data, estimating the effect of government

capital is difficult, if not impossible. What one would want to estimate
would be a production function of the form

Y˙ Y = A˙ A + a( P˙ P ) + t( S˙ S ) + b( L˙ L) + e,

but we have no separate information about private and public capital

stocks, P + S = K. In Appendix 3.2 we perform some heroics that permit
us to estimate instead an equation of the form

Y˙ Y = A˙ A + a( K˙ K ) + W( S˙ Y ) + b( L˙ L) + e,

where W can be interpreted as the marginal product of government

capital stock when private capital and public capital grow at the same
rate.
Note .first that whereas government capital expenditures (GXPK-
TOTL), S, account for a relatively large share of public expenditures in
poor countries, they add up to almost nothing in absolute terms: $46
per capita in countries with incomes under $1,000, and $109 in
those with incomes between $1,000 and $2,000. In countries between
$3,000 and $4,000, per capita public capital expenditures already
amount to $190; in those with incomes above $6,000, they are $369.
Table A3.1.1, which we put in Appendix 3.1 because interpreting the
results requires understanding how they were derived, shows that gov-
ernment capital expenditures have no effect on growth in poor coun-
tries, but they do have a sizable positive effect in wealthy countries.
Unfortunately, the interpretation of the coefficient W is ambivalent:
Either the marginal product of public capital increases as total capital
per worker (CAPW = K/L) rises or private capital stock grows in the
more-developed countries faster than public capital stock. And even if
W is a valid measure of the marginal product of public capital, it is not
clear why it should increase as countries become more developed. One
interpretation would be that governments in poor countries simply allo-
cate capital expenditures unwisely, but the almost monotonic relation
between W and K/L makes that farfetched. The interpretation we find
more plausible is that the complementarity between private investment
and public investment, the latter largely in infrastructure, increases as
the private capital stock increases. A road contributes to development
only to the extent that there something to be transported.
Ideally, we would want to see if public capital stock plays different
roles under the two regimes. Unfortunately, the sample that contains

165
 Political Regimes and Economic Growth

information about public capital expenditures (N = 1,366) is extremely

unbalanced: There are very few observations of wealthy dictatorships
and equally few of poor democracies. Hence, whereas going through
our standard steps, that is, correcting for selection and estimating the
production functions separately for the two regimes, shows that public
capital plays no role under dictatorships and a positive role under
democracies, this is just an effect of sample composition. In turn, split-
ting the sample between poor and wealthy countries is not feasible
given its imbalance. Yet for the purpose of our argument it is sufficient
to know that in poor countries the state plays little, if any, productive
role, but in wealthy countries it does.
In sum, poor countries are too poor to afford a strong state, and
without an effective state there is little difference any regime can make
for economic development. Investment is low in poor democracies, but
it is not any higher in poor dictatorships. The labor force grows rapidly
in both. Development is factor-extensive: Poor countries benefit almost
nothing from technical change. Clearly, this does not imply that all poor
countries are the same or even that regimes may not make a differ-
ence for other aspects of people’s lives; indeed, we show later that they
do. But not for economic development in poor countries.

Wealthy Countries
Once countries reach some level of development – somewhere
between $2,500 and $3,000, that of Algeria in 1977, Mauritius in 1969,
Costa Rica in 1966, South Korea in 1976, Czechoslovakia in 1970, or
Portugal in 1966 – patterns of economic development under democ-
racy and dictatorship diverge. In countries with incomes above that
threshold, regimes do make a difference for how resources are used,
for how much people produce and how much they earn. To show that
this is true, we have to go through a number of steps, piecing together
bits of evidence derived from disparate sources and based on distinct
samples. Yet we think that the overall conclusion is overwhelming.
Dictatorships appear not to change their ways whether they are in
poor or wealthy countries. Although in more-developed economies they
rely somewhat less on growth of the labor force and somewhat more
on the growth of capital stock, they use the inputs in almost the same
way and get little benefit from productivity growth. Democracies, in
turn, exhibit different patterns in poor and wealthy countries. Their
capital stock grows somewhat slower, and the labor force much slower,
when they are affluent; they use labor more productively and benefit

166
Poor and Wealthy Countries

more from productivity growth. As a result, the patterns of growth are

different in wealthier dictatorships and democracies.
Note, back in Table 3.4, that the factors that differentiate wealthy
dictatorships from wealthy democracies are the patterns, not averages.
Corrected for selection, the average rates of growth of income are again
almost identical under the two regimes. Although the observed rates
of growth are higher under dictatorships, if the two regimes had used
the same inputs in production, they would have grown at the same
rate. But they would have grown in different ways.
Growth under wealthy dictatorships is labor-extensive. The labor
force grows at a much faster pace under dictatorships, at 2.03 percent,
than under democracies, where it grows at 1.32 percent. And whereas
the elasticity of output with regard to labor is about 0.56 under dicta-
torship, it is 0.68 under democracy. Wealthier dictatorships benefit lit-
tle from technical progress, 0.33; they get most of their growth from
capital, 3.05, and in spite of employing many more workers, they get
less of it from labor, 0.85. Wealthier democracies benefit from technical
progress, 0.91, get less growth out of capital, 2.20, and more from the
labor input, 1.04, even though they employ fewer additional workers.
Because the labor force grows faster under dictatorships, and the
elasticity of output with regard to labor is lower, output per worker
(GDPW = Y/L) is also lower. In the entire sample, the observed values
are obviously lower: $5,113 under dictatorships and $14,554 under
democracies. But selection-corrected values, controlled for capital stock
and the size of the labor force, are still much lower for dictatorships:
An average worker produces $6,843 under dictatorship versus $9,054
under democracy. As we would by now have expected, the difference
in poor countries is negligible: Output per worker is $3,519 in dicta-
torships, and $3,639 in democracies. But above $3,000, the difference
is large: $13,410 versus $17,857. To harp on what may by now be
obvious, these are the outputs that would be generated under the two
regimes if they had exactly the same capital stock, labor force, and
whatever it is that we do not observe that affects both regime selec-
tion and output per worker.
Thus, dictatorships in countries with incomes above $3,000 use labor
extensively. Their labor force grows faster, the marginal worker con-
tributes less to output, and the average worker produces less than in
democracies. It merits noting that this difference is not due to labor-force
participation by women (LFPW), whose values are almost identical
under the two regimes at every income level. More women participate in
gainful activities in very poor countries, where they work in agriculture,

167
 Political Regimes and Economic Growth

Table 3.8A. Labor exploitation (labor-share sample, N = 2,061)

Average labor shares

LEVEL All Dictatorships Democracies

0–1,000 33.04 (308) 32.21 (277) 40.48 (31)

1,001–2,000 32.61 (429) 32.90 (332) 31.61 (97)
2,001–3,000 30.43 (282) 30.70 (158) 30.10 (124)
3,001–4,000 33.44 (218) 34.17 (138) 32.18 (80)
4,001–5,000 35.10 (131) 33.55 (81) 37.63 (50)
5,001–6,000 36.59 (102) 31.08 (48) 41.48 (54)
6,001– 48.94 (693) 36.43 (29) 49.59 (562)
Total 35.53 (2,061) 32.63 (1,063) 42.71 (998)

and in wealthy countries, where they are disproportionately employed

in services. Regimes have no effect on their labor-force participation:
The observed participation rates are 31.9 under dictatorships and 31.0
under democracies. Once the proportion of the labor force working in
agriculture is controlled for, selection-corrected rates are also the same.
Dictatorships simply rely on using many additional workers, of whatever
sex, even if they produce less on the average and at the margin.
Perhaps the reason is that dictatorships exploit labor.20 To examine
that possibility, we must rely on a different data source and a much
smaller sample21 (Table 3.8). The average labor share of value added

20
This is not to say that workers are not, in some sense, exploited under democracies.
We use the term in Roemer’s sense (1982): Workers under dictatorship would be better
off if they could withdraw, with the same endowments, to democracy.
21
The labor-share data come from the World Bank (1995), whose 1995 development
report (p. 234) describes this series as “derived by dividing total earnings of employ-
ees by value added in current prices to show labor’s share in income generated in the
manufacturing sector.” There are 2,061 observations, all for the period 1961–1990.
They cover countries that are, on the average, wealthier than the full sample, but in
the entire range, from $370 to $18,095. The mean is 35.5 (s.d. = 13.0), with a minimum
of 6.1 in Rwanda in 1977 and a maximum of 100.1 in the Bahamas in 1981. Note that
the unconstrained labor elasticities in Table 3.4 are 0.27 for dictatorships and 0.39 for
democracies; hence, they are close to the competitive labor share. For some developed
countries, the numbers are surprisingly small: The 1990 labor share in the United
States was 35.6 (48.5 in 1963), and in Sweden 34.8 (56.5 in 1963). For other coun-
tries they are closer to what one would expect: The 1989 labor share in France was
58.4 (68.4 in 1980), and the 1990 share in Norway was 59.0 (54.7 in 1963). Altogether,
these numbers appear reasonable.

168
Table 3.8B. Accounting for the difference in wages

Dictatorships Democracies
(N = 1,052) (N = 994)

Observed values:
(1) GDPW = Y/L 6,332 (4,858) 16,665 (8,871)
(2) LS = wL/Y 32.5 (11.42) 42.7 (13.78)
(3) WAGE = LS * (Y/L) 2,058 (1,873 7,116 (5,376)
Selection-corrected values:
LS | Y/P 33.5 (1.74) 40.0 (4.19)
LS | Y/L, OTHER (2SLS) 33.7 (3.20) 41.2 (9.29)
(4) (Y/L) | A,K,L 8,375 (4,895) 11,529 (7,858)
(5) WAGE | (Y/L) 3,733 (2,864) 5,057 (4,442)
Accounting for difference in wage:
(6) Actual output/worker (from line 1) 6,332 - 16,665 = -10,335
(7) Wages assuming same labor share (line 6 * 0.3553) 2,250 - 5,921 = -3,671
(8) Actual wages (from line 3) 2,058 - 7,116 = -5,058
(9) Difference due to labor shares (line 8 minus line 9) -192 - 1,195 = -1,387
(10) Output/worker assuming same inputs (from line 4) 8,375 - 11,529 = -3,154
(11) Wages assuming same inputs, same labor share (line 10 * 0.3553) 2,976 - 4,096 = -1,120
(12) Wages assuming same output per worker (from line 5) 3,733 - 5,057 = -1,324
Summary:
(13) Due to difference in inputs (line 7 minus line 11) -2,551 -2,551
(14) Due to difference in productivity (line 11) -1,120 -1,120
(15a) Due to difference in labor shares (line 9) -1,387
(15b) Due to difference in labor shares (line 12) -1,324
(16) Total difference in wages (lines 13 + 14 + 15) -5,058 -4,995
(17) Error (line 8 minus line 16) -63
 Political Regimes and Economic Growth

in manufacturing (LS) during the 1,052 dictatorial years was 32.5

percent, and the average during the 994 democratic years was 42.7
percent. Only a part of this difference vanishes when controlled for
output per worker: Selection-corrected labor shares are still larger for
democracies, and the estimates are tight. Hence, even if they had
generated the same output per worker, dictatorships would have paid
workers a smaller share of value added than would democracies. See
also Rodrik (1998).
Although the labor-share data concern only the manufacturing
sector, another way to see the same result is to calculate an-
nual wage rates (WAGE), multiplying output per worker by the labor
share. The observed wage under dictatorship is $2,058, whereas
under democracy it is $7,116, a difference of $5,058 in favor of
democracies. If the labor share had been the same under the two
regimes, 35.5 percent, their difference in wages would have been
$3,671. Hence, $1,387 of the observed difference is due to labor
shares.
Because a large part of the observed difference is obviously due
to income levels, in Table 3.8 we engage in some accounting. The
assumption that dictatorships and democracies have exactly the same
productive inputs yields the conclusion that the average output per
worker would be $8,375 under dictatorship and $11,529 under democ-
racy, for a difference of $3,154: As in the full sample, workers produce
less under dictatorships. If the two regimes had had the same inputs
and the same labor share, the difference in wages due only to pro-
ductivity would have been $1,120. Hence, if labor shares were the
same, the difference in inputs would account for $2,551 ($3,671 -
$1,120), and the difference in productivity would account for $1,120,
of the differences in wages. In turn, assuming that output per work-
er is the same under the two regimes yields a selection-corrected
difference of $1,324.
Hence, of the observed difference in wages (to remind, $5,058),
$2,551 is due to the fact that the regimes existed under different con-
ditions and used different productive inputs, $1,120 is due to dicta-
torships being less productive, and $1,387 is due to differences in labor
shares.
Although the number of observations is too small to permit cor-
recting for selection within subsamples, it appears again that the
gap between regimes opens up at about $3,000 of per capita income.
Whereas the labor share and average wage were much higher in India
than in China when both countries had similar per capita incomes, that

170
Poor and Wealthy Countries

pattern is not characteristic of other poor countries.22 Under $3,000,

the labor share during 767 dictatorial years was almost identical, at
32.2 percent, with the 32.0 percent during the 252 democratic years.
Between $3,000 and $8,000, the labor share under 288 years of
dictatorship was 33.7 percent, but under 323 years of democracy it
climbed to 39.6 percent. Wages offer a similar picture. Under $3,000,
an average worker earned $1,238 under dictatorship and $1,698
under democracy, but this difference is due entirely to the fact that
democracies are still somewhat wealthier in this income range. Be-
tween $3,000 and $8,000, the average wage was $4,193 under dicta-
torship, $1,556 less than the $5,649 under democracy. Some of this
difference is again due to the fact that democracies have higher output
per worker in this range than do dictatorships. But even if workers
under wealthy dictatorships had produced the same output as under
democracy, they would have earned $4,777, still $872 less.
There are few observations in this sample for which labor-share data
are available for both the entry and exit of a regime. But the impact of
transitions to dictatorship can be seen: In the eighteen cases we
observed, dictatorships immediately reduced labor share by 2.6 per-cent
of value added, from 27.4 to 24.8. The “bureaucratic-authoritarian”
regimes in Latin America were particularly brutal in redistributing
incomes from labor to capital. The Argentine military reduced the labor
share from 29.4 in 1975 to 18.1 in 1976 and to 14.2 in 1982, its last
year of existence. In Chile, labor share was squeezed from 29.6 in 1972
to 16.3 in 1973 and to 14.8 by the end of dictatorship, in 1989. Transi-
tions to democracy, however, had no immediate effect on income
shares.
To get an intuitive sense of these patterns, compare first the
neo-liberal miracle, Singapore, with the persistently social-democratic
Austria, when the two countries had the same levels of per capita
income (Figure 3.1). In 1976 Singapore had per capita income of
$5,606, and the average worker produced $13,394. The labor share
was 33.1, which meant that the average worker earned $4,433. Austria
had almost the same per capita income, $5,672, by 1963, and output
per worker was somewhat lower, $12,328. But the Austrian worker
took home 48.6 percent of what he or she produced, so that the average
wage was $5,991: $1,558 more. By 1990, Singapore had per capita

22
As India reached per capita income of $1,000 in 1984, the labor share was 50.3, and
the average wage was $1,313. China reached that level of income a year earlier, with
a labor share of 15.9 and a wage of $287.

171
Figure 3.1. Labor shares in Singapore and Austria at similar levels of per capita
income from year to year
Poor and Wealthy Countries

income of $11,698, and output per worker reached $24,396. The labor
share was 31.7, and average wage was $7,725. Austria reached the
income of $11,462 by 1987, and the average worker produced $24,366
by that time. The labor share was 54.7, so that the average wage was
$13,328: $5,603 more than in Singapore. And this illustration is not
tendentious: Among all countries in the same income range, between
$5,606 and $11,698, average output per worker was $17,143 under
dictatorship (N = 45) and $21,004 under democracy (N = 426), the aver-
age labor shares were, respectively, 34.3 and 48.9, and the average
wages were $6,882 and $10,402: a difference of $3,620.
The experience of Mexico and Portugal during the 1980s, when
the two countries had incomes in the $4,974–$6,575 range, is similar
(Figure 3.2). Mexico’s labor share hovered around 20 percent; Portugal’s
declined from 46.9 to 36.2 during that period. Still, by 1989, an average
Mexican worker earned $3,192, and a Portuguese earned $4,917.
Finally, examine the experience of Argentina, where we can observe
several regimes: dictatorship until 1972, democracy between 1973 and
1975, another dictatorship between 1976 and 1982, and a new democ-
racy after 1983. The advent of democracy in 1973 increased the labor
share by about 4 percent, and the wage went up from $3,942 in 1972
to $4,697 one year later (Figure 3.3). Wage reduction by the military
regime of 1976 was drastic: Labor share fell from 29.4 percent in 1975
to 18.1 percent, and average wage tumbled from $4,716 to $2,825.
When democracy was reestablished in 1983, labor share increased
only slightly, as the Argentine economy was buffeted by crises. By 1990,
it stood at 20.5, and average wage was $2,748: higher than the $2,155
during the last year of the military dictatorship, but well below the
Argentine historical levels.
Thus, growth under wealthier dictatorships is both labor-extensive
and labor-exploitative. The labor force grows faster under dictatorship,
the marginal worker produces less, and the average worker much less,
than under democracy. Even if they were matched for capital stock and
labor force, workers would produce less under dictatorship. And labor
earns less under dictatorship. Some of this difference is attributable to
lower output, but a large difference remains even when we account for
differences in factor endowments and even if we assume away differ-
ences in productivity. Democracy entails the freedom for workers to
associate independently of their employers and the state. And strikes
are three times more frequent under democracy than under dictator-
ship: 0.2434 per year under the former and 0.0895 under the latter.
Because, in addition, under democracy workers also vote, these results

173
Figure 3.2. Labor shares in Mexico and Portugal at similar levels of per capita
income from year to year
Figure 3.3. Argentina’s labor share by regime
 Political Regimes and Economic Growth

are not surprising. Dictatorships repress workers, exploit them, and

use them carelessly. Democracies allow workers to fight for their inter-
ests, pay them better, and employ them better.

Miracles and Disasters

A list of regimes under which economies grew at the average rate
of at least 7 percent per year, thus doubling incomes in ten years, con-
sists almost exclusively of dictatorships. Indeed, the only exception is
the somewhat peculiar case of the Bahamas. But the list of countries
in which total income declined or grew at less than 1 percent per year
during at least seven years looks almost the same: Almost all regimes
on that list are also dictatorships.23 The observed variation of perfor-
mance is simply much higher for dictatorships (s.d. = 7.04) than for
democracies (s.d. = 4.85), and so is the selection-corrected variance.
Hence, looking only at best practice leads to dangerously misleading
conclusions.
Although some countries, notably Singapore, South Korea, and
Taiwan, sustained rapid rates of growth over the entire period, glory
is transitory (Table 3.9). Of the countries that grew fastest during the
1950s – Turkey, Costa Rica, West Germany, and the Philippines – none
appears on the list of tigers in the successive periods. The next two
decades were periods of sustained growth: Gabon, Ivory Coast, Iraq,
Jordan, South Korea, Singapore, Malta, and Romania grew at the
average rates above 7 percent during the 1960s and 1970s. But growth
collapsed during the 1980s, and only four countries maintained their
rhythm of the previous decade: South Korea, Singapore, Taiwan, and
the Yemen Arab Republic. In turn, some earlier miracles turned into
disasters. In Iraq, after two decades of rapid growth, income fell at the
rate of 7.04 percent during the eight years in the 1980s for which data
are available. The same was true for Ivory Coast, which, having
sustained rapid growth between 1960 and 1980, suffered a declining
income afterward. Nigeria, having experienced spectacular growth
during the 1970s, had income growing at only 0.32 during the next
decade. Suriname grew at 8.50 during the 1970s, only to decline to a
rate of 2.27 in the 1980s. Yugoslavia grew at 7.40 during the 1970s,
and -1.25 during the next decade.

23
Among Temple’s ten miracles (1999: 116) between 1960 and 1990 there were two
democracies, Japan and Malta, and among his ten disasters there was one democracy,
Venezuela, and one country that experienced regime transitions, namely, Ghana.

176
 Miracles and Disasters

Table 3.9. Miracles and disasters (fastest- and slowest-growing regimes, over
periods of at least ten years)

Number First Growth

Country of years year rate Regime Comment

Fastest (more than 7%):

Botswana 24 1966 9.55 Dic Primary-
commodity
exporter
Nigeria 13 1966 9.24 Dic Oil exporter
Yemen A. R. 20 1970 9.19 Dic
Singapore 26 1965 9.14 Dic
South Korea 27 1961 8.66 Dic
Taiwan 39 1952 8.41 Dic
Ecuador 16 1963 7.82 Dic Oil exporter
Brazil 15 1964 7.54 Dic
Turkey 10 1951 7.49 Dic
Bahamas 10 1978 7.47 Dem
Cape Verde Islands 16 1975 7.40 Dic
Syria 30 1961 7.02 Dic One year of war
Jordan 36 1955 7.01 Dic Two years of war

Slowest (less than 1%):

Madagascar 30 1961 0.85 Dic Primary-
commodity
exporter
Comoros 16 1975 0.81 Dic Primary-
commodity
exporter
Western Samoa 11 1980 0.64 Dic
Papua New Guinea 16 1975 0.62 Dem
Chad 30 1961 0.61 Dic War, primary-
commodity
exporter
Peru 10 1980 0.48 Dem War, primary-
commodity
exporter
Uruguay 12 1973 0.48 Dic Primary-
commodity
exporter
Guyana 25 1966 0.33 Dic Primary-
commodity
exporter
Angola 15 1975 0.08 Dic War, oil exporter
Mozambique 16 1975 -0.95 Dic War

177
 Political Regimes and Economic Growth

Some of the sharp oscillations in growth were due to wars, about

which we say more in the next chapter. Another source of variance
in growth rates was reliance on the export of primary commodities
(COMEX). Countries that derived at least half of their export revenue
from primary commodities grew, in general, at a slower rate, 3.47 (s.d.
= 7.42; N = 1,346), than countries that did not, which grew at the
rate of 4.60 (s.d. = 5.58; N = 2,780). Note that commodity-exporting
countries experienced greater variance in their growth performance.
Finally, oil exporters (OIL) grew at a faster rate, 4.78 (s.d. = 7.71; N =
413), than countries that did not export oil, 4.17 (s.d. = 6.09; N = 3,713),
but the variance in growth rates was again higher among oil exporters.
Once we eliminate the years of wars and of post-war recoveries and
the countries that export oil or other primary commodities, the vari-
ance in growth rates is reduced under both regimes. But it is still
much larger, s.d. = 6.08, under dictatorships than under democracies,
s.d. = 3.87. And the lists of miracles and disasters are still populated
almost exclusively by dictatorships. Thus, looking at best practice is
misleading: The tigers may be dictatorships, but dictatorships are no
tigers.

Conclusion
The main conclusion of this analysis is that there is no trade-off
between democracy and development, not even in poor countries.
Although not a single study published before 1988 found that democ-
racy promoted growth, and not one published after 1987 concluded in
favor of dictatorships (Przeworski and Limongi 1993), there was never
solid evidence that democracies were somehow inferior in generating
growth – certainly not enough to justify supporting or even condoning
dictatorships. We hope to have put the issue to rest. There is little dif-
ference in favor of dictatorships in the observed rates of growth. And
even that difference vanishes once the conditions under which dicta-
torships and democracies existed are taken into account. Albeit in
omniscient retrospect, the entire controversy seems to have been much
ado about nothing.
Poverty appears to leave no room for politics. In countries with
incomes below $3,000, the two regimes have almost identical invest-
ment shares, almost identical rates of growth of capital stock and of
labor force, the same production function, the same contributions of
capital, labor, and factor productivity to growth, the same output per
worker, the same labor shares, and the same product wages. Poor

178
Appendix 3.1: Specification and Robustness

countries invest little, get little benefit from total factor productivity,
and pay low wages. And though a few countries have escaped this bond
of poverty, most poor countries, have remained poor. Democracy is
highly fragile in such countries, and thus most of them have dictator-
ial regimes. But regimes make no difference for growth, quantitatively
or qualitatively.
Perhaps surprisingly, affluence differentiates regimes. Wealthier dic-
tatorships invest a somewhat larger share of income, experience higher
growth of the labor force, have higher capital and lower labor elastic-
ities, derive more growth from capital input and less from labor input
and from total factor productivity, have lower output per worker, have
a lower labor share, and pay lower wages. Wealthier dictatorships
grow by using a lot of labor and paying it little. Because they repress
labor, they can pay it little; perhaps because they can pay it little, they
care less how it is used. They pay more for capital – the average rela-
tive price of investment goods is higher under dictatorships – and they
use it well. But because they rely on force to repress workers, they can
pay lower wages and use labor inefficiently.
In the end, total output grows at the same rate under the two
regimes, both in poor countries and in wealthier countries. But the
reasons are different: In poor countries, regimes simply do not matter.
In wealthier countries, their average growth rates are the same, but
the patterns of growth are different.

Appendix 3.1: Identification, Specification, and Robustness

Regimes can affect growth by influencing either the growth of factors
that enter into production or their utilization. But the effect of regimes
can be identified only if there are some conditions that are exogenous
with regard to regimes. Our general strategy, then, is to test repeatedly
whether or not the variables that enter on the right-hand side of a
growth equation are exogenous with regard to regimes, and, if they
are not, to repeat the procedure with regard to these factors until the
right-hand-side variables are all exogenous.
Because our purpose is to examine how political regimes affect eco-
nomic growth, we need to develop first an understanding of its proxi-
mate causes. These causes are economic: Whereas political factors
affect the decisions of governments and of private agents (households
and firms) concerning the allocation of resources and the distribution
of incomes, the performance of an economy results from these deci-
sions. Hence, we look first for proximate causes of growth and then for

179
 Political Regimes and Economic Growth

the factors that in turn affect these proximate causes, both the supply
of productive factors and their productivity.
To choose a growth model, we looked first for a specification that
would best account for the variations in the observed growth rates in
the entire sample, would make economic sense, and would not a priori
favor one political regime over another.
There are many such models. The statistical literature explaining
the growth of countries, via “growth regressions,” is now voluminous,
and new variables that matter are discovered almost every week. But
it turns out that few are robust. Levine and Renelt (1992) employed
virtually every variable that had previously been used in growth
models. They also listed 40 cross-sectional growth studies published
between 1980 and 1990. Only investment (in physical capital) and the
initial level of income were always robust, in the Leamer (1983, 1985)
sense. They discovered that the coefficients on other terms were fragile
and were particularly sensitive to the factor of education. Because we
do not want to prejudice the findings concerning regimes by our spec-
ification of the growth model, we are thus led to a minimalist approach,
working with only those variables that are robust and testing for
robustness by introducing additional variables.24
In the end, we use throughout what we call a “barebones” model: a
simple production function of the form Y = AF(K, L). To test robustness,
we augment this production by some other variables, listed in Table
A3.1.1. We do not use the initial level of income among the right-hand-
side variables, because that practice strikes us as mindless if the model
is estimated in growth form. If the income level affects the rate of
growth, it must influence it via the return to factors. Hence, to test the
effect of income, we estimate the production function separately for
countries within different income bands (see Table 3.4), rather than
attempting to “control” for it statistically.
We estimate the model in growth form to avoid problems due to
autocorrelation of output levels over time.25 With an apology to econ-
omists, we show how this growth form is derived.

24
This approach is not without problems. See Temple (1999).
25
An alternative would be to estimate the growth model in the form used by Mankiw,
Romer, and Weil (1992), in which the rate of convergence can be inferred from the
values of the coefficients. Both practical and theoretical considerations, however, speak
against that approach. From the practical point of view, that approach requires aver-
aging growth rates over some periods of time, whether decennia or quintennia, to avoid
problems of autocorrelation. From the theoretical point of view, it is not apparent how
to interpret the convergence coefficients (Quah 1993).

180
Appendix 3.1: Specification and Robustness

The theoretical point of departure in analyses of growth is the pro-

duction function:

Y (t ) = F (◊, ◊, ◊,). (1)

Particular models differ in specifying what productive factors enter into

the parentheses and how.
The “barebones” model is (dropping time subscripts)

Y = AF ( K , L), (2)
where K is capital stock, and L is the labor force. Let F be the Cobb-
Douglas function of the form

Y = AKaLb. (3)

Then all we need to do is take logarithms of both sides of (3), and then
the derivatives of the logs with regard to time, to get

Y˙ Y = A˙ A + a( K˙ K ) + b( L˙ L),

where AFK(K/Y ) = a is the marginal elasticity of output with regard to

capital, and AFL(L/Y ) = b is the elasticity with regard to labor.
Both a and b have, therefore, theoretical interpretations, and we
have expectations about the values of these numbers. If F is Cobb-
Douglas, then in a competitive economy the elasticities equal the
income shares, respectively, of capital and labor. The exact numbers
are matters of some controversy. The World Bank (1991: 43) estimated
the marginal elasticity of output with regard to capital to be 0.4
between 1960 and 1987. In perhaps the most careful study to date,
Islam (1995) arrived at estimates slightly over 0.4 for a sample of 96
non-oil-producing countries. In turn, the World Bank (1991: 43) esti-
mated b to be about 0.45. Our estimates of a and b are extremely
robust, not affected by the presence of other variables. In the entire
sample, the unconstrained estimate of a is 0.40, and of b a somewhat
low 0.33. Constraining the elasticities to sum up to 1 does not change
a and brings b to its complement to 1, that is, 0.60. Yet the constraint
a + b = 1 is almost always rejected by the F test. Adding human capital,
which we do not introduce into our basic specification because the data
are available for only a much smaller sample (N = 2,787), brings the
values of a and b down somewhat, but the unconstrained elasticity with
regard to human capital is very low and not statistically significant.
When the sum of the three elasticities is constrained to 1 in OLS, the
growth of human capital becomes significant, but it is still very small.

181
 Political Regimes and Economic Growth

Our data set consists of annual observations. Contrary to frequent

practice, we do not average these observations over some longer
periods (decades, quintennia), a procedure typically used to smooth
business-cycle effects. It may be a matter of dispute whether or not our
coefficients on the labor input are always of a reasonable magnitude.
But because we have to match economic performance with regime
changes, and these occur on an annual basis, we must consider growth
on the same basis. Averaging over time may smooth business-cycle
effects, but we have seen too many regressions in which political char-
acteristics have been said to have caused economic performance that
occurred earlier.
To control for the omitted variables that are constant over time or
over countries, whenever possible we have used panel methods. To
consider the endogeneity of the right-hand-side variables, we also
followed the suggestion of Temple (1999) to use their lagged values as
instruments in panel analyses. Those results, however, turned out to
yield nonsensical values for coefficients.
One specification issue is to what extent country-specific character-
istics, rather than external shocks, determine growth rates. Easterly et
al. (1993) observed that country growth rates were highly unstable over
time, whereas country characteristics used to explain growth were
much more stable. Hence, shocks, especially those to terms of trade,
play a large role in explaining growth. Nevertheless, our finding that
the entire 1950–1990 sample and various subsamples almost never
pass tests for homogeneity goes against the Easterly et al. (1993)
argument: We do find country-specific effects, and most of the time
the Hausman test indicates that these are fixed effects. Nevertheless,
as suggested by Easterly and associates, we modeled external shocks,
taking as the independent variables the average rate of growth for
the entire world during a particular year (WORLD) and the U.S. prime
rate (PRIME). The latter variable never matters once WORLD is
present in a specification, and we dropped it. As Table A3.1.1 shows,
none of the results is affected by the presence of these additional
variables.
Finally, the quality of the PWT data set is very uneven. Hence,
one factor that may affect the results is biased reporting of data.
We thus test robustness with regard to the rating of data quality,
PPPQUAL.
Table A3.1.1 shows the effect of model specification on the
selection-corrected values of growth under the two regimes. The results
are perfectly robust: Whatever variable one introduces into the speci-

182
Appendix 3.1: Specification and Robustness

Table A3.1.1. Robustness

Dictatorships Democracies
Variables:
KSG, LFG, and: Method Mean YGHAT Mean YGHAT

WORLD 2F 2.83 4.30 2.81 4.26

(4.87)a (4.40)
OPENC 2F 47.73 4.28 55.17 4.25
(4.87) (4.31)
COMEX OLS 0.40 4.30 0.21 4.24
(5.02) (4.48)
CPI 2F 20.80 4.31 46.92 4.22
(4.46) (4.08)
XR 2F 83.21 4.28 54.73 4.25
(4.87) (4.35)
PPPQUAL OLS 2.27 4.32 6.60 4.25
(5.04) (4.45)

Notes: WORLD is the average rate of growth in the world during the particular year.
OPENC is the sum of exports and imports as a proportion of GDP. COMEX is a dummy
variable indicating that a country is a primary-commodity exporter. CPI is the rate of
change in the consumer price index. XR is the nominal exchange rate (local currency to
dollars). PPPQUAL is a rating of data quality (1 = worst, 10 = best).
a
Standard errors in parentheses.

fication of the growth equations, regimes do not affect the average

growth rates. Note in particular that whereas the data quality (mea-
sured on a 1-to-10 scale) is much lower for dictatorships, which also
tend to be much poorer, the quality of the data is not correlated with
the reported growth rates and does not affect regime results.
The most difficult decision, and perhaps a dubious decision, we
made in classifying regimes was to consider as dictatorships those
countries that had regular and apparently free elections but in which
the incumbents always won the elections (see Chapter 1, the “alter-
nation” rule). Such countries grew at an exceptionally fast pace, 5.11
per annum (N = 377), as contrasted with the rate of 4.30 (N = 2,104)
for other dictatorships, and 3.95 (N = 1,645) for democracies. To test
the robustness of our conclusions with regard to this classification rule,
we reclassified these cases as democracies and replicated our standard
analysis. The results, presented in Table A3.1.2, show that this reclas-

183
 Political Regimes and Economic Growth

Table A3.1.2. Selection-corrected estimators of

the rate of growth of income (YG), under
modified classification of regimes

Regime means

Estimator Dictatorships Democracies

(selection/YG) (N = 2,030) (N = 1,961)

Biased 4.32 4.12

(7.02) (5.19)
Unconstrained OLS 4.13 4.30
(4.97) (4.73)
Constrained OLS 4.10 4.40
(5.14) (4.81)
PANEL 4.17 4.29
(4.84) (4.62)

sification does alter the results slightly in favor of democracy, but far
from enough to undermine the conclusions.

Appendix 3.2: Estimating the Contribution of

Public Investment
Assume now that output is produced in different ways by private
capital stock, P, and government capital stock, S, with

P + S = K. (1)

Let the production function be Cobb-Douglas:

Y = APaStLb. (2)

Then (2) can be written in growth form as

Y˙ Y = A˙ A + a( P˙ P ) + t( S˙ S ) + b( L˙ L). (3)
.
Our problem is that we do not know P, P, and S.
Now let
t = a X + d, (4)
where X is such that

184
Appendix 3.2: Contribution of Public Investment

P˙ P + ( S˙ S ) X = K˙ K = ( P˙ + S˙ ) ( P + S ) ,
or

X = ( S˙ S )( K˙ K - P˙ P). (5)
Substituting (5) into (4) and then into (3) yields
Y˙ Y = A˙ A + a( K˙ K ) + t( S˙ S ) + b( L˙ L ). (6)
.
Because this expression still contains S, we multiply S/S by Y/Y to
obtain

Y˙ Y = A˙ A + a( K˙ K ) + W( S˙ Y ) + b( L˙ L ), (7)

where W = d(Y/S), and a is now the elasticity of output with regard to

private capital stock.
We have all the data necessary to estimate (7), but the difficulties
with this way of proceeding are that the coefficient W need not be a
constant, and it does not have an obvious interpretation even if it is.
Note from (4) that d = t - aX, so that
W = d(Y S ) = ( t - a X )(Y S ). (8)

Given (3), t is the elasticity (dY/dS)(S/Y). Substituting (5) yields,

therefore,
W = dY dS - a( K˙ K - P˙ P )(Y S˙ ). (9)
. . .
In a steady state in which K /K = P/P = S /S, W is then the marginal
product of government capital. If public capital stock rises faster than
private, then the estimate of the marginal product is downward-biased;
if private capital stock increases faster, the bias is upward.
Now, if W is a marginal product, then we would expect its value to
fall in the stock of public capital. In fact, when we split the 1,366 avail-
able observations according to the values of total capital per worker,
we observe it to increase (Table A3.2.1). The OLS estimates increase
slowly, but the two-way fixed-effects estimates rise sharply. Therefore,
either the marginal product of public capital in fact increases with total
capital per worker or private capital grows faster than public capital
in the more-developed countries.

185
 Political Regimes and Economic Growth

Table A3.2.1. Coefficients (W) on public capital, by capital per worker

W W

K/L N OLS 2F K/L N OLS 2F

–1,000 259 4.13 1.12 1,001– 1,107 3.27 16.36

(1.452) (0.191) (1.113) (3.291)
–2,000 347 2.41 1.54 2,001– 1,019 4.85 18.79
(0.856) (0.300) (1.559) (3.569)
–3,000 415 2.83 3.48 3,001– 951 4.46 17.10
(0.951) (0.584) (1.391) (3.227)
–4,000 499 2.76 8.03 4,001– 867 4.40 21.58
(1.030) (1.627) (1.181) (3.311)
–5,000 565 3.19 8.16 5,001– 851 4.03 21.63
(1.131) (1.608) (1.045) (3.267)
–6,000 600 3.09 7.84 6,001– 766 4.10 22.40
(1.137) (1.629) (0.988) (3.260)
–7,000 640 2.42 6.92 7,001– 726 6.27 27.35
(0.879) (1.440) (1.509) (3.721)
–8,000 669 3.17 7.83 8,001– 697 2.75 20.92
(1.155) (1.647) (0.638) (2.783)

Note: All samples under the given values are homogeneous; hence the appropriate esti-
mator is OLS. All samples above the given values are heterogeneous, and random effects
are biased; hence the appropriate estimator is two-way (country and year) fixed effects.

