NÚMERO 139

STATHIS N. KALYVAS MATTHEW ADAM KOCHER

How Free is “Free Riding” in Civil Wars?

Violence, Insurgency, and the Collective
Action Problem

JUNIO 2006

www.cide.edu
Las colecciones de Documentos de Trabajo del CIDE representan
un medio para difundir los avances de la labor de investigación y
para permitir que los autores reciban comentarios antes de su
publicación definitiva. Se agradecerá que los comentarios se hagan
llegar directamente al (los) autor(es).
• D.R. ® 2006. Centro de Investigación y Docencia Económicas,
carretera México-Toluca 3655 (km 16.5), Lomas de Santa Fe, 01210,
México, D.F.
Tel. 5727•9800 exts. 2202, 2203, 2417
Fax: 5727•9885 y 5292•1304.
Correo electrónico: [email protected]
www.cide.edu

Producción a cargo del (los) autor(es), por lo que tanto el contenido

como el estilo y la redacción son su responsabilidad.
 Acknowledgements

Thanks to Abbey Steele for research assistance, and to Zeynep Bulutgil,

Eddie Camp, Sergio Galaz García, Adria Lawrence, Nicholas Sambanis,
Jonah Schulhofer-Wohl, Justin Fox. Versions of this paper were presented at
the Sawyer Seminar on Mass Killing, Center for Advanced Study in the
Behavioral Science (CASBS); the Comparative Politics Workshop,
Department of Political Science, Yale University; and the Laboratory on
Comparative Ethnic Processes (LiCEP). Thanks to all the participants for
their generous attention and thoughful commentary.
 Abstract

That rebels face a collective action problem is one of the most widely shared
assumptions in the literature on civil wars. We argue that the collective
action paradigm can be both descriptively inaccurate and analytically
misleading as applied to civil wars —which are uniquely characterized by
large-scale violence. First, while insurgent collective action may entail the
expectation of future collective benefits, public costs predominate in the
short-term. Second, the costs of nonparticipation and free-riding may equal
or even exceed those of participation. We support these points by
triangulating three different types of evidence: anecdotal evidence from
counterinsurgency operations in a number of civil wars; the unique Phoenix
dataset from the Vietnam War; and data from a regional study of the Greek
Civil War. We conclude by drawing implications for the study of civil wars.

Resumen

El supuesto problema de acción colectiva al que se enfrentan los grupos

rebeldes es una de las ideas más enraizadas en la literatura que estudia las
guerras civiles. La visión de este trabajo es que este paradigma es
descriptivamente impreciso y poco clarificador desde el punto de vista
analítico en contextos como las guerras civiles, caracterizados por la
existencia de una violencia generalizada. En primer lugar, incluso si la
acción colectiva insurgente puede introducir expectativas de recibir
beneficios colectivos en un futuro, en el corto plazo los costos públicos
predominan. En segundo lugar, los costos asociados con el recibimiento de
beneficios de alguna de las partes en conflicto sin participar a cambio (free
riding), o el abstenerse totalmente de tomar parte en el conflicto, pueden
igualar o incluso superar a los que se tendrían de tener una postura
explícitamente beligerante. La evidencia que apoya estas afirmaciones se
basa en tres fuentes diferentes de información: una parte anecdótica
obtenida de operaciones de contrainsurgencia en guerras civiles, el uso de
la base de datos Fénix (Phoenix), utilizada en la Guerra de Vietnam, e
información obtenida de un estudio a nivel regional de la Guerra Civil
Griega. El artículo concluye con la derivación de implicaciones de las
hipótesis defendidas para los estudios de Guerras Civiles.
 How Free is Free-Riding in Civil Wars ?…

Introduction

It had been by no means easy to flee into the mountains and to help set up
what, both in my opinion and in that of friends little more experienced than
myself, should have become a partisan band affiliated with the resistance
movement Justice and Liberty. Contacts, arms, money and the experience
needed to acquire them were all missing. We lacked capable men, and instead
we were swamped by a deluge of outcasts, in good or bad faith, who came
from the plain in search of a non-existent military of political organization, of
arms, or merely of protection, a hiding place, a fire, a pair of shoes.

Primo Levi
Survival in Auschwitz

The social-scientific understanding of individual participation in large-scale

mobilization is informed by theoretical arguments derived primarily from the
collective action paradigm (Olson, 1965). Even though mobilization in the
context of rebellion, insurgency, and civil war diverges from mobilization in
demonstrations and mass protest, both are similarly informed by the
collective action paradigm (Tullock, 1971). References to the collective action
problem that must be overcome if rebellion is to take place are ubiquitous.
According to a recent statement, “rebellion is a full-time commitment and it
is dangerous” (Collier, 2001: 150). Indeed, such references constitute the
foundation for one of the most influential theoretical arguments about civil
war onset: since “the collective action problem for justice-seeking rebellion
would usually be insuperable” it follows that insurgencies would tend to be
about “greed” rather than “grievance” (Collier, 2001: 150).
In this paper, we revisit the collective action paradigm as specifically
applied to the issue of insurgency.1 This paradigm rests on two pillars: first,
the “free-riding” incentive generated by the public good dimension of
insurgency and second, the risks of individual participation in insurgent
collective action. The paper’s contribution lies in its use of novel and
systematic data on violence, suggesting that while it is true that rebels run
serious personal risks in war zones, so do non-rebels. In other words, we argue
that the collective action problem applies only if insurgent collective action is
risky relative to nonparticipation (or free-riding is riskless relative to
participation).
Given the dearth of systematic data and the difficulty of observing
individual private costs and benefits, we do not test this insight directly;
rather, we approach the question using three different bodies of evidence,
each of which has shortcoming and advantages. By triangulating, we mitigate

1 In this paper, we use the terms rebellion, insurgency, and civil war inter-changeably.

DIVISIÓN DE ESTUDIOS INTERNACIONALES 1

Stathis N. Kalyvas and Matthew Adam Kocher

the problems inherent in this type of exercise. We begin by providing

anecdotal evidence from several instances of large-scale violence entailing
harsh counterinsurgency operations of repressive regimes indiscriminately
targeting entire areas. This evidence is highly suggestive, but also
unsystematic. We then proceed to examine a uniquely detailed and
systematic dataset from one of the most sophisticated programs of targeted
violence, the Phoenix Program in Vietnam. We show that even under optimal
conditions, the individual risk of nonparticipation approaches that of
participation, thus eliminating the collective action problem faced by rebel
organizers. The main advantage of the Phoenix dataset lies in its detail, but
its main limitation is its coverage of a particular group of people rather than
the general population. Last, we supply systematic regional evidence from the
Greek Civil War; the data cover all individuals in a single region, combatants
and non-combatants alike –the main problem here being external validity. All
three bodies of evidence converge in suggesting that the risks of participation
in insurgent collective action have been overestimated relative to those of
nonparticipation. In turn, this implies that nonparticipation and free-riding
are often puzzling given the costs they entail.