186
 Chapter Four

Political Instability and

Economic Growth

Introduction
Not everyone will be surprised to learn that political regimes have
no effect on average rates of economic growth. One generally held view,
made influential by Huntington (1968), is that what matters for eco-
nomic development is political stability, rather than the particular
political institutions. Any system of political institutions promotes
development as long as it maintains political order. The danger is
“political instability.”
We put “instability” in quotation marks because the concept is con-
genitally muddled. The functionalist sociology that dominated thinking
about development during most of the post-war era was constitution-
ally incapable of conceptualizing social change in other than vaguely
evolutionary terms. Under the influence of Parsons (1951), its theoret-
ical program was to search for conditions of the “equilibrium,” con-
ceptualized in a sui generis way as “functional balance.” Anything that
disturbed this balance, any abrupt change, was seen as a “breakdown,”
a limiting category. As such, it could not be conceptualized any further.
The conclusion was that things are stable when they do not change:
“We may say that the political system is stable when the impact on the
system and the environment are neutralized to the extent of keeping
them from altering . . . the structure of the political system” (Ake 1967:
100–1).
When combined with an anti-communist ideology, the “structural-
functional approach” turned “instability” into the central peril facing
the “new nations.” Instability became a specter, a harbinger of revolu-
tion. That ideological construction permeated even statistical research:
The most frequently used index of “instability” blended routine
forms of democratic activity (such as alternations in office, strikes, and

187
 Political Instability and Economic Growth

peaceful demonstrations) with violent acts (such as coups, riots, and

political assassinations).1
Given that history, we probably would do best to abandon the lan-
guage of instability altogether. But given the grip of linguistic conven-
tions, perhaps all we can do is to purge this term of its ideological
connotations. To do this, we must begin by recognizing that not all
political change constitutes “instability.” Distinctions, therefore, are
needed.
Political upheavals divert resources and energies away from pro-
duction and thus affect the contemporaneous growth of the economy.
War is most prominent among them, but one can expect that other
events of historical importance, such as regime transitions, will have
immediate economic repercussions. Even less momentous political
events, such as changes of heads of governments, may entail short-
term economic adjustments. Several studies have found that several
other political events, “socio-political unrest,” can reduce investment
or economic growth.2
Whether or not past political upheavals affect the present tempo of
development is already a more complex question, for it is by no means
obvious how the past would affect the present. One mechanism may be
that political disruptions retard the rate of accumulation of productive
factors, whether physical or human capital, and thus their current
stocks. Public infrastructure may have suffered from neglect due to
political upheavals; successive waves of emigration may have driven

1
In one of the first scales of that kind, Feierabend and Feierabend (1966) included res-
ignations of cabinet officials, peaceful demonstrations, political assassinations, mass
arrests, coups, and civil wars, assigning increasing weights to those events. Others con-
sidered turnovers of chief executives (Russett 1964: 101–4), executive transfers (Root
and Ahmed 1979), and political strikes and riots (Schneider and Frey 1985). Cukier-
man, Edwards, and Tabellini (1992) lumped together riots, political executions, changes
in the compositions of governments, unsuccessful coups and changes of governments,
and durations of governments. Alesina and Perotti (1997) commingled changes of chief
executives, changes in partisan control over governments, and changes of government
composition not leading to the replacement of the chief executive, whether or not those
changes were constitutional, and coups. For an early critique of such scales, see Hurwitz
(1973).
2
Trying to avoid business-cycle effects, cross-country studies of growth typically have
averaged each country’s observations over time and then analyzed regressions on the
averages. To remove endogeneity, some of those studies have considered the reduced
form of a simultaneous-equations model. Alesina and Perotti (1997), for example, have
studied a model in which socio-political instability affects investment, which in turn
affects instability. We are skeptical that their approach solves the problem of causality,
because events that occurred later were still considered as having affected those that
transpired earlier: A coup in 1983 entered as an independent variable for investment
in 1973.

188
Contemporaneous Political Upheavals

educated people abroad. Thus stocks of productive inputs may be lower

in countries that have experienced frequent political changes in the past.
Yet past stability may be a mixed blessing. At least with regard to democ-
racies, Olson (1982) claimed that stable political institutions allow
special interests to entrench their influence and to tear the economy
apart with their distributional demands. Hence, the effect of past politi-
cal events on the current rhythm of development is an open question.
Finally, a central claim of economic theory is that the behavior of
strategic actors, whoever they may be, is oriented by their expectations.
Hence, the question is whether or not anticipations of political changes,
whether by the political rulers or by the potential investors, affect
current economic outcomes. Faced with the prospect of losing power,
those who make political decisions may engage in pillage, rather than
pursue developmental policies that would enhance the welfare of their
successors. But they may also react differently, trying to increase the
welfare of their constituents so as to enhance their chances of surviving
in power. Private investors, in turn, may react to prospective political
changes by putting off their projects until the uncertainty resolves itself,
or, if they can, they may invest elsewhere.
Thus, the present rate of economic growth may be affected by
present, past, or anticipated political events. Within the limits of the
available data, we examine these effects one by one.

Contemporaneous Political Upheavals

Moments of dramatic political change are economically costly. We
examine in turn the effects of wars, regime transitions, changes of chief
executives, and different forms of political unrest.
The ultimate form of political upheaval, war, was quite frequent
during the period under our scrutiny.3 In one out of fourteen years,
countries suffered from war on their territory. As one would expect,
wars hurt growth: Whereas the average rate of growth during peace-
time was 4.39 percent (s.d. = 5.84; N = 3,837), during war years
economies grew at 2.15 percent (s.d. = 10.15; N = 289). Wars devas-
tated some economies, notably in Uganda after 1978, El Salvador,
Nicaragua, and Peru in the late 1980s, and Iraq after 1978. Yet other
wars had little economic effect: The standard deviation of growth rates

3
Our WAR variable includes only the wars occurring on the territory of a country, but,
in turn, it counts both foreign and civil wars. Hence, our conceptualization differs from
that used in the “democratic peace” literature.

189
 Political Instability and Economic Growth

was almost twice as large during the war years. And recovery (RECOV-
ERY) from wars was rapid: The average rate of growth during the five
years following a war was 5.98 percent (s.d. = 6.63; N = 192), and
during other years it was 4.15 (s.d. = 6.23; N = 3,934).4
Countries run by dictatorships were more likely to experience wars,
which occurred once in twelve years, as contrasted with once in every
twenty-one years under democracy. Moreover, wars caused more
damage under dictatorships, where the rates of growth were 4.68
(s.d. = 6.66; N = 2,269) during peacetime and 1.68 (N = 212) during wars,
than under democracies, where the economies grew at 3.97 (N = 1,568)
during peacetime and at 3.46 (N = 77) during wars. In turn, recoveries
were also faster under dictatorships than under democracies. Alto-
gether, then, whereas under dictatorships wars brought economic
disasters, over a longer run wars had little effect on economic growth.
Regime transitions are also economically costly. During the two
years surrounding the eighty-nine transitions,5 the rate of growth was,
on the average, as low as 2.03 percent, but in other years it was 4.33
percent. Regime transitions are momentous events that absorb atten-
tions and energies, cause some people to drastically change their lives,
and often are shrouded by extreme uncertainty. Hence, it is not sur-
prising that economies stagnate during the moments of transition.
Moreover, the share of investment in GDP was 3 percent lower during
the years of transition: 14.01 percent, as contrasted with 17.01 at other
times. Capital stock grew at a rate of 6.09 percent during transitions
and at 7.04 during other periods.
Although wars and regime transitions are both accompanied by eco-
nomic crises, not all political change constitutes “instability.” The ide-
ological connotations of this concept become most apparent when we
think of alternations of heads of governments. Changes of chief ex-
ecutives (HEADS) are just part of the normal life of democracy, out-
comes of a process that is rule-governed, repeated, and expected.
Indeed, we have seen in Chapter 2 that regular alternations in office
extend the lives of democratic regimes. In turn, changes of heads of
governments under dictatorships are, albeit with some exceptions,6

4
Referring to the examples of Japan and Germany in the aftermath of World War II, Barro
and Sala-i-Martin (1995: 176) sought to explain the rapid post-war recoveries, arguing
that whenever a war destroyed only one productive factor, the rate of return to other
factor(s) becomes larger, thus spurring rapid growth.
5
These two years are the last full year of the previous regime and the year during which
the transition occurred.
6
Principally Mexico, plus brief periods in Bolivia, Brazil, and El Salvador.

190
Contemporaneous Political Upheavals

sporadic, unruly, and often violent episodes. Death is often the only
way out of office for dictators.7 And any change of the dictator threat-
ens the survival of the dictatorship.
Numbers concerning the frequency of changes of heads of govern-
ments are eloquent. The chief executive changed once in about nine
years under dictatorship, and once in about four under democracy. The
half-life of a dictator was 8.45 years; the half-life of a democratic chief
executive was 3.96 years. One-fourth of dictators survived in power at
least 16.90 years; three-fourth of democratic rulers were gone after
7.93 years. Indeed, seven dictators in our sample ruled continuously
for at least thirty-five years; only two democratic chief executives
stayed in office nineteen years, and only one, Tage Erlander in Sweden,
twenty-three years.
As we would expect, turnover of democratic chief executives, though
much more frequent, has almost no impact on the rate of growth of
income nor on the investment share. The democratic economies grew
at the rate of 4.03 percent and invested 20.6 percent of GDP during
the 1,188 years in which no change in chief executive occurred and
grew at 3.92 percent, while investing 21.6 percent, during 345 years
when the chief executive changed once. Only when changes of gov-
ernment heads were more frequent (two changes occurred in forty-
nine years, three in six years, and four in one year) did the economy
slow down to 3.37 percent, even though investment share remained at
21.8 percent. Under dictatorship, in turn, any change of head of gov-
ernment wreaks havoc. Whereas dictatorships grew at the rate of 4.64
and invested 14.2 percent during the 2,198 years when a dictator
remained in power, growth fell to 3.29, with investment at 14.4
percent, during the 223 years when just one change occurred, and to
0.95, with investment falling to 12.9 percent, during the years when
power changed more than once (twice in forty-two years, thrice in
seven years, and five times in one year).8

7
Thirty-nine out of 351 dictators, or 11.1 percent, died while in office, as contrasted with
25 out of 441, or 5.7 percent, leaders of democratic governments. Dictators are partic-
ularly vulnerable to being airborne: Bartholomy Boganda in the Central African Repub-
lic in 1959, Abdel Salam Muhammad Aref in Iraq in 1966, René Barrientos in Bolivia
in 1969, Omar Torrijos, who was the éminence grise in Panama, in 1981, Samora
Machel in Mozambique in 1986, and Muhammed Zia ul-Haq in Pakistan in 1988 all died
under such circumstances.
8
These findings are consistent with the results of Alesina, Özler, Roubini, and Swagel
(1996), who distinguished among coups and “major” and “regular” government
changes, to discover that growth is affected by the first two, but not by routine gov-
ernment alternations.

191
 Political Instability and Economic Growth

Table 4.1. Selection-corrected effects of socio-

political unrest on growth

Variable Dictatorships Democracies

N 2,341 1,552
Method OLS 2F
STRIKES
Mean 0.09 0.24
Coefficient -0.2308 -0.1182
(Standard error) (0.1694) (0.1302)
RIOTS
Mean 0.51 0.78
Coefficient -0.1518 -0.0194
(Standard error) (0.0462) (0.0349)
DEMONSTRATIONS
Mean 0.49 0.88
Coefficient -0.1443 -0.0123
(Standard error) (0.0500) (0.0349)
UNREST
Mean 1.08 1.90
Coefficient -0.0842 -0.0114
(Standard error) (0.0248) (0.0180)
Predicted growth 4.33 4.29
(Standard error) (5.10) (4.38)
Cost of unrest -0.1166 -0.0158
(Standard error) (0.3600) (0.0487)

Notes: UNREST is the sum of STRIKES, DEMONSTRATIONS,

and RIOTS. F tests always indicate OLS for dictatorships and
two-way fixed effects for democracies. The selection model
is based on lagged per capita income (LEVLAG), lagged sum
of past transitions to dictatorship (STRAL), and an index of
religious fractionalization (RELDIF). The growth model
includes the growth of capital stock (KSG), growth of the
labor force (LFG), and the respective UNREST variables.

Finally, “socio-political unrest” (UNREST),9 though more frequent

under democracy, affects growth only under dictatorship (Table 4.1). If
strikes, anti-government demonstrations, and riots are considered as
constituting “instability,” dictatorships are much more “stable” than

9
UNREST is nothing else but what was called MOBILIZATION in Chapter 2, namely, the
sum of general strikes (STRIKES), anti-government demonstrations (AGDEMONS), and
riots (RIOTS). The very terminology reveals the ideological assumptions: What is seen

192
Past Instability

democracies: Strikes are almost three times as frequent in the latter,

and demonstrations and riots almost twice as frequent. Yet they reduce
growth only in dictatorships; democracies somehow live with “unrest”
without any economic consequences. When the frequencies of these
phenomena are added together, to form an index of “unrest,” the effect
on growth under dictatorships is almost eight times larger than under
democracies, and even though unrest is half as frequent under
dictatorships, its cost in terms of growth is more than seven times
greater.
Hence, political upheavals affect contemporaneous economic per-
formance, and they are endogenous to the two political regimes and,
in turn, affect them differently. Wars are more frequent and more dev-
astating under dictatorships, which also recover faster from the
destruction they cause. Regime transitions are accompanied by eco-
nomic crises that are deep but also short-lived. Although more fre-
quent, alternation in office has little effect on economic performance
under democracy, but it throws dictatorships into economic disarray.
Finally, again more frequent under democracy, strikes, demonstrations,
and even riots have no effect on their economies, whereas they are
economically costly under dictatorships.

Past Instability
Past political upheavals may affect present economic performance
by slowing the accumulation of stocks of productive inputs. We first
examine the impact of the past on the present by studying the effects
of the number of past regime transitions (STTR = STRA + STRD)10 and
of the number of changes of chief executives during the life of a regime
(ACCHEAD).
Past regime transitions were found to have no visible effect on
current growth rates except for the few cases in which a country had
experienced more than four regime changes in its past. As Table 4.2
shows, the average growth rate among the 2,884 instances in which a
country had experienced no regime transition in the past was more or
less the same as in the 1,171 cases in which the number of past

as “mobilization” in the literature on transitions to democracy becomes “unrest” in the

literature on political instability. We follow these usages to better relate our findings to
the respective bodies of literature.
10
Remember that to discount somewhat the effect of the distant past, we code any number
of regime transitions before 1950 as one.

193
 Political Instability and Economic Growth

Table 4.2. Growth of income (YG), by the number of past regime

transitions (STTR) and the current regime (REG)

Average growth rate

All Dictatorships Democracies

Number of past
Transitions YG N YG N YG N

0 4.20 2,884 4.36 1,858 3.90 1,026

1 4.63 660 4.80 363 4.42 297
2 4.43 321 5.60 117 3.76 204
3 3.86 142 3.84 89 3.90 53
4 4.43 48 3.71 21 4.98 27
5 2.39 30 1.93 18 3.09 12
6 2.14 22 3.69 7 1.42 15
7 2.19 4 -0.54 1 3.10 3
8 0.38 7 0.38 7 — 0
9 -0.45 8 — 0 -0.45 8

transitions ranged from one to four.11 Only in the seventy-one cases in

which the past was exceptionally stormy was growth slower.12 This
pattern holds for both regimes.13
Past changes of chief executives during the life of a particular regime
again appear to matter only in those cases in which government insta-
bility was exceptionally high (Table 4.3). The rate of growth among the
regimes that experienced one to five changes of government heads in
the past, 4.28 percent, was almost the same as that among those gov-
erned by the founding ruler, 4.54. Only when the number of past
changes was greater than five did growth slow down, and, again, that
effect was more pronounced for the dictatorships. When introduced

11
Note that studies that average over time and analyze relationships among those aver-
ages lump together past and future instability. If we were to proceed in that fashion,
we would want to know the average rates of economic growth of countries distinguished
by the number of regime transitions during the entire period 1950–1990. The result
would be the same: Only extreme instability, six or more regime changes, was associ-
ated with lower growth rates.
12
These cases include two years in Sudan, nine each in Guatemala and Honduras, twenty-
three years in Peru, and twenty-eight in Argentina.
13
When introduced into our basic growth model, past regime transitions have no impact
on current growth rates, whether in OLS or in fixed-effects panels.

194
Past Instability

Table 4.3A. Growth of income (YG), by the number of past changes of

chief executives (HEADS) and the current regime (REG)

Average growth rate

All Dictatorships Democracies

Number of past
changes of heads YG N YG N YG N

0 4.54 1,473 4.59 1,114 4.38 359

1–5 4.28 2,068 4.44 1,246 4.02 822
>5 3.41 534 3.05 110 3.50 424

Table 4.3B. Growth of income (YG), by the frequency of past changes of

chief executives (TURNOVER) and the current regime (REG)

Average growth rate

All Dictatorships Democracies

Frequency
Turnover of changes YG N YG N YG N

£0.1 >10 years 4.44 2,010 4.46 1,580 4.33 430

0.1–0.33 3–10 years 3.84 1,433 4.19 654 3.54 779
>0.33 <3 years 4.64 632 5.06 236 4.40 396

into (as always, selection-corrected) regression analysis, past changes

of heads play no role whatsoever under either regime, regardless how
one estimates the model. Even a dummy variable for more changes
than five is statistically irrelevant.
Because we are considering the numbers of changes of heads of gov-
ernment during the life of each regime, and because regimes differ in
their longevity, another way to consider the past changes is to mea-
sure the frequency of such changes relative to the age of the regime
(TURNOVER). An eye inspection of the data indicates that growth
is slower under both regimes when the accumulated frequency of
changes is between 0.1 and 0.33, meaning that heads of government
change every ten to three years. Regimes in which chief executives

195
 Political Instability and Economic Growth

change less or more frequently grow faster. Yet, again, turnover of

heads proves to be irrelevant in statistical analyses, even when we
introduce a non-linear effect.
All that one can conclude is that past instability, whether of regimes
or of heads of governments, has no effect on present rates of growth,
perhaps unless the instability is extreme. Although contemporaneous
political upheavals do retard, somewhat, accumulation of the physi-
cal capital stock (and perhaps also of human capital stock, which
we have not studied), these effects must wane as time passes. One
plausible explanation is that investment is simply postponed until
things quiet down and then implemented at levels that compensate for
the waiting. Remember that even the effects of wars do not linger for
a long time.
Our findings are thus consistent with those of Londregan and Poole
(1990), but go against the results of Alesina and Perotti (1997). The
former find that poverty increases the probability of coups and that
once a coup occurs, future coups are more likely, and yet coups do not
affect growth. Hence, Londregan and Poole speak of a “coup trap.” The
latter authors also find that past instability propagates instability, but
claim that past instability retards growth: Hence Alesina and Perotti
find a “poverty trap.” We find that past regime transitions and past
turnover of chief executives increase the probability that such changes
will occur again, but that they do not affect growth. Hence, ours is an
“instability” but not a “poverty” trap.
What, then, of the rival hypothesis, due to Olson (1982), namely, that
long-lasting democratic regimes ossify interest-group influence, thus
increasing pressures for distribution at the cost of development? If
that hypothesis is true, then the age of the regime (AGER), a proxy
for interest-group influence, should have a negative effect on growth
(Table 4.4).
Although several studies, as summarized by Mueller (1989: chap.
16), have contended that older democracies grow slower, we do not
find such an effect. When we replicate those studies, considering, as
they do, only democracies in the OECD countries, we find that the coef-
ficient on regime age is -0.0095, with a t value of -3.409, highly sig-
nificant. In the full sample, however, the effect of regime age vanishes.
We think that the reason for this discrepancy is the following: It is by
now generally accepted that among the developed countries the rate
of growth tends to decline in per capita income. In turn, as democra-
cies in these countries become older, they also have higher income
levels. Hence, the convergence effect is mistaken for the effect of regime

196
 Past Instability

Table 4.4. Effect of regime age (AGER) on the growth of income (YG);
selection-corrected panel estimates

Dictatorships: OLS estimates

OLS without group dummy variables
Ordinary least-squares regression: weighting variable = ONE; dependent variable Y;
mean = 4.42613; s.d. = 6.9409
Model size: observations = 2,396; parameters = 5; degrees of freedom (d.f.)
= 2,391
Residuals: sum of squares = 40,480.7; standard deviation = 4.11466
Fit: R2 = 0.64916; adjusted R2 = 0.64857
Model test: F(4, 2,391) = 1,106.00; probability value = 0.00000
Diagnostic: log-likelihood = -6,786.5524; restricted (b = 0) log-likelihood =
-8,041.3516; Amemiya probability criterion = 16.966; Akaike information
criterion = 5.669
Panel data analysis of Y (one-way): unconditional ANOVA (no regressors)
Source Variation Degrees of freedom Mean square

Between 9,728.20 100 97.2820

Residual 105,652 2,295 46.0359
Total 115,381 2,395 48.1757

Standard
Variable Coefficient error (s.e.) z = b/s.e. P[|Z| ≥ z] Mean of X

l -1.161860 0.33903 -3.427 0.00061 0.7599E-01

KSG 0.4002971 0.60358E-02 66.320 0.00000 7.236
LFG 0.3113418 0.43562E-01 7.147 0.00000 2.284
AGER -0.2594649E-02 0.24524E-02 -1.058 0.29006 33.80
Constant 0.9945064 0.17062 5.829 0.00000

Test statistics for the classical model

Model Log-likelihood Sum of squares R2

(1) Constant term only -8,041.35156 0.115381E+06 0.0000000

(2) Group effects only -7,935.82966 0.105652E+06 0.0843140
(3) X variables only -6,786.55239 0.404807E+05 0.6491553
(4) X and group effects -6,755.64571 0.394497E+05 0.6580909
(5) X individual and time effects -6,723.90605 0.384183E+05 0.6670304

(continued)

197
Table 4.4 (continued)

Hypothesis tests

Likelihood ratio test F tests

Comparison Chi-squared d.f. Prob. value F Num. Denom. Prob. value

(2) vs. (1) 211.044 100 0.00000 2.113 100 2,295 0.00000
(3) vs. (1) 2,509.598 4 0.00000 1,105.995 4 2,391 0.00000
(4) vs. (1) 2,571.412 104 0.00000 31.255 104 2,291 0.00000
(4) vs. (2) 2,360.368 4 0.00000 976.139 4 2,291 0.00000
(4) vs. (3) 61.813 100 0.99860 0.599 100 2,291 0.99940
(5) vs. (4) 63.479 38 0.00589 1.592 38 2,253 0.01251
(5) vs. (3) 125.293 139 0.79201 0.870 139 2,253 0.85678

Democracies: two-way (country and year) fixed-effects estimates

Least squares with group dummy variables and period effects
Ordinary least-squares regression: weighting variable = ONE; dependent variable Y;
mean = 3.91555; s.d. = 4.8232
Model size: observations = 1,595; parameters = 115; degrees of freedom
(d.f.) = 1,480
Residuals: sum of squares = 14,350.0; standard deviation = 3.11383
Fit: R2 = 0.61276; adjusted R2 = 0.58293
Model test: F(114, 1,480) = 20.54; probability value = 0.00000
Diagnostic: log-likelihood = -4,015.2140; restricted (b = 0) log-likelihood =
-4,772.3475; Amemiya probability criterion = 10.395; Akaike information
criterion = 5.179
Estimated autocorrelation of e(i, t): 0.059719
Standard
Variable Coefficient error (s.e.) z = b/s.e. P[|Z| ≥ z] Mean of X

l -0.4256338E-01 0.24952 -0.171 0.86455 -0.1141

KSG 0.3448888 0.85873E-02 40.162 0.00000 6.489
LFG 0.4015232 0.11343 3.540 0.00040 1.595
AGER 0.9236399E-02 0.15375E-01 0.601 0.54802 36.33
Constant 0.6969873 0.61231 1.138 0.25500

Test statistics for the classical model

Model Log-likelihood Sum of squares R2

(1) Constant term only -4,772.34752 0.370820E+05 0.0000000

(2) Group effects only -4,686.83567 0.333116E+05 0.1016764
(3) X variables only -4,126.84596 0.165060E+05 0.5548778
(4) X and group effects -4,079.73793 0.155593E+05 0.5804095
(5) X individual and time effects -4,015.75267 0.143597E+05 0.6127591

(continued)
198
 Past Instability

Table 4.4 (continued)

Hypothesis tests

Likelihood ratio test F tests

Comparison Chi-squared d.f. Prob. value F Num. Denom. Prob.value

(2) vs. (1) 171.024 71 0.00000 2.428 71 1,523 0.00000

(3) vs. (1) 1,291.003 4 0.00000 495.513 4 1,590 0.00000
(4) vs. (1) 1,385.219 75 0.00000 19.459 75 1,519 0.00000
(4) vs. (2) 1,214.195 4 0.00000 462.555 4 1,519 0.00000
(4) vs. (3) 94.216 71 0.03411 1.302 71 1,519 0.04942
(5) vs. (4) 127.971 38 0.00000 3.256 38 1,481 0.00000
(5) vs. (3) 222.187 110 0.00000 2.012 110 1,481 0.00000

Selection-corrected predicted values of YG

Variable Mean s.d. Cases

Dictatorships 4.3283 4.8856 3,991

Democracies 4.2176 4.3353 3,991

age. Once convergence is removed, by including democracies in less-

developed countries, the impact of regime age vanishes.14
Finally, given what we already know about the effect of past changes
of chief executives, unsurprisingly we find that the duration of tenure
of a particular head of government (AGEH) has no impact on the
current rate of growth under either regime.
To summarize again, past instability, whether of regimes or of gov-
ernments, does not matter for economic growth one way or another.
Although we found some descriptive indications that extremely fre-
quent changes of regimes may reduce growth, statistical analyses
lead to the conclusion that growth of the economy is not affected
by past regime changes, nor past government changes, nor the age
of the current regime, nor the length of tenure of the current chief
executive.
14
Our confidence in this interpretation is somewhat mitigated by the finding that once
the fixed country and year effects are considered (and they are indicated by the F test),
the effect of per capita income on the rate of growth turns out to be positive, even when
the sample is limited to countries with incomes above $3,000. Unfortunately, regime
age and per capita income are highly correlated (0.76), and we could not find an instru-
ment for LEVEL that would not be correlated with AGER.

199
 Political Instability and Economic Growth

Expectations
Unless effects of the past cumulate or unless the past is used to fore-
cast the future, there is no reason that past instability should affect
current economic performance. Political and economic actors make
decisions looking ahead, not behind. What matters for their decisions
are the consequences their actions will generate. And this means that
these decisions are based on beliefs about the future, expectations.
But how can we determine what people expect? Note that the fact
that past instability does not affect current performance implies that
the relevant actors do not extrapolate the past into the future.15 How,
then, do they form expectations? Our strategy is to assume that they
know all that we can learn from studying the observable patterns, and
probably much more. We have observed, for example, that democra-
cies survive unconditionally in wealthy countries, and we impute the
same belief to the protagonists. We have learned through statistical
analyses that past turnover of heads of governments makes it less likely
that the current chief executive will survive long in office: Again, we
impute this belief to the actors. Undoubtedly, agents’ expectations are
also based on factors we cannot observe, so that they are much more
refined than our imputations. Nevertheless, if we can predict with rea-
sonable accuracy whether or not regimes or heads of governments will
survive the current year, we have at least some basis on which to
impute expectations to the actors.
Because earlier we subjected regime stability to extensive analysis,
we can base our imputations of expectations concerning regime sur-
vival on those findings. Using the dynamic probit model, described in
Chapter 2, we calculate the probability that a regime will not survive
the current year, conditional only on variables independent of anyone’s
actions.16 The probability that a democracy will not survive the current
year is PDA, and that for a dictatorship is PAD.
To construct expectations concerning the survival in power of
heads of governments, we proceed in a similar manner. We estimate

15
Clague, Keefer, Knack, and Olson (1997: 102) take the past duration of a regime as the
proxy for the expectation of its life. Yet we know from Chapter 2 that, conditioned on
per capita income, hazard rates for democracies are constant with regard to regime
age, whereas for dictatorships they increase somewhat in duration. Hence, this is not
a good proxy.
16
They are (lagged one year whenever appropriate): the sum of past regime transitions
(STTR), the proportion of other democracies in the region (ODRP), the index of reli-
gious fractionalization (RELDIF), the proportion of Moslems (MOSLEM), a dummy for
post-colonial status (NEWC), and a dummy for the OECD countries.

200
Expectations

(fixed-effects panel) regressions in which the dependent variable is

the number of changes of chief executives during a particular year
(HEADS) and calculate the predicted probability that a change of a chief
executive will occur during this year (HEADHAZ). Note again that all
the predetermined variables are exogenous with regard to economic
performance and that they are all different from those used to deter-
mine regime transition probabilities.17
To assess the impact of expectations on current economic growth,
we study a model in which probabilities predicted for the next year are
assumed to affect the behavior of the economy this year. Hence, the
statistical model is
Y˙ (t ) Y (t ) = A A + a K˙ (t ) K (t ) + b L˙ (t ) L(t ) + gHAZ(t + 1) + Ql(t ) + e(t ),
where HAZ stands either for the probability of regime transitions
(EXPPDA, EXPPAD) or for the probability of changes of chief
executives (EXPHAZH), and all other variables are the same as in
Chapter 3.18
Here is the intuition. Political and economic actors in a particular
country make decisions at t anticipating the likelihood that, given the
conditions independent of their actions, the regime and/or the current
head of government will survive or fall at t + 1. Given exogenous con-
ditions at t, rulers – whoever makes public policy – face the choice
between engaging in pillage and running away or making every effort
to enhance the welfare of the relevant political actors so as to survive
in power. When rulers face a high exogenous probability that they will
be forced out of power or that the regime will collapse, they will extract
all they can: At least that is the general line of the economic analyses
of the predatory state (Olson 1993, 1997; McGuire and Olson 1996). In
turn, when the exogenous probability of survival is high, rulers may
be better off extracting low rents and maximizing their chances of
remaining in power. Hence, the empirical hypothesis is that insecure

17
They include the degree to which the military influences politics (CIVMIL), lagged past
turnover of heads of governments (TLAG), and two dummy variables indicating whether
or not either legislative or presidential elections took place this year (LEGELEC, PRE-
SELEC).
18
Though adding a correction for regime selection, this approach follows Londregan and
Poole (1990, 1992), as well as Cukierman et al. (1992) and Alesina et al. (1996). Note
that the PAD and PDA variables measure the probability of a regime dying during the
next year given the conditions this year, and the instrument for regime selection,
LAMBDA, is the marginal probability that the regime would have been different this
year had the present conditions been slightly different. Hence, these variables are inde-
pendent, and in fact uncorrelated.

201
 Political Instability and Economic Growth

rulers, those with a high probability of being forced out of office, will
generate bad economic performance. We should thus observe that the
higher the hazards facing the regime and the head of government, the
lower should be the rate of growth.
Table 4.5 presents results of statistical analyses that include both
regime and head hazards (the results are the same when these two
threats are considered separately). Threats to regimes have no effect

Table 4.5. Effects of expected hazards to regime (EXPHAZR) and to heads of

governments (EXPHAZH) on the growth of income (YG); selection-corrected
two-way (country and year) fixed-effects estimates

Dictatorships: OLS estimates

OLS without group dummy variables
Ordinary least-squares regression: weighting variable = ONE; dependent variable Y;
mean = 4.50829; s.d. = 6.9608
Model size: observations = 2,312; parameters = 6; degrees of freedom (d.f.) =
2,306
Residuals: sum of squares = 39,722.4; standard deviation = 4.15038
Fit: R2 = 0.64526; adjusted R2 = 0.64449
Model test: F(5, 2,306) = 838.89; probability value = 0.00000
Diagnostic: log-likelihood = -6,568.0204; restricted (b = 0) log-likelihood =
-7,766.0517; Amemiya probability criterion = 17.270; Akaike information
criterion = 5.687
Panel data analysis of Y (one-way): unconditional ANOVA (no regressors)
Source Variation Degrees of freedom Mean square

Between 9,672.29 100 96.7229

Residual 102,302 2,211 46.2698
Total 111,975 2,311 48.4530

Standard
Variable Coefficient error (s.e.) z = b/s.e. P[|Z| ≥ z] Mean of X

l -0.9847312 0.36536 -2.695 0.00703 0.7450E-01

KSG 0.3987071 0.61948E-02 64.361 0.00000 7.416
LFG 0.3074352 0.46609E-01 6.596 0.00000 2.267
EXPHAZH -1.398653 0.53845 -2.598 0.00939 0.1456
EXPPAD -0.9620452 2.7833 -0.346 0.72961 0.2271E-01
Constant 1.153648 0.16993 6.789 0.00000

(continued)

202
 Expectations

Table 4.5 (continued)

Test statistics for the classical model

Model Log-likelihood Sum of squares R2

(1) Constant term only -7,766.05163 0.111975E+06 0.0000000

(2) Group effects only -7,661.61895 0.102302E+06 0.0863792
(3) X variables only -6,568.02038 0.397224E+05 0.6452561
(4) X and group effects -6,534.99719 0.386037E+05 0.6552466
(5) X individual and -6,506.03398 0.376485E+05 0.6637770
time effects

Hypothesis tests

Likelihood ratio test F tests

Prob. Prob.
Comparison Chi-squared d.f. value F Num. Denom. value

(2) vs. (1) 208.865 100 0.00000 2.090 100 2,211 0.00000
(3) vs. (1) 2,396.063 5 0.00000 838.893 5 2,306 0.00000
(4) vs. (1) 2,462.109 105 0.00000 29.702 105 2,206 0.00000
(4) vs. (2) 2,253.244 5 0.00000 738.928 5 2,206 0.00000
(4) vs. (3) 66.046 100 0.99552 0.639 100 2,206 0.99779
(5) vs. (4) 57.926 37 0.01546 1.487 37 2,169 0.03001
(5) vs. (3) 123.973 138 0.79866 0.866 138 2,169 0.86427

Democracies: two-way (country and year) fixed-effects estimates

Least squares with group dummy variables and period effects
Ordinary least-squares regression: weighting variable = ONE; dependent variable Y;
mean = 4.04603; s.d. = 4.7533
Model size: observations = 1,482; parameters = 111; degrees of freedom (d.f.) =
1,371
Residuals: sum of squares = 13,268.4; standard deviation = 3.11093
Fit: R2 = 0.60318; adjusted R2 = 0.57134
Model test: F(110, 1,371) = 18.94; probability value = 0.00000
Diagnostic: log-likelihood = -3,727.1336; restricted (b = 0) log-likelihood =
-4,412.5516; Amemiya probability criterion = 10.403; Akaike information
criterion = 5.180
Estimated autocorrelation of e(i, t): 0.064337

(continued)

203
 Political Instability and Economic Growth

Table 4.5 (continued)

Standard
Variable Coefficient error (s.e.) z = b/s.e. P[|Z| ≥ z] Mean of X

l -0.8218807E-02 0.25353 -0.032 0.97414 -0.1162

KSG 0.3407804 0.89056E-02 38.266 0.00000 6.698
LFG 0.4057808 0.12025 3.374 0.00074 1.605
EXPHAZH 0.2288663E-01 0.40964 0.056 0.95545 0.2859
EXPPDA 7.133481 8.7859 0.812 0.41683 0.2423E-01
Constant 0.9317520 0.33272 2.800 0.00510

Test statistics for the classical model

Model Log-likelihood Sum of squares R2

(1) Constant term only -4,412.55157 0.334608E+05 0.0000000

(2) Group effects only -4,341.49816 0.304014E+05 0.0914347
(3) X variables only -3,821.90677 0.150787E+05 0.5493623
(4) X and group effects -3,787.81487 0.144007E+05 0.5696255
(5) X individual and -3,727.67390 0.132781E+05 0.6031756
time effects

Hypothesis tests

Likelihood ratio test F tests

Prob. Prob.
Comparison Chi-squared d.f. value F Num. Denom. value

(2) vs. (1) 142.107 67 0.00000 2.124 67 1,414 0.00000

(3) vs. (1) 1,181.290 5 0.00000 359.872 5 1,476 0.00000
(4) vs. (1) 1,249.473 72 0.00000 17.952 72 1,409 0.00000
(4) vs. (2) 1,107.367 5 0.00000 335.077 5 1,409 0.00000
(4) vs. (3) 68.184 67 0.43675 0.990 67 1,409 0.50093
(5) vs. (4) 120.282 37 0.00000 3.135 37 1,372 0.00000
(5) vs. (3) 188.466 105 0.00000 1.772 105 1,372 0.00001

on the rate of growth under either regime. But threats to dictators

cause economic performance to deteriorate, whereas similar threats
to democratic chief executives have no effect whatsoever. Even though
changes of chief executives are relatively infrequent in dictatorships,

204
Expectations

the average cost of dictators’ insecurity is to lower the annual growth

rate by 0.20, but under democracies this cost is zero.19
We cannot tell whether this difference is due to the institutional con-
straints or to motivations of rulers. One could think that both dictators
and heads of democratic governments would have predatory motiva-
tions – they seek to extract rents – but dictators have more influence
over the economy than do individual democratic rulers, who are con-
strained by various checks and balances. Democratic rulers can manip-
ulate the economy somewhat – at least some studies have found
evidence for electoral business cycles (for a review, see Alesina 1995)
– but the institutional constraints are severe.
Yet it is also conceivable that public officials in democracies are more
likely to be public-spirited, motivated to selflessly serve their countries.
Motivations of political rulers have an endogenous component as well as
an exogenous component. People who want to become wealthy do not
seek office under democracy, because holding democratically elected
office is not the best way to get rich. A Somoza or a Marcos could not
stash billions in Swiss banks under democracy: They would have been
thrown out of office or would have ended up in jail or both. Hence, one
could expect that the accountability mechanisms characteristic of each
regime separate the types of people who seek power. The accountability
mechanism is a menu of contracts that specify sanctions ( jail, out of
office, in office) depending on outcomes (theft, bad but legal perfor-
mance, good performance). Hence, only people for whom the particular
contracts are incentive-compatible will take them. True, as always when
information is asymmetric, rents are unavoidable. But some people who
would accept a contract “in office whatever” would not accept one saying
“in office only if rents are low.” Thus, even if the distribution of motiva-
tions is the same in all countries, the recruitment of types to power is
endogenous with regard to political regimes.
Yet, whatever the source of the difference between regimes, the fact
is that institutions matter: Threats to dictators produce the effect that
is generally expected, that of “instability,” but similar threats to demo-
cratic rulers do nothing of the sort.
The effect of expected political changes on investment is a more
complicated topic, even if typically awash in banalities.20 Most students

19
The average cost of insecurity is the product of the average incidence of unrest and its
regression coefficient on growth.
20
Eager to demonstrate that political instability reduces investment, several studies have
proceeded as if only the variance and not the expected return determined the optimal
rate of investment. Yet, if investment is reversible or if it cannot be delayed, then the

205
 Political Instability and Economic Growth

of political instability believe that it reduces investment. Indeed, the

typical chain of reasoning proceeds from political instability (some-
times seeing instability as caused in turn by income inequality) to
investment and only then to growth: The effect of instability on
growth is seen as being mediated by its effect on investment. The rit-
ualistically repeated claim is that political instability threatens the secu-
rity of property rights, inducing uncertainty, which in turn reduces
investment.
Without additional distinctions, that story makes little sense. Con-
sider expectations that the political leader may change. Investors may
expect that when the head of government changes, so may policy:
Hence, expectations of a change of chief executive would induce uncer-
tainty among investors. They may also expect that insecure rulers are
more likely to engage in pillage. If investment entails fixed costs, uncer-
tainty may cause investors to postpone their projects until it resolves
itself or, if they can, to reallocate them elsewhere. But the fear of
increased rent extraction by the ruler is more likely to affect the uti-
lization of the already installed capacity than to affect investment: A
ruler who is about to grab all he can and run away will not take with
him a newly constructed building or a newly purchased machine, but
rather the output from the already installed factories. Hence, the fear
of pillage by an outgoing ruler should lead private actors to reduce pro-
duction, not investment. Indeed, diverting resources to investment may
be a way of protecting one’s liquid resources. This is why we find that
where rulers can pillage – and this is true only of dictatorships – the
rate of growth of output does fall. But because uncertainty and fear of
confiscation may work in opposite directions, the effect on investment
of expectations that dictators will change cannot be foretold. And,
indeed, we learn that threats to chief executives have no effect on
investment under either regime (Table 4.6).
Threats to regimes are different matters. What one would expect
about their effect on investment would depend on what one thinks of
the impact of political regimes on the security of property rights: a topic
shrouded in ideologically motivated confusion. One recently popular
argument is that democracies safeguard property rights. North was

amount of optimal investment will depend only on the expected rates of return. If invest-
ment is irreversible (i.e., assets can be sold only at a price different from that at which
they were purchased) and if it can be postponed, the amount of optimal investment
will decline as the value of waiting (which in turn increases in uncertainty) becomes
greater. Yet even if optimal investment declines under most circumstances as uncer-
tainty increases, it still does increase in the expected return. See Pindyck (1991).

206
Table 4.6. Effects of expected hazards to regime (EXPHAZR) and to heads of
governments (EXPHAZH) on the share of investent in GDP (INV ); selection-
corrected, AR1, one-way (country) fixed-effects estimates

Dictatorships
Least squares with group dummy variables
Ordinary least-squares regression: weighting variable = ONE; dependent variable Y;
mean = 4.53340; s.d. = 3.7121
Model size: observations = 2,211; parameters = 108; degrees of freedom (d.f.) =
2,103
Residuals: sum of squares = 15,023.2; standard deviation = 2.67277
Fit: R2 = 0.50667; adjusted R2 = 0.48157
Model test: F(107, 2,103) = 20.19; probability value = 0.00000
Diagnostic: log-likelihood = -5,255.5781; restricted (b = 0) log-likelihood =
-6,036.6940; Amemiya probability criterion = 7.493; Akaike information
criterion = 4.852
Estimated autocorrelation of e(i, t): 0.683293
Standard
Variable Coefficient error (s.e.) z = b/s.e. P[|Z| ≥ z] Mean of X

l -0.3571177E-01 0.26036 -0.137 0.89090 0.9196E-02

LEVEL 1.397619 0.22572 6.192 0.00000 0.6364
PINV -0.3100040E-02 0.17128E-02 -1.810 0.07030 34.24
WORLD -0.1149058 0.41293E-01 -2.783 0.00539 0.8947
PRIME 0.1120425 0.29213E-01 3.835 0.00013 2.899
EXPHAZH -0.1277436 0.48291 -0.265 0.79137 0.5240E-01
EXPPAD -23.36819 8.8062 -2.654 0.00796 0.7952E-02

Test statistics for the classical model

Model Log-likelihood Sum of squares R2

(1) Constant term only -6,036.69400 0.304525E+05 0.0000000

(2) Group effects only -5,297.85375 0.156088E+05 0.4874368
(3) X variables only -5,708.17003 0.226237E+05 0.2570840
(4) X and group effects -5,255.57810 0.150232E+05 0.5066678

Hypothesis tests

Likelihood ratio test F tests

Prob. Prob.
Comparison Chi-squared d.f. value F Num. Denom. value

(2) vs. (1) 1,477.681 100 0.00000 20.066 100 2,110 0.00000
(3) vs. (1) 657.048 7 0.00000 108.906 7 2,203 0.00000
(4) vs. (1) 1,562.232 107 0.00000 20.185 107 2,103 0.00000
(4) vs. (2) 84.551 7 0.00000 11.711 7 2,103 0.00000
(4) vs. (3) 905.184 100 0.00000 10.639 100 2,103 0.00000

(continued)
Table 4.6 (continued)

Democracies
Least squares with group dummy variables
Ordinary least-squares regression: weighting variable = ONE; dependent variable Y;
mean = 6.00054; s.d. = 2.9618
Model size: observations = 1,413; parameters = 73; degrees of freedom (d.f.) =
1,340
Residuals: sum of squares = 6,409.38; standard deviation = 2.18704
Fit: R2 = 0.48253; adjusted R2 = 0.45473
Model test: F(72, 1,340) = 17.35; probability value = 0.00000
Diagnostic: log-likelihood = -3,073.2221; restricted (b = 0) log-likelihood =
-3,538.6734; Amemiya probability criterion = 5.030; Akaike information
criterion = 4.453
Estimated autocorrelation of e(i, t): 0.717687
Standard
Variable Coefficient error (s.e.) z = b/s.e. P[|Z| ≥ z] Mean of X

l 0.2002355E-01 0.17690 0.113 0.90988 -0.3422E-02

LEVEL 0.2083183 0.10095 2.064 0.03905 1.856
PINV -0.7306690E-02 0.23773E-02 -3.074 0.00212 23.77
WORLD 0.1191477 0.40349E-01 2.953 0.00315 0.8116
PRIME 0.5693934E-01 0.33139E-01 1.718 0.08576 2.393
EXPHAZH 0.1648749 0.20816 0.792 0.42832 0.8292E-01
EXPPDA 13.30201 10.248 1.298 0.19429 0.6666E-02

Test statistics for the classical model

Model Log-likelihood Sum of squares R2

(1) Constant term only -3,538.67337 0.123861E+05 0.0000000

(2) Group effects only -3,088.31220 0.654775E+04 0.4713633
(3) X variables only -3,365.31538 0.969103E+04 0.2175895
(4) X and group effects -3,073.22208 0.640938E+04 0.4825347

Hypothesis tests

Likelihood ratio test F tests

Prob. Prob.
Comparison Chi-squared d.f. value F Num. Denom. value

(2) vs. (1) 900.722 65 0.00000 18.478 65 1,347 0.00000

(3) vs. (1) 346.716 7 0.00000 55.819 7 1,405 0.00000
(4) vs. (1) 930.903 72 0.00000 17.355 72 1,340 0.00000
(4) vs. (2) 30.180 7 0.00009 4.133 7 1,340 0.00019
(4) vs. (3) 584.187 65 0.00000 10.555 65 1,340 0.00000

Notes: LEVEL is per capita income; PINV is the relative price of investment goods; WORLD is
the average rate of growth (of per capita income) in the world during the year; PRIME is the U.S.
interest rate.
Expectations

perhaps the first to argue that security of property rights is critical for
growth (North and Thomas 1973; North 1990). According to North and
Weingast (1989: 803): “The more likely it is that the sovereign will alter
property rights for his or her own benefit, the lower the expected
returns from investment and the lower in turn the incentive to invest.
For economic growth to occur the sovereign or government must
not merely establish the relevant set of rights, but make a credible
commitment to them” 21 Yet North was never explicit about the insti-
tutions that would embody that commitment: We could find only one
passage in his book on institutions in which he explicitly identified
those institutions as democratic (1990: 109). This connection was pos-
tulated by Olson (1991: 153), who claimed that a dictator (“autocrat”)
cannot credibly commit himself: “If he runs the society, there is no one
who can force him to keep his commitments.” But Olson as well fails
to explain how democratic institutions could provide such a credible
commitment. His followers merely observe that “for democracies, new
leaders are constrained in their ability to alter the property and con-
tract rights established in previous administrations. Consequently,
investors are concerned with the expected longevity of democracy more
than with the expected tenure of a leader. Autocrats, on the other hand,
are relatively unconstrained in their ability to disregard previously
established rights” (Clague et al. 1997: 102).
The property-rights literature treats the state as the only source of
potential threat. But property rights can be threatened by private
actors: Capitalist property is threatened by organized workers, land-
lords’ property by landless peasants. It is by no means clear that the
villain is necessarily “the ruler.” Indeed, one liberal dilemma is that a
strong state may be a potential threat to property, but a strong state is
required to protect property from private encroachments.
Democracy in the political realm exacerbates the threat to property
from the propertyless by equalizing the right to influence the alloca-
tion of resources. Distributions of consumption caused by the market
and those voted on by citizens must differ, because democracy offers
those who are poor, oppressed, or otherwise miserable as a conse-
quence of the initial distribution of endowments an opportunity to find
redress via the state. Endowed with political power in the form of uni-
versal suffrage, those who suffer as a consequence of private property
will attempt to use that power to expropriate the riches.
21
North and Weingast discovered that in seventeenth-century England, democracy did
secure property rights – a finding not particularly surprising given that only the prop-
ertied enjoyed political rights.

209
 Political Instability and Economic Growth

The economic consequences of democracy were at the center of

debates concerning the rights to vote and to organize during the first
half of the nineteenth century. Conservatives agreed with socialists that
democracy, specifically universal suffrage and the freedom to form
unions, must threaten property. The Scottish philosopher James Mack-
intosh predicted in 1818 that if the “laborious classes” gained fran-
chise, “a permanent animosity between opinion and property must be
the consequence” (cited in Collini, Winch, and Burrow 1983: 98). David
Ricardo was prepared to extend suffrage only “to that part of them [the
people] which cannot be supposed to have an interest in overturning
the right to property” (Collini et al. 1983: 107). Thomas Macaulay, in
his speech on the Chartists in 1842 (1900: 263), pictured universal suf-
frage as “the end of property and thus of all civilization.” Eight years
later, Karl Marx expressed the same conviction that private property
and universal suffrage are incompatible (1952: 62). According to his
analysis, democracy inevitably “unchains the class struggle”: The poor
would use democracy to expropriate the riches; the rich would be
threatened and would subvert democracy, typically by “abdicating”
political power to the permanently organized armed forces. As a result,
either capitalism or democracy would crumble. The combination of
democracy and capitalism was thus seen as an inherently unstable
form of organization of society, “only the political form of revolution of
bourgeois society and not its conservative form of life” (1934: 18), “only
a spasmodic, exceptional state of things . . . impossible as the normal
form of society” (1971: 198). Hence, in that view it is democracy, not
dictatorship, that imperils property.
Our findings offer a surprisingly resounding resolution of that con-
troversy. A 10 percent increase in the expected probability that a dic-
tatorship will fall (EXPPAD) causes the investment share to decline by
1.37 to 2.55.22 In turn, whereas the coefficients on the threats to
democracies are not significant unless one considers that growth and
investment are simultaneously determined by some exogenous vari-
ables (SURE estimator), they are invariably positive and also quite
large. Hence, it is clear what investors think about the security of their
property under each regime. If our measure of investors’ expectations

22
Different estimators yield different values for this effect, but they are always negative
and large. See Appendix 4.1. As one would expect, when we exclude the 147 (out of
2,163) observations of communist regimes and thus consider only those economies in
which there existed private property to be threatened and in which investment was to
a large extent a result of private decisions, the effect of threats to dictatorships is even
stronger.

210
Conclusion

is accurate – we should not ignore the caveat, because this analysis

entails several assumptions – they will reduce investment when dicta-
torships are likely to fall, and most likely will increase their stakes when
democracies are about to die.
The association of democracy with secure property rights is so wide-
spread that economists routinely use the former as a proxy for the
latter. Indeed, Barro (1989: 22) could find in the entire world only three
dictatorships – Chile, South Korea, and Singapore – that were not
hostile to private property. Yet most dictatorships repress labor and
keep wages down (see Chapter 3). On the average, dictatorships tax at
rates no higher than those under democracy: Observed tax rates are
lower under dictatorships, and selection-corrected rates are almost the
same under the two regimes (Cheibub 1998). On the average, dicta-
torships are less likely to nationalize private firms: Between 1960 and
1979, there were 85 acts of nationalization in 326 democratic years,
one in every 3.83 years, and 406 acts of nationalization in 1,255
authoritarian years, or one in every 3.09 years (Kobrin 1980).23
Thus, whereas some dictatorships are hostile to capital – the MNR
in Bolivia in the 1950s, the Peruvian military regime of 1968, the San-
dinista regime in Nicaragua, “socialist” dictatorships in Africa, and the
short-lived military regime in Portugal come to mind – most non-
communist dictatorships have been and continue to be resolutely com-
mitted to defending private property from encroachment by those
without it. Hence, investors seem to know what they are doing. It is
not “instability” they fear, but democracy.

Conclusion
Although much more frequent under democracies, political insta-
bility affects economic performance only under dictatorships.
Only wars are more frequent in dictatorships, and they cause more
short-term damage to the economy. Changes of chief executives, though
much less frequent in dictatorships, are economically costly only in these
regimes. The same is true of various forms of “socio-political unrest”:
Strikes, anti-government demonstrations, and riots occur more often in
democracies, but they retard growth only in dictatorships.
Effects of the past are generally short-lived, so that past instability has
almost no effect on current economic performance. Unless the instabil-
ity is extreme, economies catch up quickly, so that effects of the past wane.

23
We thank Stephen Kobrin for sharing his data.

211
 Political Instability and Economic Growth

Whether because of institutional constraints or because of the moti-

vations of those who govern democracies, expected changes in the
heads of governments have no economic consequences under democ-
racies. But under dictatorships, economic growth slows down signifi-
cantly when the tenure of rulers is threatened. Finally, the effect of
threats to the stability of regimes on investment shows that investors
fear democracy and hope to find a safe haven in many dictatorships:
The prospective demise of a dictatorship causes investors to flee, but
its advent makes them flock.
Thus, we find that political instability retards growth exclusively
under dictatorships. Huntington and other “realists” got it all wrong.
Studies of political instability have been guided by the hypothesis that,
to state it in the words of Alesina and Perotti (1997: 21), “what influ-
ences growth is not so much the type of regime (dictatorship or democ-
racy) but regime instability, that is, the propensity to coups and major
changes of government.” That hypothesis is logically incoherent, for
“political instability” cannot even be defined independently of political
institutions. The notion that alternation in office or other manifestations
of popular opposition, whether strikes or anti-government demonstra-
tions, constitute “instability” under democracy is just ludicrous. Such
phenomena are frequent in democracies precisely because democracy
is a system in which people are free to express their dissatisfaction with
governments. And they are rare in dictatorships because it is to prevent
such forms of expression that dictatorships are established and main-
tained. Democracies are inherently “unstable”: The phenomena that
constitute anomalies, breakdowns of rule, under dictatorship are just
essential, definitional features of democracy. Changes of chief execu-
tives in democracies occur consequent to elections or other regularized
procedures, but under most dictatorships the only way that a ruler can
be changed is by a coup. To discover, as Alesina at al. (1996) do, that
coups reduce growth, but regular alternations in office do not, is not to
find that some forms of “political instability” affect development, but
only that some political events that constitute instability under dictator-
ship do not under democracy.
Once it is understood that the same political phenomena have dif-
ferent meanings under different systems of political institutions, it is
not surprising that economic actors react to them differently. Under
dictatorships, whenever the regime is threatened, whenever rulers in
fact change or are expected to change, whenever workers muster the
courage to strike or masses of people to demonstrate their opposition

212
Appendix 4.1: Estimating the Impact of Expectations

to the government, the economy suffers. Under democracy, everyone

knows that the government will change from time to time, that workers
may strike, and people may express their dissatisfaction with the gov-
ernment in a variety of ways. Hence, when such phenomena do tran-
spire, they evoke at most an economic yawn.

Appendix 4.1: Estimating the Impact of Expectations on

Economic Performance
Although the qualitative results are surprisingly robust, estimates of
the impact of expectations on growth and investment are somewhat
sensitive to the estimating methods. Indeed, we are surprised that the
results are so stable, because (1) growth and investment share may be
interdependent in different ways, (2) except for growth under dicta-
torships, both series have fixed effects, and (3) the investment series
is highly autocorrelated. Because simultaneity, fixed effects, and auto-
correlation can be treated in different ways, we provide the results of
applying different estimators:
1. Panel for growth, YG, only. These are the same results as in Table
4.5 in the text. F tests indicate that the dictatorial sample is homo-
geneous, but the democratic sample has strong country and year
effects.
2. Panel for investment, INV, only. Same as Table 4.6. Because both
series have fixed effects and because both are autocorrelated, we
estimate one-way (country) fixed effects with first-order autocorre-
lation, 1F-AR1.
3. Because investment is autoregressive, as an alternative to AR1 esti-
mates, we lag investment and correct for autocorrelation using one-
way fixed effects and the Hatanaka estimator, 1F-HAT.
4. One way to consider interdependence between growth and invest-
ment is by estimating two-stage least squares, with predicted invest-
ment and predicted growth as instruments. Given the fixed effects,
we estimate a 1F model with AR1 for investment. Hence, the model
is 2SLS-1F-AR1.
5. Finally, instead of thinking of growth and investment as reciprocally
influencing each other, we can consider them as jointly determined
by the predetermined variables, a SURE regression model. We allow
for different values of the autocorrelation (LIMDEP Model 1).
Because this model cannot be estimated with country effects, we use
regional dummies. Hence, the model is SURE-AR1, with regional
dummy variables.

213
 Political Instability and Economic Growth

Table A4.1.1. Alternative estimators of the effects of hazards on growth and

investment

Rate of Growth (YG) Investment (INV)

Model N EXPHAZH EXPPJK EXPHAZH EXPPJK

PANEL
Dictatorship (OLS) 2,312 -1.3987 -0.9620
(0.5385) (2.7831)
Democracy (2F) 1,482 0.0229 7.1334
(0.4096) (8.7850)
1F-AR1
Dictatorship 2,211 -0.1277 -23.3682
(0.4829) (8.8062)
Democracy 1,413 0.1649 13.3020
(0.2082) (10.2480)
1F-HATANAKA
Dictatorship 2,211 -0.1892 -25.5281
(0.4887) (8.3104)
Democracy 1,413 0.2826 10.1023
(0.2948) (6.8620)
2SLS-1F-AR1
Dictatorship 2,211 -1.9042 -10.3226 -0.1712 -19.8435
(0.6721) (6.2826) (0.4889) (8.6305)
Democracy 1,413 0.0180 2.1543 0.1748 14.2092
(0.4177) (8.3964) (0.2087) (10.1490)
SURE-AR1
Dictatorship 2,312 -1.2968 -4.5280 -0.4194 -13.6541
(0.5655) (4.0328) (0.5038) (5.6299)
Democracy 1,482 0.0403 1.8544 0.3668 24.1056
(0.3891) (4.6827) (0.2359) (6.3859)

The results in Table A4.1.1 show that all of the qualitative results
are robust, except for the impact of threats to democracy on invest-
ment. All models show that whenever dictators expect to be thrown
out, growth declines sharply, but threats to democratic chief executives
do not affect growth. Threats to either regime have no impact on
growth. All the models agree, again, that threats to chief executives
have no consequences for investment and that the higher probabilities
of dictatorships falling strongly reduce investment. The impact of
threats to the survival of democracy is always to increase investment,

214
Appendix 4.1: Estimating the Impact of Expectations

but the coefficients are significant only in the SURE model. Otherwise,
the probability that these effects will be positive ranges between 10
and 19 percent.
The growth and investment models in Table A4.1.1 are the same as
in Tables 4.5 and 4.6, respectively. We show only the coefficients on
the expected hazards to heads and to regimes. All the estimates are
selection-corrected. Standard errors are in parentheses.

215
 Chapter Five

Political Regimes and Population

Total Income, Population, and Per Capita Income

Although total income grew faster under dictatorships (at the rate
of 4.42) than under democracies (3.95), the observed average rate of
growth of per capita income was higher under democracy: Per capita
income grew at the rate of 2.46 under democracy and at the rate of
2.00 under dictatorship. The same is true of consumption. Whereas
total consumption increased at the rate of 4.24 under dictatorship and
3.92 under democracy, per capita consumption grew faster under the
latter, at 2.43 per year as opposed to 1.82 under dictatorship.
Because what matters for individual well-being is the growth of
each person’s income and consumption, rather than the development
of the aggregate economy, the impact of regimes on the growth of per
capita income and consumption is what we ultimately care about.
Indeed, Lucas (1988: 3) saw the problem of economic development as
“simply the problem of accounting for the observed pattern, across
countries and across time, in levels and rates of growth of per capita
income.”
That is why most studies take the growth of income per capita as
the variable to be explained. Yet even if per capita income is the correct
focus, treating it as the single explanandum confounds the effects of
two social processes that are to some extent independent: the growth
of total output and the dynamics of. population.1 Note that the rate of
growth of per capita income [G = (Y/ P)/(Y/P)] is the difference between

1
To see the econometric problems entailed in taking per capita income . as the
. dependent .
variable,
. substitute
. . the production-function formulation in place of Y /Y: (Y/ P)/(Y/P) = A/A
+ aK/K + bL/L - P/P. To estimate this equation, one either (1) must assume that the labor
force and the
. population. grow. at the same .rate (as did Przeworski and Limongi 1997),
so that (Y/ P)/(Y/P) = A/A + aK/K + (b - 1)L/L, with an obvious bias if the assumption

216
Total Income, Population, and Per Capita Income

.
the rate of growth of .total income (YG = Y/Y) and the rate of growth of
population (POPG = P/P):
.
(Y P )(Y P ) = Y˙ Y - P˙ P .
Yet growth economists tend either to relegate population growth
to the status of an exogenous datum or to ignore it altogether. As
Barro and Sala-i-Martin (1995: 308) observed, “most modern theo-
ries of growth have assumed that the rate of population growth is an
exogenous constant.” Even textbooks on endogenous growth theory
rarely discuss population growth, with some notable exceptions cited
later.
Because the effects of political regimes, as well as of other factors,
on the growth of the economy and on the growth of population need
not be the same, taking the growth of per capita income as the depen-
dent variable obscures the autonomous contributions of these two
processes. Indeed, the most surprising finding in our entire study is
that regimes matter more for the growth of population than for the
growth of income. In the end, regimes have more to do with demog-
raphy than with economics.
Consider some basic facts: Under dictatorships, population grew
at the rate of 2.42 percent per annum, and under democracies at the
rate of 1.49: At those rates the population will double in thirty-four
years under dictatorship, but only in forty-seven years under democ-
racy. Decomposing the observed rates of growth of per capita income
yields

Regime G = YG - POPG

Dictatorships 2.00 = 4.42 - 2.42

Democracies 2.46 = 3.95 - 1.49
Difference -0.46 = 0.47 - 0.93

Correcting for regime selection attenuates these differences. As we

saw in Chapter 4, our best selection-corrected estimates of the rate of
growth of GDP were 4.30 for dictatorships and 4.24 for democracies.
Our best estimates of selection-corrected rates of population growth
(Table 5.1B, Hatanaka estimator) are 2.18 for dictatorships and 1.59
for democracies. Using these estimates gives

. .
L/L = P/P does not hold, or (2) must assume that the population grows at a constant
exogenous rate (as did Mankiw et al. 1992; Helliwell 1994), or omit labor force and pop-
ulation from the specification altogether (as did Barro 1997).

217
 Political Regimes and Population

Regime G = YG - POPG

Dictatorships 2.12 = 4.30 - 2.18

Democracies 2.65 = 4.24 - 1.59
Difference -0.53 = 0.06 - 0.59

If the regimes had been matched for the exogenous factors that
determine selection as well as for those that drive the growth of total
income and of population, per capita incomes would have grown at
a faster rate under democracies: They would have doubled in about
thirty-five years under dictatorship and in about twenty-six years under
democracy. And note that almost all of this difference is due to differ-
ences in the growth of population, rather than of total income.
Population, therefore, deserves to be studied even if all one wants
to explain is the dynamic of per capita income. Yet the demographic
patterns that cause population to grow slowly or quickly – birth and
death rates, fertility, infant mortality – have profound consequences for
well-being in their own right. They are sources of joys and sorrows;
they represent the quality and the length of lives; they shape the life
prospects of children. They particularly affect women: their choices,
their status within the household, and within the society at large.
The dynamic of population is a complex process. The rate of popu-
lation growth is the difference between birth rates and death rates, and
again regimes may affect these rates differently. Hence, having estab-
lished that population grows faster under dictatorships, we decompose
population growth and examine the effect of regimes on various aspects
of demographic behavior. We then return to the impact of demography
on the growth of per capita incomes.

Political Regimes and Population Growth

Population was found to grow faster under dictatorships than under
democracies. We were bewildered by this fact, and the remainder of
this chapter seeks to explain it. But first we need to establish that it is
a fact.
Table 5.1A shows the observed rates of population growth by levels
of per capita income.2 In countries with per capita incomes under

2
Average income is not the best measure of the wealth of a country for the purposes of
analysis of demographic patterns, because it obscures patterns of distribution. Two
countries may have the same average income, but have different proportions of the pop-
ulation under some particular income levels. If what matters for demographic patterns

218
Political Regimes and Population Growth

Table 5.1A. Observed rates of population growth (POPG), by per capita

income (LEVEL)

Rate of population growth

Proportion
LEVEL dictatorships All Dictatorships Democracies

0–1,000 0.9273 2.641 2.645 2.583

1,001–2,000 0.7472 2.581 2.582 2.578
2,001–3,000 0.6207 2.165 2.290 2.960
3,001–4,000 0.5874 1.906 2.038 1.719
4,001–5,000 0.5424 1.581 1.880 1.228
5,001–6,000 0.4308 1.176 1.526 0.911
6,001– 0.0812 1.022 1.463 0.983
Total (N = 4,126) 2.049 2.423 1.486

Note: All cell entries are based on at least 68 observations.

$2,000, population grows at almost exactly the same rate under the
two regimes. As incomes become larger, population growth slows
down. But the rates at which it falls are sharply different under the
two regimes. Below $2,500, population grew at the rate of 2.58 under
dictatorship (N = 1,851) and 2.47 under democracy (N = 417). Between
$2,500 and $5,000, the gap opened up: The rate of population growth
under dictatorship was 2.10 (N = 478), and 1.61 (N = 348) under
democracy. Above $5,000, population grew at the rate of 1.50 (N = 152)
under the former, and 0.97 (N = 880) under the latter.
With caveats concerning regime selection, this finding is reinforced
by a diachronic analysis. In the countries that were ruled by dictator-
ships during the entire period of our study, population grew at the rate
of 2.43 per annum. In the countries that were observed only as democ-
racies, population increased at the rate of 1.12. This difference is enor-
mous, but it is largely due to the levels of per capita income at which
those regimes were stable. To assess the diachronic impact of regimes,

are absolute income levels, then the country with a more skewed distribution will exhibit
behavior characteristic of a poorer country with a less skewed distribution. When a data
set was sufficiently large, we added a measure of skewness, simply the share of the
median quintile, to all the other variables: This measure always matters, and always in
expected directions. In most cases, however, the overlap between income distribution
and other data is too small to permit statistical analyses.

219
 Political Regimes and Population

Table 5.1B. Selection-corrected estimators of population growth (POPG)

Regime means Regime effect

Dictatorships Democracies Constant Individual

Estimator (N = 2,396) (N = 1,595) (N = 3,991)

Biased 2.42 1.46 0.2238 0.0460

(1.57) (1.11) (0.0745) (0.2964)
OLS-AR1 2.12 1.70
(0.64) (0.52)
HATANAKA 2.18 1.59
(0.59) (0.63)
PANEL 1F/2F 2.43 1.51
(0.85) (0.33)
Unobservable 0.2458 0.0500
(0.0806) (0.2957)
Observable 0.2182 0.0462
(0.0759) (0.3297)

LFPW sample N = 2,076 N = 1,254

Biased 2.49 1.40

(1.50) (1.10)
PANEL 1F/2F 2.50 1.22
(0.80) (0.62)

EDT sample N = 1,745 N = 1,042

Biased 2.45 1.42

(1.38) (1.07)
PANEL 2F/2F 2.37 1.32
(0.19) (0.31)

LFPW and EDT sample N = 1,692 N = 1,015

Biased 2.48 1.42

(1.38) (1.07)
PANEL 2F/2F 2.66 1.28
(0.71) (0.52)

(continued)

220
 Political Regimes and Population Growth

Table 5.1B (continued)

FEMSEC sample N = 1,002 N = 648

Biased 2.49 1.23

(1.24) (1.09)
OLS-AR1 2.17 1.67
(0.67) (0.71)

LFPW, EDT, FEMSEC N = 770 N = 459

Biased 2.51 1.28

(0.87) (1.05)
OLS-AR1 2.35 1.53
(0.41) (0.55)

Notes: Probit equations include the lagged values of LEVEL, STRA, and RELDIF plus the exoge-
nous variables in the LFG equations. These are LEVEL, NEWC, BRITCOL, CATH, PROT, MOSLEM,
COMEX, plus LFGLAG in the Hatanaka estimator, plus the variables defined under each sample.
Standard errors in parentheses.

we examine the countries that experienced regime transitions, thus

controlling the fixed-effects characteristics of particular countries.
Examine the columns in Table 5.2 that exclude the communist dicta-
torships, which had exceptionally slow population growth, as discussed
later. In dictatorships that were to give way to democracies, popula-
tion grew at the rate of 2.33, and the democracies that followed those
dictatorships had a rate of 2.27. In turn, in democracies that were to
be followed by dictatorships, population grew at the rate of 2.36, but
once those regimes became dictatorships, population grew at the rate
of 2.40. These differences are not large, but the fact that population
grows faster in dictatorships that follow democracies is astonishing,
given the general trend for population growth to slow down with
time. Hence, except for the communist regimes, the effect of regimes
is diachronic, not just cross-sectional.
As always, given that dictatorships and democracies exist under dif-
ferent conditions, not only at different levels of development, one must
test whether it is indeed regimes and not those conditions that account
for the difference. Table 5.1B presents several estimates of the popu-
lation growth expected under the two regimes matched for per capita

221
 Political Regimes and Population

Table 5.2. Observed diachronic effects of regimes on population growth

Excluding
Entire sample communist

Regime N POPG N POPG

Stable dictatorship 1,709 2.52 1,603 2.63

(1.57) (1.55)
Stable democracies 1,104 1.12 1,104 1.12
(0.98) (0.98)
Dic preceding Dem 607 2.10 532 2.33
(1.42) (1.35)
Dem following Dic 371 2.24 366 2.27
(1.03) (1.01)
Dem preceding Dic 290 2.36 290 2.36
(2.05) (2.05)
Dic following Dem 425 2.40 425 2.40
(1.63) (1.63)

income (LEVEL), the extent of reliance on exports of primary com-

modities (COMEX), colonial legacy (NEWC, BRITCOL), frequencies of
Catholics (CATH), Protestants (PROT), and Moslems (MOSLEM), level
of education (EDT), labor-force participation of women (LFPW), rates
of female enrollment in secondary education (FEMSEC), and in some
analyses the lagged rate of population growth. Whatever statistical
model one uses – correcting for autocorrelation, with or without a
lagged dependent variable, a fixed-effect panel model, or Heckman’s
(1988) instrumental variables with constant effects – a difference
remains.
In sum, all the evidence indicates that regimes affect the rate of pop-
ulation growth. The observed rates of growth of population are higher
under dictatorship at each level of per capita income, and much higher
in countries with incomes above about $2,500. Correcting for other
exogenous variables does not erase this difference. Finally, even though
the diachronic effects are small, regime transitions affect population
growth regardless of their direction. Thus, the finding is robust. We
find it surprising.
The rate of growth of population (POPG) is itself a composite of three
rates: rate of births (BIRTHS), rate of deaths (DEATHS), and rate of net

222
Political Regimes and Population Growth

Table 5.3A. Observed rates of population growth under dictatorship

and democracy (small sample)

Rate of population growth

LEVEL All Dictatorships Democracies

267–500 2.47 (2.28) 2.47 (2.28) —

501–1,000 2.54 (2.46) 2.54 (2.47) 2.47 (2.35)
1,001–1,500 2.59 (2.22) 2.63 (2.21) 2.48 (2.28)
1,501–2,000 2.52 (2.25) 2.49 (2.21) 2.65 (2.40)
267–2,000 2.54 (2.33) 2.54 (2.33) 2.54 (2.33)
267–2,500 2.52 (2.31) 2.52 (2.31) 2.52 (2.33)
267–3,000 2.49 (2.28) 2.51 (2.30) 2.43 (2.16)
267–1,000 2.52 (2.43) 2.53 (2.43) 2.47 (2.35)
1,001–2,000 2.56 (2.23) 2.57 (2.21) 2.55 (2.33)
2,001–3,000 2.33 (2.05) 2.34 (2.12) 2.31 (1.94)
3,001–4,000 1.93 (1.76) 2.03 (1.85) 1.82 (1.66)
4,001–5,000 1.21 (1.33) 1.27 (1.34) 1.13 (1.32)
5,001–6,000 0.83 (0.82) 0.80 (0.86) 0.85 (0.81)
6,001–18,095 0.68 (0.86) 1.01 (1.45) 0.66 (0.82)
Total (N = 1,493) 1.54 (1.52) 2.22 (2.08) 1.00 (1.08)

Note: The rate of population growth is calculated as the difference between birth and
death rates. The rates of population growth for the same observations in the original data
file (POPG) are in parentheses.

migration. Population may grow slower in democracies because their

birth rates are lower, because their death rates are higher, or because
people migrate to dictatorships. Hence, we need to do some more
decomposing.
The data concerning birth and death rates are available only for a
much smaller sample and come from a different source (World Bank).
In Table 5.3A we show the “domestic” rates of population growth,
derived as the difference between birth and death rates. These domes-
tic rates differ from the rates of population growth shown in Table 5.1A
(and in parentheses in Table 5.3A for the same sample) by net migra-
tion. It is apparent that migration occurs from poor to rich countries:
Regardless of their regimes, in countries with per capita incomes under
$4,000 the rates of domestic population growth are higher than the

223
 Political Regimes and Population

rates including migration. Wealthier countries tend to attract migrants,

again regardless of their regimes. Nevertheless, the overall structure
of this sample preserves most patterns of the larger one: Population
grows faster under dictatorships in all but one income band, poor coun-
tries differ little regardless of regimes, and the rate of domestic popu-
lation growth falls faster as income increases in wealthier democracies
than in wealthier dictatorships.
To assess the impact of regimes, we must be careful that they are
not observed at different stages of the demographic transition from
high birth and death rates to low birth and death rates. In Western
Europe, this process moved from annual death rates in the range of
20 to 50 per thousand and birth rates in a very similar range, with
population growing at under 1 percent per annum, during the eigh-
teenth century, to death rates under 10 per thousand and birth rates
again only slightly higher, with population growing at rates under 0.5
percent during recent years. Because death rates fell first and at a
steeper rate than birth rates, during the transition the population grew
at much higher rates. That process was slow and long in Western
Europe, and rates of growth of population rarely exceeded 2 per-
cent. Fertility rates did not decline immediately when death rates fell,
exhibiting varying degrees of “inertia.” Moreover, crude birth rates
continued to increase well after fertility decline because of the increase
in the proportion of women of childbearing age – a phenomenon
referred to as “demographic momentum” (Merrick 1994: 90).3 The
result of the combined effects of inertia and momentum was that total
population increased by a factor ranging from 1.62 in France between
1785 and 1970 to 3.83 in Sweden between 1810 and 1960. (This
summary is based on Livi-Baci 1997: chap. 4).
Figure 5.1 shows a stylized picture of the demographic transition
that combines assumptions about the period preceding our observa-
tions (in this case, pre-1960) with data derived from our demographic
sample. An income of $250 per capita (as always, 1985 PPP USD) is
generally considered to be the absolute subsistence minimum. The
picture is drawn assuming that during the earlier period, before ad-
vances in medicine and the development of artificial methods of birth
control, death rates in countries at that minimum hovered around 30
per thousand, and birth rates around 45, yielding a population growth

3
Note that birth rates (BIRTHS) are numbers of births per thousand people, and fertility
rates (FERTIL) are numbers of children per woman during her lifetime.

224
Figure 5.1. Demographic transitions by per capita income and regime
 Political Regimes and Population

of 1.5 percent.4 The lowest income observed in our sample was $267,
the highest death rate was 32.4, and the highest birth rate 57.2. The
50 annual observations of countries with incomes below $500 show an
average death rate of 20.3 and an average birth rate of 45.0. During
the period of our scrutiny, poor countries were already in the grip of
transition: Death rates fell sharply with income, followed by somewhat
lagging birth rates. As a result, the poor countries we observed had
very high rates of population growth, over 2.5 percent. Death rates
become stationary in countries with incomes over $2,000, but both fer-
tility inertia and demographic momentum (discussed later) continue
to operate, so that birth rates fall less rapidly. They decline slowly as
countries get richer. In countries with incomes above $5,000, the tran-
sition appears to be completed, and the population grows at less than
1 percent per year.
The notable feature in Figure 5.1 is that democracies have lower
death rates and lower birth rates at almost all income levels. Hence,
the reason the population grows at a slower rate under democracies
is not that they have higher death rates. In fact, they have lower death
rates, but also lower birth rates, sufficiently lower for their population
to grow slower.
Examine Table 5.3B. The observed birth rate (BIRTHS) under dic-
tatorships was 36.2 per thousand people, and under democracies it
was 19.5. And, except for democracies with incomes under $1,000, for
which there are only eleven observations (six from India alone), birth
rates are higher under dictatorship at every income level. Hence, while
birth rates decline with per capita income, the difference between
regimes is not due solely to their incomes. More children are born
under dictatorships, even when they are matched for income with
democracies.
Death rates (DEATHS) were also higher under dictatorships, 14.1
per thousand, as contrasted with 9.4 under democracies. The death
rate in dictatorships with incomes under $2,000 was 15.9, and in
democracies with such incomes it was 12.9. The death rate was also
higher under dictatorships with incomes between $2,000 and $5,000.
Only in countries with incomes above $5,000 were these rates about
the same under the two regimes: 9.54 under dictatorships and 9.36

4
Estimates of pre-transition birth and death rates vary. It is generally thought that they
were about equal, at the rate of 45 per thousand, until the eighteenth century. Recent
research, however, has emphasized that they were more variable over time and less
homogeneous than previously thought. The estimates we use are consistent with the
findings of Livi-Baci as well as Crook (1997: 60).