Violence and the Collective Action Problem

The collective action problem has been the dominant paradigm for the
analysis of political groups and group action in sociology, economics, and
political science since it was introduced by Mancur Olson in the 1960s. Since
the basic outlines of this framework are so widely known, we treat it only
briefly. Individuals value many goods that can be produced only through
collective action. Collective goods are non-rival and non-excludable. That I
enjoy such a good does not in anyway limit your ability to benefit from it; if
the good is provided, everyone can take advantage of it. Political goods like
democracy, the rule of law, or collective defense are classic examples. Yet
individuals also value purely personal goods, such as the time, opportunity
cost, or risk involved in acting collectively. In other words, the benefits of
collective action are public, while the costs are borne privately. The choice of
each individual to work for the collective benefit or not usually has no bearing
on its provision. Under these circumstances, every person’s best move is to
stay home and let someone else work for the public benefit, in short to “free-
ride”. If everyone reasons as they should, public goods will be systematically
underprovided, even when everyone wants them. Olson’s logic makes
collective action for public goods a puzzle: under quite general conditions, it
is irrational. Yet collective action is common, even where and when it is not
enforced or promoted by the public power. A considerable portion of the
empirical literature since Olson has focused on the use of selective incentives

2 CIDE
 How Free is Free-Riding in Civil Wars ?…

and other mechanisms that explain how the collective action problem is
overcome.
There has been much work on the conditions under which the collective
action problem does or does not apply. The bulk of this work applies primarily
to mass mobilization in the context of demonstrations, social protest, and
nonviolent contentious actions and social movements (Tarrow, 1998), i.e., in
contexts short of mass violence.
Nevertheless, the collective action paradigm also dominates current
understandings of insurgent collective action. Tullock’s application (1971) was
the first to challenge prevailing understandings in the revolutions literature of
peasant grievances automatically translating into rebellion. This approach was
further elaborated by Lichbach (1994), applied empirically by Popkin (1979),
and critically extended by Wood (2003). The application of the collective
action framework to the study of insurgent mobilization differs from standard
applications in one key respect: the attention it pays to violence and the
subsequent shift from an exclusive focus on the public good and free-riding
aspect of rebellion to the costs associated with insurgent participation.
Rebel groups typically claim public goods as their goals: secession,
autonomy, democratization, and redistribution appear to be the most
common. Even putting aside the question of how authentic and widely shared
are these aims, they are subject to high levels of uncertainty and very long
time-horizons.2 The ills of civil war, death of self or family, economic
collapse, and forced relocation, are immediate and affect non-rebels as well
as rebels. Given that by 1999, the typical post-WWII civil war lasted 16 years
(Fearon, 2004), and a great many concluded with the victory of the state, the
expected value of the proffered public goods must be considered small
relative to the public ills of civil war. Hence it is not surprising that unlike
collective mobilization and protest, the study of insurgent mobilization has
highlighted the costs of individual participation: in a demonstration is
generally much less costly than in an armed insurgency (Collier, 2001). Thus,
the central implication of the collective action paradigm for the study of
insurgent action is obvious: rebel activists face tremendous obstacles in
starting and sustaining insurgencies.
It is unclear whether the collective action problem applies primarily to the
formation of a core political movement or to massive popular recruitment by
this movement —first movers or late joiners.3 Our focus, in the context of this
paper, is on the latter. It has been argued that first movers are high-risk
political entrepreneurs with strong motivations for whom the collective action

2 Note, however, that it seems entirely logical, for example, that the order that exists at t-1 (pre-insurgency) is a

public good that individuals prefer not to sacrifice, and not joining the insurgency de facto contributes to the
provision of the former good. Scholars have not considered order adequately.
3 This is not the same distinction between onset and duration of civil war. In the empirical literature, onset is

computed on the basis of fatalities, a computation that assumes already large rebel organizations.

DIVISIÓN DE ESTUDIOS INTERNACIONALES 3

Stathis N. Kalyvas and Matthew Adam Kocher

problem does not apply, Elster (1989) calls them “everyday Kantians”.4 The
world does not lack Che Guevaras ready to launch insurgencies and as likely to
fail. What it does lack is followers willing to take the necessary risks; it is the
success of entrepreneurs in recruiting followers that produces insurgencies. In
Collier’s (2001: 143) formulation: “The Michigan Militia was unable to grow
beyond a handful of part-time volunteers, whereas the FARC in Colombia has
grown to employ around 12,000 people.”
The puzzle, then, is to explain extant participation given both highly
costly sanctions and the posited public good nature of rebellion. There are
two types of responses. A first body of research looks for private selective
incentives powerful enough to overcome the cost of participation. Actual
members of rebel groups enjoy several types of benefits that are both rival
and excludable, such as “loot” —but also power and security. The enormous
literature on “selective incentives” within the collective action tradition
provides ample evidence that rebels usually receive such private material
incentives for participation above and beyond any anticipated public goods.5
The voluminous civil war literature on “greed” and looting of the past decade
is a case in point. A second body of research strays out of the narrow
rationalist paradigm by emphasizing non-rational behavior (Muller and Opp
1986), pointing to “opportunity structures” (Tilly 1978; Brockett 2005),
stressing “in-process” benefits (Wood, 2003), and underscoring social ties
(Gould, 1995; Petersen, 2001).
In this paper we take a look at the collective action paradigm by granting
it its assumptions. Our analysis challenges its automatic applicability to cases
of civil war, which are uniquely characterized by large-scale violence. In
many, if not most, instances of large-scale violence, obstacles to collective
action are much lower than assumed. This is so for two reasons. First, while
insurgent collective action may entail the expectation of future collective
benefits, in fact public costs predominate in the short-term. In other words,
the classic collective action formulation of collective benefits vs. private
costs misdescribes the nature of violent conflict. Second, the costs of
nonparticipation and free-riding may equal or even exceed those of
participation. In other words, it may be safer to join in insurgent collective
action or the difference between participation and nonparticipation in terms
of risk incurred may be too small to be relevant, thus priming decisions along
dimensions other than risk. Hence, we argue, the collective action paradigm
can be both descriptively inaccurate and analytically misleading as applied to
civil wars.

4 According to Elster (1989: 192), “everyday Kantianism” states that “one should cooperate if and only if universal

cooperation is better for everybody than universal defection” (192).

5 Selective incentives is the term of art used for the individual side-payments that are dispensed by organizations

to overcome the Olsonian logic of collective action. See Lichbach (1996; 1995: 215–238) for evidence of the
widespread existence of the private benefits that accrue to rebels.

4 CIDE
 How Free is Free-Riding in Civil Wars ?…

The literature has hinted at some of these points. Several scholars have
attempted to understand how large-scale violence shapes individual
incentives for participation (e.g. Lichbach, 1987; Mason, 1989; Mason and
Krane, 1989; Maranto and Tuchman, 1992; Wood, 2003; Brocket, 2005).
Because the focus has been on individuals choosing or not to join a rebel
movement, these studies have focused on the role of government repression.
The findings, however, are inconclusive and ambiguous. Lichbach and Gurr
(1981) suggest a U-shaped relationship between repression and dissent, where
higher levels of repression lead to more resentment and bring new groups into
fold, increasing rebellion. In Why Men Rebel, in contrast, Gurr (1970:270)
proposed a concave function, where anger over initial repression gives way to
fear. Indeed, the effect repression will have on the organization and
maintenance of an insurgent group could be to dampen the recruitment
possibilities by raising the costs for individuals to join and the group to
continue —as posited in most studies. For instance, Tullock (1971:90) assumes
that, while the risk of injury or death is a constant cost for individuals, if the
government increased the punishment of collaborators if revolution fails, then
the net expected utility of the individual rebel decreases. Hence, repression
only leads to a lower likelihood of rebellion. Lichbach (1987: 269), quoting
Greene (1974: 112), considers the opposite outcome: violence used by a
government against its own citizens may be seen as arbitrary, which would
tend to “lower the government’s legitimacy and raise the society’s
revolutionary potential”. As a result, Lichbach concludes, “the apathetic
become politicized, the reformers become radicalized, and the
revolutionaries redouble their efforts. Thus when the government follows a
policy of coercion, the policy itself may become the target of dissent by new
challenging groups, thereby spreading conflict and engulfing the entire
nation” and repression “radicalizes” previous “free-riders” to the revolution.
In this formulation of the effects of repression, Lichbach does not consider
the violence of repression per se, which may alter an individual’s cost-benefit
calculation. Wood (2003) argues that the insurgents in El Salvador successfully
framed the government violence as illegitimate and thus mobilized additional
supporters.
Partly as a result of the absence of conclusive findings, the
overwhelmingly shared assumption of the collective action literature is that
participation in violent collective action is “risky”, and this risk is usually
conceptualized as the principal individual (expected) cost paid by rebels: in
Gould’s (1995: 204) formulation, “while activists might have little trouble
persuading a casual acquaintance to sign a petition, they would have great
difficulty convincing such a person to risk injury, death, or imprisonment.”
While it is undoubtedly true that rebels run serious personal risks in war
zones, so do non-rebels. The relevant question, therefore, is whether violent
collective action is risky relative to nonparticipation. Because of the obvious