226
Table 5.3B. Observed birth and death rates under dictatorship and democracy

All Dictatorships Democracies

LEVEL Births Deaths Births Deaths Births Deaths

0–1,000 43.7 (246) 18.5 (246) 43.7 (235) 18.5 (235) 43.8 (11) 19.1 (11)
1,001–2,000 39.1 (229) 13.4 (229) 39.6 (179) 14.0 (179) 37.0 (50) 11.5 (50)
2,001–3,000 33.0 (132) 9.9 (129) 33.9 (81) 10.8 (79) 31.7 (51) 8.6 (50)
3,001–4,000 27.9 (112) 8.5 (110) 29.5 (61) 9.3 (57) 26.0 (51) 7.6 (53)
4,001–5,000 21.6 (88) 9.5 (88) 23.1 (50) 10.4 (50) 19.7 (38) 8.4 (38)
5,001–6,000 18.1 (90) 9.8 (90) 18.2 (30) 10.3 (30) 18.1 (60) 9.6 (60)
6,001– 16.1 (606) 9.3 (603) 19.0 (34) 8.9 (33) 15.9 (572) 9.3 (570)
Total 26.9 (1,503) 11.5 (1,495) 36.2 (670) 14.1 (663) 19.5 (833) 9.4 (832)

Note: Birth and death rates are per 1,000; numbers of observations in parentheses.
 Political Regimes and Population

under democracies. Hence, except for relatively wealthy countries,

more people die under dictatorship at every income level.5
The lower death rates under democracy are to some extent due to
lower infant mortality (INFMORT), lower again at every income level.
Table 5.3C shows that the difference between regimes is large: Ex-
cept for very poor countries, for each 1,000 children born, between
seven and twenty more die under dictatorship. But decomposing the
mortality rates (Table 5.3D) shows that adults also die at higher rates
under dictatorship: Of the average deaths under dictatorships, 14.1 per
thousand, 3.4 deaths were of infants, and 10.7 of adults, whereas
under democracy, of the 9.4 deaths per thousand, infants accounted
for 0.5, and adults for 8.9. Infant mortality constitutes a large share of
total mortality in poor countries, but as both birth rates and mortality
of infants fall with income, it is a small source of total mortality in
richer ones. Yet infant mortality rates persist at high levels even in
wealthy dictatorships: They are as high in dictatorships with incomes
above $6,000 as in democracies with incomes between $4,000 and
$5,000.
The effect of these patterns on life expectancy (LIVES) is startling.6
People live longer under democracy. And the difference is large, very
large: about four years longer within each band of per capita income.
In democracies with incomes between $3,000 and $4,000, people live
almost as long as those in dictatorships which have twice that income.
Moreover, regime differences in life expectancy are similar for males
and females: Men live 66.2 years under democracy and 50.8 under dic-
tatorship, and women 71.5 years under democracy and 54.2 under dic-
tatorship. Hence, there is no reason to think people live shorter lives
under dictatorships because they send young boys to die in wars. Lives
are simply shorter under dictatorship.
Thus, there is no doubt that regimes affect birth rates, death rates,
and life expectancy. Observed population growth is lower under democ-
racies because they experience both fewer births and fewer deaths:
With the birth rate at 1.95 per hundred and the death rate at 0.94,
democracies in this sample had populations growing at an average rate

5
Democracies with incomes above $6,000 are, on the average, wealthier (Y/P = $14,429)
than dictatorships in this income range (Y/P = $7,322). They also have more people
over age 65, as discussed later. This explains the higher death rates in wealthy
democracies.
6
It may appear strange to speak of life expectancy under a particular regime when indi-
viduals outlive regimes. But life expectancy is only a summary statistic for the current
age-specific death rates.

228
Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9780511804946.006
Downloaded from https://www.cambridge.org/core. London Metropolitan University, on 02 Oct 2017 at 16:34:42, subject to the Cambridge

Table 5.3C. Observed infant mortality (INFMORT) and life expectancy (LIVES) under dictatorship and democracy

All Dictatorships Democracies

LEVEL INFMORT LIVES INFMORT LIVES INFMORT LIVES

0–1,000 131.1 (239) 46.5 (226) 130.5 (228) 46.4 (215) 142.9 (11) 47.2 (11)
1,001–2,000 97.4 (207) 53.1 (183) 101.7 (162) 52.2 (144) 81.6 (45) 56.3 (39)
2,001–3,000 69.3 (120) 60.8 (93) 75.4 (70) 59.2 (59) 60.9 (50) 63.6 (34)
3,001–4,000 45.6 (101) 65.5 (74) 52.1 (50) 64.2 (44) 39.3 (51) 67.3 (30)
4,001–5,000 34.2 (82) 67.3 (44) 39.9 (48) 65.0 (25) 26.1 (34) 70.2 (19)
5,001–6,000 25.4 (82) 70.1 (41) 29.6 (31) 68.6 (18) 22.8 (51) 71.3 (23)
6,001– 14.9 (586) 72.9 (255) 26.0 (27) 67.6 (18) 14.4 (559) 73.2 (239)
Total 55.1 (1,417) 60.2 (916) 93.6 (616) 53.3 (521) 25.4 (801) 69.3 (395)

Note: Infant mortality per 1,000; numbers of observations in parentheses.

 Political Regimes and Population

Table 5.3D. A decomposition of death rates under dictatorship and

democracy

All Dictatorships Democracies

LEVEL INFMORT Rest INFMORT Rest INFMORT Rest

0–1,000 5.7 12.8 5.7 12.8 6.3 12.8

1,001–2,000 3.8 9.6 4.0 9.9 3.0 8.5
2,001–3,000 2.3 7.6 2.6 8.2 1.9 6.7
3,001–4,000 1.3 7.2 1.5 7.8 1.0 6.6
4,001–5,000 0.7 8.8 0.9 9.5 0.5 7.9
5,001–6,000 0.5 9.3 0.5 9.8 0.4 9.2
6,001– 0.2 9.1 0.5 8.4 0.2 9.1
Total 1.4 10.0 3.4 10.7 0.5 8.9

Notes: Deaths due to infant mortality are calculated by multiplying birth rates by rates of
infant mortality. “Rest” is the remainder from death rates.

of 1.00 percent, but dictatorships, with the birth rate at 3.62 and the
death rate at 1.41, had populations increasing at 2.22 percent.
Note that although the rates of growth of population implied by these
birth and death rates are exactly the same for the two regimes in coun-
tries with incomes under $2,500, 2.52 percent, regimes make a dif-
ference even in poor countries. Both birth and death rates are higher
under dictatorships: Aggregating all observations under $2,500 yields
birth rates (per hundred) of 4.11 for dictatorships and 3.67 for democ-
racies, and death rates of 1.60 for dictatorships and 1.15 for democ-
racies. In countries with incomes above $2,500, the difference in death
rates almost vanishes: 0.99 for dictatorships and 0.92 for democracies.
But the difference in birth rates increases: 2.57 for dictatorships and
1.73 for democracies. The gap in rates of population growth among
wealthier countries – 1.57 for dictatorships and 0.82 for democracies
– is thus due almost entirely to the difference in birth rates.
These facts do not cease to amaze us, but one must be careful not
to attribute to regimes effects that are in fact caused by the conditions
under which they exist. We have already seen that these differences
are not due to per capita income; they occur at almost every income
level. They are also not due to the timing of regimes: Because birth
rates and death rates decline in time, and life expectancy increases,

230
Political Regimes and Population Growth

the observed patterns would have been biased against dictatorships if

they had been more frequent during the earlier periods. But that is not
the case: The distribution of regimes over time is almost the same in
this sample.7 Hence, timing does not bias the observed patterns.
Controlling for selection statistically is difficult with these data. First,
as Figure 5.1 shows, they are highly non-linear: The changes in all the
variables are extremely rapid at low income levels, but at higher levels
there is almost no change. At least with regard to death rates, a
logarithmic transformation does not linearize all the convexity, so that
even log-linear regressions generate errors that are correlated with per
capita income, biasing the results somewhat in favor of democracy.
Second, the observations are highly autocorrelated. Third, there are
strong country and period effects. Finally, the effects of selection are
strong, so that selection-corrected values are sensitive to the specifi-
cation of selection mechanisms.
Yet, though the exact magnitudes may be off, regimes affect births,
deaths, and durations of lives independently of the conditions under
which they exist. We provide the details of selection-based analyses
later, when we try to explain these patterns. Taking only per capita
income and year (or fixed time effects) confirms that if the two regimes
had had the same income levels and had existed at the same time, dic-
tatorships would have had higher birth rates, higher death rates, and
shorter lives. Corrected for selection, the differences between regimes
with regard to birth rates are in the range of 5.5 to 11.8 per thousand;
the differences with regard to death rates range from 1.2 to 4.4; and
the differences in life expectancy range from 5.8 to 18.7 years in favor
of democracy (Table 5.4).
Perhaps even more persuasive are the diachronic effects of regime
transitions. As Table 5.5 shows, transitions from dictatorship to de-
mocracy were associated with birth rates falling from 29.2 to 26.3. One
might think that because birth rates have generally fallen over time,
that decline was simply due to the timing of the regimes. But transi-
tions from democracy to dictatorship were associated with an increase
of birth rates from 34.6 to 36.3, even though those dictatorships
followed democracies in time. Fertility exhibits the same pattern.
Whereas in dictatorships that were to give way to democracies an
average woman had 4.1 children, in the democracies that succeeded

7
Between 1960 and 1974, dictatorships constituted 0.42 of regimes; between 1975 and
1990, they made up 0.47 of regimes. Hence, if anything, dictatorships made up a larger
proportion of observations during the latter period.

231
 Political Regimes and Population

Table 5.4. Selection-corrected rates of births and deaths and of life

expectancy

Selection-corrected predicted

Biased BIRTHS Log-BIRTHS Log-BIRTHS

Regime BIRTHS (OLS-AR1) (OLS-AR1) (2F PANEL)

Dictatorship 35.44 27.04 26.19 29.00

N = 589 (12.60) (11.16) (9.00) (4.75)
Democracy 18.69 21.50 20.05 17.23
N = 781 (7.89) (4.73) (3.77) (0.18)

Selection-corrected predicted

Biased DEATHS Log-DEATHS Log-DEATHS

Regime DEATHS (OLS-AR1) (OLS-AR1) (2F PANEL)

Dictatorship 13.52 11.03 10.39 12.61

N = 582 (5.90) (3.97) (3.02) (0.42)
Democracy 9.28 9.39 9.17 8.24
N = 780 (2.19) (0.56) (0.51) (1.54)

Selection-corrected predicted

Biased LIVES Log-LIVES Log-LIVES Log-Gap Log-Gap

Regime LIVES (OLS-AR1) (OLS-AR1) (2F PANEL) (OLS-AR1) (2F PANEL)

Dictatorship 53.99 60.95 61.24 53.01 60.99 58.77

N = 440 (10.17) (11.94) (14.33) (0.03) (9.34) (4.73)
Democracy 70.28 67.34 67.06 71.70 67.95 70.66
N = 343 (6.64) (4.09) (4.28) (2.18) (3.95) (0.94)

Notes: The selection equation is estimated by dynamic probit, with STRA, RELDIF, and LEVEL.
The exogenous variables for the demographic variables are LEVEL and YEAR in OLS and LEVEL
in the two-way fixed-effects (2F) panel. Log columns indicate that the dependent variable was
estimated in logs, and the predicted values were exponentiated. Gap = (85 - LIVES)/85, where
LIVES stands for life expectancy.

them the fertility fell to 3.7. In turn, whereas fertility in democracies

that preceded dictatorships was 5.0, in the dictatorships that followed
them the fertility increased to 5.3. Transitions from dictatorships to
democracies were accompanied by death rates falling from 11.8 to 9.9,
and the transitions from democracies to dictatorships again contradict
temporal patterns: Death rates rose from 12.0 to 12.5. Finally, life

232
 Explanations

Table 5.5. Observed diachronic effects of regimes on births, fertility, deaths,

and life expectancya

Regime N BIRTHS N FERTIL N DEATHS N LIVES

Stable dictatorships 441 38.9 408 5.6 434 15.0 361 51.3
Stable democracies 681 17.7 572 2.4 682 9.3 291 71.7
Dic preceding Dem 185 29.2 177 4.1 185 11.8 124 59.6
Dem following Dic 112 26.3 114 3.7 110 9.9 82 64.1
Dem preceding Dic 61 34.6 52 5.0 61 12.0 43 55.8
Dic following Dem 115 36.3 103 5.3 115 12.5 99 55.9

a
Because some regimes both preceded and followed alternative regimes, they are double-counted in this table.

expectancy increased in democracies that followed dictatorships, rising

to 64.1 from 59.6, and it remained stagnant in dictatorships that fol-
lowed democracies, at 55.9. As a result of those changes in birth and
death rates, the rate of (domestic) population growth declined from
1.74 in dictatorships preceding democracies to 1.65 in the democra-
cies that succeeded them, but it increased from 2.26 in democracies
that did not survive to 2.38 in the dictatorships that followed them. The
effects in dictatorships that followed democracies are astounding,
because they hurl themselves against the general trends for birth rates
and death rates to decline and for life expectancy to increase with time.
And on the demographic scale, these are huge differences.

Explanations

A Methodological Digression
The causes of these differences between regimes are not easy to
determine. Before proceeding with explanations, we need to pause and
consider what it means to “explain” differences between regimes.
We have observed that the two kinds of regimes differ in their demo-
graphic patterns. Moreover, this difference does not disappear when
we match regimes for per capita income and timing. Suppose that we
find that when the regimes are matched for some other exogenous con-
ditions, then the difference vanishes, as it does with regard to the rate
of growth of total income. That would mean that the difference is
caused not by regimes, but rather by the conditions under which they

233
 Political Regimes and Population

exist, something we previously failed to detect. We then would have

explained the observed difference.
Suppose, however, that the list of plausible and available exo-
genous variables is exhausted, and the difference persists. We then
would conclude that the observed difference is not due to exogenous
conditions but is an effect of regimes. But how can this difference be
explained?
Matching regimes for the observed exogenous conditions will not
eliminate the difference between them when the effects of those condi-
tions are not the same under the two regimes. Regimes may process
exogenous conditions in distinct ways. An example that is now classic
concerns nascent famines (Sen 1981): When lives are threatened, the
free press and other voices raise alarms in democracies, putting pres-
sure on the government to act, whereas in dictatorships the calamity
remains obscured from public scrutiny and takes its toll. In regression
terms, the coefficients on the exogenous variables are different: The
effect of the proportion of the labor force in agriculture on the death rate
should not be different from zero in democracies, but positive in dicta-
torships. If the share of the labor force employed in agriculture is exoge-
nous with regard to regimes, the counterfactual matching of regimes is
valid, but the difference in death rates persists because of the distinct
ways in which regimes respond to a large agricultural sector.
Still, the coefficients on all the exogenous variables may be the same
under the two regimes, and the difference between them may remain
when they are matched for the values of these variables. Algebraically,
this means that the regression constants must differ. This may indicate
that we have omitted some additional exogenous variable(s). Such
omission will bias the coefficients on the included variables if the
omitted variable is correlated with them in the sample, or it will bias
the intercept estimator (unless the mean of the observations of the
omitted variable is zero). Although statistical tests are available, even
if the model passes these tests, the possibility that it is misspecified
should always be kept in mind as the rival hypothesis. Yet if the model
does pass these tests, there is a strong presumption that there is some-
thing about regimes that, on the average, makes them different.
Sometimes the constants have a theoretical interpretation that can
provide a handle for an explanation. For example, we interpreted the
intercepts of the production function as the rate of technical progress,
and we told a story in which wealthy dictatorships pay little attention
to it because they compete by repressing workers and keeping wages
down. Often, however, such interpretations are far from obvious. For

234
Explanations

all we know, for example, life under dictatorships may be so grim that
what kills people may be despair: Despair can kill. Clearly, chagrin is
not something we can systematically observe. In the end, then, explain-
ing regime differences may inevitably entail simply telling intuitively
plausible stories, as always subject to rival interpretations.

Deaths, Infant Mortality, and Life Expectancy

Mortality began to decline in Europe in the second half of the eigh-
teenth century, as a result of the reduced frequencies of epidemics and
of famines and because of practices that reduced the spread of in-
fectious diseases (Livi-Baci 1997: chap. 4). Sustained and substantial
reductions in mortality, however, date only to the second half of the
nineteenth century (Easterlin 1998: 8). Such gains accelerated until the
middle of the twentieth century. Notwithstanding the war, the period
of maximum progress, due to pharmacological discoveries, was
between 1930 and 1950. The reduction in mortality extended during
that period to twenty-one developing countries (Coale and Hoover
1958; Ray 1998: 312).8 Subsequently, mortality declined and life
expectancy increased in almost all countries. Indeed, whereas per
capita incomes tended to diverge during the post-war era, mortality
rates differ much less now than they did some fifty years ago.
Although life expectancy increased for all age brackets, the reduc-
tion in mortality and the corresponding gains in life expectancy
were concentrated among young people. Livi-Baci (1997: 122–3) cites
studies according to which about two-thirds of the gains in life ex-
pectancy have resulted from a decline in mortality during the first
fifteen years of life.
Easterlin (1998: chap. 6) emphasizes that reductions in mortality
were largely due to independent technological innovations in medicine
and public health, rather than being merely consequences of improved
standards of living. Nevertheless, as Figure 5.1 shows, during the
period under our scrutiny, reductions in mortality were extremely rapid
as incomes increased from very low levels, whereas in countries with
per capita incomes over $2,000, further increases in incomes had
almost no effect.
Because the demographic data sample contains few poor democra-
cies (only 61 observations under $2,000) and equally few wealthy
dictatorships (63 observations over $5,000), the functions relating
8
According to Crook (1997: chap. 6), the decline in mortality in colonies was largely due
to health measures undertaken by the colonial powers in order to protect their soldiers.

235
 Political Regimes and Population

mortality to per capita income are very different for the two regimes.
Hence, to correct for effects of selection, we use a method, due to
Heckman (1988), that allows us to analyze all the observations together
and examine regime effects as intercept and slope dummies. Concep-
tually, this method is equivalent to what we have done with regard to
the growth of output, where we also cited some Heckman (1988) esti-
mators along with Heckman (1976, 1979) methods.
Mortality rates differ between regimes even when we assume that
their populations have the same age structure. Note that in the modern
(post-demographic-transition) order of things, people are more likely to
die as they get older. Hence, regimes in countries with relatively more
people over age 65 (OVER65) should be expected to have higher mor-
tality rates. In fact, democracies have a larger proportion of old people,
8.31 percent (N = 1,299), than dictatorships, where old people consti-
tute 3.77 percent (N = 1,984) of the population. Hence, the bias in the
observed world goes against democracies, which nevertheless have
lower observed mortality. Yet the ordering of deaths by age is a new phe-
nomenon: In the pre-modern era, the probability that children would
die before their parents was very high (Livi-Baci 1997: 118). Hence, in
very poor countries one should expect that mortality will be higher if
there is a higher proportion of people under age 15 (UNDER15). Indeed,
mortality is positively correlated with the proportion of old people under
democracy, and with the proportion of young people under dictatorship.
And the population is younger in dictatorships, where young people
constitute 42.35 percent of the population, compared with 31.32 in
democracies. Hence, in this respect, the sample is biased against dicta-
torships, which have higher observed mortality.
Correcting for the age structure of the population and for per capita
income – assuming they are the same under the two regimes – does
not eliminate the difference in mortality rates. But this correction may
be insufficient if large numbers of deaths are associated with child-
birth. According to the World Health Organization, as cited by Lloyd
(1994: 208), half a million women die in childbirth each year. And
when we use our birth and infant mortality rates, we arrive at about
another half a million infants who die during the first year of life, with
higher rates in poor countries and under dictatorship. Finally, a rough
calculation shows that about 45 million people die, on the average,
each year in the world as a whole. Hence, the 1 million deaths asso-
ciated with childbirth account for approximately 2.2 percent of all
deaths: a significant share.
Introducing fertility rates into the analysis shows that they are

236
Explanations

important in determining mortality under both regimes, somewhat

more under dictatorships. As Table 5.6A shows, mortality falls sharply
with income and with time, and in both cases faster in dictatorships.
A higher fertility rate is an important cause of mortality under both
regimes, only slightly higher under dictatorship. The results concern-
ing the structure of the population should be viewed with caution,
because the age variables are highly collinear. Yet even when regimes
are matched for age structure, per capita income, and fertility, mor-
tality rates under dictatorship remain higher, by 1.86 per thousand.
Mortality rates depend on various conditions that are subject to gov-
ernment policies. We broach this subject separately because there are
good reasons to think that social expenditures are endogenous with
regard to regimes,9 something we cannot test given the paucity of the
data. The proportion of GDP spent on social expenditures by the central
government – a sum of central-government spending on education,
social security and welfare, health, housing, and recreation and culture
– is much higher in democracies, 18.9 percent (N = 569), than the 9.2
percent (N = 549) in dictatorships. Almost the entire difference is due
to expenditures on social security and welfare, which consume 10.4
percent of GDP in democracies and 2.1 percent in dictatorships. The
expenditures on education are almost the same, respectively 3.9 and
4.1 percent of GDP, but expenditures on health are twice as large in
democracies, 3.3 percent, as compared with 1.7 percent in dictator-
ships. Expenditures on housing and culture are small and are the same
under the two regimes, and we ignore them in what follows. And note
that if government expenditures are exogenous with regard to regimes,
then we can treat the situation as ceteris paribus, but if they are not,
we should not correct for them.
How does government spending on various aspects of social policy
affect mortality? The combined sample that contains data for demo-
graphic variables and government spending is even smaller, but the
results appear quite robust. Statistical analysis shows that government
social expenditures reduce mortality, approximately equally under the
two regimes. When expenditures are disaggregated, it turns out that
expenditures on social security and welfare have no effect on death
rates under democracy, but reduce them somewhat under dictatorship.
Educational expenditures sharply decrease mortality under both
regimes, confirming the claim of Livi-Baci (1997: 172) that “improved

9
Moreover, the overlap between data sets that contain mortality, fertility, and government
expenditures is too small to use these variables together.

237
 Political Regimes and Population

Table 5.6A. Effect of regimes on the logarithm of death ratesa

AR1 model: e(t) = re(t - 1) + u(t)

Initial value of r = 0.73515
Maximum iterations = 20
Iterations = 4; sum of squares = 15.614; log-likelihood = 933.177909
Final value of r = 0.85452
Durbin-Watson: e(t) = 0.29096
Standard deviation: e(t) = 0.21852
Standard deviation: u(t) = 0.11350
Durbin-Watson: u(t) = 2.07924
Autocorrelation: u(t) = -0.03962
N(0, 1) used for significance levels
Standard
Variable Coefficient error (s.e.) z = b/s.e. P[|Z| ≥ z] Mean of X

Constant 5.510743 1.6478 3.344 0.00083

LEVEL -0.1274850E-04 0.37101E-05 -3.436 0.00059 5,648
YEAR -0.1818836E-02 0.84144E-03 -2.162 0.03065 1,976
UNDER15 -2.043796 0.40408 -5.058 0.00000 0.3266
OVER65 3.507457 0.70420 4.981 0.00000 0.7853E-01
FERTIL 0.2121669 0.12648E-01 16.775 0.00000 3.787
Dictatorship (compared with democracy)
REG 15.32860 2.2074 6.944 0.00000 0.4359
LEVEL -0.4565579E-04 0.81470E-05 -5.604 0.00000 895.6
YEAR -0.7053947E-02 0.11352E-02 -6.214 0.00000 861.1
UNDER15 -2.381716 0.52383 -4.547 0.00001 0.1749
OVER65 -4.366002 1.0122 -4.314 0.00002 0.1972E-01
FERTIL 0.2924777E-01 0.15896E-01 1.840 0.06577 2.308
Control for regime selection
EBAR -0.2715914 0.57594E-01 -4.716 0.00000 -0.4390E-02

Regime differences in log-DEATHSb

Variable Mean Standard deviation Minimum Maximum Cases

Democracy 9.7404 1.9504 5.2084 16.5867 1,225

Dictatorship 11.5973 4.3235 3.7727 23.7727 1,225

Note: The variables for dictatorship are intercept and slope dummies, X * REG, where REG = 1
if the regime is a dictatorship. Hence, the coefficients indicate differences from democracy.
a
Selection on unobservables; individual regime effect; Heckman (1988) estimator.
b
Expected values of death rates if regimes had been matched for all independent variables.

238
Explanations

education . . . appears to be a necessary prerequisite to improved san-

itary conditions.” Health expenditures have the same effect, indepen-
dently of education, again in accordance with Livi-Baci’s observation
that “those countries which have had particular success in combating
death are those in which government policy has allocated sufficient
human and economic resources to the health sector.”
As Table 5.6B shows, if social expenditures are assumed to be
endogenous with regard to regimes, the expected mortality under
dictatorship will be 10.95, and under democracy 9.20: a difference of
about two deaths per thousand. If these expenditures are in fact an
effect of regimes, we should stop here. But we can also ask whether
or not dictatorships still would have had higher mortality rates had
they spent as much on each social policy as democracies. It turns out
that had the regimes been matched for all the exogenous factors listed
earlier (except fertility), and had they spent exactly the same amounts
on various social policies, the mortality rates would have been identi-
cal: 8.83 per thousand.
In sum, exposed to the same exogenous conditions, dictatorships
would have had much higher mortality rates. If we assume public expen-
ditures to have been endogenous, mortality would have been still higher
under dictatorship. Only when public expenditures are treated as exoge-
nous with regard to regimes does the difference in mortality vanish.10
We will not retrace all the steps with regard to infant mortality. The
effect of regimes on infant mortality is much more pronounced. Regard-
less of the specification of the model and the method of estimation,
selection-corrected average infant mortality is much higher under dic-
tatorships. Perhaps surprisingly, educational expenditures have little,
if any, impact on infant mortality. Health expenditures, in turn, do
reduce it sharply under both regimes, and their effect is greater than
on gross mortality.11 When regimes are matched for all the exogenous

10
One set of variables that may have been omitted comprises the environmental conditions.
Communist dictatorships have been particularly notorious for environmental catastro-
phes. Yet democracies, from the United States to Italy to India, have also had their share
of environmental disasters. Without systematic data, we hesitate to go further.
11
When absolute amounts of per capita expenditures are used, health expenditures do not
matter under democracy, and they reduce mortality somewhat under dictatorship. This
finding may indicate threshold effects. One prenatal visit and a vaccination may be suffi-
cient to increase rates of survival, and because an average dictatorship spends only $33
on health, the statistical analysis may be showing the difference between those dictator-
ships that are below and above some threshold. But if that were all, we would not have
observed the statistical relationship between expenditures as shares and mortality
among democracies, which spend, on the average, $291. Because our concern is to assess
the impact of regimes, rather explain infant mortality, we leave it at that.

239
 Political Regimes and Population

Table 5.6B. Effect of regimes on the logarithm of death ratesa

AR1 model: e(t) = re(t - 1) + u(t)

Initial value of r = 0.70035
Maximum iterations = 20
Iterations = 4; sum of squares = 8.792; log-likelihood = 421.977027
Final value of r = 0.88048
Durbin-Watson: e(t) = 0.23904
Standard deviation: e(t) = 0.25772
Standard deviation: u(t) = 0.12218
Durbin-Watson: u(t) = 2.17600
Autocorrelation: u(t) = -0.08800
N(0, 1) used for significance levels
Standard
Variable Coefficient error (s.e.) z = b/s.e. P[|Z| ≥ z] Mean of X

Constant 7.201781 3.1244 2.305 0.02117

LEVEL -0.9422765E-05 0.49916E-05 -1.888 0.05906 7,798
YEAR -0.3432290E-02 0.15599E-02 -2.200 0.02779 1,980
UNDER15 3.481486 0.43837 7.942 0.00000 0.2949
OVER65 10.73364 0.92944 11.548 0.00000 0.9101E-01
GXPDSSEC -0.8482766E-03 0.35967E-02 -0.236 0.81355 9.668
GXPDEDUC -0.2733571E-01 0.73131E-02 -3.738 0.00019 4.027
GXPDHLTH -0.1931424E-01 0.71485E-02 -2.702 0.00690 3.050
Dictatorship (compared with democracy)
REG 12.87115 4.4377 2.900 0.00373 0.3003
LEVEL -0.7954322E-04 0.11181E-04 -7.114 0.00000 768.4
YEAR -0.5620413E-02 0.22293E-02 -2.521 0.01170 594.5
UNDER15 -2.692637 0.58188 -4.627 0.00000 0.1205
OVER65 -2.698238 1.6588 -1.627 0.10382 0.1201E-01
GXPDSSEC -0.1186252E-01 0.64192E-02 -1.848 0.06460 0.7222
GXPDEDUC -0.4307885E-02 0.98115E-02 -0.439 0.66061 1.229
GXPDHLTH 0.9076060E-02 0.15080E-01 0.602 0.54726 0.4987
Control for regime selection
EBAR -0.2131209 0.69873E-01 -3.050 0.00229 -0.6662E-03

(continued)

240
 Explanations

Table 5.6B (continued)

Regime differences in log-DEATHSb

Variable Mean Standard deviation Minimum Maximum Cases

Expenditures endogenous
Democracy 9.1993 1.7295 5.1996 15.5567 606
Dictatorship 10.9521 2.6055 4.8773 19.0811 606

Expenditures exogenous
Democracy 8.8305 1.8020 4.6046 15.5567 606
Dictatorship 8.8300 2.5699 3.7642 15.8390 606

Note: The variables for dictatorship are intercept and slope dummies, X * REG, where REG = 1 if the regime
is a dictatorship. Hence, the coefficients indicate differences from democracy.
a
Selection on unobservables; individual regime effect; Heckman (1988) estimator.
b
Inner product of exogenous variables and dictatorship dummies.

conditions, including fertility and social spending, the predicted rate of

infant mortality is 41.5 under dictatorships and 31.2 under democra-
cies: a difference of 10.3.
Finally, given what we have learned about gross and infant mortal-
ity, we would expect that differences in life expectancy would not dis-
appear when regimes are matched for different conditions and even
for social spending. When regimes are matched for their per capita
income, time, and age structure, the expected difference in life ex-
pectancy is 5.0 years. When they are also matched, in a smaller
sample, for the proportion of the labor force in agriculture, the differ-
ence is 4.9 years. Agriculture shortens life under both regimes, but its
effect is low when controlled for income and other factors. Finally,
when we assume, perhaps invalidly, that regimes have the same social
spending, the difference is still 3.4 years. The effects of social spend-
ing on life expectancy are positive and identical under the two regimes:
Both educational and health expenditures lengthen lives. Given that the
effects of exogenous conditions and of social spending are the same,
we are left with an unexplained difference.

Fertility and Births

Prior to demographic transition, European birth rates ranged from
a low of about 30 to a high above 45 per thousand. The poor countries

241
 Political Regimes and Population

in our sample were close to the upper limit. According to Livi-Baci

(1997: 176), “this is due primarily to higher levels of nuptiality: poor
country age at marriage (or the age at which reproductive union is
established) has traditionally been low, with almost no one remaining
unmarried, unlike the situation in the West.”12 The average birth rate
in the “more developed countries” between 1990 and 1995 was 12.6
(Livi-Baci 1997: table 5.2).
Birth rates depend on the proportion of women of childbearing age
in the population and on their individual fertility. The first factor is a
cumulative effect of demographic history; the second is a decision or,
as the case may be, a non-decision. If dictatorships are experiencing
the second phase of demographic transition, in which fertility is falling
or already has fallen to a stable level, but the population is young as a
result of high fertility in the past, then they will have high birth rates
because of conditions they inherited. Hence, to see if birth rates are an
effect of regimes, we must isolate the effect of fertility from that of age
structure.
Because the birth rate is the product of the proportion of women of
fertile age and their fertility (per year), we can decompose the regime-
specific birth rates into the part due to demographic momentum and
the part due to fertility. This is easier said than done, because the age
category in our data set includes all people between ages 15 and 65,
and because fertility rates are per lifetime rather than per year. Hence,
we must make some assumptions: Appendix 5.1 explains how we
arrived at the relevant estimates. But the conclusion is clear: Regimes
differ in their fertility rates, not in their age structure. The proportion
of women of childbearing age in the population we estimate to be 19
percent for democracies and 21.5 percent for dictatorships. Observed
fertility differences, in turn, are large: 2.7 per woman under democ-
racy, and 5.2 under dictatorship. If we assume the reproductive age to
last 25 years, between ages 15 and 40, then birth rates per woman
per year are 0.089 under democracy and 0.142 under dictatorship.
Combining these estimated age distributions with the annual fertility
rates yields birth rates higher under dictatorship by 13.6 per thousand.
Thus, we calculate that if the age structure were the same under the
two regimes, the dictatorial birth rate would be between 10 and 11
(per thousand) higher than under democracy. Hence, the difference in
birth rates is not due to “momentum” but to “inertia.”

12
But see Dasgupta (1995: 1,890) for evidence that social mechanisms of fertility control
have been used in some parts of Africa.

242
Explanations

Even though the relationship between birth rates, on the one hand,
and fertility and age structure, on the other hand, is purely algebraic,
we also tested the conclusions of these calculations by regressing the
logarithm of birth rates on the logarithms of fertility and of the pro-
portion of the population between ages 15 and 65, which is the bracket
for which we have data. If birth rates are due to momentum, then age
structure should matter. It does not: The coefficients on fertility are high
and significant under both regimes, and the coefficients on the age
structure have high errors. Birth rates are higher under dictatorships
because of higher fertility, not because of age structure.
To explain the regime differences in the birth rates, we must there-
fore explain the differences in fertility. Note that because death is
something that, except for suicide or euthanasia, happens regardless
of one’s will, it is a rare province to have escaped the attention of micro-
economists. Fertility, however, became a subject of the “new economic
demography” (Schultz 1974).
Before we review economic theories of fertility decisions, some back-
ground information will be useful. Fertility depends on the age at which
women establish reproductive unions, on the proportion of women who
ever enter such unions, and on the duration of postpartum sexual absti-
nence: These are social mechanisms that regulate it. Fertility can be
also regulated by traditional methods of birth control, coitus interrup-
tus and abortion. Finally, it can be regulated by contraceptive devices.
A total fertility rate (TFR) of seven children per woman is generally
considered to be the norm of “natural fertility,” that is, the number of
children a woman would bear in her lifetime if fertility were not reg-
ulated by any methods. The highest known fertility rate for a human
population is that of French Canadians born before 1860, 11.4 per
woman (Livi-Baci 1997: 61–6). Social mechanisms of regulation reduce
fertility to five per woman, but to cite Livi-Baci again (1997: 126), vol-
untary fertility control was the decisive factor in fertility decline. Fer-
tility rates in Europe during the first half of the twentieth century fell
to three per woman; in several present-day European countries, they
are one per woman.
Thus, different mechanisms cause fertility reduction from different
levels. Dasgupta (1995: 1,880) observed that “the factors that would
influence the drop in the total fertility rate in a society from, say, 7 to
5 should be expected to be different from those that would influence
the drop from 5 to 3 in that same society.” Whether or not fertility is
a matter of decision depends on the availability of the technology for
regulating it. This technology, which according to Livi-Baci (1997: 183)

243
 Political Regimes and Population

consisted of traditional methods, appeared in the rural areas of France

in the late eighteenth century and spread to other European countries
during the second half of the nineteenth century. Artificial methods of
birth control were illegal in the West until World War II.
Although the data are sparse, Table 5.7 shows that the use of con-
traception is much higher under democracy. More than 50 percent of
women use contraception in democratic countries; only one in four in
dictatorships. Moreover, the frequency of contraception is again higher
under democracy at each level of income, and the differences are large.
According to Sen (1994b, 1995), consensual methods of family plan-
ning are more effective than coercive ones, and dictatorships are more
likely to rely on the latter. Yet we suspect that the use of coercive
methods is rare (Livi-Baci 1997: 183–4). Most dictatorships either
oppose all methods of population control or rely on consensual
methods. This was certainly true of Eastern Europe, where the most
prevalent method of population control seems to have been abortion.
Hence, even if Sen’s comparison of Kerala with China does show that
consensual methods are more effective, we see no reason to think that
regimes differ systematically in the methods of family planning they
use.
Moreover, there is strong evidence that the mere availability of
contraception seems to have little effect on fertility (Pritchett 1994).
Although the use of contraception is strongly related to fertility, a prior
question is why people do or do not use contraceptive methods. And
there is a fair degree of consensus that what drives fertility decisions
is the demand for children. Contraception is a technology that enables
people to make choices, but people may decide against using it even
when it is available.13
People want to have children for many reasons. They are sources
of pleasure and of self-confirmation. Yet viewing children as an auto-
nomous value does not explain why fertility should vary across popu-
lations and across conditions. If we want to have as many children as
possible, then why does fertility decline as incomes increase?14 If we

13
This is not to say that access to contraception is universal. It is estimated that more
than 100 million couples (about 15% of couples in which the wife is of reproductive
age) would wish to limit their fertility but do not have access to contraception (Cassen
1994: 6). According to Lloyd (1994: 191), 16% of births are unwanted in countries
where fertility rates average above 6, 30% in countries where fertility rates average
between 4 and 6, and 25% where fertility falls below 4.
14
In the classic model of Becker and Lewis (1973), as rendered by Razin and Sadka (1995:
chap. 3), parents care about their own consumption, c, but also about the number, n,
and the “quality,” z, of children, maximizing u(c, z, n), with all partial derivatives

244
Table 5.7A. Observed patterns of contraception and fertility, by per capita income
and regimea

All Dictatorships Democracies

LEVEL CONTRACEPTION FERTILITY CONTRACEPTION FERTILITY CONTRACEPTION FERTILITY

0–1,000 10.4 (73) 6.1 (239) 8.3 (60) 6.1 (228) 19.9 (13) 6.2 (11)
1,001–2,000 23.0 (85) 5.7 (208) 21.9 (65) 5.8 (164) 26.3 (20) 5.5 (44)
2,001–3,000 36.9 (59) 4.7 (112) 31.3 (35) 5.1 (69) 45.1 (24) 4.1 (43)
3,001–4,000 51.8 (31) 3.8 (92) 48.8 (15) 4.0 (53) 54.6 (16) 3.5 (39)
4,001–5,000 53.7 (20) 3.0 (76) 44.6 (9) 3.2 (50) 61.1 (11) 2.6 (26)
5,001–6,000 51.3 (13) 2.6 (82) 48.2 (11) 2.5 (30) 68.5 (2) 2.7 (52)
6,001– 68.4 (48) 2.2 (526) 60.5 (4) 2.6 (27) 69.1 (44) 2.1 (499)
Total 35.0 (329) 3.8 (1,335) 24.8 (199) 5.2 (621) 50.7 (130) 2.7 (714)

a
Proportion of women practicing contraception, fertility per woman. Numbers of observations in
parentheses.
 Political Regimes and Population

want to have some fixed number of children, then why doesn’t fertil-
ity respond immediately to declines in infant mortality? If we want to
have as many children as other people do, why do some couples sud-
denly begin to reduce fertility?
In light of recent economic theory, as excellently summarized by Ray
(1998: chap. 9), demand for children depends on three factors: (1) their
insurance value, (2) externalities in child-rearing, and (3) the costs,
including opportunity costs, of child-rearing.
Dasgupta (1995: 1,894) cited a simulation by May and Heer (1968)
according to which an average Indian couple in the 1960s who wanted
to be 95 percent sure of having a son surviving when the father reached
the age of 65 had to bear 6.3 children. Several features of that example
merit attention. First, parents derive support from children during their
old age.15 Let the probability that any one child will support the parents
when they are old be p. This probability depends on the chance that
the newborn child will survive until the parents are old (most impor-
tantly, child mortality), on the chance that the child will earn enough
income to support himself and the parents, and on social norms that
make children responsible for their aged parents. Second, how many
children people will want will also depend on the alternatives other
than children’s support: parents’ assets, but also non-contributory old-
age insurance schemes. Let the probability of deriving support from
sources other than one’s offspring be q. The total number of children
a couple will then want so as to be 95 percent sure of support in old
age is the n that solves (1 - q)(1 - p)n = 0.05.16 Note that as q gets larger,
the number of children required for insurance declines. Assuming p =
0.3, the requisite number of children is 6.5 when q = 0.1, and it is 1.9
when q = 0.7.17

positive, subject to the non-linear budget constraint c + zn £ y, where y is income. The

effect of income on fertility is not clear in this model, depending on various elasticities
of substitution in the parents’ utility function.
15
According to the World Bank (1984: 52), 80–90% of parents interviewed in Korea, the
Philippines, Thailand, and Turkey counted on receiving economic assistance from their
children in old age.
16
This reasoning is based on the work of Ray (1998: 311), with some modifications. The
equation is derived as follows: Because p is the probability of being supported by
any one child, (1 - p)n is the probability that no child will support the parents; hence
s = [1 - (1 - p)n] is the probability that at least one child will. The probability of
being supported either by the state or by one’s own assets is q. Hence, the probability
of being supported by one source or another is qs + q(1 - s) + (1 - q)s. Simplifying and
setting the right-hand side to 0.95 yields 1 - (1 - q)(1 - p)n = 0.95 or (1 - q)(1 + p)n
= 0.05.
17
The insurance or “old-age security” hypothesis states that if parents care about chil-
dren only as sources of support in old age, then the availability of other sources of

246
Explanations

The number of children required for insurance against old age also
declines as the probability of being supported by any single one of them
increases. When q = 0.1, the requisite number of children is 6.5 with
p = 0.3, but it is 2.9 with p = 0.6. This example introduces the issue of
gender bias (Dasgupta 1995; Ray 1998). We used the masculine gender
in the preceding paragraph deliberately: In many societies, only sons
are educated, and only sons acquire the capacity to earn income.
Hence, if insurance were to be derived equally from sons and daugh-
ters, p would be doubled, increasing, say, from 0.3 to 0.6, and with
almost no outside sources of insurance (q = 0.1), optimal fertility would
decline from 6.5 to 2.9.
Looking ahead to statistical analyses, we will thus expect that fer-
tility will be low when infant mortality (INFMORT) is low and when
government spending on health (GXPDHLTH) and female school enroll-
ments (FEMSEC) are high, all increasing p, and when government
spending on social security (GXPDSSEC) is high, thus increasing q.
Clearly, these are not the only considerations implied by this theory,
but these are the only insurance-related factors we can systematically
observe.
The second set of factors affecting fertility encompasses various
kinds of externalities. Some occur within the household or, in cultures
where the term “household” does not have much meaning (Dasgupta
1995), within a broader and looser unit. Even within a household, the
costs and benefits of children need not be the same for all household
members. The woman of reproductive age may bear all the costs, but
the man or his mother (Sen 1995: 6) may reap some of the benefits.
Hence, if the power within the household rests with people other than
the childbearing woman, the decision may be to have more children
than the woman might desire. Basu, as cited by Sen (1995: 6), argues
that “the real pity is often not that men wield so much domestic power,”
but that it is “during the prime reproductive years that female power

support will lower the number of children they (or the society on the average) will have,
where other sources include (1) investment (lending at a positive rate of return) or (2)
a social security scheme whereby the current productive generation is taxed to subsi-
dize the subsistence of the beyond-the-productive-age generation. As the assumptions
indicate, this hypothesis is not very robust. The relationship between the availability
of alternative sources of support and a reduction of fertility does not or need not hold
if (1) parents care about the quantity and quality of children per se, (2) they can borrow
and not only lend, and (3) the social security scheme includes lump-sum transfers to
children – analysis based on Razin and Sadka (1995). Hence, on prior grounds the
validity of this hypothesis appears doubtful, and it is surprising that it holds as well as
it does in democracies, as discussed later.

247
 Political Regimes and Population

is at its lowest.” Moreover, both Dasgupta (1993, 1995) and Ray (1998)
emphasize that externalities may extend beyond the narrow repro-
ductive unit. If grandparents or other kin bear some of the costs of
rearing children, fertility will be higher. According to Ray (1998: 319),
some public services – public education, but also subsidized housing
and health services – may also lower the private costs of having chil-
dren and lead to higher fertility rates.
Unfortunately, we do not have systematic information about family
structure. What we know are the public expenditures on education
(GXPDEDUC) and the frequencies of three religions, namely, Catholics
(CATH), Protestants (PROT), and Moslems (MOSLEM). Clearly, cultural
factors that shape the family structure are not limited to religions, but
one would expect that religions would affect fertility and that some of
that impact would assume the form of division of labor within the
reproductive units.
Finally, fertility should depend on the cost of having children,
including opportunity costs. As a result of capital accumulation and
technical progress, the cost of producing market goods and service
falls relative to the cost of raising children (Barro and Sala-i-Martin
1995: 313). Hence, the costs of children rise with income, wages,
or other measures of opportunity costs. Because women bear most
of the burden of child-rearing, the opportunity costs are higher if
they engage in gainful activities and if they are educated. Galor
and Weil (1996) argued that increases in capital per worker raise
women’s relative wages and that, in turn, this reduces fertility
by raising the costs of children more than it raises the household
income.18
Hence, with costs in mind, fertility should depend again on labor-
force participation of women (LFPW), on income (LEVEL), on the level
of educational attainment (EDT), and, if Galor and Weil are correct, on
capital per worker (CAPW).
Notwithstanding Dasgupta’s observation that “a general theory of
fertility behavior is not currently available” (1995: 1881), this economic
model goes a long way toward explaining the observed patterns. When
we apply it to study the entire set of observations, regardless of

18
In their model, an increase in the income of the male will increase the demand
for children, whereas an increase in the income of the female will have both an
income effect and a substitution effect, so that equal increases of both will have
an ambiguous effect. But if women’s wages increase by more than men’s, then the
cost of raising children will grow faster than household income, and fertility will
decline.

248
Explanations

regimes, all the variables have the signs predicted by the theory, and
they are almost always statistically significant.19 Moreover, in various
specifications and the corresponding subsamples, they invariably
account for at least 86 percent of variance. This is strong performance
for any theory.
Capital per worker, which is supposed to raise women’s relative
wages and thus opportunity costs, according to Galor and Weil (1996),
is negatively associated with fertility. It is also, however, highly corre-
lated with per capita income, so one cannot distinguish the effect of
average income from the effect of women’s relative wages. When used
together with income, the coefficient on capital per worker flips signs
from one specification to another, whereas the coefficient on income
remains robustly negative. Given this collinearity, we do not use capital
per worker in the analysis of regimes.
Given the paucity of data, we cannot organize the analysis of regimes
by theoretical considerations – insurance value, externalities, and
opportunity costs – but only by sample size. We use two samples:
without (sample A; N = 552) and with (sample B; N = 276) social expen-
ditures. Table 5.7B shows the observed patterns of fertility by income
levels and summarizes the results of regression analyses, with vari-
ables ordered by theoretical considerations.
The insurance variables are much more robust for democracies,
where infant mortality (INFMORT), female secondary-school enroll-
ment rates (FEMSEC), social security expenditures (GXPDSSEC), and
expenditures on health (GXPDHLTH) have robust signs. Female enroll-
ment rates negatively affect fertility under dictatorship. Infant mortality
increases fertility under dictatorships only when public expenditures
are not considered. And expenditures on social security and on health
have no effect on fertility under dictatorship. Hence, insurance con-
siderations are much more powerful under democracy. Later we dis-
cuss why this may be so.
Externalities play the same role under the two regimes. Public edu-
cational expenditures behave in the way postulated by Ray (1998: 319),
that is, higher expenditures are associated with higher fertility. Hence,
it seems that public education is a source of externality in child-rearing.
To the extent to which the three religions for which we have data (and
implicitly the fourth category, namely, all others) affect the household
structure, one could also think that they indicate externalities in

19
We could not use female illiteracy because the overlap between data on fertility and on
illiteracy consists of only 12 observations.

249
 Political Regimes and Population

Table 5.7B. Summary of selection-corrected regressions of fertility, by

regime

Sample A Sample B

Variable All Dic Dem All Dic Dem

Constant -5.78 -5.84 2.66 -5.60 -16.97 10.99

Insurance:
INFMORT +++ +++ +++ +++ 0 +++
FEMSEC --- --- --- --- -- ---
GXPDSSEC --- 0 ---
GXPDHLTH -- 0 ---
Externalities:
GXPDEDUC ++ ++ +++
CATH 0 ++ + +++ ++ ++
PROT +++ ++ +++ +++ ++ ++
MOSLEM 0 0 0 ++ ++ -
Opportunity costs:
LFPW --- --- --- --- 0 0
LEVEL --- 0 -- --- - 0
EDT -- 0 - - -- 0
YEAR +++ ++ 0 ++ +++ --
N 552 246 306 276 94 182
R2 0.86 0.79 0.74 0.91 0.87 0.84
FERTILITY:
Observed 3.67 5.21 2.44 3.13 5.18 2.07
Predicted 3.89 3.48 2.77 2.31

Notes: Pluses and minuses indicate signs of coefficients. Three signs indicate significance
at the 0.01 level, two at the 0.05 level, and one at the 0.10 level; a zero means the coef-
ficients are not statistically significant. Selection equation includes STRA, RELDIF, ODWP,
NEWC, and BRITCOL. The estimators are OLS-AR1, separately for each regime.

child-rearing. The coefficients, however, make little sense and, except

for a positive impact of Protestantism, are unstable.
Finally, opportunity costs are important under both regimes. Labor-
force participation of women matters under both regimes, but only
when social expenditures are not considered, suggesting that the
opportunity costs of women working outside the household are lower
when the state picks up educational and health expenditures. High per

250
Explanations

capita income is associated with lower fertility in the entire sample,

but its impact is much weaker when the regimes are distinguished.
Finally, the level of education of an average member of the labor force
seems to have a weak negative effect on fertility, although this result
is not robust.20
The effect of labor-force participation by women (LFPW) requires
a separate comment. The relationship between female participa-
tion and fertility is non-linear under both regimes, but in different
ways: Under dictatorships, it is U-shaped, but under democracy it is
strictly convex to the origin. To put it in words, under dictatorship,
fertility is high when the labor force contains either few or many
women; under democracy, it is high when fewer women are in the
labor force. Because female participation is itself U-shaped with regard
to the proportion of the labor force in agriculture, what this must mean
is that fertility declines in both regimes as the role of agriculture
declines, and female participation with it, but as the society becomes
more services-oriented and more women again enter into the labor
force, fertility increases under dictatorship, but it continues to decline
under democracy. And what this, in turn, implies is that women in
wealthy dictatorships are doubly exploited: As workers, they receive
lower wages than under democracy (see Chapter 3); as mothers, they
continue to work outside the household while bringing up children.
Indeed, according to Lloyd (1994: 193), “time devoted to children is
largely responsible for differences in total work time between the
sexes.”
Female secondary-school enrollment rates increase with income
under both regimes, but here the picture is more nuanced.21 As Table
5.8 shows, both male and female secondary enrollments are higher in
democracies with incomes up to $3,000 and above $5,000 than in dic-
tatorships with such incomes. But between $3,000 and $5,000 both
male and female enrollment rates are higher in dictatorships. In part,
this is due to communist dictatorships: Enrollments in Eastern Europe
were high and equal across genders. Yet removing communist coun-
tries from the sample does not quite obviate this pattern, because Latin

20
The general evidence about the impact of female educational attainment on fertility is
overwhelming (Sen 1995: 17). If we are not getting strong results, it is probably because
our educational-attainment variable (EDT) does not distinguish sexes.
21
The rates of primary-school enrollment are higher under democracy, for females and
males, but many of them are above 100 percent, meaning that the school population
is larger than the relevant age bracket, which implies in turn that there is a lot of recidi-
vism. Hence, these rates cannot be unambiguously interpreted.

251
Table 5.8. Observed enrollment ratios, by gender, under dictatorship and democracy

All Dictatorships Democracies

LEVEL Females Males N Females Males N Females Males N

Primary school
0–1,000 52.8 72.0 (453) 52.8 71.7 (228) 52.8 77.9 (24)
1,001–2,000 76.2 89.7 (361) 75.4 89.7 (287) 79.3 89.6 (74)
2,001–3,000 95.3 103.0 (184) 92.6 102.7 (124) 109.0 103.6 (60)
3,001–4,000 103.1 104.8 (142) 101.7 103.7 (87) 105.3 108.6 (55)
4,001–5,000 99.7 102.9 (90) 96.0 100.2 (58) 106.4 107.9 (32)
5,001–6,000 105.1 106.2 (94) 102.8 104.1 (48) 107.5 108.4 (46)
6,001– 102.2 101.4 (405) 103.8 109.2 (29) 102.1 102.0 (376)
Total 83.2 92.0 (1,729) 73.6 86.8 (1,062) 98.5 100.3 (667)
Secondary school
0–1,000 9.0 17.2 (420) 8.7 16.5 (396) 13.7 29.2 (24)
1,001–2,000 22.8 30.2 (340) 22.0 30.0 (269) 26.0 30.7 (71)
2,001–3,000 37.0 41.6 (176) 34.5 41.1 (122) 42.5 42.9 (54)
3,001–4,000 49.1 52.8 (141) 51.7 55.8 (87) 45.1 47.5 (53)
4,001–5,000 56.4 62.0 (82) 57.8 65.9 (53) 53.9 54.0 (28)
5,001–6,000 65.8 67.5 (91) 61.9 65.4 (47) 70.0 69.6 (43)
6,001– 82.2 80.8 (413) 64.2 69.3 (39) 84.1 82.0 (373)
Total 41.8 46.2 (1,663) 26.2 33.3 (1,013) 66.1 66.3 (650)

Secondary-school enrollment ratios by region

Region Females Males N

Africa 11.4 19.0 522

South Asia 22.0 35.9 79
East Asia 54.8 66.9 32
Southeast Asia 40.8 44.3 82
Oceania 31.3 34.7 21
Middle East and
North Africa 36.3 50.4 167
Latin America 39.4 37.1 241
Caribbean 55.5 52.7 79
Eastern Europe 77.3 78.1 65
Industrialized 83.1 82.8 375

Note: Enrollment ratios are ratios of population enrolled to the relevant age categories.
 Political Regimes and Population

American countries, which are frequent in this income range, have

somewhat higher enrollments for females than for males. Hence, it is
not necessarily true that dictatorships educate fewer women: They do
so in some parts of the world, but not in others.
Note that education enters into the determination of fertility in three
ways. Public expenditures on education encourage fertility by lowering
the cost of raising children, but female enrollment in secondary schools
lowers fertility. A higher educational level of the labor force seems to
reduce fertility, but these results are weaker because we cannot dis-
tinguish genders.
What, then, is the impact of regimes on fertility? We have seen that,
to some extent, regimes have different ways of processing the condi-
tions under which they operate. Specifically, insurance incentives
operate more strongly under democracy, but externalities and oppor-
tunity costs seem to have the same effect under the two regimes.
Assessing the net impact of regimes is difficult, because several of
the variables that shape fertility may or may not be exogenous
with regard to regimes. We have seen that infant mortality is higher
under dictatorships matched for exogenous conditions. We may also
suspect that public expenditures are shaped by regimes. Nevertheless,
even if we assume that all these factors are exogenous, fertility appears
to be higher under dictatorships: 3.89 versus 3.48 when social ex-
penditures are not considered, and 2.77 versus 2.31 when they are
included.
Why, then, would regimes affect fertility even when they face the
same conditions? The most plausible explanation for this finding is
policy stability.22 At least with regard to social security expenditures,
under democracy people know that it is unlikely that these could be
drastically reduced: Citizens know that they can vote politicians out
of office if the latter curtail these expenditures, and they can expect
that, knowing this, elected politicians will not curtail them. Hence, even
if governments come and go, this is one budget item that is close to
sacrosanct. In turn, when dictators change, fiscal policy can take any
turn, and even when dictators remain in place, policy directions can
change.
This argument can be generalized. Fertility is something intimate, a
matter for private decisions. As Drèze (1992: 158) observed, “for most
of us, ‘adding a new person to the world’ is first and foremost adding a

22
We owe this suggestion to James Vreeland.

254
Explanations

new person to the family.”23 Except for outright coercion, such as invol-
untary sterilization, governments affect fertility decisions only by
changing the incentives for the individual reproductive units. This
system of incentives, as we have seen, consists of two sets of compo-
nents: those that affect the likelihood that the members of the current
reproductive cohort will be able to sustain themselves in old age without
the aid of their children, and those that affect the prospects that their
offspring will be able and willing to support them. Reproductive deci-
sions are made with a very long time horizon, about forty years. Hence,
the uncertainty they entail is very high: It is difficult to plan for forty
years from now. Moreover, the decision not to have the next child
becomes irreversible long before the uncertainty resolves itself: The lag
between the end of the reproductive age and that of the productive age
is also large. And what is at stake is vital. Hence, there are good reasons
for people to exhibit a high degree of risk aversion, perhaps to the point
of using maximin as the criterion.
This structure of reproductive decisions implies that people will
desist from having the next child only if they are very sure that they
will be supported in the distant future by their already living children
or by their private assets or by the government. Any threat to stabil-
ity, economic or political, will lead reproductive units to hoard children:
to have as many as possible, given the costs. And planning for a distant
future is much more risky under dictatorship; we have cited the evi-
dence throughout this book. As regimes, dictatorships are much less
durable than democracies. Once we distinguish successive dictator-
ships, say ayatollahs replacing a shah, we find that the expected lives
of dictatorships are much shorter than those of democracies. Just
looking at the incidences of closings and reopenings of legislatures
within authoritarian regimes shows that the expected life for regimes
we call “autocracies,” those without legislatures, is shorter than ten
years, and the expected life for dictatorships with legislatures is about
twenty-five years, whereas the expected life for democracies is about
forty years (see Tables 1.5 and 1.6 in Chapter 1). Any change of
the chief executive threatens the current authoritarian regime, but
under democracy such changes are a part of everyday life (Chapter 2).
And, as a result, economic performance under dictatorships is much
more volatile than under democracies: The average duration-weighted

23
The private nature of fertility decisions has been a persistent theme for Sen (e.g.,
1995).

255
 Political Regimes and Population

standard deviation of the rate of growth of total income is 6.94 during

the spells of dictatorship, and 4.82 during the spells of democracy
(Chapter 3).24
Hence, even if during a particular year under dictatorship the
economy is booming, old people are receiving pensions, and younger
people’s daughters are receiving secondary education, there is little
assurance that growth will be sustained forty years hence, that one will
receive a pension, or that one’s daughters will find employment. And
because children are the least risky asset that people can accumulate,
parents hoard children, rather than target their optimal number. For
the effect of uncertainty on hoarding versus targeting, see Ray (1998:
314).25

Human Capital, Fertility, and Economic Growth

Societies can accumulate children, human capital, or physical
capital. But at any particular moment they cannot accumulate every-
thing: People are constrained by their resources. Hence, people must
choose what to accumulate. If children, human capital, and physical
capital are produced with different technologies and at different costs,
and if they combine differently in production, the decisions people
make will depend on the value they attach to each outcome and the
relative rates of return to their eventual investments. These rates of
return will depend on the relative magnitudes of the stocks: how many
children people already have, how educated the labor force is, and how
many machines there are.
At this level of abstraction, no more can be said. One might think
that if a society confronted a situation in which there were machines
that could not be operated because people did not have the necessary
skills, then people would invest in education, and, vice versa, when
people could not use their skills because there were not enough
machines, they would invest in physical assets. But the situation is
complicated by the fact that people have a third alternative, namely, to
have many children, not educate them, and not invest in physical
assets. Moreover, when people are poor and cannot borrow to educate

24
A question arises at this point as to why investments in physical capital stock do not
differ between regimes and yet investment in children is motivated by uncertainty.
25
Note that this explanation illuminates the reasons that fertility was very low under com-
munism. Those regimes did offer income security, mainly through full employment, and
they educated women on par with men. And they were expected by everyone, eminent
U.S. political scientists included, to last forever.

256
Human Capital, Fertility, and Economic Growth

their children, they may have no choice but to have many children.
Hence, entire societies may find themselves in a trap, a low-level equi-
librium. In this trap, people are poor; because they are poor, they have
more children rather than educate them or accumulate physical assets;
because they do not accumulate either human or physical capital, they
remain poor, and so forth.
What exactly entraps societies in poverty is far from clear. But we
find evidence in favor of one story, told by Becker, Murphy, and Tamura
(1990). Theirs is a complicated model. They have to make so many
assumptions that listing them would require us to summarize half of
their article. The key idea is that human capital is accumulated with
increasing returns, so that when human capital is abundant, investing
in the education of the children one already has (“quality”) is more ben-
eficial than having additional children (“quantity”). As a result, soci-
eties with a low stock of human capital tend to have higher fertility.
The low-level trap arises when countries do not have enough human
capital to operate the available stock of physical capital.
If a country is caught in this trap, either (1) both human and capital
stocks will decline or (2) human capital will decline but physical
capital will be accumulated.26 The story is in some ways counterintu-
itive. The reason the stock of physical capital will decline when it is
high relative to human capital is intuitively clear: Machines cannot be
fully exploited when there are no skills to operate them, so that the
return to physical investment is low. But whether or not the stock of
human capital will decline will depend on whether a higher physical
capital stock will raise the cost of the time spent investing in human
capital more than it will increase the utility of future consumption,
which will be true only under some conditions that we cannot observe
directly (Figure 5.2).
If Becker, Murphy, and Tamura (B-M-T) are correct, then one should
observe the following: With each point in the space constituted by H
(on the horizontal axis) and K (on the vertical axis), both observed when
the country entered our sample, one can associate the average rate of
growth of per capita income during the period the country was
observed. A B-M-T trap exists if one can draw through this space an
appropriately sloped (convex, passing through the origin) line, M(H),
that represents the largest physical capital stock that can be fully
26
In our data set, which includes the level of education of an average member of the
labor force (EDT), which we take as the measure of human capital, H, there are 40
annual observations (out of 2,900) in which both K and H decline, and 222 observa-
tions in which K increases and H declines.

257
 Political Regimes and Population

Figure 5.2. Growth of countries in and out of the trap

exploited given the level of human capital. Points above this line, that
is, points for which K(H) > M(H), should then be characterized by poor
economic performance, and points below it by good performance. Note
that if long-term economic growth were random with regard to com-
binations {H, K}, then for every line drawn through this space, the dif-
ference in performance associated with points above and below it
would be zero. Hence, if we can draw a line such that performance
above it is poorer than below it, then the conclusions of the B-M-T
model are supported. Indeed, if economic performance depended
only on the stocks of human and physical capital, then only instances
of failure would be above this line, and only instances of success
below it.
The evidence strongly supports the existence of such a trap. Drawing
the line M(H) that maximizes the difference in average performance
between countries above and below it,27 we find that in the twenty-one
countries in the trap, per capita incomes grew in the long run at the
average rate of 0.62 percent, while the ninety-two countries below it

27
We used brute force, taking a quadratic function that passes through the origin, iter-
ating over the two parameters, and calculating averages above and below the line.

258
Human Capital, Fertility, and Economic Growth

had the rate of 2.58, a difference of -1.96. Among the twenty-one coun-
tries initially observed in the trap region, seven ended up with per
capita incomes lower than those with which they began, and five more
had incomes growing at less than 1 percent per year. Among the
ninety-two countries that began below the trap region, only six declined
over the period, two of them (Burundi and Mozambique) ravaged by
civil wars. Moreover, in accordance with the theory, the average fertil-
ity at the end of the period, some twenty to thirty years later, was 5.96
among the countries that were originally in the trap, and 3.81 among
the countries that were not.
Comparing some trajectories of countries in the H, K space is most
telling. Figure 5.3A shows the time paths of H and K for some coun-
tries that began in the trap, and Figure 5.3B portrays some paths for
countries that started outside of the trap. Whereas human capital was
accumulated by all countries, the stock of physical capital remained
stagnant or declined over time in Benin, the Central African Republic,
Gabon, Madagascar, Colombia, and Iran – countries that started with
educational levels insufficient to operate their physical capital. Gabon,
with its heavy reliance on oil, experienced a rapid accumulation of
physical capital, but it turned out to be transitory even as the stock of
human capital increased. In turn, except for India, countries that
started with levels of education high relatively to their stocks of phys-
ical capital increased both stocks. Brazil, which started with a rela-
tively low level of education but also of physical capital, accumulated
physical capital very rapidly, but then experienced a sharp drop. Most
startling, however, are the experiences of South Korea and Taiwan, two
countries that in the early 1960s had extremely low levels of physical
capital and high levels of education. In both countries, the accumula-
tion of physical capital was simply explosive.
What causes what is less apparent. But, contrary to Barro (1989),
we find that in our data set increased fertility has a clear impact of
reducing the growth of per capita income (Table 5.9). In the model that
includes on the right-hand side the growth of productive inputs, each
additional child reduces the growth of per capita income by 0.43. Given
that the average fertility is about four, and the average rate of growth
of per capita incomes is 2.46, this means that, ceteris paribus, in coun-
tries in which women have two children, per capita incomes double in
about twenty-one years, and in countries in which fertility is six, they
double in about forty-three years, almost a generation later.
Most of the observations in the trap are of poor countries, some with
extremely low levels of education, almost none. Democracies, as we

259
Figure 5.3A. Countries in the trap
Figure 5.3B. Countries out of the trap
 Political Regimes and Population

Table 5.9. Impact of fertility on the growth of per capita income

Ordinary least-squares regression: weighting variable = ONE; dependent variable is G;

mean = 2.46091; standard deviation = 5.2782
Model size: observations = 1,087; parameters = 5; degrees of freedom = 1,082
Residuals: sum of squares = 14,630.4; standard deviation = 3.67718
Fit: R2 = 0.51643; adjusted R2 = 0.51464
Model test: F(4, 1,082) = 288.88; probability value = 0.00000
Diagnostic: log-likelihood = -2,955.3139; restricted (b = 0) log-likelihood =
-3,350.2010; Amemiya probability criterion = 13.584; Akaike information criterion =
5.447
Autocorrelation: Durbin-Watson statistic = 2.24911; r = -0.12456

AR1 model: e(t) = re(t - 1) + u(t)

Initial value of r = -0.12456
Maximum iterations = 20
Final value of r = -0.12562
Durbin-Watson: e(t) = 2.25124
Standard deviation: e(t) = 3.67738
Standard deviation: u(t) = 3.64825
Durbin-Watson: u(t) = 2.00264
Autocorrelation: u(t) = -0.00132

Variable Coefficient Standard error z = b/s.e. P[|Z| ≥ z] Mean of X

(s.e.)

Constant 1.133814 0.23336 4.859 0.00000

KSG 0.3855428 0.11020E-01 34.986 0.00000 6.894
LFG 0.5458592E-01 0.86468E-01 0.631 0.52786 1.826
EDTG 0.9638270E-01 0.22294E-01 4.323 0.00002 2.879
FERTIL -0.4310592 0.60930E-01 -7.075 0.00000 3.962
r -0.1256196 0.30104E-01 -4.173 0.00003

know, are rare among such countries. Yet it is striking that only a single
observation of a poor democracy, Papua New Guinea, is to be found
among the trap cases, which consist of several African dictatorships
with very low education, Nepal, three oil-producing dictatorships with
higher educational levels, namely, Gabon, Iran, and Iraq, and two
wealthier democracies, Colombia and Venezuela (Table 5.10).
Because countries caught in the trap were almost exclusively ruled
by dictatorships, the impact of regimes on fertility can be questioned:
The observed difference between regimes may be due to the fact that
dictatorships were caught in the demographic trap, while democracies
enjoyed levels of human capital that allowed them to escape it. Because
only three democracies were initially in the trap, and two of them were

262
Human Capital, Fertility, and Economic Growth

Table 5.10. Countries in the trap and their long-term growth

Country Year K/L H G FERTIL REG

Angola 1975 379 0.514 -2.733893 6.4 Dic

Benin 1960 539 0.515 -0.058952 6.5 Dic
Burkina Faso 1960 88 0.075 0.133211 7.0 Dic
Central African
Republic 1961 210 0.305 -0.561102 5.8 Dic
Ethiopia 1960 43 0.095 0.823698 7.0 Dic
Gabon 1961 1,917 1.190 2.939512 5.5 Dic
Gambia 1965 284 0.320 1.571987 6.5 Dic
Guinea 1966 128 0.030 1.600147 6.5 Dic
Ivory Coast 1961 242 0.220 1.409908 6.8 Dic
Madagascar 1961 2,462 1.400 -1.787089 6.5 Dic
Mali 1961 320 0.155 0.150148 7.0 Dic
Mauritania 1961 784 0.065 0.357157 6.8 Dic
Senegal 1961 688 0.315 0.457079 6.5 Dic
Sudan 1971 940 0.780 -0.201483 6.4 Dic
Tunisia 1961 1,109 0.610 3.480121 4.3 Dic
Colombia 1960 4,414 1.875 2.376248 2.9 Dem
Venezuela 1960 14,994 2.520 -0.064273 3.5 Dem
Iran 1960 3,348 0.800 1.400212 6.3 Dic
Iraq 1960 8,529 0.240 1.255016 6.2 Dic
Nepal 1961 183 0.090 1.826954 6.3 Dic
Papua New Guinea 1975 1,750 1.170 -1.422171 5.3 Dem

Notes: “Year” is the first year for which education data were available. K/L is capital per
worker. H is EDT: years of education completed by an average member of the labor force.
G is the average rate of growth of per capita income between the initial year and 1990
or the closest year for which data were available. FERTIL is fertility per woman during
the last year data were available. REG is regime during the period.

relatively wealthy, even the fact that the twenty dictatorships caught in
the trap ended with an average fertility of 6.33, and the three democ-
racies with 3.87, is not very telling. But the trajectories of countries
that were initially not in the trap testify again to the effect of regimes:
The forty-three dictatorships ended with an average fertility of 5.13,
and the forty-eight democracies with a fertility of 2.64.28 And even if

28
These numbers do not add up to the total of countries, because in two cases fertility
data were not available.

263
 Political Regimes and Population

those democracies initially had much higher stocks of both human and
physical capital, the B-M-T theory asserts that the dictatorships, even
though poorer, would have reduced their fertility and grown over the
long run. Yet we find that although this theory finds strong support
in the data, regimes still have a profound impact on demographic
behavior.
Note that we now have seen two distinct explanations for fertility.
In the insurance model, parents want to have children only because
they offer a potential source of future support. In the B-M-T model,
parents want to have children because they derive satisfaction from
having children. These models are not rivals; both motives can play a
role. Moreover, they both imply that fertility should be higher in poor
countries. And both models are strongly supported by the data. Hence,
we are willing to believe that both insurance and altruistic considera-
tions lead people in poor countries to have more children. Neverthe-
less, even in poor countries political regimes affect fertility. And this
difference, we believe, is due to the insecurity inherent in dictatorships.
It is this insecurity that leads people to invest in the least risky asset,
namely, children. Hence, political regimes do make a difference even
when countries are matched for the exogenous conditions that drive
fertility in both models.

Conclusion
The overall picture of life and death under the two political regimes
is startling. Even if we assume that regimes face the same conditions,
democracies have lower birth rates and lower death rates. Women in
democracies have fewer children. More children survive to adulthood.
As adults, they live longer, years longer.
It is useful to look at these findings in reverse order. Birth rates are
higher under dictatorship because they have higher fertility, not
because of different age structures. And they would have higher fertil-
ity even if they faced the same conditions as democracies, most likely
because policy uncertainty leads parents to hoard children rather than
to target their number.
The regime difference in death rates is lower than that between birth
rates. The observed differences in infant mortality are larger, and so
are differences corrected for selection. And the effect of regimes on life
expectancy is astonishing: The observed difference is enormous at each
income level, and this difference again does not vanish when regimes
are matched for exogenous conditions.

264
Appendix 5.1: Calculating Age Structure

Regardless of regime, poverty forces women to procreate, it exposes

them to the tragedy of seeing their infants die, and it exposes all people
to disease and premature death. The effects of poverty are overwhelm-
ing: People in very poor countries, with incomes below $1,000, live
46.5 years on the average, but in countries with incomes above $6,000,
they live 72.9 years. Poor people have many more children, much
higher infant mortality, and much higher death rates. But political
regimes matter, and they matter even in poor countries: For several
aspects we have considered, their effect is equivalent to a difference of
$1,000 in per capita income.
The effect of political regimes on the lives of women is glaring.
Remember that the rates of labor-force participation do not differ
between regimes. What this means is that under dictatorship women
engage in gainful activities outside the household as frequently as
under democracy. But in addition they bear many more children, see
more of them die, and are themselves more likely to die. And, as Sen
(1995: 16) observed,29 “one of the most important facts about fertility
and family size is that the lives that are most battered by over-frequent
child birth are those of the women who bear these children. . . . It is
not only the case that at least half a million women die every year from
maternity-related causes through afflictions that are entirely pre-
ventable, but also hundreds of millions of women are shackled invol-
untarily to a life of much drudgery and little freedom because of
incessant child bearing and rearing.”
Because the difference in birth rates is larger than that in death
rates, population grows slower under democracies, again matched with
dictatorships for exogenous conditions. And because the difference
between regimes in population growth is larger than the difference in
the growth of total output, per capita incomes grow faster under
democracy.

Appendix 5.1: Calculating Age Structure

For the sample that contains data on age distribution and birth
rates, we know that the age distribution is

29
See also Cassen’s overview of the literature: “The clearest evidence of negative effects
of population growth under high fertility [is seen] at the individual and household levels.
Mothers exposed to large numbers of pregnancies have a high risk of dying; many of
these deaths are from unsafe abortions. Children with large [numbers] of brothers and
sisters will be more likely to be deprived in various ways. Girls suffer in particular; and
once they fail to be educated, the scene is set for intergenerational transmission of
poverty and high fertility” (1994: 4).

265
 Political Regimes and Population

Democracy Dictatorship

N 816 569

0–15 0.27 0.40

15–65 0.63 0.55
65– 0.10 0.05

Birth rate (per hundred) 1.92 3.69

We want to know the proportion of the population of childbearing

age, which we take to be between ages 15 and 40.
Inspecting the annual data shows that this age structure is
almost invariant over time. Hence, to standardize everything to
100, suppose that during each year the population of each of the
two regimes consists of 100 persons distributed in the foregoing
proportions.
Now, we also know that 1.92 persons are born each year under
democracy, and 3.69 under dictatorship. Because 15 years later there
will be 27 persons under age 15 in democracy, and 40 persons in dic-
tatorship, it must be true that

15(1 - i ) B = P1,

where (1 - i) is the survival rate for people under age 15, B stands for
births (per hundred), and P1 is the number of people under age 15.
Solving this equation yields

1-i i

Democracy 0.94 0.06

Dictatorship 0.73 0.27

This may appear to be an enormous difference, but note that infant

mortality (i.e., mortality during the first year of life) is 2.54 per hundred
under democracy, and 9.36 under dictatorship. This implies that of the
1.92 children born each year under democracy, 0.0488 will die during
the first year, 0.0664 will die during the next fourteen years, and
1.8048 will survive beyond the age of 15 and enter into the 15–65
cohort. Under dictatorship, in turn, of the 3.69 children born each year,
0.3388 will die in the first year, 0.6386 will die during the next four-
teen years, and 2.6937 will enter into the 15–65 cohort.
If the number entering each year into the 15–65 cohort is E, then it
must be true in turn that

266
Appendix 5.1: Calculating Age Structure

50(1 - d ) E = P2 ,

where P2 is the number of persons in this cohort. Solving this equation

for each regime yields

1-d d

Democracy 0.70 0.30

Dictatorship 0.41 0.59

Again, the difference in these survival rates is huge. But note that
the observed life expectancy is 69.3 years under democracy and 53.3
under dictatorship.
Now, if the chance of reaching the age of 65 conditional on having
survived until 15 is 0.70 for democracy and 0.41 for dictatorship, the
chance of surviving any next year within this age cohort is 0.993 under
democracy and 0.982 under dictatorship. This in turn implies that the
chance of reaching the age of 40, having reached 15, is 0.84 under
democracy and 0.64 under dictatorship. Because under democracy
1.8048 persons will enter this age cohort, which consists of 25 years,
the total population at risk is 45, and under dictatorship, where 2.6937
will enter, the population at risk is 67. Multiplying the populations at
risk by the survival probabilities yields an estimate of 38 persons
between ages 15 and 40 under democracy, and 43 under dictatorship.
Hence, the age structure is

Democracy Dictatorship

0–15 0.27 0.40

15–40 0.38 0.43
41–65 0.25 0.12
65– 0.10 0.05

Note that these estimates of the 15–40 bracket are biased toward
low numbers, because we assume that the death rate is uniform in the
entire 15–65 bracket. But the bias goes in the same direction under
the two regimes, and the shares of the population of childbearing age
are very similar under the two regimes.
To see what these numbers imply, examine their consequences for
birth rates. Given the observed rates of fertility per woman, and assum-
ing, as we do, that the reproductive period lasts 25 years, the birth
rates per woman per year are 0.089 under democracy and 0.142 under
dictatorship. Because women of reproductive age are about half of the

267
 Political Regimes and Population

population that age, the birth rate (per hundred) implied by the esti-
mated age distribution is 1.69 for democracies, as compared with 1.92
observed, and 3.05 for dictatorships, as compared with 3.69 observed.
Hence, we are underestimating birth rates because we have underes-
timated the populations in the childbearing cohorts. But we see that
the differences in birth rates between regimes are due to different fer-
tility rates, not to the age structure of the population. If the proportion
of women of childbearing age in the entire population had been the
same under democracy as under dictatorship, 21.5 percent, the esti-
mated birth rate under democracy would have been 1.91, as compared
with 3.05 under dictatorships with the same age structure. If the age
distribution under dictatorship had been that estimated for democracy,
the birth rate under dictatorship would have been 2.70, as compared
with 1.69 under democracy. Hence, what differentiates regimes is fer-
tility, not age structure, “inertia” rather than “momentum.”