DIVISIÓN DE ESTUDIOS INTERNACIONALES 5

Stathis N. Kalyvas and Matthew Adam Kocher

obstacles in tackling this question directly, we formulate and supply evidence

for the following claim: insurgent participation is much less dangerous,
relative to non-participation, than posited by the collective action paradigm
and generally thought. We suspect that, in actual fact, there is considerable
variation across the landscape of violent cases, and perhaps even greater
variation over space and time within cases. It could very well be, for instance,
that in certain instances non-participation is, on average, more dangerous
than participation.
A significant obstacle to the development of a research program about the
effect on violence on collective action has been the dearth of systematic
data—especially high quality disaggregated data. In fact, this question has not
been examined systematically within the collective action literature, even
though on it rests a central, vital assumption of the paradigm. Often, it is
ignored completely.
Consider, for instance, two excellent recent additions to the literature on
dissidence and collective violence, both of which use a micro-level research
design and take their point of departure from the collective action problem.
Wood (2003: 8-10) asserts that the insurgent participation of the Salvadoran
agricultural collectivists she studied was “risky,” citing evidence that
campesinos constituted a very high proportion of the victims of state violence.
Wood notes that: “The vast majority (more than 85%) of the serious acts of
violence… were carried out by state agents or those acting under the direction
of state agents against alleged supporters of opposition organizations
(emphasis added).” While the evidence is sufficient to establish that being a
poor Salvadoran farmer was very risky during this period, there is no way to
judge from the evidence how participants in collective action differed from
non-participants in their average risk.6 Wood (2003: 12) further notes that
“class differences among the campesinos of the case-study areas do not
explain differences in their participation. The evidence presented here from
the case-study areas show that participants in the insurgency came from a
variety of poor rural class backgrounds. The many campesinos who joined
government networks and civil patrols or served as government informants
came from equally diverse economic backgrounds.” In other words, rural class
position is not a reliable proxy for participation in dissident collective action.
Hence, aggregated data cannot be used to reliably assess the relative risk of
participation: knowing that many campesinos died is not informative regarding
the risk of being a rebel.
Similarly, Petersen (2001) stipulates that participants in the anti-
occupation violence of wartime and post-war Lithuania ran enormous risks.

6 Furthermore, Wood (2003: 151-3) reports that in the regions of El Salvador where she conducted ethnographic

research, a stalemate had emerged and there was generally low violence against civilians exercised by both sides,
with the insurgents having an advantage in the exercise of local control. During this period, she estimates that about
one third of the peasants in the area she studied collaborated with the insurgents.

6 CIDE
 How Free is Free-Riding in Civil Wars ?…

Much like Wood, he cites only aggregated figures on victimization: “Hundreds

of thousands deported, tens of thousands killed, 200,000 Jews murdered, and
so on” (2001: 302). Were Soviet counterinsurgents effective in identifying and
selecting guerrillas and their supporters from the Lithuanian population?
Petersen’s own evidence suggests that Soviet counterinsurgent violence was
indiscriminate in the extreme.
We suspect the reason for this common elision is a lack of careful
analytical and descriptive attention to the nature of warfare in civil war. War
differs from other types of violence, in both quantitative and qualitative
terms. It is not simply that more people die in wars, but that they may be
targeted in ways that differ from other violent contexts. In riots or violent
protests, we reasonably suppose that staying home dramatically attenuates
individual risk.7 When war passes through an area, the risk of being victimized
is distributed much more widely.
Perhaps more important is that risk varies across specific types of warfare.
In conventional warfare, typical in interstate contexts, combatants almost
always run greater risks than civilians. Conventional warfare exhibits
frontlines that signal clearly to non-combatants where risk is maximized. Even
when armies do not particularly care to discriminate, civilians can increase
their security by moving away from the battlefield, while soldiers are
concentrated precisely at the point of decision where individual risk reaches
its maximum.
By contrast, wars fought in irregular fashion tend to implicate civilians in a
way that is much more direct and consequential. The reason is that rebel
fighters, as well as the spies and agents of both the state and the rebels, hide
among the civilian population. This feature of irregular war, also known as the
“identification problem” (Kalyvas, 2006), is present in the following remark of
John Kerry from his experience in Vietnam (in Brinkley, 2003: 50): “Wherever
I went and young Vietnamese men would look at me I grew scared. There
really was no way to tell who was who. You could be in a room with one and
not know whether he was really a Charlie [i.e. a Vietcong] or not. It became
easy to sense the distrust that must exist in the outlying areas. How could one
really fight in the fields and know whether at any time the men beside you
were not going to turn tail and train their guns on you? Whom did you begin to
trust and where did you draw the line?”
A consequence of the identification problem is that armed actors may
target the civilian population indiscriminately, i.e. using a series of very rough
“profiles”, such as a person’s ethnicity, locality, sex, or age (Kalyvas, 2006).8

7 Quite obviously, this claim holds only for those who choose whether or not to participate, not for those who

may be targeted by rioters during ethnic riots and pogroms.

8 The difference between selective or targeted and indiscriminate violence lies in the level of target selection:

individualized selection makes for selective violence whereas aggregation (village, region, ethnic group, etc.)
produces indiscriminate violence.

DIVISIÓN DE ESTUDIOS INTERNACIONALES 7

Stathis N. Kalyvas and Matthew Adam Kocher

If this is the case, then shunning participation in the rebellion and free-riding
may actually prove deadlier than joining it, since the rebels may be in a
position to offer a degree of protection.9 Adding the benefit of risking one’s
life with a modicum of dignity and honor, as well as a sense of choice in one’s
fate (as opposed to being killed in an undignified way without even the
possibility of acting), makes the difference even starker.
The danger that truly indiscriminate violence may “push the population
into the arms of the enemy” has been widely appreciated in the literature on
insurgency and civil war (see the following section). Yet even violence that is
intended to discriminate between combatants and civilians may be
sufficiently inaccurate as to create similar pressures. We explore this
possibility in a later section by focusing on a “hard case”, an instance in which
there is substantial evidence of an intent to discriminate, yet violence fell
disproportionately on the innocent.
Whether violence is indiscriminate or merely poorly selective in intent,
there are good reasons to suppose that it is perversely selective in effect. The
reason is that rebel combatants have access to skills, resources, and networks
that should promote their survival relative to non-combatants. Rebel
organizations warn their members of approaching raids, provide safe-houses,
bunkers, escape routes, and food caches, and train their members in
concealment, evasion, and survival. In some civil wars, particularly in African
countries with large proportions of the population living close to the
subsistence line, combatants may be the only people in a position to avoid
war-induced famine.
If large-scale violence entails public ills and private benefits, then the
theoretical and empirical puzzle is precisely the converse of the one raised by
the collective action problem: if participation is rational for everyone, why
don’t we see more of it? Once violence becomes sufficiently generalized, we
should expect to see precisely the opposite of what the collective action
literature predicts. Rather than organizations chasing individuals, offering
incentives to get them over the risk threshold to cooperation, we should
observe that organizations reject willing candidates. Below we present some
evidence to that effect.

9 In Gross’s formulation (1979: 212) about the German occupation of Poland: “One would expect that

noncompliance with German demands carried such drastic penalties that scarcely anyone would dare to defy them.
But full compliance was impossible; terror continued and even intensified with time. The population quickly
recognized the new logic of the situation: whether one tried to meet German demands or not, one was equally
exposed to violence… It makes no sense, in the context of random punishment, to style one’s life according to the
possibility of being victimized, any more than it makes sense to orient all of one’s everyday acts to the possibility of
an accident.”