268
 Conclusion

To make sense of the forest of numbers that populate these pages,

we first need a summary of the central findings. And although our
question concerns the impact of political regimes, one would be blind
not to note first the grip on people’s lives of sheer poverty. Although
regimes do make a difference for material welfare, their effect pales in
comparison with that of scarcity.
In every aspect we have examined, the differences between poor
and rich countries have been enormous. For one, even if democracies
do occasionally spring up in poor countries, they are extremely fragile
when facing poverty, whereas in wealthy countries they are impreg-
nable. Hence, poor people are much more likely to be ruled by dicta-
tors. Obviously, poverty means that people consume less. They also live
shorter lives, have more children, see more of them die and fewer of
them become educated, and are more likely to suffer from collective
violence. However one thinks of well-being, people with low incomes
lead poor lives.
Moreover, while with regard to mortality rates and life expectancies
the gap between poor and rich countries has been closing, the dispar-
ities of incomes and of fertility rates have increased. The coefficient of
variation (the ratio of the standard deviation to the mean) of death rates
fell from 0.44 when the countries were first observed (“entry” year) to
0.40 during the last year each country was observed (“exit” year). The
coefficient of variation of life expectancy declined from 0.23 to 0.17.
But the coefficient of variation of fertility increased from 0.33 to 0.46,
and the coefficient of variation of per capita income rose from an
already enormous 0.89 to 1.05. Whereas the multiple of per capita
income of the richest to the poorest country when they were first
observed was 40.4, at the end of the period it was 57.9.
These disparities of income and fertility increased because countries

269
 Conclusion

that were richer to begin with developed further, while many countries
that were poor remained poor. Of the eighty-three countries with per
capita incomes under $2,000 when first observed, fifty-seven had
remained equally poor or had become even poorer some thirty or forty
years later. Of the fifty-two countries that began with higher incomes,
all but seven at least doubled their income, and none declined over the
long run.
Hence, poverty can trap societies in its grip. One way poverty binds
is that when a society is poor, so is the state, and when the state is
poor it cannot extract resources and provide public services required
for development. Another trap occurs when the initial stock of human
capital is low relative to the physical capital stock; people tend to have
more children, and high fertility sharply reduces the growth of per
capita income.
Yet the bonds of poverty are not inexorable. Some countries, notably
Taiwan, South Korea, Thailand, Japan, Singapore, Portugal, Greece,
and Malta, grew spectacularly, at least quadrupling their per capita
incomes, with all the benefits that development brings. Of this list, two
(Taiwan and Singapore) were ruled by dictators during the entire
period and one (South Korea) during most of it, two were democracies
throughout (Japan and Malta), and the remaining three (Thailand,
Portugal, and Greece) experienced both regimes. Thus, although such
spectacular successes are rare to begin with, there is nothing to indi-
cate that it takes one regime or the other to generate them. But, in any
case, to evaluate the impact of regimes, one must look at their entire
record, not just their best record.
Political regimes have no impact on the growth of total income when
countries are observed across the entire spectrum of conditions. Con-
trary to widespread concerns, democracies do not reduce the rate of
investment even in poor countries. Indeed, it appears that when coun-
tries are poor there is little that governments can do, so that it makes
little difference for economic growth whether rulers are elected or hold
power by force. In wealthier countries, the patterns of growth are
not the same. Dictatorships rely on growth of the labor force and on
keeping wages low, whereas democracies pay higher wages, use labor
more effectively, and benefit more from technical progress. But though
growth under wealthier dictatorships is more labor-extensive and
labor-exploitative than growth under wealthier democracies, so that
the functional distributions of income are different, the average rates
of growth of total income are about the same.
The economic effects of political instability differ across regimes.

270
Conclusion

Indeed, the concept of political stability is not transportable across

regime types. The same phenomena that constitute instability in dic-
tatorships – changes of rulers, strikes, demonstrations – are just part
of everyday life in democracies. In dictatorships, any change or expec-
tation of change of leaders, any manifestation of political opposition,
is economically costly. But though such phenomena are much more
frequent in democracies, they are so routine that they do not affect
the economy.
Thus, we did not find a shred of evidence that democracy need be
sacrificed on the altar of development. The few countries that devel-
oped spectacularly during the past fifty years were as likely to achieve
that feat under democracy as under dictatorship. On the average, total
incomes grew at almost identical rates under the two regimes. The only
difference is that wealthier dictatorships have a more unequal func-
tional distribution of income: They pay lower wages. But these
findings also imply that the recently heralded economic virtues of
democracy are yet another figment of the ideological imagination.
Democracy has other virtues, but, at least with regard to the growth
of total economies, political regimes are not what matters.
Yet per capita incomes grow faster in democracies. The reason is
that democracies have lower rates of population growth. In spite of
rapid diffusion of medical advances, death rates remain somewhat
higher under dictatorship, and life expectancies are much shorter. But
the main reason population grows faster under dictatorships is that
they have higher birth rates, and the difference in birth rates is due to
higher fertility, not to the age structure of the population. We are not
certain why fertility is higher in dictatorships, but there are indications
that it is the policy instability inherent in dictatorships that induces
families to procreate.
Women are particularly affected by dictatorships. They participate in
gainful activities at the same rates as they do in democracies, and, as
workers, they get lower wages. But they also have more children, see
more of them die, and are themselves more likely to die in childbirth.
These findings add up to a bleak picture of dictatorships. Although
democracies are far from perfect, lives under dictatorships are grim
and short. Dictatorships are regimes in which political rulers accede
to power and maintain themselves in power by force. They use force
to prevent people from expressing their opposition and to repress
workers. Because they rule by force, they are highly vulnerable to any
visible signs of dissent. They are successful economically only if they
are “stable,” that is, if no one expects that the dictator will be changed

271
 Conclusion

or the dictatorship will be abolished. Because in dictatorships the

policies depend on the will, and sometimes whim, of a dictator, they
exhibit high variance of economic performance: Some generate mira-
cles, some disasters, and many generate both. Because their policies
and their performances are so unpredictable, they do not allow people
to plan their lives over a longer horizon, and thus they induce house-
holds to hoard the least risky asset, namely, children. In the end, per
capita incomes grow slower and people live shorter lives in dictator-
ships. Thus, whereas scarcity makes lives destitute, regimes do make
some difference, not only for political liberty but also for material
well-being.
Many of our aggregate numbers reveal political repression, mater-
ial destitution, and mass violence. Yet, in spite of the persistence of
occasional massacres, in spite of poverty increasing in large parts
of the globe, in spite of widespread dissatisfaction with the function-
ing of democracy even in wealthy countries, there is no doubt that the
past five decades have witnessed historically unprecedented political
and material progress. When we first observed these countries,
41 percent of the world’s population lived under democracy; when we
last observed them, 48 percent did, and even more since then. Per
capita income of an average inhabitant of the globe was $1,878 when
first observed, and $4,292 around 1990: 2.3 times higher. Of each one
thousand people, 15.3 died during the entry year, but only 9.7 died
during the exit year. An average person could expect to live 52.3 years
when first observed, and 64.7 when last observed, 12 years longer.
True, as countries developed they did not necessarily become similar:
Wages in authoritarian Singapore were still much lower than in demo-
cratic Austria when they became equally wealthy. But both political
progress and material progress are evident.
Because China and India add up to one-half of all of us, they deserve
special attention. China still remains under authoritarian rule, and
India has maintained a democratic regime against all odds. Per capita
income in China rose from $470 in 1961 to $1,324 in 1990, and in
India per capita income increased from $559 in 1951 to $1,262 in
1990. Death rates fell in China from 14.2 per thousand in 1961 to 6.6
in 1987; in India they declined from 22.2 in 1962 to 11.4 in 1987. Life
expectancy in China was 40.5 years in 1961, and 68.7 in 1987,
whereas in India it was 45.5 in 1962 and 57.8 in 1987. Finally, fertil-
ity tumbled in China from 4.7 in 1961 to 2.1 in 1990, and in India it
went down from 6.5 in 1962 to 4.2 in 1987. Hence, progress was not
limited to countries that were already wealthy.

272
Conclusion

Africa is a source of mitigation in any account such as ours. A recent

study shows that a dummy variable for this continent is negative and
significant in all growth regressions. Africa has experienced slower
growth of incomes, slower social development, more corruption and
bureaucracy, and more frequent civil wars than the rest of the less-
developed countries (Collier and Gunning 1999). Yet even this conti-
nent shows some progress, albeit slow and spotty. The average African
country had a per capita income of $747 when first observed, and $966
when last observed. Death rates have fallen from 22.1 to 15.8 per
thousand, and life expectancy has increased from 41.8 to 50.6 years.
In turn, fertility has remained almost unchanged, 6.4 for the first and
6.2 for the last observations. And only 2 percent of Africans lived
under democracy as of our last observation, in or before 1990.
Ours has been a strange century, one in which mass murder has
repeatedly punctuated the progress of democracy and of material
welfare. About 160 million people, roughly 10 percent of the world’s
population in 1900, were killed in wars in the same hundred years
during which democracy became an almost universal value of mankind
and technical innovations multiplied incomes and extended lives. Thus,
what the future may bring is far from clear. The tools of social science
give us little with which to look ahead. Certainly, extrapolating trends
is a risky, if not dubious, undertaking. Yet, writing at the eve of a mil-
lennium, one cannot but wonder what the future may bring.
Let us first put together all that we learned and then peek ahead.
Here are the pieces we have:

1. The probability that a dictatorship will die and a democracy will

be established is pretty much random with regard to per capita
incomes, about 2 percent each year. But the probability that, once
established, a democracy will survive increases steeply and mo-
notonically as per capita incomes get larger. Indeed, democracy is
almost certain to survive in countries with per capita incomes above
$4,000.
2. Total national incomes grew at an average rate of 4.23 percent per
annum. Incomes grew slowly in poor countries; growth rates were
highest at income levels between $2,000 and $3,000, declining again
at higher incomes. The observed rates of growth were 4.42 under
dictatorships and 3.95 under democracies, but when countries were
matched for exogenous conditions, the growth rates under the two
regimes were almost the same.
3. Rates of population growth were much lower in wealthier countries.
But at every income level, population grew faster under dictator-

273
 Conclusion

ships. As a result, the observed rates of population growth were 2.42

percent per annum for dictatorships and 1.46 percent for democra-
cies. When the regimes were matched for exogenous conditions, a
large part of that difference remained: Our best guess is that
population would have increased at a rate of 2.18 under dictator-
ships and at 1.59 under democracies had they faced the same
conditions.
4. Finally, the incidence of war on the territory of a country was much
lower in wealthier countries. We also know independently that
democracies do not war with each other. Although wars not only kill
but also impede growth, the recovery from wars is rapid, so that
they do not have long-term economic effects.

Given what we know about the past forty years or so, we can do
two things. First, we can see if this schematic description is sufficient
to reconstruct the patterns we observed. And then we can venture into
speculations, by using these patterns to look ahead. What we want to
know is how many countries will be ruled by democracies in the year
2030, what will be their incomes and their populations, and how many
will suffer from wars.
Let us first see how well can we reproduce the patterns we observed.
Starting with per capita incomes, population, and regimes observed
during the entry years, we can use the dynamic patterns summarized
earlier to predict values for the same variables for the exit years (see
Appendix C.1). Because the dynamic of regime transitions is proba-
bilistic, we will not get unique answers. Thus, the expected values are
based on repeated runs (Table C.1).
Note that the fit between simulated and observed values is very
close, except for the incidence of war. The average per capita income
observed at the exit year was $4,292, and the average predicted value
is $4,300. The proportion of world population living under democracy
is slightly underpredicted by simulation, and the predicted number of
dictatorships is just one more than observed. The predicted rate of
growth of population is a slight overestimate, so that the total world
population at the exit year is predicted as 5,101 million, just slightly
above the observed value. In turn, the number of countries experienc-
ing wars on their territories is grossly underpredicted by the simula-
tion. The reason is that the 1980s were exceptionally war-ridden, so
that the probabilities of war were much higher at the end of the period.
But, except for war, this simple simulation seems sufficient to recon-
struct the observed patterns (Table C.2).
Given that our predictions for the year 2030 are based on a

274
Table C.1. Average values of selected indicators observed or predicted during
various years

Year Per capita Proportion Number of Population Total Countries

Source income democratic dictatorships growth population in war

Entry
Observed 1,878 0.41 84 1.63 2,764 6
Exit
Predicted 4,300 0.44 75 1.75 5,101 8.4
Observed 4,292 0.48 74 1.70 5,068 16
2030
Predicted 11,199 0.67 45 1.23 9,376 5.3

Notes: All the numbers refer only to the 135 countries in our data set. Predicted numbers are based
on simulation. “Proportion democratic” is the proportion of the world population living under democ-
racy. “Population growth” is the rate of growth of the world population. “Total population” is the
population of the world, in millions. “Countries in war” is the number of countries experiencing
wars on their territories.
 Conclusion

Table C.2. Distribution of countries in entry year, exit year, and 2030

Entry year Exit year 2030

LEVEL Number Wars Dic Number Wars Dic Number Wars Dic

0–1,000 48 5 41 34 7 34 12 3 11
1,001–2,000 45 1 30 23 4 13 15 2 13
2,001–3,000 13 0 8 20 3 13 8 1 6
3,001–4,000 9 0 2 12 2 6 10 1 6
4,001–5,000 8 0 2 7 0 2 4 0 3
5,001–6,000 5 0 1 4 0 2 3 0 1
6,001–7,000 4 0 0 5 0 0 4 0 1
7,001–10,000 3 0 0 8 0 2 17 0 0
10,001–20,000 22 0 2 26 0 1
20,001–30,000 11 0 1
30,001–40,000 5 0 1
40,001–50,000 4 0 0
50,001– 16 0 0

Total 135 6 84 135 16 74 135 7 44

Minimum level 226 312 523

Maximum level 9,121 18,073 71,211
Multiple 40.4 57.9 136.2

Notes: Numbers for year 2030 are based on one simulation, closest to the average predicted values.
LEVEL is income per capita. “Number” is the number of countries in each range of per capita income.
“Wars” is the number of countries experiencing wars on their territories. “Dic” is the number of
dictatorships.

mechanical extrapolation, caveats would be redundant. But one caveat

could invalidate all our optimistic forecasts: depletion of non-renew-
able resources or severe degradation of the environment. Unfortu-
nately, systematic data concerning environmental degradation are not
available. But there are good grounds to fear that if these past trends
were also mechanically extrapolated, our future would appear much
less rosy. So, with this one caveat, foolishly but valiantly, we present
the results:
1. Average world per capita income will be much higher – our guess is
about 2.5 times higher – than the 1990 level. Incomes will grow
faster than in the past, because there will be fewer poor countries
that will grow exceptionally slowly. Some per capita incomes we
predict to be extremely high. Obviously, if rates of growth decline as
high incomes get even larger, our prediction is grossly overopti-

276
Conclusion

mistic. But income convergence among the developed countries is

just a description of the past. Whether or not increasing returns will
set in at the level of entire economies, as distinct from particular
firms or sectors, is still an open question.
2. Given that dictatorships will be disappearing, for various reasons,
and that per capita incomes will be higher, more democracies will
survive, so that most of humanity will live under democracy. But
quite a few dictatorships will be left, mostly in poor countries and
mostly in Africa. Because India has remained a democracy against
all odds, we repeatedly predict it as a dictatorship.
3. Given that population growth declines as incomes become larger and
that it is lower in democracies, population growth will slow down.
The population of the 135 countries in our sample will be about 9.4
billion in 2030, which is not far from more sophisticated predictions:
The United Nations forecast for the entire world in 2025 is 8.5 billion
(Livi-Baci 1997: 222).
4. Even though more countries will be rich and more will be democ-
racies, we still predict quite a few wars, all in poor countries – in
fact, all in Africa. Given that we did not anticipate that violent con-
flicts would surge in the 1980s, this is not even an extrapolation:
just a memento mori.
5. Finally, consistently with other projections (Quah 1996), we believe
that if the past patterns continue, income disparities among coun-
tries will continue to increase. Because total income grows slowly in
very poor countries and population grows rapidly, per capita income
increases very slowly. Countries that already have medium income
levels grow rapidly, and their population growth slows down sharply
as they become richer. In turn, already wealthy countries grow
somewhat slower, but they begin at high levels and have low popu-
lation growth. Hence, distribution of per capita incomes flattens with
time, and disparities increase.

Hence, if the patterns we have observed persist, the world will be

better, much better. More people will be living in democracies; they will
be wealthier; and they will be enjoying all the benefits that wealth
brings, probably including great improvements in public health and
medical technology. But not all of us will enjoy this progress: Poverty
will still be widespread, dictators will still repress, and wars will still
ruin lives.
Hence, optimism does not entail quiescence. We have learned that
the bonds of poverty are difficult to break, that poverty breeds dicta-
torships, and that dictatorships make lives miserable. We also know

277
 Conclusion

that if we allow the patterns of the past to persist, most people will be
much better off, but many will remain within the double grip of poverty
and dictatorship – if we allow the patterns of the past to persist.

Appendix C.1: Simulation

This simulation is based on the rates of growth of total income pre-
sented in Table 3.6, the probabilities of regime transition by regime
from Table 2.3, the rates of population growth by regime from Table
5.1A (modified, see later), and the probabilities of war by per capita
income and regime, not shown in the text.
Two simulations were conducted. The purpose of the first was to
validate the model. Entry-year values were taken as observed, and the
variables were iterated until the exit year. Thus the exit-year values are
based on the number of years each country was observed.
The second simulation was used to generate extrapolated values.
Observed exit values were used to start, and each country was iterated
40 times, to generate predictions for the year 2030 or close to it (the
earliest exit year was 1987).
The two random elements in the simulation are regime transitions
and wars. Because the predicted values have very low variance, the
averages reported in the text are based in both cases on ten runs.
Observed rates of population growth persistently overpredicted the
total population observed by the exit year and underpredicted the
average per capita income. To calibrate these rates, we uniformly sub-
tracted 0.004 from all the growth rates presented in Table 5.1A. This
was sufficient to generate results very close to observed values.
The program for the simulation, in BASIC, is available on request.

278
 Appendix I

Selection Model

Selection as a Problem of Identification

The problem raised by non-random selection is how to make infer-
ences from what we observe to what we do not.1
What we want to know is the probability that some outcome vari-
able Y will assume the specific value yi for observations i = 1, . . . , N,
each characterized by a vector of exogenous factors xi. Hence, the prob-
ability of interest is Pr{Y = yi|X = xi}, which, to simplify the notation, we
will write most of the time as P(y|x).
Now, think of the entire sample as consisting of cases observed in
states d = 0, 1. Let the probability that di = 1 depend on some factors
zi = {xi, vi}, where v are those features of i that affect only its states but
not the outcomes under each state, and x are the variables that affect
the outcome and, perhaps, selection into states. In our context, di = 1
if country i is observed under a dictatorship, and di = 0 if it is observed
as a democracy.2 The dependent variable Y is any among those exam-
ined earlier. We will use economic growth as the example.
Because the problem is symmetric, let us work with Y1, which we
observe only if d = 1, so that

yi1 if di = 1
yi1 =
unobserved if di = 0. (1)

1
This entire section follows Manski (1995). A note on our terminology is needed: Statis-
ticians tend to speak of endogenous sampling rather than of non-random selection as
the source of selection bias. We depart from this terminology for two reasons: (1) We
want to emphasize that selection is not a matter of the sample frame chosen by the
researcher but of the mechanisms by which the entire set of observations is produced
by “nature.” (2) We reserve the terms “endogenous” and “exogenous” selection for the
mechanisms by which these observations are produced.
2
Note that in order to lighten the notation, we ignore time.

279
 Appendix I: Selection Model

The probability distribution of y1 in the entire sample is

P ( yi1 xi ) = P ( yi1 xi , di = 1)P (di = 1 zi ) + P ( yi1 xi , di = 0)P (di = 0 zi ). (2)

What do we know? We observe the selection probability P(d = 1|z),

the censoring probability P(d = 0|z), and the distribution P(y1|x, d = 1)
of y1 conditional on selection. We do not observe, however, the distri-
bution of y1 for cases observed as d = 0, P(y1|x, d = 0). This last prob-
ability concerns counterfactual observations: It is the probability that
an observation with X = xi, observed under d = 1, would have had a
particular value of Y1, Y1 = yi, had it been observed under d = 0. To put
it in terms of our example, P(y1|x, d = 0) is the probability that a country
observed as a democracy would have had a particular rate of growth
had it been a dictatorship. And because we do not observe counter-
factuals, we do not know what this distribution is.
Given that we cannot observe P(y1|x, d = 0), the probability P(y1|x)
in (1) is not identified. Moreover, the same is true of the expected value
of y given x:

E ( y1 x) = E ( y1 x, d = 1)P (d = 1 z ) + E ( y1 x, d = 0)P (d = 0 z ), (3)

and, again, we do not know E(y1|x, d = 0).

Manski (1995: 25) shows that even without making any assump-
tions, the P(y|x) is bounded. Suppose that we want to know the
probability that Y1 £ yi, P(Y1 £ y|x). Then, setting P(Y1 £ y|x, d = 0) as
alternately zero and one, we can set the bounds on P(Y1 £ y|x) as

P (Y1 £ y x, d = 1)P (d = 1 z) £ P (Y1 £ y x) £

P (Y1 £ y x, d = 1) P (d = 1 z) + P (d = 0 z). (4)

Hence, the upper bound on the probability of interest, of Y1 being less

than or equal to y, is then equal to the censoring probability, P(d = 0|z).
Because the mean of a random variable is not a continuous function
of its distribution function, however, no bounds can be established with
regard to the expected value (Manski 1995: 26).
Because the same is true for y0, the observational structure is

yi1 if di = 1, unobserved otherwise,

yi =
yi0 if di = 0, unobserved otherwise. (5)

We want to know two distributions, P(y1|x) and P(y0|x), and their

expected values. The first is the distribution of y in the entire sample
if all cases are observed as d = 1; the second is the distribution of y in
the entire sample if all cases are observed as d = 0. In terms of our

280
Statistical Models of Selection

example, y1 is growth under dictatorship, and y0 under democracy.

Obviously, if we want to know the impact on y of being in states d, we
want to determine the difference between these two distributions, con-
ditional on x. For example, we may want to estimate the difference of
the expected values, E(y1|x) - E(y0|x), or the probability that growth
under democracy would have been higher than that observed under
dictatorship, P(y0 > y1|x), or the difference between the probabilities
that an economy would decay under either regime, P(y0 < 0|x) - P(y1 <
0|x). The question about the impact of regimes on growth concerns
such differences.
We already know that we cannot identify P(yj|x): For each d = j, j =
0, 1, we observe only the cases with d = j, but not those with d π j. We
observe P(d|x), P(y1|x, d = 1), and P(y0|x, d = 0), but not P(y1|x, d = 0)
and P(y0|x, d = 1). Hence, we face underidentification. The same is true
of the expected values E(y1|x) and E(y0|x).
To see that underidentification is caused by non-random selection,
assume that being under d = j has nothing to do with the outcome yj.
Then the conditional probability of each outcome is the same whether
the cases are observed under either of the two states j = 0, 1, so that
P(yj|x) = P(yj|x, d = j), and the model is fully identified. This is what
experimentalists assume and quasi-experimentalists try to emulate.
To conclude, in the presence of non-random selection we face the
problem of how to identify the distribution of cases we do not ob-
serve. As always, identification requires additional assumptions. And,
as always, stronger assumptions impose narrower bounds on the dis-
tribution, but require more theory, or faith, as the case may be.
Manski’s (1995) general recommendation is that we should use weaker
identifying assumptions and be satisfied with interval estimates.
Because he discusses various identifying assumptions that arise natu-
rally in the context of selection, we will not pursue this route further.

Statistical Models of Selection

To approach issues entailed in estimating the distributions P(yj|x),
assume that selection into j = 0, 1 is governed by some unobserved
random variable d*, such that the observed state is d = 1 when d* > 0,
and d = 0 otherwise, so that P(d = 1) = P(d* > 0) and P(d = 0) = 1 -
P(d = 1).
The joint density of (yj, d) in the sample is, for any i,
f ( y j , d ) = P (d π j ) f ( y j x, d = j )P (d = j ), j = 0, 1, (6)

281
 Appendix I: Selection Model

that is, the probability that yj is not observed, times the density of yj
given that it is observed, times the probability that it is observed.
The density f(y1|x, d = 1) is the joint density of (y1, d*|z) given that
d = 1 or d* > 0:

f ( y1 x, d = 1) = f ( y1, d * z, d * > 0) = f1 ( y1 x)P (d * > 0 y1 ) P (d * > 0) , (7)

with -• £ y1 £ •, 0 £ d* £ •, where f1(y1|x) is the marginal density of

y1, and P(d* > 0|y1) is the conditional probability of being in the state
d = 1 given y1 (Pudney 1989). Substituting into (6) yields

f ( y1 d ) = P (d = 0)* f1 ( y1 x)P (d = 1 y1 ). (8)

Similarly for y0.

The likelihood of yi1, yi0, di, zi in the sample is thus
i= N
L = ’ Ldii1 L1i 0-di , (9)
i =1

where Lij = f(yij, di) is given by (8).

To determine this likelihood, we must determine the probabilities
P(di = j) and the marginal densities fij(yij|xi), j = 0, 1. In other words, we
need to identify the mechanisms of selection and of performance con-
ditional on selection. This is what follows.
Selection is a mechanism that allocates observations i = 1, . . . , N to
states j = 0, 1, . . . , J - 1, in our case J = 2. Selection can be a conse-
quence of a deliberate decision of the researcher: This is what the argu-
ments about research design are all about. It may result from some
irredeemable defect of the sampling process: One cannot phone people
who have no phones. Or it may be generated by the world around,
independently of any operations by the researcher. Only the last situ-
ation is of interest here.
Consider the effect of political regimes on economic growth. If a
country has an authoritarian regime, we observe growth under dicta-
torship; if it is a democracy, we observe growth under democracy. In
none of such situations do we observe the consequences of states that
were not realized. Hence, we do not observe the entire population, but
only a sample.
It may seem strange to refer to the “population” as including unob-
served cases. But if we are to compare the effects of states j on the per-
formance of an individual characterized by an xi, we must allow this
individual to be potentially observable under the full range of j = 0, 1,
. . . , J - 1. That is, even if in fact we observe the case i only in state

282
Statistical Models of Selection

di = j, we must conjure the possibility of this case having been observed

as di π j, and vice versa. We must thus think in terms of a “super-
population, consisting of a continuum of ‘potential’ individuals.” The
actual sample is then regarded as having been drawn by “nature” from
this super-population (Pudney 1989: 45).
If the actual population is a random sample of the potential one, any
multivariate combination (xi, di) can be drawn with a positive proba-
bility. This property of random sampling is what allows us to isolate
the effect of d on y given x, that is, to “control” for x the effect of d on
y. If the sample is random, we can “match” pairs (xi, di = 0) and (xi, di
= 1). But if sampling is not random, there will be some observations
without a “match”: Some of the xi’s will not have support. And increas-
ing the sample size will not increase the probability of finding a match.
In the presence of non-random selection, quasi-experimental compar-
isons fail regardless of the number of observations. As a result, we will
be unable to tell whether or not democracies would have grown as fast
had they faced the same conditions as the observed dictatorships.
How, then, does “nature” do its work? Let us first describe the selec-
tion mechanisms generically and then distinguish some types. Assume
that each case can be characterized by some proclivity to be in the state
j = 0, 1, conditional on z = {v, x}. Let the proclivity to be in state d = 1
be d*. Then, thinking in regression terms, we can write
d* = Za + u, (10)
where Z are the factors observed by the investigator, and u are those
that are not observed. Suppose further that we standardize the dis-
tribution in such a way that the observed state d is related to the
unobserved proclivity by
1 if d* > 0 fi za > -u fi -za < u,
d=
0 otherwise. (11)
The probability that d = 1 is then given by
- za
P (d = 1) = P (d * > 0) = Ú
-•
u du = F ( - za ), (12a)

where F is some distribution that satisfies the axioms of probability

theory, and
P (d = 0) = 1 - F ( - za ). (12b)
Selection is endogenous if an individual chooses d to maximize the
utility associated with the outcomes of the choice, that is, solves the

283
 Appendix I: Selection Model

problem maxj U(y|x). The solution to this maximizing problem is to

choose
1 if U ( y1 x) > U ( y0 x),
d=
0 if U ( y1 x) £ U ( y0 x).
Hence the probability that an individual will choose d = 1 is the prob-
ability that U(y1|x) > U(y0|x).3 Assuming that each utility depends on
factors observable by the investigator, as well as some that are not, the
utility of being in state j is U(yj|x) = Xbj + uj. Then P[U(y1|x) > U(y0|x)] =
P[U(y1|x) - U(y0|x) > 0] = P[(xb1 - xb0) > (u0 - u1)] = P[x(b1 - b0) > u0 -
u1] = P(xb > -u), where b = b1 - b0 and u = u1 - u0. This is the same as
(12), except that the only variables that enter into purely endogenous
selection are those that determine the eventual outcomes, that is, x.
Selection is exogenous if state d occurs not because of the conse-
quences associated with the choice but because of some features vi.
Then the probability that an individual will behave in the way d = 1
will depend on the characteristics v, so that
P (d = 1) = P (va + u > 0) = P ( -va < u),
which is again the same as (12), except that the variables that affect
the outcome do not enter this model. To use the language of Gambetta
(1987), selection is driven by “push” rather than “pull” factors.
To evaluate the relative strengths of these competing explanations,
we first need to construct a general model in which both explanations
might play roles. The state under which a particular country is
observed depends, then, on the strength of the factors V that push the
country toward d = 1 or the utility that the country derives from the
outcomes in state d = 1. Hence,

d * = vg + [U ( y1 - y0 )]d + u and P (d = 1) = P (d * > 0), (13)

where g and d indicate the respective forces of the “push” and the “pull”
factors.4

3
Or, in the multinomial case, that U(yj|x) > maxkπjU(yk|x).
4
For an early example of such a model, see Lee (1978). In some cases, this model can
be interpreted in an alternative manner, namely, Vg can be viewed as the “transition
cost” of moving from state d = 0 to d = 1. For example, if Y stands for the time needed
to commute to work by bus or by car, and V stands for family income, an individual will
move from bus to car if the income given up to buy the car is worth less than the utility
of the time saved. Hence, the interpretation of the model depends on the substantive
meaning attached to the V variables. Yet if V becomes simply “psychic cost,” the eco-
nomic hypothesis becomes tautological.

284
Statistical Models of Performance

Thus, if countries are democratic or authoritarian because of their

exogenous characteristics, such as level of economic development,
education, or urbanization (Lipset 1960), selection is exogenous. If
countries have regimes under which they grow faster economically,
selection is endogenous. Our analyses invariably show that the factors
that enter on the right-hand side of the performance equation are not
statistically significant in the selection equation. Hence, throughout the
book, we treat selection as exogenous.

Statistical Models of Performance

Assume now that y is a function of x, with the population density
f ( y|x). Write the regression equations as
yi0 = xib0 + ui0, (14a)
yi1 = xib1 + ui1. (14b)
The expected value of y1 for the selected sample is, for any i,

E ( y1 x, Y1 is observed) = E ( y1 x, d = 1) = E ( y x, d * > 0)
= E ( y x, - za < u) = xb1 + E (u1 d * > 0), (15)

where E(u1|x, d* > 0) is the conditional expectation of u1 given that y1

is observed. The expected value for the population is

E ( y1 x) = xb1. (16)

Similarly for y0.

Hence, the values expected in the sample differ from the population
values by the conditional expectations of errors given that the values
are observed. If selection is not random, then uj, j = 0, 1, are corre-
lated with u in the selection equation (10), so that cov(uuj) π 0.
The expected value for the selected samples can be derived as
follows. Note first that if the uj’s are correlated with u, then
E (u j d ) = s ju E (u d ), (17)
where sju is a regression coefficient of uj on u. In turn, to estimate the
expected values E(u|d* > 0) and E(u|d* £ 0), we can use the fact that
the expected value of the truncated normal distribution is

E (u d * > 0) = E (u - za < u) = f ( - za ) [1 - F ( - za )] = f ( za ) F ( za ) = l1 .
(18a)

285
 Appendix I: Selection Model

In turn,

E (u d * < 0) = E (u - za > u) = - f ( - za ) F ( - za ) = - f ( za ) [1 - F ( za )] = l 0 ,
(18b)

where the lj’s are the inverse Mill ratios or the hazard rates.
From (12) we know that Y1 is observed when -za < u, so we can
write the expected values in the observed sample as

E ( y1 d = 1) = xb1 + s1u l1, (19a)

and, similarly for y0,

E ( y0 d = 0) = xb 0 + s0u l 0. (19b)

Note that if (14) is estimated on the basis of the observed sample,

the variable lj is omitted from the specification. Hence, selection is a
source of omitted-variable bias (Heckman 1979).
We can now also understand why controlling (“matching”) for the
variables that enter into both the selection and outcome equations may
in fact exacerbate, rather than attenuate, the selection bias (Achen
1986). Following Heckman (1988), let us distinguish first between selec-
tion on observables and on unobservables. Selection on observables
occurs when the expected covariance E(uju) π 0, but once the observed
variables Z are controlled, it vanishes, so that E(uju|z) = 0. Selection is on
unobservables when E(uju) π 0 and E(uju|z) π 0, so that controlling the
factors observed by the investigator does not remove the covariance
between the errors in the outcome equations and the selection equa-
tions. Now, note that the regression coefficient sju = cov(uju)/var(uj). If
selection is on unobservables, controlling for some variable x in the
outcome equation may reduce the error variance uj without equally
reducing the covariance uju. Hence, the coefficient on the omitted vari-
able will be larger, and the bias will be exacerbated. Suppose that coun-
tries are more likely to have dictatorial regimes when they are poor or
when their leaders are unenlightened and that the rate of growth is
lower when per capita income is lower or leaders are unenlightened. The
researcher can observe per capita incomes, but not enlightenment. Then
controlling for per capita income in the outcome equation will reduce the
error variance, but not the covariance due to enlightenment. The obser-
vations will be matched for per capita income, but not for enlightenment:
There will be no democracies with unenlightened leaders at low levels.
Hence, the observed difference in growth will be due purely to the unob-
served difference in enlightenment, not to regimes.

286
Estimating Selection-Corrected Models

This completes the development of the model. We have learned that

the expected values for the observed cases are biased because they
covary with the variable that determines which cases are observed. If
selection is exclusively on observables, this bias can be corrected by
traditional controlling techniques. But if it is on unobservables, such
controls would only exacerbate the bias.

Estimating Selection-Corrected Models of Performance

Correcting for selection bias is not uncontroversial. Several
researchers have found that corrections for selection are not robust,
but are highly sensitive to relatively minor changes in assumptions
about distributions (Goldberger 1983). Others have found that some
estimation methods fail to correct for this bias and may even exacer-
bate it (Stolzenberg and Relles 1990). We seem to be in the quandary
that was spelled out by Heckman (1988: 7): “Different selection
methods will produce similar estimates of program or institution
impacts only if there is no problem of selection bias. The variability in
estimates across selection models indicates that selection is a
problem.” So if there is no selection bias, different methods for cor-
recting it will be robust; if there is a bias, they need not be. The alter-
native ways out of this dilemma are to rely solely on experiments – a
solution that, regardless of its merits, is not feasible in the context of
comparative research – or to ignore the problem altogether, or to strug-
gle with the problem by using a variety of approaches. As Heckman
(1988: 6) emphasized, “there is no single ‘correct’ way to eliminate
selection bias.” But there are alternative approaches, and the differ-
ences among them are instructive.
There are two basic approaches to parametric estimation of selec-
tion models, plus a number of variations. The first is maximum likeli-
hood; the second is Heckman’s (1976) two-step method.
To be able to use maximum likelihood, we must assume joint dis-
tributions of (yj, d*, z) and derive the respective marginal distributions,
fj(yj|x), as well as the conditional distributions P(d = j|yj), j = 0, 1, . . . ,
J - 1. As long as J = 2 and the observations are independent, identi-
cally distributed (i.i.d.), and normally distributed, this is not problem-
atic. The marginal distribution is

fij ( yij xi ) = s -j 1F[( yij - xib j ) s j ], (20a)

the probability P(di = j|yij) is

287
 Appendix I: Selection Model

{ 1
P (di = j yij , zi ) = F [ zia + r ju ( yij - xib j ) s j ] [1 - r] 2
}, (20b)

and the probability P(di π j) is

P (di π j zi ) = 1 - F( zia ). (20c)
To get maximum-likelihood estimates (MLEs), all we need to do is to
substitute (20) into (8) and maximize the likelihood given by (9), each
j at a time. The MLE is consistent and efficient, although sensitive to
departures from normality.
Yet under several conditions maximum-likelihood estimation is dif-
ficult. This is true when J > 2,5 when the performance series is auto-
correlated, or when it exhibits fixed or random effects: problems we
often encountered in this study. The main problem is that P(d = j|yj, z)
may be intractable. In those situations, we rely on Heckman’s two-step
method, which is consistent but not fully efficient.
The advantage of Heckman’s method is that its consistency does not
require yj, d* to be jointly normal, provided that d* is normally dis-
tributed and the errors in equation (22) are distributed independently
of d*, but not that they are normal (Amemiya 1984: 33). Heckman’s
method is to estimate by maximum likelihood the selection part of the
model
k
L = ’ P (di = j ), (21)
i

where k is a (J - 1) vector of dummy variables that assume the value

k = 1 when d = j, and k = 0 otherwise. Estimating (17) will yield â,
Fij = Fj(ziâ), and lij’s. Equations (15) can then be written as linear
regressions:
yij = xibj + sju lij + eij, (22)
where eij = yij - E(yij|di = j), so that E(e) = 0.
Discussions of variance and of the properties of this estimator can
be found in the work of Amemiya (1985) and Greene (1990).
Heckman (1988) discussed alternative estimators, based on a
control-function approach. Although we used those estimators in
Chapter 5, we refer the reader to the original source.

5
Lee (1983) assumes that the variables uij and ei, which is a transformation of ui, have
well-defined marginal distributions and that they also have an i.i.d. joint distribution,
which will be true in case of multinomial logit. Yet it turns out that this assumption is
highly restrictive, because it implies that all the regression coefficients sju in (18) must
have the same sign (Schmertmann 1994).