8 CIDE
 How Free is Free-Riding in Civil Wars ?…

Large-scale Violence in Civil War and Military Occupation

The counterinsurgency campaigns waged by the German and Japanese armies

in occupied territories during World War II have been widely described as
indiscriminate instances of mass violence. A combination of factors accounted
for this practice, including the fact that these armies were fighting a total
war, were racist, and lacked the resources for less violent forms of
pacification.
Few observations enjoy more currency among historians than the
futility of the German and Japanese anti-partisan reprisals. “Whatever the
purpose of the German policy of reprisals”, a study of occupied Greece points
out, “it did little to pacify Greece, fight communism, or control the
population. In general, the result was just the opposite. Burning villages left
many male inhabitants with little place to turn except guerrilla bands. Killing
women, children, and old men fed the growing hatred of the Germans and the
desire for vengeance” (Condit, 1961: 268). German observers in neighboring
Yugoslavia “frankly concluded that rather than deterring resistance, reprisal
policy was driving hitherto peaceful and politically indifferent Serbs into the
arms of the partisans” (Browning, 1990: 68). Nazi reprisals produced a similar
effect all over occupied Europe (Mazower, 1998: 179),10 while Japanese
reprisals had similar effects in Asia.11 Kalyvas (2006) identifies several
mechanisms underlying the counterproductive effects of indiscriminate
violence, including emotional reactions and norms of fairness, an ambiguous
structure of incentives, reverse discrimination, the production of selective
incentives for rivals, and the systematic overestimation of the strength of ties
between political actors and civilians.
An example of how such violence inadvertently produces private goods
to be used by the rebels is the case of rebel protection. Such protection
emerges as a private good only because of indiscriminate violence. As this
violence escalates, so does the value of protection against it. Survival-
maximizing civilians will be likely to join and/or collaborate with an
organization that credibly offers them protection against an indiscriminate
rival. In El Salvador the power of the revolutionary organization often was its
ability to provide security for its members. When asked why he joined, a
Salvadoran insurgent answered that he “had no choice… It was a matter of
survival. Those were the days when not to go meant getting killed” (Anderson
2004: 222). A former Muslim rebel in the Southern Philippines remarked that

10In the Soviet Union (Cooper, 1979), Poland (Lotnik, 1999: 87), Bosnia (Gumz, 2001: 1037), Italy (Klinkhammer

1997: 83), and France (Kedward, 1993:190).

11In China (Li, 1975:209-10; 231), the Philippines (McCoy, 1980:215), Malaya (Kheng, 1983), Burma (Tucker,

2001), and Vietnam (Herrington, 1997: 21).

DIVISIÓN DE ESTUDIOS INTERNACIONALES 9

Stathis N. Kalyvas and Matthew Adam Kocher

he “joined because of the violence created by the Ilaga (Christian fighters);

because there was no place safe during the trouble at that time” (in McKenna,
1998: 183). In occupied France, “when the acts of reprisals are added to the
indiscriminate round-ups and the residue of Vichy collaborationism, the
pressure on the population in a multitude of localities to look to the maquis as
a place of refuge, or as a receptive and mobilizing organization, was high”
(Kedward, 1993: 190). Historians of the French resistance link its development
to German forced labor dragnets: rather than being sent to Germany as
industrial laborers, many Frenchmen opted to join the resistance movement —
an option they had not considered before the German initiative. As a result,
participation in the French maquis took off only after the Germans began to
recruit laborers (Kedward, 1993).
Under such circumstances, non-participation and free-riding can be
extremely costly.12 An implication is that many rebel organizations often
welcome and even provoke state reprisals.13 What is more, rebel organizations
with the capacity to provide civilians with protection from state violence can
decide whether to turn it into a public good available to all or, preferably, a
private good available only to those individuals or communities who opt to
collaborate with them.14 The latter option turns indiscriminate violence into
an extremely counterproductive weapon: the decision by insurgents not to
protect a village that is unfriendly to them amounts to exposing it to state
violence: in other words, using one’s enemies as one’s own enforcers.15

The Phoenix Program

Although the evidence is very suggestive it remains highly unsystematic. This

is an exceptionally difficult problem, because civil war tends to degrade data
collection capabilities, and because it usually involves systematic attempts to
conceal the identities of insurgent (and sometimes incumbent) members. Even
systematic knowledge of the identities of the victims of large scale violence,
which is now available for some civil wars, is not sufficient for these purposes.
Short of extremely detailed ethnographic/historical work it can be very

12 On this point see Tone (1994: 78), Stoll (1993: 20), Mason and Krane (1989), and Davis (1988: 23).
13 See Aussaresses (2001: 62) and Hayden (1999: 39; 57). International sympathy caused by atrocities adds an
additional benefit for insurgents.
14In Japanese-occupied China, the communists were able to teach peasants how to face Japanese raids following

the “run for shelter under enemy attack” paofan method. By inducing collective discipline and eliminating free-
riding, they were able to turn peasants into a disciplined group; in turn, the peasants won safety, which they could
not have achieved on their own (Wou 1994: 231). Similar tactics have been used in many places, including such
methods as in-site hiding through the building of underground community tunnels (Vietnam), bunkers (Lithuania), or
foxholes and caves (Latin America) (Wickham-Crowley 1991: 43; Lansdale 1964: 85).
15An interesting twist is this: as a sanction for tax evasion, the Vietcong sent offenders for “reeducation” in

hamlets which were shelled by the government army (Elliott 2003: 873).

10 CIDE
 How Free is Free-Riding in Civil Wars ?…

difficult to distinguish civilians from combatants based on the facts of

individual cases.
In this section we examine a unique data source from the Vietnam War to
try to improve our grasp on the rate of victimization of participants in violent
collective action relative to the larger population for at least one civil war. By
analyzing the individualized targeting data left behind by the notorious US
Phoenix Program (and making some fairly cautious assumptions), we derive an
estimate of the proportions of Vietcong guerrillas and civilians killed.
Although Phoenix accounted for only a fraction of the overall violence of the
war (in particular, it covers only state violence), there is a reasonable
argument to be made that it represented the best attempt of the US and
South Vietnamese governments to target the Vietcong selectively and avoid
civilian casualties. Consistent with the qualitative literature on the subject,
we find that Phoenix was wildly inaccurate, killing or otherwise victimizing
numerous civilians for every legitimate Vietcong member.
The Phoenix Program was a joint intelligence-gathering and coordination
system designed to identify and “neutralize” clandestine agents of the
Vietcong in South Vietnam. This program emerged as a response to the need
for discrimination in targeting. As a CIA operative recalls in his memoirs, by
1971 the war was transformed into “one in which whom we killed was far
more important than how many we killed” (Herrington 1997: 69).
In essence, the Phoenix Program was a clearinghouse for information
gathered by numerous military and police organizations operating as part of
the US alliance. The intelligence gathered by Phoenix could then be used to
more accurately target individuals for capture or assassination. Alleged
Vietcong agents were identified by name, alias, date and place of birth.
Where available, fingerprints and photographs were compiled, and an
organizational profile was assembled. The information was then widely
disseminated to the various military forces, police organizations, and official
militia forces involved in prosecuting the war. In practice, very few
individuals were killed or captured by forces directly tasked to Phoenix. The
system mostly just kept track of what happened to identified persons at the
hands of other organizations. In theory, Phoenix was oriented toward
identifying and disposing of the most important Vietcong agents, “executive
cadre at all levels of the communist apparatus” (Thayer, 1985: 208).
At the end of the war, Phoenix left behind a database, the National Police
Infrastructure Analysis Subsystem II – NPIASS-II (NARA: n.d.), containing a
summary of all the information held on each individual, including a record of
each person’s status as of the close of the program: captured, killed,
defected, or “at large.” The data are completely unique. We are unaware of
any other civil war combatant that has left behind a systematic (if partial)
record of its intended victims and their eventual fates. The Phoenix Program
identified 73,697 individuals as members of the clandestine Vietcong

DIVISIÓN DE ESTUDIOS INTERNACIONALES 11

Stathis N. Kalyvas and Matthew Adam Kocher

infrastructure in South Vietnam. By the end of the war 15,438 people, or

about 21% of those selected, had been killed. We use this and other
information in the database to attempt an estimate of the percentage of
Phoenix’s victims that was innocent.
Although the full extent of the suspicion against each individual is
unknown, at the beginning of 1971 Phoenix began to self-evaluate the quality
of its evidence using a single binary variable. An individual identified by three
or more independent sources, or by an “irrefutable source”, as a Vietcong
agent was labled “confirmed”, while those under a lesser degree of suspicion
were labled “unconfirmed”. We do not know what counted as a source, nor do
we know the standard of confirmation for independence among sources. Just
about 11% of all the persons identified in the Phoenix list met this standard of
confirmation (Table 1).