288
Estimating Selection-Corrected Models

Any method for correcting selection effects yields unbiased estimates

of the coefficients in the outcome (performance) equations (15). With
these coefficients, we can generate the unbiased values, observed and
counterfactual, of the yij’s for each individual i in each state j. All we
need to do is to estimate
yij|xi = xib̂j. (23)
These are the values of yi expected in each state j for an individual
characterized by xi. Because for each i = 1, . . . , N the exogenous con-
ditions xi are the same across the states, these expected values indi-
cate the effect of being in state j on the outcome yij, which is what we
wanted to know all along. Now we have matched cases: For each xi,
we observe the outcome or performance yi under each j. We can there-
fore calculate what would have been the difference between regimes if
the entire sample had been observed under both regimes.
Finally, one last complication. In the entire foregoing analysis, we
have matched cases by creating the observations we were missing, that
is, by constructing counterfactuals. Each of the y~j|x values tells us what
the value of the outcome yi would have been had the case i, observed
with a particular set of features or under a particular combination of
conditions xi, been observed under each of the states in which it was
not. But this procedure entails the assumption that the variables x are
exogenous with regard to the states j. If they are endogenous, then the
counterfactuals are no longer in the modus potentialis but in the modus
irrealis: If j affects x, then the counterfactual premise “had xi been the
same under a different j” is no longer possible. Suppose that political
regimes affect the rate of growth of the labor force – which they in
fact do – which in turn enters into the list of the variables explaining
growth. Then we cannot conjure a counterfactual in which the labor
force would grow at the same rate under the two regimes. Hence, when
we calculate the values corrected for selection, but not when we gen-
erate unbiased estimators, we should correct only for those x variables
that are exogenous with regard to j. But this, in turn, calls for testing
the exogeneity of each x, and may lead to a regress.

289
 Appendix II

Codebook

All of the variables used in this study are listed here in alphabetical
order. Information about sources is given in brackets. Variables with
no citation of source may be (1) coded by the authors, (2) derived from
existing variables, or (3) matters of common knowledge. The names
for all variables are in capital letters.

ACCHEAD: Number of changes of chief executives (HEADS) accumu-

lated during the life of a particular political regime as defined by
REG.
AGEA: Age in years of the current regime as classified by AUT.
AGEH: Number of years chief executive has been in power. The year the
chief executive comes to power is coded 1. In cases in which the chief
executive changed more than once during one year (HEADS > 1),
AGEH = 1. Also, AGEH = 1 for the first year in countries that became
independent after 1950 (even in cases where the chief executive
served as a prime minister or governor general of the colony). AGEH
is missing for Switzerland, Uruguay (1951–1966), and Yugoslavia
after 1980 because of their collective executives. [Banks, Day, and
Muller 1997; da Graça 1985; Bienen and Van de Walle 1991]
AGEI: Age in years of the current regime as classified by INST.
AGER: Age in years of the current regime as classified by REG. The year
in which the regime comes into existence is coded as 1. When applic-
able, ages were extended back as far as 1870.
AGETOGO: Number of years before a regime transition (as classified by
REG). The transition year is coded as 0. Countries that never experi-
enced a transition during the 1950–1990 period, and the last regime
in each country, are coded as missing.
ASPELL: Number of successive spells of political regimes as classified
by AUT. A spell is defined as years of continuous rule under the same
regime.

291
 Appendix II: Codebook

AUT: Classification of political regimes in which dictatorships are

distinguished by the existence of a legislature (elected or appointed):
coded 0 if democracy; 1 if bureaucracy (dictatorship with a leg-
islature); 2 if autocracy (dictatorship without a legislature). Coding
adjusted for transitional regimes. A transition year is coded as the
regime that emerges in that year.
BIRTHS: Crude birth rate per 1,000 people. [World Bank 1994, series
374]
BRITCOL: Dummy variable coded 1 for every year in countries that were
British colonies at any time after 1918, and coded 0 otherwise.
CAPW: Capital per worker, defined as KS/LF.
CATH: Percentage of Catholics in the population. [LPS 1993/94]
CIVMIL: Civil–military relations: coded 1 if a government was controlled
by a non-military component of the nation’s population; 2 if an out-
wardly civilian government was effectively controlled by a military
elite; 3 if there was direct rule by the military, usually (but not
necessarily) following a military coup d’état; 4 if other (all regimes
not falling into one or another of the foregoing categories, including
instances in which a country, save for reasons of exogenous influence,
lacked an effective national government). [Banks 1996]
COMEX: Dummy variable coded 1 for countries that are primary-
commodity exporters, as defined by the International Monetary Fund
(IMF), and coded 0 otherwise. [IMF 1994]
CONTRACEPTION: Percentage of female population using contraception.
[World Bank 1994, series 369]
CPI: Rate of inflation: annual rate of growth of the consumer price index.
[IMF 1994]
DEADLOCK: Dummy variable coded 1 for the cases in which the largest
legislative party controls more than one-third but less than one-half
of seats (33.3 < SEATS £ 50), and coded 0 otherwise.
DEATHS: Crude death rate per 1,000 people. [World Bank 1994, series
375]
DEMADIC: Dummy variable coded 1 for all the years of a democracy
that follows a dictatorship (as defined by REG), and coded 0
otherwise.
DEMBDIC: Dummy variable coded 1 for all the years of a democracy
that precedes a dictatorship (as defined by REG), and coded 0
otherwise.
DEMONSTRATIONS: Any peaceful public gathering of at least 100
people for the primary purpose of displaying or voicing their opposi-
tion to government policies or authority, excluding demonstrations of
a distinctly anti-foreign nature. [Banks 1996]

292
Appendix II: Codebook

DICADEM: Dummy variable coded 1 for all the years of a dicta-

torship that follows a democracy (as defined by REG), and coded 0
otherwise.
DICBDEM: Dummy variable coded 1 for all the years of a dictatorship
that precedes a democracy (as defined by REG), and coded 0
otherwise.
DIVIDED: Classification of political regimes in which dictatorships are
distinguished by the number of formal powers: coded 0 if democracy;
1 if dictatorship with a legislature or at least one political party; 2 if
dictatorship with executive only. A transition year is coded as the
regime that emerges in that year.
EDT: Cumulative years of education of the average member of the labor
force (Bhalla-Lau-Louat series). [Bhalla 1994]
EDTG: Percentage annual rate of growth of EDT.
EFFPARTY: Number of effective parties, defined as 1/(1 - F), where
F = party fractionalization index. [Banks 1996]
ELF60: Index of ethnolinguistic fractionalization, 1960. [Easterly and
Levine 1997]
EXPHAZH: HEADHAZ led one year.
EXPPAD: PAD led one year.
EXPPDA: PDA led one year.
EXSELEC: Mode of effective executive selection: coded 1 if direct elec-
tion (election of the effective executive by popular vote or election
of committed delegates for the purpose of executive selection); 2 if
indirect election (selection of the effective executive by an elected
assembly or by an elected but uncommitted electoral college); 3 if non-
elected (any means of executive selection not involving a direct or indi-
rect mandate from an electorate). [Banks 1996, but modified and
completed where appropriate]
FEMSEC: Gross enrollment ratio, secondary, female. [World Bank 1994,
series 301]
FERTIL: Total fertility rate (births per woman). [World Bank 1994, series
373]
FLAGC: Dummy variable coded 1 for the first year each country is
observed, and coded 0 otherwise.
FLAGE: Dummy variable coded 1 for the last year each country is
observed, and coded 0 otherwise.
FLAGR: Dummy variable coded 1 in the first year of each regime spell
(as classified by REG),and coded 0 otherwise.
G: Annual rate of growth of per capita income (LEVEL).
GDPW: Real GDP per worker, 1985 international prices. [Heston and
Summers 1993; hereafter, PWT 5.6]

293
 Appendix II: Codebook

GINI: Gini index of income distribution. [Deininger and Squire 1996]

GLAG: G lagged one year.
GXPDEDUC: Central-government expenditure on education as a share
of GDP (both at factor cost, current national currency). [World Bank
1994, series 450/series 9]
GXPDHLTH: Central-government expenditure on health as a share of
GDP (both at factor cost, current national currency). [World Bank
1994, series 451/series 9]
GXPDSSEC: Central-government expenditure on social security and
welfare as a share of GDP (both at factor cost, current national
currency). [World Bank 1994, series 452/series 9]
GXPKTOTL: Central-government total capital expenditure as a share of
GDP (both at factor cost, current national currency). [World Bank
1994, series 462/ series S9]
HEADHAZ: Predicted probability that a change of chief executive will
occur during the current year, given CIVMIL, TLAG, LEGELEC, and
PRESELEC.
HEADS: Number of changes of the chief executive in each year. Chief
executives are presidents in presidential democracies, prime minis-
ters in parliamentary and mixed democracies, and whoever are the
effective rulers in dictatorships (designated explicitly as “dictators,”
or as “heads of military juntas,” “presidents,” “leaders of the ruling
party,” “executors of the state of emergency,” or “kings”). Contrary to
Bienen and Van de Walle (1991), we did not exclude acting or provi-
sional governments, on the assumption that one cannot distinguish
between cases in which heads attempted to consolidate power and
failed and cases in which heads did not try to do so. HEADS is
coded as missing for Switzerland, Uruguay from 1951 to 1966, and
Yugoslavia after 1980, cases of a collective executive with specific
rules for rotation of the chief executive. For Portugal, the president
was considered to be the chief executive between 1976 and 1982, and
the prime minister between 1983 and 1990. [da Graça 1985; Bienen
and Van de Walle 1991; Banks 1996]
INCUMB: Consolidation of incument advantage. Dummy variable coded
1 if (1) the regime year qualifies as a democratic regime and (2)
at some time during their current tenure in office the incumbents
(person, party, military hierarchy) unconstitutionally closed the lower
house of the national legislature and rewrote the rules in their favor.
INEQ: Ratio of income shares of the top quintile to the bottom quintile
of income recipients. [Deininger and Squire 1996]
INFMORT: Infant mortality rate per 1,000 live births. [World Bank 1994,
series 379]

294
Appendix II: Codebook

INST: Classification of political regimes in which democracies are

distinguished by the type of executive: coded 0 if dictatorship; 1 if
parliamentary democracy; 2 if mixed democracy; 3 if presidential
democracy. A transition year is coded as the regime that emerges in
that year.
INST2: Identical to INST except for the cases that resulted in a transi-
tion from above (by the incumbent). In those cases the incumbent rule
(INCUMB) for classification of democracies and dictatorships was not
applied, and the regime was classified as parliamentary, mixed, or
presidential. [Carlson 1998]
INV: Real gross domestic investment (private and public) as a percent-
age of GDP (both in 1985 international prices). [PWT 5.6]
INVLAG: INV lagged one year.
KS: Capital stock per capita. Capital-stock data come from PWT 5.6 for
1,578 observations in 62 countries. The remaining cases are gener-
ated according to the following procedure: (1) Regress (in logs) output
on capital stock and labor force for the observations for which data
are available, using a fixed-effects model. (2) Regress the country-
specific fixed effects on initial LEVEL and regional dummy variables,
and fill for the remaining 73 countries. (3) Use the estimated coeffi-
cients from (1) and the predicted fixed effects from (2) to fill both for
the missing years in countries for which some data are available and
for all years in countries for which they are not. Retain the original
observations whenever available, and fill with the estimates when
they are not.
KSG: Growth of KS.
LAGREG: Lagged REG. In this case, transition years are coded as the
regime that dies. For instance, there was a transition from democ-
racy to dictatorship in Argentina in 1955. In that year, LAGREG = 0.
LEGELEC: Number of elections held for the national lower chamber in
a given year. [Mostly taken from Banks 1996, complemented by
Keesing’s Contemporary Archives]
LEGSELEC: Legislative selection: coded 0 if no legislature exists
(includes cases in which there is a constituent assembly without ordi-
nary legislative powers); 1 if there is a non-elected legislature (exam-
ples include the selection of legislators by the effective executive, or
on the basis of heredity or ascription); 2 if there is an elected legisla-
ture (legislators, or members of the lower house in a bicameral
system, are selected by means of either direct or indirect popular elec-
tion). [Banks 1996, but modified and completed where appropriate]
LEVEL: Real GDP per capita, 1985 international prices, chain index.
[PWT 5.6]

295
 Appendix II: Codebook

LEVLAG: LEVEL lagged one year.

LEVSQR: LEVEL squared.
LF: Labor force, derived from PWT 5.6.
LFAGRIC: Labor in agriculture (% of total labor force). [World Bank
1994, series 309]
LFG: Labor-force growth; annual rate of growth of LF.
LFGLAG: LFG lagged one year.
LFLAG: LF lagged one year.
LFPW: Labor force, female (% of total). [World Bank 1994, series 308]
LIVES: Life expectancy at birth, total (in years). [World Bank 1994,
series 387]
LS: Labor share of value added in manufacturing. [World Bank 1994,
series UM VAD WAGE ZS]
MA2GL: Two-year moving average of G, lagged one year.
MA3GL: Three-year moving average of G, lagged one year.
MAJORITY: Dummy variable coded 1 for the cases in which the largest
legislative party controls more than one-half of seats (SEATS > 50),
and coded 0 otherwise.
MINORITY: Dummy variable coded 1 for the cases in which the largest
legislative party controls up to one-half of seats (SEATS £ 50), and
coded 0 otherwise.
MOBILIZATION: Sum of STRIKES, DEMONSTRATIONS, and RIOTS.
MOBILIZE: Classification of political regimes in which dictatorships are
distinguished by the presence of political parties: coded 0 if democ-
racy; 1 if mobilizing dictatorship (with parties); 2 if exclusionary dic-
tatorship (without parties). A transition year is coded as the regime
that emerges in that year.
MOSLEM: Percentage of Moslems in the population. [LPS 1993/94]
NEWC: Dummy variable coded 1 for every year in countries that became
independent after 1945, and coded 0 otherwise.
ODRP: Other democracies in the regions, percentage. Percentage of
democratic regimes (as defined by REG) in the current year (other
than the regime under consideration) in the REGION to which the
country belongs. For example, in 1980 Kenya had an authoritarian
regime. The number of democracies in the region equaled 4, and the
total number of regimes in the region equaled 44. ODRP for Kenya
1980 then equaled 4/44. In turn, Austria had a democratic regime in
1980. The number of other democracies in the region was 24, and
the total number of regimes in the region was 25. ODRP for Austria
in 1980 then equaled 24/25.
ODWP: Other democracies in the world, percentage. Percentage of
democratic regimes (as defined by REG) in the current year (other

296
Appendix II: Codebook

than the regime under consideration) in the world. Constructed in the

same way as ODRP.
OIL: Dummy variable coded 1 if the average ratio of fuel exports to total
exports in 1984–1986 exceeded 50%, and coded 0 otherwise. [IMF
1994]
OPENC: Exports and imports as a share of GDP (both in 1985 interna-
tional dollars). [PWT 5.6]
OVER65: Proportion of people over 65 years of age. [World Bank 1994,
series 356–8, 364–9]
PAD: Probability that a dictatorship will not survive in the current year,
given STTR, ODRP, RELDIF, MOSLEM, NEWC, and a dummy variable
for OECD countries.
PARTY: Number of political parties in a given year: coded 0 if no parties
(political parties were banned, or elections were held on a non-
partisan basis, or incumbents used their electoral victory to establish
non-party rule by banning all parties, or the current term in office
ended up in a later year in the establishment of non-party rule); 1 if
one party (the share of seats in the lower house of the national legis-
lature held by the largest party was 100%, or there was only one party
list presented to voters, or incumbents used the electoral victory to
establish one-party rule by banning all opposition parties or forcing
them to merge with the ruling party, or the current term in office
ended up in a later year in the establishment of one-party rule); 2 if
more than one party. [Banks 1996, but modified and completed where
appropriate]
PDA: Probability that a democracy will not survive in the current year,
given STTR, ODRP, RELDIF, MOSLEM, NEWC, and a dummy variable
for OECD countries.
PINV: Price level of investment. [PWT 5.6]
POP: Population, in thousands. [PWT 5.6]
POPG: Annual rate of growth of population. [PWT 5.6]
PPPQUAL: Data quality (ranging from 0 = worst to 10 = best) of Penn
World Tables, version 5.5, by country. [Appendix A:2 in Summers and
Heston 1991]
PRESELEC: Number of presidential elections held in a given year.
[Mostly taken from Banks 1996, complemented by Keesing’s Contem-
porary Archives]
PRIME: US prime rate. [IMF 1994]
PROT: Percentage of Protestants in the populaton. [LPS 1993/94]
RECOVERY: Dummy variable coded 1 for the first five years after the
end of a war (WAR).
REG: Dummy variable coded 1 for dictatorships and 0 for democracies.

297
 Appendix II: Codebook

A transition year is coded as the regime that emerges in that year.

For instance, there was a transition from democracy to dictatorship
in Argentina in 1955. In that year, REG = 1.
REGION: Region of the world, coded 1 if Latin America and Caribbean;
2 if Middle East; 3 if Eastern Europe; 4 if Africa; 5 if South Asia; 6 if
East Asia; 7 if OECD. [World Bank 1994]
REGION2: Region of the world, coded 1 if sub-Saharan Africa; 2 if South
Asia; 3 if East Asia; 4 if Southeast Asia; 5 if Pacific islands/Oceania;
6 if Middle East/North Africa; 7 if Latin America; 8 if Caribbean and
non-Iberic America; 9 if Eastern Europe/Soviet Union; 10 if industrial
countries.
RELDIF: Index of religious fractionalization, based on CATH, PROT,
MOSLEM, and other religions.
RIOTS: Number of violent demonstrations or clashes of more than 100
citizens involving the use of physical force. [Banks 1996]
RSPELL: Number of successive spells of political regimes as classified
by REG. A spell is defined as years of continuous rule under the same
regime.
SEATS: Percentage of seats in the lower house of the national legisla-
ture that are held by the largest political party or party list. [Based on
Banks 1996, but modified and completed where appropriate]
SIZE: POP * LEVEL of each country.
STRA: The sum of past transitions to authoritarianism (as defined by
REG) in a country. If a country experienced one or more transitions
to authoritarianism before 1950, STRA was coded 1 in 1950.
STRAL: STRA lagged one year.
STRD: The sum of past transitions to democracy (as defined by REG) in
a country. If a country experienced one or more transitions to democ-
racy before 1950, STRD was coded 1 in 1950.
STRIKES: Number of strikes of 1,000 or more industrial or service
workers that involved more than one employer and were aimed at
national-government policies or authority. [Banks 1996]
STTR: Sum of STRA and STRD.
T: Cumulative years of life of a regime from the first observation.
TINC: Regime transitions (including transitions by incumbents), as
coded by INST2: coded 0 for years in which there was no transition;
1 for years in which TTI = 1; and 2 for years in which there was a
regime transition led by the current incumbent. TINC includes all the
transitions in TTI (except transitions from one type of democracy to
another, two of them in Brazil and one in France), although five cases
in which TTI = 1 were recoded as TTI = 2.
TLAG: TURNOVER lagged one year.

298
Appendix II: Codebook

TTI: Dummy variable coded 1 if a regime transition (as classified by

INST) occurred at any time during the current year, and the regime
at the end of the year was different from the regime at the beginning
of the year, and coded 0 otherwise.
TTR: Dummy variable coded 1 if a regime transition (as classified by
REG) occurred at any time during the current year and the regime at
the end of the year was different from the regime at the beginning of
the year, and coded 0 otherwise.
TTRLAG: TTR lagged one year.
TTRLEAD: TTR led one year.
TURNOVER: Rate of turnover of chief executives per year of life of a
regime. Defined as ACCHEAD/T, where T is the cumulative years of
life of the regime from the first observation.
TYPEII: Dummy variable coded 1 if the “alternation” rule for classify-
ing regimes as democracies or dictatorships applies, and coded 0
otherwise.
UNDER15: Proportion of people under 15 years of age. [World Bank
1994, series 356–8, 364–9]
UNREST: The same as MOBILIZATION.
UNSTABLE: Dummy variable coded 1 for all the years in a country that
experienced at least one regime transition (as defined by REG)
between 1950 and 1990.
WAGE: Annual wage rate, defined as GDPW * LS.
WAR: Dummy variable coded 1 when there was a war of any type (inter-
national or civil) on the territory of a country, and coded 0 otherwise.
[Singer and Small 1994]
WORLD: Average annual rate of growth of LEVEL for all the countries
in the world, weighted by SIZE.
XR: Exchange rate (national currency relative to the U.S. dollar). [PWT
5.6]
YEAR: Calendar year.
YG: Annual rate of growth of real GDP (LEVEL), defined as G + POPG.
YGLAG: YG lagged one year.

299
 References

Achen, Christopher H. 1986. The Statistical Analysis of Quasi-Experiments.

Berkeley: University of California Press.
Ahmad, Zakaria Haji. 1988. Malaysia: Quasi Democracy in a Divided Society. In
Democracy in Developing Countries: Asia, edited by L. Diamond, J. J. Linz, and
S. M. Lipset, pp. 347–83. Boulder, CO: Lynne Rienner Publishers.
Ake, Claude. 1967. A Theory of Political Integration. Homewood, IL: Dorsey Press.
Alesina, Alberto. 1995. Elections, Party Structure, and the Economy. In Modern
Political Economy, edited by J. S. Banks and E. A. Hanushek, pp. 145–70.
Cambridge University Press.
Alesina, Alberto, Sule Özler, Nouriel Roubini, and Phillip Swagel. 1996. Political
Instability and Economic Growth. Journal of Economic Growth 1(June):189–211.
Alesina, Alberto, and Roberto Perotti. 1996. Income Distribution, Political Instabil-
ity, and Investment. European Economic Review 40:1203–28.
1997. The Politics of Growth: A Survey. In Government and Growth, edited by
V. Bergström, pp. 11–60. Oxford University Press.
Almond, Gabriel A., and Sidney Verba. 1963. The Civic Culture: Political Attitudes
and Democracy in Five Nations. Princeton, NJ: Princeton University Press.
Alvarez, Michael. 1997. Presidentialism and Parliamentarism: Which Works?
Which Lasts? Ph.D. dissertation, Department of Political Science, University of
Chicago.
Amemiya, Takeshi. 1984. Tobit Models: A Survey. Journal of Econometrics 24:3–61.
1985. Advanced Econometrics. Cambridge, MA: Harvard University Press.
Arat, Zehra F. 1988. Democracy and Economic Development: Modernization Theory
Revisited. Comparative Politics 21(1):21–36.
Arndt, Heinz Wolfgang. 1987. Economic Development: The History of an Idea.
University of Chicago Press.
Atkinson, A. B. 1970. On the Measurement of Economic Inequality. Journal of
Economic Theory 2:244–63.
Banks, Arthur S. 1996. Cross-National Time-Series Data Archive (magnetic tape).
Binghamton, NY: Center for Social Analysis, State University of New York at
Binghamton.
Banks, Arthur S., Alan J. Day, and Thomas C. Muller (eds.). 1997. Political Hand-
book of the World, 1997. Binghamton, NY: CSA Publications.
Barro, Robert J. 1989. Economic Growth in a Cross Section of Countries. In NBER
Working Paper no. 3120. Cambridge, MA: National Bureau of Economic
Research.

301
 References

1990. Government Spending in a Simple Model of Economic Growth. Journal of

Political Economy 98(5):103–25.
1997. Determinants of Economic Growth. Cambridge, MA: MIT Press.
Barro, Robert J., and Xavier Sala-i-Martin. 1995. Economic Growth. New York:
McGraw-Hill.
Becker, Gary S., and Gregg H. Lewis. 1973. On the Interaction between the Quan-
tity and Quality of Children. Journal of Political Economy 81:279–88.
Becker, Gary S., Kevin M. Murphy, and Robert Tamura. 1990. Human Capital,
Fertility, and Economic Growth. Journal of Political Economy 98(5):12–37.
Benhabib, Jess, and Andres Velasco. 1996. On the Optimal and Best Sustainable
Taxes in an Open Economy. European Economic Review 40(1):134–54.
Bhalla, Surjit S. 1994. Freedom and Economic Growth: A Virtuous Cycle? Paper
presented at the Nobel Symposium Democracy’s Victory and Crisis. Uppsala
University, August 27–30.
Bienen, Henry, and Nicolas Van de Walle. 1991. Of Time and Power: Leadership
Duration in the Modern World. Stanford, CA: Stanford University Press.
Bobbio, Norberto. 1989. Democracy and Dictatorship. Minneapolis: University of
Minnesota Press.
Bollen, Kenneth A. 1980. Issues in the Comparative Measurement of Political Democ-
racy. American Sociological Review 45:370–90.
1993. Liberal Democracy: Validity and Method Factors in Cross-National Mea-
sures. American Journal of Political Science 37:1207–30.
Bollen, Kenneth A., and Robert W. Jackman. 1989. Democracy, Stability, and
Dichotomies. American Sociological Review 54:438–57.
Burkhart, Ross E., and Michael S. Lewis-Beck. 1994. Comparative Democracy: The
Economic Development Thesis. American Political Science Review 88(4):903–10.
Burns, J. H. 1969. J. S. Mill and Democracy, 1829–61. In Mill, edited by J. B.
Schneewind. London: Macmillan.
Cardoso, Fernando Henrique. 1979. On the Characterization of Authoritarian
Regimes in Latin America. In The New Authoritarianism in Latin America,
edited by David Collier, pp. 33–60. Princeton, NJ: Princeton University Press.
Carey, John M. 1997. Institutional Design and Party Systems. In Consolidating the
Third Wave Democracies: Themes and Perspectives, edited by L. Diamond,
M. F. Platter, Y.-H. Chu, and H.-M. Tien, pp. 67–92. Baltimore: Johns Hopkins
University Press.
Carlson, Rolf Eric. 1998. Presidentialism in Africa: Explaining Institutional Choice.
Ph.D. dissertation, Department of Political Science, University of Chicago.
Cassen, Robert. 1994. Overview: Population and Development: Old Debates and New
Conclusions. In Population and Development: Old Debates, New Conclusions,
edited by R. A. C. Cassen, pp. 1–28. Washington, DC: Overseas Development
Council.
Cheibub, José Antonio. 1998. Political Regimes and the Extractive Capacity of
Governments: Taxation in Democracies and Dictatorships. World Politics 50(3):
349–76.
Cheibub, José Antonio, and Adam Przeworski. 1997. Government Spending and
Economic Growth under Democracy and Dictatorship. In Understanding Democ-
racy: Economic and Political Perspectives, edited by A. Breton, G. Galeotti,
P. Salmon, and R. Wintrobe, pp. 107–24. Cambridge University Press.
Clague, Christopher, Philip Keefer, Stephen Knack, and Mancur Olson. 1997. Insti-
tutions and Economic Performance: Property Rights and Contract Enforcement.
In Institutions and Economic Development: Growth and Governance in Less-

302
References

Developed and Post-Socialist Countries, edited by C. Clague, pp. 67–90.

Baltimore: Johns Hopkins University Press.
Coale, Ansley J., and Edgard M. Hoover. 1958. Population Growth and Economic
Development in Low Income Countries: A Case Study of India’s Prospects.
Princeton, NJ: Princeton University Press.
Collier, Paul, and Jan Willem Gunning. 1999. Explaining African Economic Perfor-
mance. Journal of Economic Literature 37(1):64–111.
Collini, Stefan, Donald Winch, and John Burrow. 1983. That Noble Science of
Politics. Cambridge University Press.
Coppedge, Michael. 1992. (De)institutionalization of Latin American Party Systems.
Paper read at the XVII International Congress of the Latin American Studies
Association, 24–27 September, 1992, Los Angeles.
Coppedge, Michael, and Wolfgang H. Reinicke. 1990. Measuring Polyarchy. Studies
in Comparative International Development 25:51–72.
Cosslett, Stephen R. 1991. Semiparametric Estimation of a Regression Model with
Sampling Selectivity. In Nonparametric and Semiparametric Methods in Econo-
metrics and Statistics, edited by W. A. Barnett, J. Powell, and G. Tauchen,
pp. 175–98. Cambridge University Press.
Coulon, Christian. 1988. Senegal: The Development and Fragility of Semidemocracy.
In Democracy in Developing Countries: Africa, edited by L. Diamond, J. J. Linz,
and S. M. Lipset, pp. 141–78. Boulder, CO: Lynne Rienner Publishers.
da Graça, John V. 1985. Heads of State and Government. New York University Press.
Cramer, J. S. 1986. Econometric Applications of Maximum Likelihood Methods.
Cambridge University Press.
Crook, Nigel. 1997. Principles of Population and Development. Oxford University
Press.
Cukierman, Alex, Sebastian Edwards, and Guido Tabellini. 1992. Seigniorage and
Political Instability. American Economic Review 82(3):537–55.
Dahl, Robert A. 1971. Polyarchy. New Haven, CT: Yale University Press.
1990. Transitions to Democracy. Paper read at the symposium Voices of Democ-
racy, March 16–17, University of Dayton Center for International Studies.
Dasgupta, Partha. 1993. An Inquiry into Well-being and Destitution. Oxford
University Press.
1995. The Population Problem: Theory and Evidence. Journal of Economic Liter-
ature 33(December):1879–902.
Deininger, Klaus, and Lyn Squire. 1996. A New Data Set Measuring Income Inequal-
ity. World Bank Economic Review 10(3):565–91.
de Schweinitz, Karl, Jr. 1959. Industrialization, Labor Controls, and Democracy.
Economic Development and Cultural Change 7(4):385–404.
Diamond, Larry. 1992. Economic Development and Democracy Reconsidered. In
Reexamining Democracy: Essays in Honor of Seymour Martin Lipset, edited by
G. Marks and L. Diamond, pp. 93–139. Newbury Park, CA: Sage Publications.
Diamond, Larry, and Juan J. Linz. 1989. Introduction: Politics, Society, and Democ-
racy in Latin America. In Democracy in Developing Countries: Latin America,
edited by L. Diamond, J. J. Linz, and S. M. Lipset, pp. 1–58. Bouder, CO: Lynne
Rienner Publishers.
Drèze, Jean. 1992. From the “Value of Life” to the Economics and Ethics of Popu-
lation: The Path Is Purely Methodological. Recherches Economiques de Louvain
58.
Drèze, Jean, and Amartya Sen. 1989. Hunger and Public Action. Oxford University
Press.

303
 References

Easterlin, Richard A. 1998. Growth Triumphant: The Twenty-First Century in

Historical Perspective. Ann Arbor: University of Michigan Press.
Easterly, William, Michael Kremer, Lant Pritchett, and Lawrence H. Summers. 1993.
Good Policy or Good Luck? Country Growth Performance and Temporary
Shocks. Journal of Monetary Economics 32(3):459–83.
Easterly, William, and Ross Levine. 1997. Africa’s Growth Tragedy: Policies and
Ethnic Divisions. Quarterly Journal of Economics 112(November):1203–50.
Eisenstadt, S. N. 1968. The Protestant Ethic Theses in the Framework of Sociolog-
ical Theory and Weber’s Work. In The Protestant Ethic and Modernization: A
Comparative View, edited by S. N. Eisenstadt, pp. 3–45. New York: Basic Books.
Fearon, James D. 1991. Counterfactuals and Hypothesis Testing in Political Science.
World Politics 43:169–95.
Feierabend, Ivo K., and Rosalind L. Feierabend. 1966. Aggressive Behaviors Within
Polities, 1948–1962: A Cross-National Study. Journal of Conflict Resolution
10(September):249–71.
Figueiredo, Argelina Cheibub, and Fernando Limongi. 2000. Presidential Power,
Legislative Organization, and Party Behavior in Brazil. Comparative Politics
32(2):151–70.
Findlay, Ronald. 1990. The New Political Economy: Its Explanatory Power for LDCs.
Economics and Politics 2(2):193–221.
Frank, Robert H. 1997. The Frame of Reference as a Public Good. The Economic
Journal 107(November):1832–47.
Freedom House. 1992. Freedom in the World: Political Rights and Civil Liberties.
New York: Freedom House.
Galenson, Walter. 1959. Labor and Economic Development. New York: Wiley.
Galor, Oded, and David N. Weil. 1996. The Gender Gap, Fertility, and Growth.
American Economic Review 86:374–87.
Gambetta, Diego. 1987. Were They Pushed or Did They Jump? Cambridge Univer-
sity Press.
Gastil, Raymond D. 1980. Freedom in the World: Political Rights and Civil
Liberties. New York: Freedom House.
1990. Freedom in the World: Political Rights and Civil Liberties. New York:
Freedom House.
Gellner, Ernest. 1991. Civil Society in Historical Context. International Social Science
Journal 129:495–510.
Goldberger, Arthur. 1983. Abnormal Selection Bias. In Studies in Econometrics, Time
Series, and Multivariate Statistics, edited by S. Karlin, T. Amemiya, and
L. Goodman, pp. 67–84. Stamford, CT: Academic Press.
Goodman, Nelson. 1979. Fact, Fiction, and Forecast. Cambridge, MA: Harvard
University Press.
Greene, William H. 1990. Econometric Analysis. New York: Macmillan.
Gurr, Ted Robert. 1990. Polity II: Political Structures and Regime Change,
1800–1986. Ann Arbor, MI: Inter-University Consortium for Political and Social
Research.
Han, Sung-Joo. 1988. South Korea: Politics in Transition. In Democracy in Devel-
oping Countries: Asia, edited by L. Diamond, J. J. Linz, and S. M. Lipset,
pp. 267–304. Boulder, CO: Lynne Rienner Publishers.
Hausman, Daniel M. n.d. Causation and Counterfactual Dependence Reconsidered.
Unpublished paper, University of Wisconsin, Madison.
Heckman, James J. 1976. The Common Structure of Statistical Models of
Truncation, Sample Selection, and Limited Dependent Variables and a Simple

304
References

Estimator for Such Models. Annals of Economic and Social Measurement

5:475–92.
1978. Dummy Endogenous Variables in a Simultaneous Equation System. Econo-
metrica 46:931–59.
1979. Sample Selection Bias as a Specification Error. Econometrica 47:153–61.
1988. The Microeconomic Evaluation of Social Programs and Economic Institu-
tions. In Chung-Hua Series of Lectures by Invited Eminent Economists, no. 14.
Taipei: The Institute of Economics, Academia Sinica.
1990. Selection Bias and Self-selection. In The New Palgrave: Econometrics, edited
by J. Eatwell, M. Milgate, and P. Newman, pp. 201–24. New York: Norton.
Helliwell, John. 1994. Empirical Linkages Between Democracy and Economic
Growth. British Journal of Political Science 24(December):225–48.
Heston, Alan, and Robert Summers. 1993. Penn World Tables 5.6. Available on
diskette from the National Bureau of Economic Research, Cambridge, MA.
Hirschman, Albert O. 1981. The Social and Political Matrix of Inflation: Elaborations
on the Latin American Experience. In Essays in Trespassing: Economics to
Politics and Beyond, edited by Albert O. Hirschman, pp. 177–207. Cambridge
University Press.
Holm, John D. 1988. Botswana: A Paternalistic Democracy. In Democracy in Devel-
oping Countries: Africa, edited by L. Diamond, J. J. Linz, and S. M. Lipset,
pp. 179–216. Boulder, CO: Lynne Rienner Publishers.
Huber, Evelyne, Dietrich Rueschemeyer, and John D. Stephens. 1993. The Impact
of Economic Development on Democracy. Journal of Economic Perspectives
7(3):71–86.
Huntington, Samuel P. 1968. Political Order in Changing Societies. New Haven, CT:
Yale University Press.
1991. The Third Wave: Democratization in the Late Twentieth Century. Norman:
University of Oklahoma Press.
1993. The Clash of Civilizations? Foreign Affairs 72(3):22–49.
Huntington, Samuel P., and Jorge I. Dominguez. 1975. Political Development. In
Macropolitical Theory, edited by F. I. Greenstein and N. W. Polsby, pp. 1–114.
Reading, MA: Addison-Wesley.
Huntington, Samuel P., and Joan M. Nelson. 1976. No Easy Choice: Political
Participation in Developing Countries. Cambridge, MA: Harvard University
Press.
Hurwitz, Leon. 1973. Contemporary Approaches to Political Stability. Comparative
Politics 5(3):449–63.
IMF. 1994. International Financial Statistics. Washington, DC: International Mone-
tary Fund.
Inkeles, Alex. 1990. Introduction. Studies in Comparative International Develop-
ment 25:1–16.
Islam, Nazrul. 1995. Growth Empirics: A Panel Data Approach. Quarterly Journal
of Economics 110(4):1127–70.
Jackman, Robert W. 1973. On the Relation of Economic Development to Democra-
tic Performance. American Journal of Political Science 17:611–21.
1974. Political Democracy and Social Equality: A Comparative Analysis. American
Sociological Review 39:29–45.
Jackson, Karl D. 1988. The Philippines: The Search for a Suitable Democratic
Solution, 1946–1986. In Democracy in Developing Countries: Asia, edited by
L. Diamond, J. J. Linz, and S. M. Lipset, pp. 231–68. Boulder, CO: Lynne Rienner
Publishers.

305
 References

Johnson, Norman Lloyd, and Samuel Kotz. 1970. Distribution in Statistics:

Continuous Univariate Distributions, 2 vols. Boston: Houghton Mifflin.
Jones, Mark P. 1995a. Electoral Laws and the Survival of Presidential Democracies.
Notre Dame, IN: Notre Dame University Press.
1995b. A Guide to the Electoral Systems of the Americas. Electoral Studies
14(1):5–21.
Kaldor, Nicolas. 1956. Alternative Theories of Distribution. Review of Economic
Studies 23:83–100.
Kamiński, Marek. 1997. Jak komuniści mogli zachować wᐗadzȩ w 1989 roku. Rzecz
. .
o (nie)kontrolowalnej odwilzy, sondazach opinii publicznej i ordynacji wybor-
czej. Studia Socjologiczne 2:5–35.
Kavka, Gregory S. 1986. Hobbesian Moral and Political Theory. Princeton, NJ:
Princeton University Press.
Keesing’s Contemporary Archives. Several years. London: Keesing’s Limited.
Kelsen, Hans. 1945. General Theory of Law and State. Cambridge, MA: Harvard
University Press.
King, Gary. 1989. Unifying Political Methodology: The Likelihood Theory of Statis-
tical Inference. Cambridge University Press.
Kobrin, Stephen J. 1980. Foreign Enterprise and Forced Divestment in LDCs.
International Organization 34(1):65–88.
La Palombara, Joseph. 1963. Bureaucracy and Political Development. Princeton,
NJ: Princeton University Press.
Leamer, Edward E. 1983. Let’s Take the Con Out of Econometrics. American
Economic Review 73(1):31–43.
1985. Sensitivity Analyses Would Help. American Economic Review 75(3):308–13.
Lee, L. F. 1978. Unionism and Wage Rates: A Simultaneous Equations Model With
Qualitative and Limited Dependent Variables. International Economic Review
19:415–33.
1982. Some Approaches to the Correction of Selectivity Bias. Review of Economic
Studies 49:355–72.
1983. Generalized Econometric Models with Selectivity. Econometrica 51:507–12.
Levine, Ross, and David Renelt. 1992. A Sensitivity Analysis of Cross-Country
Growth Regressions. American Economic Review 82(4):942–63.
Lewis, Bernard. 1993. Islam and Liberal Democracy. Atlantic Monthly 271(2):89–98.
Lewis, David. 1973. Conterfactuals and Comparative Possibility. Journal of Philo-
sophical Logic 2:418–46.
Lijphart, Arend (ed.). 1992. Parliamentary versus Presidential Government. Oxford
University Press.
Linz, Juan J. 1975. Totalitarian and Authoritarian Regimes. In Handbook of Politi-
cal Science: Macropolitical Theory, edited by F. I. Greenstein and N. W. Polsby,
pp. 175–373. Reading, MA: Addison-Wesley.
1984. Democracy: Presidential or Parliamentary. Does It Make a Difference?
Manuscript, Department of Political Science, Yale University.
1990a. The Perils of Presidentialism. Journal of Democracy 1(1):51–69.
1990b. The Virtues of Parliamentarism. Journal of Democracy 1(4):84–91.
Lipset, Seymour Martin. 1959. Some Social Requisites of Democracy: Economic
Development and Political Legitimacy. American Political Science Review
53(1):69–105.
1960. Political Man: The Social Bases of Politics. Garden City, NY: Double day.
1981. Political Man: The Social Bases of Politics, expanded edition. Baltimore:
Johns Hopkins University Press.