TABLE 1: PHOENIX PROGRAM RESULTS BY CONFIRMATION STATUS FROM JANUARY 1971

CONFIRMED VCI UNCONFIRMED VCI
STATUS
KILLED 4.53% 20.32% 18.16%
(366) (10,341) (10,707)
CAPTURED 1.35% 32.21% 27.98%
(109) (16,392) (16,501)
DEFECTED 0.21% 15.33% 13.26%
(17) (7,801) (7,818)
AT LARGE 93.91% 32.14% 40.60%
(7,587) (16,355) (23,942)
KILLED + CAPTURED 5.88% 52.53% 46.14%
(475) (26,733) (27,208)
TOTAL 100.00% 100.00% 100.00%
(8,079) (50,889) (58,968)
PR(Χ2) = 0.000

A simple cross-tabulation of confirmation and status tells a remarkable story

of capricious violence. Table 1 compares the eventual fates of those in the
confirmed and unconfirmed categories. From the beginning of 1971 until the
time the database was closed, about 4.5% of those in the confirmed category
had been killed, while roughly 94% remained at large. By contrast, 20% of the
unconfirmed had been killed, additional large percentages had been captured
or had defected to the government – roughly 34% and 18% respectively – while
only 25% remained at large. In other words, Phoenix divided its pool of
supposed Vietcong agents into two categories, one of high suspicion and one
of low suspicion. In a truly awesome process of perverse selection, those
under low suspicion were almost five times more likely to be killed than those
under high suspicion. The unconfirmed were nearly twenty-four times more
likely to be captured than the confirmed, while nearly 94 out of every 100

12 CIDE
 How Free is Free-Riding in Civil Wars ?…

highly suspicious individuals escaped the US and South Vietnamese net

entirely. Of those killed, over 96% came from the unconfirmed category.
In somewhat different terms, the odds ratio of killed/at large in the
unconfirmed category to killed/at large in the confirmed category is
approximately 31.88, while the same odds ratio for killed + captured is about
26.14. That is, an individual had close to 32 times greater odds of being
killed, and 26 times greater odds of being captured or killed, in the
unconfirmed category.
The simplest and most plausible explanation for these data is that the
confirmation process was at least reasonably successful at distinguishing real
Vietcong agents from innocents. To elaborate the Phoenix Program had two
selection mechanisms; the first divided the population into two groups: those
on the list and those off the list and the second mechanism divided those on
the list into confirmed and unconfirmed. It seems unreasonable to assume
that the US/GVN directed more effort toward locating and “neutralizing”
unconfirmed persons than confirmed ones.16 A reasonable (though
conservative) assumption is that effort levels were identical for the two
groups.
How, then, can we explain the much higher survival and escape rates for
confirmed persons? The most sensible assumption to make is that
“confirmation” was an effective selection procedure, in the sense that it
chose real Vietcong members from the larger list in much higher proportion to
their total numbers than it did innocents. Real Vietcong agents had a range of
organizational resources they could draw upon to avoid capture or
assassination at the hands of counterinsurgency forces, while incorrectly
identified innocents were completely exposed. Unconfirmed persons were
victimized at higher rates not because counterinsurgents wanted it that way,
but rather because it was easier to find them and to subdue or kill them. A
further point in favor of this view is that virtually nobody in the confirmed
category defected, while defection was common among the unconfirmed,
precisely what we would expect if the confirmed category were largely
composed of actual, committed Vietcong.
Given this general interpretation of the data, we use a simple
mathematical model to estimate the ratio of civilians to Vietcong victimized
by Phoenix. We make three key assumptions. First, we assume that the
individuals in the Phoenix database can be partitioned into two stable and
mutually exclusive groups: Vietcong (V) and innocents (I). It follows from this
assumption that the “confirmed” and “unconfirmed” categories can likewise

16 According to NPIASS-II, Phoenix was aware of the “current address” of nearly 64% of confirmed persons, but

less than 1% of unconfirmed persons. Likewise, nearly 23% of confirmed persons, but less than 1% of unconfirmed
persons, were the subjects of individualized arrest warrants. These data strongly suggests that significantly greater
effort was oriented toward those under higher suspicion.

DIVISIÓN DE ESTUDIOS INTERNACIONALES 13

Stathis N. Kalyvas and Matthew Adam Kocher

be partitioned into Vietcong and innocents, resulting in four integer-valued

variables and two equations as follows:

Vu ≡ number of Vietcong in unconfirmed category

Iu ≡ number innocents (civilians) in unconfirmed category
Vc ≡ number of Vietcong in confirmed category
Ic ≡ number innocents (civilians) in confirmed category

(1) Vu + Iu = 50889

(2) Vc + Ic = 8079

The equations follow analytically from the first assumption and the totals
of confirmed and unconfirmed persons in the data. The first assumption is the
most controversial, and we return to a more detailed defense of it below.
Second, we assume a constant proportion of victims for Vietcong (pvn) and
a constant proportion for innocents (pin), in each case independent of their
confirmation status.17 In other words, we stipulate that nothing about the
process of confirmation itself affected rates of victimization, but rather that
these were determined by the characteristics of the actors. Note that this is a
conservative assumption, since we would expect more effort to be directed
toward those under greater suspicion. The assumption is operationalized in
equations (3) and (4):

(3) pvn (Vc ) + pin (Ic ) = 475 / 366 0 < pvn , pin < 1

(4) pvn (Vu ) + pin (Iu ) = 26733 /10341 0 < pvn , pin < 1

Here we give two totals (once again derived from the data) for each
equation, the first for captured + killed, the second only for killed. Below we
give estimates based on both sets of totals. The inequalities follow from the
definition of pvn and pin as proportions.
Third and finally, we assume that the odds ratio of innocents/Vietcong in
the unconfirmed group to innocents/Vietcong in the confirmed group is the
same as the odds ratio derived from Table 1 of victimized/at large in the
unconfirmed group to victimized/at large in the confirmed group. This
assumption is summarized in equation (5):

17 pvn stands for “proportion of vietcong neutralized” and pin stands for “proportion of innocents neutralized”.

14 CIDE
 How Free is Free-Riding in Civil Wars ?…

Iu / 50889
Vu / 50889 I V
(5) = u c = 26.1400
Ic / 8079 IcVu
Vc / 8079

Once again, the four denominators are derived from the data, while the
odds ratio compares captured + killed/at large across the confirmed and
unconfirmed categories.18
Equation (5) is easier to grasp intuitively if we rearrange the terms as
follows:

Iu I
(5*) = 26.14 c
Vu Vc

What this equation says is that the ratio of innocents to Vietcong in the
unconfirmed category is more than 26 times greater than the same ratio for
the confirmed category. On its face, it makes considerable sense to assume a
higher ratio of innocents to Vietcong in the unconfirmed category. Since
“confirmation” was a more rigorous selection process, we should not be
surprised to see that it resulted in more “hits”, that is more correctly
identified Vietcong. The question is: how much higher? Equation (5) implies
that the entire difference in the rate of victimization between the confirmed
and unconfirmed categories is accounted for by the composition of these
categories. This assumption exploits and operationalizes the argument we
defended above, namely that the reason “confirmed” persons were much less
likely to be victimized than unconfirmed persons is precisely because they
were much more likely to be real Vietcong agents.19
Returning to the first assumption, that membership and non-membership
are well-defined, stable, and exclusive categories is an important simplifying
stipulation that we need to make in order to gain leverage. The Vietcong was
a highly sophisticated military/bureaucratic structure, differentiated
according to both ranks and functions. Following from this insight, another
possible interpretation of the data is that important Vietcong agents were

18 Why should we use killed + captured, instead of just killed? Both capture and assassination required that an

individual be physically located and identified by forces sufficient to take action against him. In fact, this is a
conservative assumption. Since defection was a good way to avoid being killed or captured (and sentenced to
prison), we might expect innocent but threatened individuals to take this option at higher rates than would highly
committed Vietcong agents.
19 It is worth pointing out in this connection that were we to assume an odds ratio of 1, that is were we to

assume that the confirmed and unconfirmed categories had approximately the same composition, there are no
possible solutions that satisfy the remainder of our system of equations. An odds ratio slightly higher than 5 is the
lowest possible value consistent with the remaining equations.

DIVISIÓN DE ESTUDIOS INTERNACIONALES 15

Stathis N. Kalyvas and Matthew Adam Kocher

able to escape capture or assassination, while rank and file members were
killed at rates similar to, or only slightly lower than, innocents. This
interpretation, that Phoenix was good at locating rank and file Vietcong
agents but poor at getting high ranking agents, is the one adopted by Thayer
(1985).20 Perhaps important agents were more likely to be confirmed, but less
likely to be caught, than rank and file participants. Although the latter
supposition seems quite plausible, the former seems less so. If the physical
security of Vietcong was better protected at higher levels of the
organizational chart, we would expect their identities to be more closely
guarded secrets as well. Consequently, we would expect, if anything, that
high ranking agents were less likely to be confirmed.
In any case, it would require more fine-grained data than we have
available to evaluate this supposition rigorously. NPIASS-II does, however,
include some information about the beliefs held by Phoenix managers
themselves regarding the importance of the people on the list. The data tell
us, for instance, whether or not a suspect was believed to be a full or
probationary member of the Communist Party. Also, Phoenix includes agents’
supposed “echelon,” or the scale of operations in which they were involved
(e.g. region, province, district, village). Individuals operating at a higher scale
were generally assumed to be more important.
Forty percent of those on the Phoenix list were believed to be Communist
Party members. According to the data, roughly the same percentage of
confirmed and unconfirmed persons were party members. Controlling for
party membership has little effect on our main results. While among full party
members, the inverse relationship between neutralization and confirmation is
weaker than among ostensible non-members, it is much stronger among
probationary party members. The inverse relationship is strongest among
those whose party membership is unknown. The bivariate data on party
membership and “neutralization” are equally confusing. Full party members
were less likely to be captured or killed than non-party members, but more
likely to be killed. Probationary party members were more likely to be killed
or captured than full members, non-members, or those whose membership
status was unknown. Similarly, controlling for echelon does not affect the
results. While there are differences in the relationship between confirmation
and neutralization across the echelon categories, they are small and do not
correspond to a clear pattern either. The data suggest at best a very weak
relationship between perceived importance and the rates at which individuals
were captured or killed.

20As an important Defense Department analyst during the Vietnam War, Thayer had full access to all the data we
use here, as well as much that remains classified. His book does not consider the possibility we suggest, perhaps
because it was simply inconceiveable to him that the Phoenix Program was not merely unproductive, but actually
counterproductive. Although it is true that the majority of people “neutralized” by Phoenix were believed to be
low-ranking, this result follows straightforwardly from the pyramidal form of any military or bureaucratic
organization.

16 CIDE
 How Free is Free-Riding in Civil Wars ?…

Returning to summarize our model, we have the following 5 equations in 6

variables:

Vu ≡ number of Vietcong in unconfirmed category

Iu ≡ number innocents (civilians) in unconfirmed category
Vc ≡ number of Vietcong in confirmed category
Ic ≡ number innocents (civilians) in confirmed category
pin ≡ proportion of innocents victimized
pvn ≡ proportion of Vietcong victimized

(1) Vu + Iu = 50889
(2) Vc + Ic = 8079

(3) pvn (Vc ) + pin (Ic ) = 475 / 366 0 < pvn , pin < 1
(4) pvn (Vu ) + pin (Iu ) = 26733 /10341 0 < pvn , pin < 1

Iu / 50889
Vu / 50889 I V
(5) = u c = 26.1400
Ic / 8079 IcVu
Vc / 8079

Obviously, these equations cannot be solved for unique solutions. Instead,

we graph all possible solutions that are consistent with the definitions and
inequalities (rather than admit only solutions with an odds ratio of exactly
26.14, we admit all solutions within a 1-unit band around this value). The
mathematical details of how we obtained our solution set can be found in the
appendix; we also subject our results to a “robustness check,” that shows our
results hold even under weakened assumptions.
Figures 1 and 2 contain all solutions based on the number of individuals
victimized (killed + captured). Figure 1 graphs the proportion of innocents
victimized (pin) against the difference Iu – Vu (the number of innocents in the
unconfirmed category minus the number of Vietcong in the unconfirmed
category, which must sum to a constant) and against the difference Vc – Ic
(Vietcong in the confirmed category minus innocents in the confirmed
category, which once against must sum to a constant). In Figure 2 we graph
the proportion of Vietcong victimized (pvn) against the same two quantities. In
Figures 3 and 4, we repeat the same graphs, but in this case using only the
totals for persons killed.

DIVISIÓN DE ESTUDIOS INTERNACIONALES 17

Stathis N. Kalyvas and Matthew Adam Kocher

Table 2 exhibits the two extreme solutions of Figures 1 and 2, along with
an approximately intermediate solution.21 Although there are thousands of
possible combination of values consistent with the inequalities and
definitions, all fall within a fairly narrow range of possibility. On one extreme,
the model is consistent with all but one victim innocent. At the other
extreme, only 689 out of a total 27208 captured or killed are Vietcong. In the
former case, the proportion of actual insurgents victimized approaches zero.
In the latter case, the proportion of innocents victimized approaches unity.
The intermediate solution, though far from unique, suggests more realistic
values: 1% of actual Vietcong captured or killed, and about 86% of innocents
victimized.
Although we cannot pin down a precise solution, making some simple,
conservative, and highly plausible assumptions, we estimate that the Phoenix
Program victimized at least 38 innocents for every 1 actual Vietcong agent (the
intermediate solution is about 78 innocents for every 1 Vietcong).
Two interesting subsidiary observations are worth noting. First, our findings
are consistent with the belief that the Phoenix Program was pretty good at
identifying its enemies: over all solutions, 40 – 55% of the individuals selected
into Phoenix are estimated to have been Vietcong. Of those subjected to the

21 In this context, there are various ways to conceive of an intermediate solution. We chose a solution in which

the number of innocents vs. Vietcong victimized is intermediate between the extremes.

18 CIDE
 How Free is Free-Riding in Civil Wars ?…

more rigorous process of “confirmation,” we estimate that 92 – 96% were

Vietcong. Identifying them was one thing, catching them another entirely.
Second, for all solutions, more Vietcong were victimized in the unconfirmed
category than in the confirmed category. These two observations suggest that,
while Phoenix’s results may have been inaccurate, the program was by no
means irrational, either in its selection procedures or in its targeting of
“unconfirmed” persons.

TABLE 2: HOW MANY INNOCENTS WERE VICTIMIZED? EXTREME AND INTERMEDIATE SOLUTIONS
MAXIMUM INTERMEDIATE MINIMUM
CONFIRMED INNOCENTS (IC) 624 443 310
CONFIRMED VIETCONG (VC) 7455 7636 7769
UNCONFIRMED INNOCENTS (IU) 35133 30875 26209
UNCONFIRMED VIETCONG (VU) 15756 20014 24680
PROPORTION OF INNOCENTS VICTIMIZED (PVN) 0.761 0.856 0.999
PROPORTION OF VIETCONG VICTIMIZED (PIN) 0.000027 0.012 0.021
TOTAL INNOCENTS VICTIMIZED 27207 26864 26519
TOTAL VIETCONG VICTIMIZED 1 344 689
TOTAL VICTIMIZED 27208 27208 27208

DIVISIÓN DE ESTUDIOS INTERNACIONALES 19

Stathis N. Kalyvas and Matthew Adam Kocher

In Table 3, we repeat the exercise using figures for killed only, showing
the extreme and intermediate solutions from Figures 3 and 4. The solution set
in this case is considerably larger. At one extreme, the data and assumptions
are consistent with no actual Vietcong killed, despite having more than 13,000
real agents in the pool of suspects. At the other extreme, nearly 1,900 of the
dead are actual Vietcong agents, while over 8,800 are innocent. In other
words, the most optimistic (i.e. most accurately selective) scenario is that
about 4.7 innocent persons were killed for every Vietcong agent. In the
intermediate case, we have about 10.3 innocents killed for every rebel
participant.
Beyond these raw results, the data give us some reason to suppose that
the pressure of indiscriminate or poorly selective violence fell
disproportionately on certain sectors of the population. Table 4 shows logistic
regression estimates of killed/captured and killed on confirmation status, age,
and sex.