306
References

Livi-Baci, Massimo. 1997. A Concise History of World Population, 2nd ed. Oxford:
Blackwell.
Lloyd, Cynthia B. 1994. Investing in the Next Generation: The Implications of High
Fertility at the Level of the Family. In Population and Development: Old Debates,
New Conclusions, edited by R. A. C. Cassen, pp. 181–202. Washington, DC:
Overseas Development Council.
Londregan, John B., and Keith T. Poole. 1990. Poverty, the Coup Trap, and the
Seizure of Executive Power. World Politics 42:151–83.
1992. The Seizure of Executive Power and Economic Growth: Some Additional
Evidence. In The Political Economy of Business Cycles and Growth, edited by
A. Cukierman, Z. Hercovitz, and L. Leiderman, pp. 51–79. Cambridge, MA: MIT
Press.
LPS. Leksykon Państw Swiata 1993/94. Warszawa: Real Press.
Lucas, Robert E. 1988. On the Mechanics of Economic Development. Journal of
Monetary Economics 22:3–42.
Macaulay, Thomas B. 1900. Complete Writings, vol. 17. Boston: Houghton Mifflin.
(Originally published 1:42.)
McDonald, Ronald H., and J. Mark Ruhl. 1989. Party Politics and Elections in Latin
America. Boulder, CO: Westview Press.
McGuire, Martin C., and Mancur Olson. 1996. The Economics of Autocracy and
Majority Rule. Journal of Economic Literature 34:72–97.
Macpherson, C. B. 1966. The Real World of Democracy. Oxford: Clarendon
Press.
Maddala, G. S. 1983. Limited-Dependent and Qualitative Variables in Econometrics.
Cambridge University Press.
Maddison, Angus. 1995. Monitoring the World Economy, 1820–1992. Paris: Orga-
nization for Economic Co-operation and Development.
Mainwaring, Scott. 1993. Presidentialism, Multipartism, and Democracy: The
Difficult Combination. Comparative Political Studies 26(2):198–228.
Mainwaring, Scott, and Matthew Soberg Shugart. 1997. Conclusion: Presidentialism
and the Party System. In Presidentialism and Democracy in Latin America,
edited by S. Mainwaring and M. S. Shugart, pp. 394–439. Cambridge Univer-
sity Press.
Mankiw, N. G., P. Romer, and D. Weil. 1992. A Contribution to the Empirics of
Economic Growth. Quarterly Journal of Economics 107:407–38.
Manski, Charles F. 1995. Identification Problems in the Social Sciences. Cambridge,
MA: Harvard University Press.
Marks, Gary, and Larry Diamond (eds.). 1992. Reexamining Democracy: Essays in
Honor of Seymour Martin Lipset. Newbury Park, CA: Sage Publications.
Marx, Karl. 1934. The Eighteenth Brumaire of Louis Bonaparte. Moscow: Progress
Publishers.
1952. The Class Struggle in France, 1848 to 1850. Moscow: Progress Publishers.
1971. Writings on the Paris Commune, edited by H. Draper. New York: Interna-
tional Publishers.
May, D. A., and D. M. Heer. 1968. Son Survivorship Motivation and Family Size in
India: A Computer Simulation. Population Studies 22:199–210.
Merrick, Thomas W. 1994. Population Dynamics in Developing Countries. In Popu-
lation and Development: Old Debates, New Conclusions, edited by R. A. C.
Cassen, pp. 79–105. Washington, DC: Overseas Development Council.
Mill, John Stuart. 1991. Representations on Representative Government. Buffalo,
NY: Prometheus. (Originally published 1861.)

307
 References

Montesquieu. 1995. The Spirit of the Laws, translated by Thomas Nugent. New York:
Hafner Press. (Originally published 1748.)
Moore, Barrington, Jr. 1966. Social Origins of Dictatorship and Democracy: Lord
and Peasant in the Making of the Modern World. Boston: Beacon Press.
Mueller, Dennis C. 1989. Public Choice II: A Revised Edition of Public Choice.
Cambridge University Press.
Mueller, John. 1992. Democracy and Ralph’s Pretty Good Grocery: Elections, Equal-
ity, and the Minimal Human Being. American Journal of Political Science
36:983–1003.
Muller, Edward N. 1988. Democracy, Economic Development, and Income Inequal-
ity. American Sociological Review 53:50–68.
Mutis, Alvaro. 1996. Adventures of Maqroll: Four Novellas. New York: Horper-
Collins.
North, Douglass. 1990. Institutions, Institutional Change, and Economic Perfor-
mance. Cambridge University Press.
North, Douglass C., and Robert P. Thomas. 1973. The Rise of the Western World: A
New Economic History. Cambridge University Press.
North, Douglass C., and Barry W. Weingast. 1989. The Evolution of Institutions
Governing Public Choice in 17th Century England. Journal of Economic History
49:803–32.
O’Donnell, Guillermo. 1973. Modernization and Bureaucratic-Authoritarianism:
Studies in South American Politics. Berkeley: Institute of International Studies,
University of California.
1979. Postscript: 1979. In Modernization and Bureaucratic-Authoritarianism:
Studies in South American Politics. Berkeley: Institute of International Studies,
University of California.
O’Donnell, Guillermo, and Philippe C. Schmitter. 1986. Transitions from Authori-
tarian Rule: Tentative Conclusions about Uncertain Democracies. Baltimore:
Johns Hopkins University Press.
Olson, Mancur, Jr. 1963. Rapid Growth as a Destabilizing Force. Journal of
Economic History 23(December):529–52.
Olson, Mancur. 1982. The Rise and Decline of Nations: Economic Growth, Stagfla-
tion, and Social Rigidities. New Haven, CT: Yale University Press.
1991. Autocracy, Democracy, and Prosperity. In Strategy and Choice, edited by
R. J. Zeckhauser, pp. 131–57. Cambridge, MA: MIT Press.
1993. Dictatorship, Democracy, and Development. American Political Science
Review 87(3):567–76.
1997. The New Institutional Economics: The Collective Choice Approach to
Economic Development. In Institutions and Economic Development: Growth
and Governance in Less-Developed and Post-Socialist Countries, edited by
C. Clague, pp. 37–64. Baltimore: Johns Hopkins University Press.
Ozbudun, Ergun. 1988. Turkey: Crises, Interruptions, and Reequilibration. In
Democracy in Developing Countries: Asia, edited by L. Diamond, J. J. Linz, and
S. M. Lipset, pp. 187–230. Boulder, CO: Lynne Rienner Publishers.
Parsons, Talcott. 1951. The Social System. Glencoe, IL: Free Press.
Pasinetti, Luigi. 1961. Rate of Profit and Income Distribution in Relation to the Rate
of Economic Growth. Review of Economic Studies 29(October):267–79.
Pindyck, Robert S. 1991. Irreversibility, Uncertainty, and Investment. Journal of
Economic Literature 29(3):1110–48.
Pritchett, Lant H. 1994. Desired Fertility and the Impact of Population Policies.
Population and Development Review 20:1–55.

308
References

Przeworski, Adam. 1975. Institutionalization of Voting Patterns, or Is Mobilization

the Source of Decay? American Political Science Review 69:49–67.
1990. The State and the Economy under Capitalism. Chur, Switzerland: Harwood
Academic Publishers.
1991. Democracy and the Market. Cambridge University Press.
1997. Bᐗȩdy: Informacja a przejście do demokracji. Studia Socjologiczne 2:51–66.
Przeworski, Adam, José Antonio Cheibub, and Fernando Limongi. 1997. Culture and
Democracy. In World Culture Report, pp. 157–87. Paris: UNESCO.
Przeworski, Adam, and Fernando Limongi. 1993. Political Regimes and Economic
Growth. Journal of Economic Perspectives 7(3):51–70.
1997. Development and Democracy. In Democracy’s Victory and Crisis, edited by
A. Hadenious. Cambridge University Press.
Przeworski, Adam, Bernard Manin, and Susan C. Stokes (eds.). 1999. Democracy
and Accountability. Cambridge University Press.
Pudney, Stephen. 1989. Modelling Individual Choice: The Econometrics of Corners,
Kinks, and Holes. Cambridge University Press.
Quah, Danny T. 1993. Galton’s Fallacy and Tests of the Convergence Hypothesis. In
Endogenous Growth, edited by T. M. Andersen and K. O. Moene, pp. 37–54.
Oxford: Blackwell.
1996. Twin Peaks: Growth and Convergence in Models of Distribution Dynamics.
The Economic Journal 106:1045–55.
Ram, Rati. 1986. Government Size and Economic Growth: A New Framework and
Some Evidence from Cross-Section and Time-Series Data. American Economic
Review 76:191–203.
Rama, Martin. 1993. Empirical Investment Equations for Developing Countries. In
Striving for Growth after Adjustment: The Role of Capital Formation, Regional
and Sectoral Studies, edited by Luis Serven and Andres Solimano, pp. 107–43.
Washington, DC: World Bank.
Rao, Vaman. 1984. Democracy and Economic Development. Studies in Compara-
tive International Development 19(4):67–81.
Ray, Debraj. 1998. Development Economics. Princeton, NJ: Princeton University
Press.
Razin, Assaf, and Efraim Sadka. 1995. Population Economics. Cambridge, MA: MIT
Press.
Riker, William. 1965. Democracy in America, 2nd ed. New York: Macmillan.
Robinson, James. 1995. Theories of “Bad Policy”. Policy Reform 1:1–17.
Rodrik, Dani. 1998. Democracies Pay Higher Wages. In NBER Working Paper no.
6364. Cambridge, MA: National Bureau of Economic Research.
Roemer, John E. 1982. A General Theory of Exploitation and Class. Cambridge, MA:
Harvard University Press.
Root, F. R., and A. Ahmed. 1979. Empirical Determinants of Manufacturing Direct
Investment in Developing Countries. Economic Development and Cultural
Change 27(4):751–7.
Rose, Leo E. 1988. Pakistan: Experiments with Democracy. In Democracy in Devel-
oping Countries: Asia, edited by L. Diamond, J. J. Linz, and S. M. Lipset, pp.
105–42. Boulder, CO: Lynne Rienner Publishers.
Roy, A. D. 1951. Some Thoughts on the Distribution of Earnings. Oxford Economic
Papers 3:135–46.
Russett, Bruce. 1964. Deaths from Domestic Group Violence per 1,000,000 Popula-
tion, 1950–1962. In World Handbook of Political and Social Indicators, edited
by B. E. A. Russett. New Haven, CT: Yale University Press.

309
 References

Rustow, Dankwart A. 1970. Transitions to Democracy: Toward a Dynamic Model.

Comparative Politics 2(3):337–64.
Sah, Raaj K., and Joseph Stiglitz. 1988. Committees, Hierarchies and Polyarchies.
The Economic Journal 98(June):451–70.
Schmertmann, Carl P. 1994. Selectivity Bias Correction Models in Polychotomous
Sample Selection Models. Journal of Econometrics 60:101–32.
Schmitter, Philippe C. 1991. The International Context for Contemporary Democra-
tization: Constraints and Opportunities upon the Choice of National Institutions
and Policies. In East-South System Transformations, working paper no. 8,
University of Chicago.
Schneider, F., and B. S. Frey. 1985. Economic and Political Determinants of Foreign
Direct Investment. World Development 13:161–75.
Schultz, T. W. (ed.). 1974. Economics of the Family: Marriage, Children and Human
Capital. Chicago: National Bureau of Economic Research.
Schumpeter, Joseph A. 1942. Capitalism, Socialism, and Democracy. London:
George Allen & Unwin.
Sen, Amartya. 1981. Poverty and Famines: An Essay on Entitlement and Depriva-
tion. Oxford University Press.
1988. The Concept of Development. In Handbook of Development Economics,
edited by T. N. Srinivasan. Amsterdam: Elsevier Science.
1993. The Economics of Life and Death. Scientific American May, 40–7.
1994a. Freedoms and Needs. The New Republic, January 10–7, 31–7.
1994b. Population: Delusion and Reality. The New York Review of Books
41(15):1–8.
1995. Population Policy: Authoritarianism versus Cooperation, International
Lecture Series on Population Issues. New Delhi, India: The John D. and Cather-
ine T. MacArthur Foundation.
Shugart, Matthew Soberg, and John M. Carey. 1992. Presidents and Assemblies:
Constitutional Design and Electoral Dynamics. Cambridge University Press.
Singer, J. David, and Melvin Small. 1994. Correlates of War Project: International
and Civil War Data, 1816–1992 [computer file]. Ann Arbor, MI: J. David Singer
and Melvin Small [producers], 1993. Ann Arbor, MI: Interuniversity Consortium
for Political and Social Research [distributor], 1994.
Starr, Harvey. 1991. Democratic Dominoes: Diffusion Approaches to the Spread
of Democracy in International System. Journal of Conflict Resolution 35:356–
81.
Stepan, Alfred, and Cindy Skach. 1993. Consitutional Frameworks and Democratic
Consolidation: Parliamentarism Versus Presidentialism. World Politics 46(1):1–
22.
Stolzenberg, Ross M., and Daniel A. Relles. 1990. Theory Testing in a World of Con-
strained Research Design. Sociological Methods and Research 18:395–415.
Summers, Robert, and Alan Heston. 1991. The Penn World Table (Mark 5): an
Expanded Set of International Comparisons, 1950–1988. Quarterly Journal of
Economics May:327–68.
Temple, Jonathan. 1999. The New Growth Evidence. Journal of Economic Litera-
ture 37(1):112–56.
Tetlock, Philip E., and Aaron Belkin. 1996. Counterfactual Thought Experiments in
World Politics: Logical, Methodological, and Psychological Perspectives. In Coun-
terfactual Thought Experiments in World Politics: Logical, Methodological, and
Psychological Perspectives, edited by P. E. Tetlock and A. Belkin, pp. 1–38.
Princeton, NJ: Princeton University Press.

310
References

Therborn, Goran. 1977. The Rule of Capital and the Rise of Democracy. New Left
Review 103(May–June):3–41.
Vanhanen, Tatu. 1992. The Process of Democratization. New York: Crane, Russak.
Wantchekon, Leonard. 1996. Political Coordination and Democratic Instability.
Unpublished manuscript, Yale University.
Weber, Max. 1958. The Protestant Ethic and the Spirit of Capitalism. New York:
Scribner. (Originally published 1904.)
Weffort, Francisco. 1992. Qual Democracia? São Paulo: Companhia das Letras.
Whitehead, Laurence. 1991. The International Dimension of Democratization: A
Survey of the Alternatives. Paper read at the XV congress of the International
Political Science Association, Buenos Aires, Argentina.
World Bank. 1984. World Development Report 1984. Washington, DC: World Bank.
1991. World Development Report 1991. Washington, DC: World Bank.
1994. World Data 1994. World Bank Indicators on CD-ROM. Washington, DC:
World Bank.
1995. World Development Indicators 1995. Washington, DC: World Bank. CD-
ROM.
1997. World Development Indicators 1997. Washington, DC: World Bank. CD-
ROM.
Zavoina, R., and W. McKelvey. 1975. A Statistical Model for the Analysis of Ordinal
Level Dependent Variables. Journal of Mathematical Sociology 4(Summer):
103–20.
Zielinski, Jakub. 1997. The Polish Transition to Democracy: A Game-Theoretic
Approach. European Archives of Sociology 36:135–58.
Zimmerman, Ekkart. 1987. Government Stability in Six European Countries During
the World Economic Crisis of the 1930s: Some Preliminary Considerations.
European Journal of Political Research 15:23–52.
1988. Economic and Political Reactions to the World Economic Crisis of the 1930s
in Six European Countries. International Studies Quarterly 32:305–34.

311
 Author Index

Ahmad, Zakaria Haji, 21 Coulon, Christian, 26n

Ahmed, A., 188n Crook, Nigel, 226n, 235n
Ake, Claude, 187 Cukierman, Alex, 188n, 201n
Alesina, Alberto, 188n, 191n, 196,
201n, 205, 211 daGraça, John V., 291
Almond, Gabriel A., 126n Dahl, Robert A., 15, 16, 33, 34, 35n,
Amemiya, Takeshi, 137, 288 79n, 101
Arat, Zehra F., 79n Dasgupta, Partha, 5, 6n, 42n, 243,
Atkinson, A. B., 121 246, 247, 248
Day, Alan J., 291
Banks, Arthur S., 19, 19n, 35n, de Schweinitz, Karl, 2, 142
291, 292, 293, 294, 295, 297, Deininger, Klaus, 117, 294
298 Diamond, Larry, 79, 89, 112
Barro, Robert, J., 143n, 144, 164, Dominguez, Jorge I., 142
190n, 211, 217, 217n, 248, 259 Drèze, Jean, 144, 254
Becker, Gary S., 244n, 257
Belkin, Aaron, 9 Easterlin, Richard A., 235
Benhabib, Jess, 164n Easterly, William, 182, 293
Bhalla, Surjit S., 293 Edwards, Sebastian, 188n
Bienen, Henry, 291, 294 Eisenstadt, S. N., 126
Bobbio, Norberto, 15n, 32
Bollen, Kenneth A., 55, 56, 56n, 57, Feierabend, Ivo K., 188n
79n Feierabend, Rosalind L., 188n
Burrow, John, 210 Figueiredo, Argelina C., 136
Findlay, Ronald, 144
Cardoso, Fernando Henrique, 18 Frank, Robert H., 6n
Carey, John M., 31, 134 Freedom House, 20, 34, 79n
Carlson, Rolf Eric, 295 Frey, B. S., 188n
Cassen, Robert, 244n, 265n
Cheibub, José Antonio, 126, 163n, Galenson, Walter, 2, 142
164n, 211 Galor, Oded, 248, 249
Clague, Christopher, 200n, 209 Gambetta, Diego, 284
Coale, Ansley J., 235 Gastil, Raymond D., 34, 56
Collier, Paul, 273 Gellner, Ernest, 126
Collini, Stefan, 210 Goldberger, Arthur, 287
Coppedge, Michael, 34, 55, 56 Greene, William H., 288

313
 Author Index

Gunning, Jan Willem, 273 Macaulay, Thomas, 210

Gurr, Ted Robert, 56 McDonald, Ronald H., 26n
McGuire, Martin C., 201
Han, Sung-Joo, 24, 26n McKelvey, W., 79
Heckman, James J., 147n, 156, 222, Macpherson, C. B., 14n
236, 286, 287, 288 Maddison, Angus, 103
Heer, D. M., 246 Mainwaring, Scott, 129n, 134
Helliwell, John, 217n Manin, Bernard, 33
Heston, Alan, 293 Mankiw, N. G., 180n, 217n
Holm, John D., 23 Manski, Charles F., 279, 281
Hoover, Edgard M., 235 Marks, Gary, 79
Huber, Evelyne, 89 Marx, Karl, 210
Huntington, Samuel P., 3, 34, 35, 40, May, D. A., 246
45, 92, 98, 99, 109, 126, 142, 143, Merrick, Thomas W., 224
187, 212 Mill, J. S., 34n, 125
Hurwitz, Leon, 188n Montesquieu, 15, 32, 126
Moore, Barrington, Jr., 103
IMF, 292, 297 Mueller, Dennis, 196
Inkeles, Alex, 56 Mueller, John, 34n
Islam, Nazrul, 181 Muller, Edward N., 33
Muller, Thomas C., 291
Jackman, Robert W., 33, 57, 79n Murphy, Kevin M., 257
Jackson, Karl D., 26n Mutis, Alvaro, 163n
Jones, Mark P., 134
Nelson, Joan M., 3, 35, 98
Kaldor, Nicolas, 143n North, Douglass C., 144, 206, 209,
Kamiński, Marek, 116 209n
Kavka, Gregory S., 32
Keefer, Philip, 200n O’Donnell, Guillermo, 18, 34, 89, 97,
Kelsen, Hans, 15 98, 99, 99n
Knack, Stephen, 200n Olson, Mancur, 109, 144, 189, 196,
Kobrin, Stephen J., 211, 211n 200n, 201, 209
Ozbudun, Ergun, 25
La Palombara, Joseph, 2 Özler, Sule, 191n
Leamer, Edward E., 180
Lee, L. F., 284n, 288n Parsons, Talcott, 187
Levine, Ross, 180, 293 Pasinetti, Luigi, 143n
Lewis, Bernard, 126 Perotti, Roberto, 188n, 196, 211
Lewis, Gregg H., 244n Pindyck, Robert S., 206n
Lijphart, Arend, 30n Poole, Keith T., 196
Limongi, Fernando, 126, 136, 178, 216 Pritchett, Lant H., 244
Linz, Juan J., 18, 32, 112, 129 Przeworski, Adam, 16, 33, 34, 97,
Lipset, Seymour Martin, 11, 78, 79, 89, 116, 126, 144, 164n, 178,
99, 101, 103, 126, 285 216
Livi-Baci, Massimo, 224, 226n, 235, Pudney, Stephen, 58, 283
236, 237, 238, 242, 243, 244, 277
Lloyd, Cynthia B., 236, 244n, 251 Quah, Danny T., 160, 180n, 277
Londregan, John B., 196
LPS (Leksykon Państw Swiata), 292, Ram, Rati, 164
296, 297 Rama, Martin, 147
Lucas, Robert E., 216 Rao, Vaman, 143, 143n

314
Author Index

Ray, Debraj, 235, 246, 246n, 247, 248, Stepan, Alfred, 134
249, 256 Stephens, John D., 89
Razin, Assaf, 244n, 247n Stiglitz, Joseph, 144n
Reinicke, Wolfgang H., 55, 56 Stokes, Susan C., 33
Relles, Daniel A., 287 Stolzenberg, Ross M., 287
Renelt, David, 180 Summers, Robert, 293
Riker, William, 33 Swagel, Phillip, 191n
Robinson, James, 144
Rodrik, Dani, 170 Tabellini, Guido, 188n
Roemer, John E., 168n Tamura, Robert, 257
Romer, P., 180n Temple, Jonathan, 176n, 180n, 182
Rose, Leo E., 26n Tetlock, Philip E., 9
Roubini, Nouriel, 191n Therborn, Goran, 89
Rueschemeyer, Dietrich, 89 Thomas, Robert P., 209
Ruhl, J. Mark, 26n
Russett, Bruce, 188n van de Walle, Nicolas, 291, 294
Rustow, Dankwart A., 103 Vanhanen, Tatu, 34
Velasco, Andres, 164n
Sadka, Efraim, 244n, 247n Verba, Sidney, 126n
Sah, Raaj K., 144n
Sala-i-Martin, Xavier, 143n, 190n, 217, Wantchekon, Leonard, 97
248 Weber, Max, 126
Schmertmann, Carl P., 288n Weffort, Francisco, 33
Schmitter, Philippe C., 18, 83n, 97 Weil, David N., 180n
Schneider, F., 188n Weingast, Barry, 209, 209n
Schultz, T. W., 243 Whitehead, Laurence, 83n
Schumpeter, Joseph A., 15 Winch, Donald, 210
Sen, Amartya, 4, 5, 6, 144, 234, 244, World Bank, 168n, 181, 246n, 292,
247, 251, 255n, 265 293, 294, 296, 297
Shugart, Matthew Soberg, 31, 134
Skach, Cindy, 134 Zavoina, R., 79
Squire, Lyn, 117, 294 Zielinski, Jakub, 97
Starr, Harvey, 83n Zimmerman, Ekkart, 128n

315
 Country Index

Afghanistan, 116 Botswana Democratic Party (BDP),

Africa (also African), 4, 18, 36, 48, 23
109, 142, 211, 242n, 262, 273, Masire, Quett Ketumile, 23
277 Brazil, 2, 7, 22, 31, 47, 49, 50n, 96,
Albania, 36n 103, 120n, 190n, 259, 298
Algeria, 42n, 166 Figueiredo, João Batista, 22
Angola, 36 Neves, Tancredo, 22
Antigua and Barbuda, 36n Brunei, 36n
Argentina, 3, 17, 26, 48, 50n, 87, 98, Bulgaria, 42, 47, 94, 97, 103, 106
101, 106, 131, 171, 173, 194 Burkina Faso, 42n, 50n, 52n
Illia, Arturo, 17 Lamizana, Sangoulé, 52n
Malvinas Islands, 89 Burma, 83
Videla, Jorge, 19 Burundi, 42n, 259
Viola, Roberto, 18
Armenia, 36n Cameroon, 42n
Asia, 26, 142 Cape Verde Islands, 36, 149
Australia, 21n Central African Republic, 42n, 191n,
Austria, 94, 106, 171, 173, 296 259
Azerbaijan, 36n Boganda, Bartholomy, 191n
Central America, 4
Bahamas, 24, 168n, 176 Ceylon, 83
Bahrain, 36, 77n Chad, 42n
Bangladesh, 26, 36, 42, 47, 87 Chile (also Chilean), 3, 7, 8, 16, 31n,
Barbados, 24, 42n 46n, 47, 50n, 87, 95, 106, 120n,
Belarus, 36n 127, 171, 211
Belgium, 35n, 94 Allende, Salvador, 17
Belize, 36, 87 El Mercurio, 17
Benin, 42n, 50n, 259 Pinochet, Augusto, 3
Bhutan, 36n China, 2, 3, 47, 98, 170, 244, 272
Bolivia, 26, 29, 47, 48, 50n, 190n, Deng Tsao Ping, 19n
191n, 211 Mao Zedong, 144
Barrientos, René 191n Colombia, 26, 26n, 47, 106, 259, 262
Movimiento Nacionalista Gómez, Laureano, 26n
Revolucionario (MNR), 211 Communist countries, 87, 116
Bosnia and Herzegovina, 36n Comoro Islands, 36
Botswana, 23, 25, 26, 28, 42n Congo, 42n, 47, 83, 129

316
Country Index

Costa Rica, 166, 176 Haiti, 47

Croatia, 36n Honduras, 26, 26n, 35, 47, 48, 50n,
Cuba, 36n 194
Cyprus, 36n National Party, 26n
Czech Republic, 36n Hungary, 42, 47, 94, 116
Czechoslovakia, 36n, 42, 47, 96, 116,
166 Iceland, 94
India (also Indian), 21, 83, 87, 103,
Democratic Republic of Korea (North 120n, 226, 239n, 246, 259, 272,
Korea), 36n 277
Dominica, 36n Kerala, 244
Dominican Republic, 24, 29, 47, 120n Indonesia, 47
Iran, 47, 259, 262
East Asia, 77 Iraq, 176, 189, 191n, 262
Ecuador, 26, 47, 50n Aref, Abdel Salam Muhammad, 191n
Egypt, 26, 28 Ireland, 94
El Salvador, 17, 26, 35, 47, 189, 190n Israel, 17n, 48
Romero Bosque, Pío, 17 Italy, 17n, 94, 106, 239n
Eritrea, 36n Ivory Coast, 42n, 176
Estonia, 36n
Ethiopia, 36n, 46 Jamaica, 7, 20, 24, 42n, 87
Europe (also European), 126, 235, 241, Japan (also Japanese), 23, 24, 25, 26,
243, 244 27, 48, 87, 120n, 162, 176n, 190n,
Eastern Europe, 48, 77, 244, 251 270
Western Europe (also Western Liberal Democratic Party (LDP), 24,
European), 15, 16n, 34, 35n, 48, 26
103, 224 Jordan, 50n, 52n, 176

Finland, 16, 106, 120n Kazakhstan, 36n

France, 19, 31, 35n, 46, 49, 94, 168n, Kenya, 42n, 296
224, 244, 298 Kiribati, 36n
Kuwait, 42n, 50n, 77n
Gabon, 42n, 94, 176, 259, 262 Kyrgyzstan, 36n
Gambia, 26, 28
Georgia, 36n Laos, 36, 47, 52n
Germany, 16, 17n, 106, 190n Souvanna Phouna, 52n
East Germany, 47, 94 Latin America (also Latin American), 4,
West Germany, 87, 94, 176 26, 34, 40, 42, 48, 48n, 87, 126,
Ghana, 36, 48, 49, 87, 129, 176n 127, 129, 171, 251
Great Britain (also England, United Latvia, 36n
Kingdom), 16, 23, 99, 103, 126n, Lebanon, 36n
209n Lesotho, 42n
Blair, Tony, 23 Liberia, 46
Churchill, Winston, 1 Libya, 36n
Conservative Party, 23 Lithuania, 36n
Greece, 47, 50n, 87, 96, 106, 131, 162, Luxembourg, 46
270
Grenada, 24, 47, 50n Macedonia, 36n
Guatemala, 26, 35, 48, 50n, 194 Madagascar, 20, 42n, 259
Guinea, 36, 101n National Front of the Malagasy
Guyana, 26, 28, 42n Revolution (FNDR), 20

317
 Country Index

Madagascar (cont.) Paraguay, 47

Vanguard of the Malagasy Revolution Persian Gulf, 36, 50
(AREMA), 20 Peru, 4, 48, 50n, 87, 106, 189, 194,
Malawi, 42n 211
Malaysia (also Malays), 21, 22, 25, 26, Fujimori, Alberto, 3–4, 87
28, 36, 87, 94 Philippines, 22, 26, 26n, 42, 47, 50n,
Barisan Nasional (BN), 22 83, 120n, 129, 176, 246n
Maldives, 36n Aquino, Benigno, 26n
Mali, 42n Kilusang Bagong Lipuna (KBL, New
Malta, 42n, 162, 176, 176n, 270 Society Movement), 26n
Mauritania, 42n Marcos, Ferdinand, 22, 26n, 205
Mauritius, 7, 24, 42n, 48, 83, 120n, Poland, 17n, 20, 42, 47, 96, 106,
166 116
Mexico (also Mexican), 17, 26, 27, 28, Portugal, 47, 57n, 87, 96, 103, 106,
47, 57n, 87, 94, 173, 190n 162, 166, 173, 211, 270, 294
Partido Revolucionario Institucional Salazar, Antonio de Oliveira, 57n
(PRI), 17, 25, 26
Middle East (also Middle Eastern), 4, Qatar, 36, 50, 77n
48
Moldova, 36n Romania, 2, 47, 116, 149, 176
Mongolia, 3 Russia, 3, 47, 127
Morocco, 36, 50n, 52n Rwanda, 42n, 168n
Mozambique, 36, 191n, 259
Machel, Samora, 191n São Tomé and Príncipe, 36n
Myanmar, 83, 48, 50n Saudi Arabia, 50, 77n
Senegal, 26, 26n, 28, 42n
Namibia, 36n Senegalese Progressive Union (UPS),
Nepal, 47, 52n, 262 26n
Mahendra, Bir Bikram Shah Deva, Senghor, Léopold Sédar, 26n
52n Seychelles, 87
Netherlands, 46, 94 Sierra Leone, 42n, 47, 83
New Zeland, 103 Singapore, 2, 27, 28, 42n, 87, 94,
Nicaragua (also Nicaraguan), 47, 160n, 162, 171, 173, 176, 211,
52–53, 56, 189, 211 270, 272
Ortega, Saavedra, Daniel, 53 Slovakia, 36n
Somoza, Anastasio, 25, 205 Slovenia, 36n
Niger, 42n Solomon Islands, 24, 87
Nigeria, 29, 42n, 47, 49, 83, 109, 129, Somalia, 36n, 42n, 47, 83
176 Somaliland, 36n
Norway, 16, 94, 103, 168n South Africa, 28, 87
South Korea, 2, 22, 24, 25, 26, 26n,
Oman, 36, 50, 77n 28, 42, 48, 49, 87, 96, 162, 176,
211, 246n, 259, 270
Pacific Islands, 77 Park, Chung-Hee, 22, 24, 25, 28
Pakistan, 26, 26n, 42, 48, 49, 83, 103, Rhee, Syngman, 27n
106, 191n Soviet Union (also USSR), 36n, 47, 94,
Haq, Muhammed Zia ul-, 191n 97, 99, 116
Khan, Ayub, 26n Spain, 47, 57n, 87, 94, 97, 106
Panama, 47, 50n, 191n Franco, Francisco, 57, 89
Torrijos, Omar, 191n Sri Lanka, 47, 83, 101, 101n, 120n
Papua New Guinea, 36, 87, 262 St. Kitts and Nevis, 36n

318
Country Index

St. Lucia, 24, 36n Uganda, 35, 42n, 47, 129, 189
St. Vincent and the Grenadines, 36n Ukraine, 36n
Sudan, 35, 36, 48, 50n, 87, 194 United Arab Emirates, 36, 50, 77n
Suriname, 36, 47, 49, 83, 131, 176 United States, 34, 45, 99, 101n, 103,
Swaziland, 42n 109, 116, 126n, 142, 160n, 168n,
Sweden, 168n, 191, 224 239n
Erlander, Tage, 191 Kissinger, Henry, 116
Switzerland, 46, 51n, 291, 294 Roosevelt, Franklin D., 1
Syria, 94 Uruguay, 47, 50n, 51n, 52, 95, 101,
101n, 291, 294
Taiwan (also Taiwanese), 2, 27, 87, 94, Bordaberry, Juan, 52
97, 98, 162, 176, 259, 270
Tajikstan, 36n Vanuatu, 87
Tanzania, 42n Venezuela, 47, 87, 103, 106, 176n, 262
Thailand, 35, 47, 48, 50n, 87, 162, Pérez Jiménez, Marcos, 87
246n, 270
Togo, 42n Western Samoa, 16n, 42n
Trinidad and Tobago, 20, 24, 42n
Tunisia, 36 Yemen Arab Republic, 42n, 176
Turkey, 25, 27, 47, 48, 50n, 87, 176, Yugoslavia, 36n, 51n, 94, 103, 106,
246n 116, 176, 291, 294
Democratic Party (DP), 25
Republican People’s Party (RPP), Zaire, 42n
25 Zambia, 42n, 149
Turkmenistan, 36n Zimbabwe, 42

319
 Subject Index

Alternation in office, 23–30 Dictatorship (also authoritarian

Autocracy, 32–33, 49–54, 255 regime), 2, 3, 13, 18–23, 33, 45,
54, 175–176
Birth rates, 222–233, 241–256, economic development and, 92–
264–265, 269–273 98
Bureaucracy, 32–33, 49–54 and economic development, 4,
142–145
Capital stock longevity of, 45–47
growth of, 149–151 types of, 31–33
public, 164–165
Chief executive, 19, 51 Economic growth, 146, 160, 162–163,
changes of, 51–54, 255 216–218, 269–277
and economic growth, 190, and age of democracy, 196–199
194–196 and the state, 162–166
Civil-military relations, 35, 136 Education
Communism (also communist and fertility, 254
countries), 3, 116, 210, 251 and political regimes, 101
Consolidation of democracy, 101–103, school enrollment, 251–254
141 Elections, 15–30
Consolidation of incumbent rule, 21–22
Contestation, 15–18, 34 Fertility, 241–265, 269–273
and contraception, 244
Death rates (also mortality), 222–241, and death rates, 236–237
264–265, 269–273 and economic growth, 259
and government expenditures, economic theories of, 246–248
237–239 regulation of, 243
Democracy, 2, 7, 13–30, 33–36, 54–55,
175–176 Government expenditures, 162–163
and economic development, 4, capital expenditures, 165
142–145 social expenditures, 237
economic development and, 3,
78–88, 98–106 Human capital, 256–259
longevity of, 45–47
types of, 30–31 Infant mortality, 228, 239–241,
Demographic transition, 224–226, 247
241–242 Investment, 146–149

320
Subject Index

Labor force in wealthy countries, 166–176

growth of, 151–153, 167 and efficiency, 153
participation of women in, 167–168, and fertility, 254–256, 263–264
251, 265 and growth of capital stock, 149–151
Legislature, 19–20, 32, 50 and growth of labor force, 151–153
Life expectancy, 228–233 and investment, 146–149
and labor shares, 170–173
Material well-being, 4–6 and material welfare, 271–272
of women, 6, 265, 271 and political instability, 193, 211–213
Migration, 222 and property rights, 209–211
Mixed democracy, 30–31, 49, 52–54, and wars, 190
129n Poor countries (also poverty), 142,
Modernization theory, 88–89, 92–98 158–166, 178–179, 257, 265,
269–270
Parliamentary democracy, 30–31, 49, Population growth, 217–268, 273–
52–54, 129–136 274
longevity of, 129, 136 Presidential democracy, 30–31, 49,
Political instability, 187–188, 270 52–54, 129–136
and economic growth, 188–189, longevity of, 129, 136
211–213 Probit, 79n, 137–139
current, 189–193
expectations of, 200–211, 213–215 Regime transitions (also transitions),
past, 193–199 40–45, 47–48, 88–92, 273
and investment, 205–211 to democracy, 40–45, 47–48, 92–98
Political parties, 20–22 to dictatorship, 40–45, 47–48, 98
Political regimes and economic growth, 112–116, 190,
classification of, 13–36, 55–59 192–193
definition of, 18 and political mobilization, 114–116
dynamics of, 88–92
and colonial legacy, 126 Selection, 9, 279–289
and cultural heterogeneity, 125–126 and counterfactuals, 9–10, 289
and economic growth, 106–112, endogenous on observables, 9,
123 284–285
and income distribution, 117–122 exogenous on observables, 9,
and international political climate, 283–285
127 on unobservables, 9, 286
and intra-regime instability, Strikes, 173, 192–193
123–125
and per capita income, 92–106, 123 Trade-off between democracy and
and political history, 127 development, 3, 142–143, 178, 271
and types of democracy, 128–136
and economic growth, 142–146, Wars, 35, 178, 189–190, 273–277
153–158, 178–179, 270–271 Waves of democratization, 40–45, 127
in poor countries, 158–166 Wealthy countries, 166–176, 179, 270

321