TABLE 3: HOW MANY INNOCENTS WERE KILLED? EXTREME AND INTERMEDIATE SOLUTIONS
MAXIMUM INTERMEDIATE MINIMUM
CONFIRMED INNOCENTS (IC) 1556 366 63
CONFIRMED VIETCONG (VC) 6523 7713 8016
UNCONFIRMED INNOCENTS (IU) 43969 28327 8749
UNCONFIRMED VIETCONG (VU) 6920 22562 42140
PROPORTION OF INNOCENTS KILLED (PIN) 0.235 0.340 0.999
PROPORTION OF VIETCONG KILLED (PVN) 0.00000075 0.031 0.038
TOTAL INNOCENTS KILLED 10707 9759 8811
TOTAL VIETCONG KILLED 0 948 1896
TOTAL KILLED 10707 10707 10707

The data suggest an interesting story of “profiling” used in the infliction of

violence. First, age has a strong association with both dependent variables.
The odds of being killed decreased by about 2.7% for every year of age, while
the odds of being captured or killed decreased by about .8% for every
additional year. The association between sex and violence is more
complicated: women had slightly higher than twice the odds of being captured
or killed once identified by Phoenix, but men had nearly 5.5 times greater
odds of being killed. In fact, the database shows that well over 50% of all the
women in the database were captured, while slightly less than 5% were killed.
These results are consistent with Jones’s (2000) argument that young men are
disproportionately targeted in episodes of mass violence, though with the
caveat that women appear to have had a greater probability of suffering non-
violent victimization. If the Phoenix Program was in fact as inaccurate as we
think, then the age and sex results suggest a particularly acute dilemma for
military age Vietnamese men during these years. In effect, we should assume

20 CIDE
 How Free is Free-Riding in Civil Wars ?…

that the denominator for calculating civilian risk is much smaller than the
general population. Old folks, young children, and women may have been
probabilistically “profiled out” in the process of selecting individuals for
lethal violence.
TABLE 4: LOGISTIC REGRESSION ESTIMATES ON KILLED/CAPTURED AND KILLED22
DV = KILLED/CAPTURED DV = KILLED
CONFIRMATION 0.064 (-54.91) 0.267 (-22.86)
SEX (MALE = 1) 0.460 (-33.54) 5.445 (36.08)
AGE IN YEARS 0.992 (-10.29) 0.970 (-25.24)
2
PSEUDO-R 0.11 0.07
N 48435 48435
SOURCE: US DEPARTMENT OF DEFENSE, NPIASS-II. ALL COEFFICIENTS IN ODDS RATIO FORM. Z-STATISTICS IN PARENTHESES.

How should we contextualize these results? There are obvious limits to our
ability to generalize, for the Vietnam War and beyond it to other wars. The
violence associated with the Phoenix Program represented a small fraction of
the total carnage of Vietnam. Far larger numbers of deaths resulted from
conventional battlefield confrontations of the US and South Vietnamese
against the Vietcong and North Vietnamese armies. Barely discriminate
bombing killed hundreds of thousands of combatants and civilians.
Nevertheless, the core of the conflict for many years consisted of the sort of
selective and individualized capture and murder that the Phoenix Program
documents. It is possible that the selective murder of Vietnamese was more
accurate before, after, and outside of Phoenix, despite the level of resources
and effort directed precisely toward improving selectivity. One reason to
doubt this supposition is that forces directly tasked to Phoenix accounted for
a relatively small share of the violence: 2% of neutralizations over a period of
a year and a half analyzed by Thayer (1985). The vast majority of deaths were
associated with the ongoing “military operations” of local paramilitary forces
(Elliott, 2003: 1137), though what these operations consisted of is difficult to
say. Thus, it is probable that Phoenix reflected more general processes of
violence that had been ongoing in southern Vietnam for many years.
We would need to know considerably more than we do about the rate of
Vietcong membership in the overall population in order to determine who was
safer overall. However, it is important to contextualize Phoenix, which
represented the best efforts of a highly sophisticated military and intelligence
bureaucracy to improve on past performance. The creators of the Phoenix
Program understood and to a certain extent regretted the low-level of
discrimination of earlier US efforts in the Vietnam War. Phoenix was the fruit
of what they learned. Thus, we have some reason to suppose that it was more

22 These results should be treated with caution due to large numbers of missing values on the age variable. Note

also that the inclusion of the age and sex variables has virtually no effect on the size of the coefficient for
confirmation.

DIVISIÓN DE ESTUDIOS INTERNACIONALES 21

Stathis N. Kalyvas and Matthew Adam Kocher

accurate than what went before. And, though we have ample reason to
question the accuracy of American efforts at discrimination, it is not clear
that we should question their competence relative to other counterinsurgency
campaigns waged by poorer and less sophisticated governments.
In comparative perspective, the Phoenix Program is probably closer in
nature to typical civil war violence than the battles and aerial bombardments
that tend to be associated in the public mind with the Vietnam War. It is
possible that American counterinsurgent forces in Vietnam were simply far
less competent than is typically the case in civil wars. If so, this specifically
American problem seems to have persisted in Iraq. According to a February
2004 confidential report by the International Committee of the Red Cross,
American military officers estimated that 70-90% of the prisoners held at Abu
Ghraib at the time were unconnected to the insurgency (ICRC 2004).23
Yet, indiscriminate and inaccurate violence is generally associated with
the poorly trained and disciplined soldiers of developing country armies,
rather than with highly professionalized first world militaries. If military
discipline or overall resources have any bearing on the capacity of an
organization to be reliably selective, then we should expect the Phoenix
Program to have been much more accurate than typical counterinsurgent
efforts. If this supposition holds up, then our data would reflect one of the
more optimistic scenarios for defending the assumptions of the collective
action problem (in other words, it is a hard case for us).
It is worth asking here why violence that is selective in intent may end up
being so inaccurate. The reason is that civil war stretches the bureaucratic
capacity of states: it demands a lot of information while straining existing
resources. States (as well as challengers) rely on individual informants who
have incentives to denounce their personal or local enemies. Phoenix was no
exception. According to one account, “The people who identified members of
the (Vietcong) shadow government often had many types of non-Communist
enemies in their area of operation, particularly if they worked in their native
areas. Like most people, they had personal enemies: the men who had
insulted their sisters, the men who had stolen their sweethearts, the farmers
who had borrowed money from their families and failed to repay it, and even
the GVN (South Vietnam Government) officials who had beaten their cousins.
Family members of these enemies also could be fair game, especially when
previous offenses had involved relatives” (Moyar, 1997: 114). A South

23 “Certain CF military intelligence officers told the ICRC that in their estimate between 70% and 90% of the

persons deprived of their liberty in Iraq had been arrested by mistake. They also attributed the brutality of some
arrests to the lack of proper supervision of battle group units” (ICRC, 2004: 8). Even the highly selective process
by which individuals were shipped to Guantanamo appears to have suffered from similar problems: it turns out that
92% of the 517 Guantanamo detainees had not been al-Qaeda fighters, while 95% of them were not captured by the
Americans themselves; some 86% were handed over in Afghanistan and Pakistan after a widespread campaign in
which big financial bounties were offered in exchange for anyone suspected of links to al-Qaeda and the Taliban
(Simpson 2006).

22 CIDE
 How Free is Free-Riding in Civil Wars ?…

Vietnamese colonel told Moyar (1997: 116) about the men of the Provincial
Reconnaissance Units, among those relying on Phoenix information in their
raids: “If they saw a beautiful girl, they tried to be her boyfriend. If they got
turned down, then they accused her of being a VCI (a member of the Vietcong
Infrastructure).” A U.S. adviser confirmed this tendency by recalling an
example: “One guy who was a source of information about the VC relieved his
family of three generations of debt. He turned in phony reports fingering as
Viet Cong people his family owed money to” (quoted in Moyar, 1997: 293).
The same tendency can be observed in many civil wars (Kalyvas, 2006). For
example, in El Salvador, many false denunciations were “enough to seal one’s
fate, since government forces seldom ought to investigate the charges and
‘innocent until proven guilty’ was not a principle recognized by the military,
security forces, or ORDEN civilian irregulars” (Binford, 1996: 107).
Obviously, armed actors are aware of this trend. As a CIA adviser in
Vietnam recalls: “There were times when I questioned a name on the blacklist
of VCI. ‘Is this guy actually VC infrastructure, or is he a political enemy or a
business enemy of the province chief or district chief of somebody else?’”
(quoted in Moyar, 1997: 122). However, given extremely stretched resources
and the need to act, they tend to err in the direction of false positives:
“Better to kill mistakenly than release mistakenly” went a Vietnamese slogan,
popular among some insurgents; for them, “justice was not an abstract ideal,
but a tool in the political struggle”; “if it came down to a conflict between
the revolution’s prestige and abstract notions of justice, it was clear which
would prevail” (Elliott, 2003: 91, 947). A U.S. commander in Iraq remarked
about Iraqi counterinsurgents that “if they shoot somebody, I don’t think they
would have remorse, even if they killed someone who was innocent” (in
Maass, 2005: 47).

The Greek Civil War

While extremely detailed and systematic, the Phoenix data covers only a
subset of individuals24 and provides little information about actual insurgent
combatants, focusing rather on alleged civilian agents and collaborators. We
conclude with data from a regional study on the Greek Civil War. Like the
Vietnam War, this was a complex war fought in different phases between 1943
and 1949 and entailing extensive foreign presence.

24 In other words, there may be a selection bias in the Phoenix database. The data give us no way to evaluate this

possibility systematically. Yet, the motivational mechanism for denunciation established in Kalyvas (2006) leads us
to believe any such bias is likely to be small. That is, we believe as a general matter that denunciation in civil war is
at best weakly correlated with guilt.

DIVISIÓN DE ESTUDIOS INTERNACIONALES 23

Stathis N. Kalyvas and Matthew Adam Kocher

Using extensive archival sources, we were able to calculate the exact toll
of violence in one region, the Argolid, located in the northwest tip of the
Peloponnese, Greece’s southern peninsula. The civil war in that area was
fought during a single year, in 1943-44, coinciding with the German
occupation of Greece. The conflict pitted members of the pro-communist
resistance army ELAS (Greek Popular National Army) against right-wing
collaborationist militias supported by the German occupation troops. Both
resisters and collaborationists recruited locally and the war had a strong
“neighbor against neighbor” aspect.

TABLE 5: REBEL COMBATANT AND CIVILIAN VICTIMIZATION RATES, ARGOLID, 1943-1944

TOTAL RURAL CIVILIAN POPULATION 45140
MILITARY AGE MEN 13542
VICTIMS OF GERMAN/COLLABORATIONIST VIOLENCE 353
PROBABILITY OF CIVILIAN VICTIMIZATION .0261
REBEL COMBATANTS 500
REBELS KILLED IN ACTION 20
PROBABILITY OF REBEL VICTIMIZATION .04

Between September 1943, when the conflict began, and September 1944,
when the Germans left, the collaborationists and Germans killed 353
individuals, slightly less than one percent of the rural population of that area
(.78%). Although a few women and children were killed during these
operations, about 90% of the victims were men of military age (18-45 years
old). Assuming that men of military age made up about 30% of the total
population, we estimate that the likelihood that a man of military age
inhabiting the rural Argolid was killed by the occupation army to be 2.61%. On
the other hand, 20 local rebel fighters (members of the 6th Regiment of ELAS--
which operated in the area) were killed in action during the same period—
about 4% of the Argolid rebels (Table 5).25
The difference between 2.61% and 4% is rather negligible and falls within
the range of measurement error. The assumption that joining the rebels was a
highly risky choice relative to not joining, is unsupported by the evidence. In
fact, this slight difference vanishes when one takes into account two
additional aspects of the violence. First, the Germans and their local
auxiliaries did not restrict themselves to killing civilians. They drove hundreds

25 The data on rebels killed in action is from Vazeos (1961: 131-4). The 6th regiment of ELAS which was active in

the Argolid and Korinthia areas and recruited primarily from these two regions reached a force of 3,500 men in
October 1944, after the occupation’s end (Vazeos, 1961: 96). However, many men were recruited after the
Germans left. An estimate of recruits from the Argolid prior to the Germans’ departure is 500. Another 15 rebel
fighters were killed during the battle of Athens, in December 1944, when the Communist resistance lost its bid for
power against the Greek government and the British. We exclude them from the analysis because after the
departure of the Germans the violence against civilians subsided and the Communists, in total control of the Greek
countryside, drafted thousands of peasants into their army.

24 CIDE
 How Free is Free-Riding in Civil Wars ?…

more to prisons and concentration camps in Greece, and slave labor camps in
Germany.26 In contrast, the Germans did not take rebel prisoners. In fact, a
significant number of the rebels killed in action were actually shot after being
captured. The problem from the Germans’ perspective was that the rebels
avoided contact and were able to hide in the mountains and escape the
dragnets. Second, violence was not uniformly distributed; some villages had
comparatively many more victims. The worst-hit villages exhibited rates of
civilian victimization approximating 10%. In these villages, it was clearly safer
for military-aged men to join an armed faction rather than stay home.
To summarize, the data from the Argolid contradict both core assumptions
of the collective action paradigm (the benefits of free-riding and the risk of
joining an insurgency). In conjunction with the anecdotal evidence from a
wide range of civil wars and the Phoenix data, these data challenge the
collective action paradigm.

26 The rosters of the local prison in the town of Nafplio show over 1,000 individuals held there during the same

period. Hundreds more were sent to a concentration camp in the neighboring town of Korinthos, while a smaller
but unspecified number were sent to slave labor camps in Germany.

DIVISIÓN DE ESTUDIOS INTERNACIONALES 25

Stathis N. Kalyvas and Matthew Adam Kocher

Conclusion

Our data question the assumption that insurgency is automatically subject to

the collective action problem. “Free riding” is not as free in civil wars as
currently thought. If the collective action paradigm has been so dominant it is
because scholars have tended to overestimate the risks faced by rebel fighters
and/or underestimate the risks faced by nonparticipants —a result of limited
attention to the dynamics of violence and of the tendency to impute
preferences over investigating them empirically. Tullock’s (1971: 93) critique
of historians and social scientists of the revolutions literature for imputing
individual motivations for participation from the observed macro ‘public good’
post-hoc, was right on target. But, in the years that followed his pathbreaking
analysis, many social scientists have likewise inferred the calculations that
supposedly lead people to join an insurgency from empirically unsupported
assumptions about comparative risk.
Our data suggest that taking the collective action problem seriously
implies that sometimes the real puzzle in civil wars is nonparticipation rather
than collective action. Some simple mechanisms suggest themselves. First, if
civil war works as we suggest, then membership in a rebel organization can be
a “club good”. That is, members of the club receive special benefits, both
material and nonmaterial, from which non-members are excluded. Consider
William McNeill’s (1947: 80–1) description of the process of joining ELAS:

In actual fact, a soldier in ELAS lived a good deal better than did the ordinary
peasant, and did not have to work with the same drudging toil. He further had
the psychological exhilaration of believing himself a hero and the true
descendant of the robber klefti who had fought in the War of Independence and
were enshrined in the Greek national tradition. Under the circumstances, many
peasant’s son found himself irresistibly attracted to the guerrilla life; and an
over abundant peasant population made recruitment easy. Fewer came from
towns; life was relatively comfortable there, and EAM had other work for
townsmen, organizing strikes or serving as propagandists among the more
illiterate peasants. From the very beginning the chief factor that limited the
number of the guerrillas was lack of weapons.

The classic club goods problems from economics are adverse selection and
moral hazard. A rebel organization provides security and resources to its
members, but it also seeks tough, ruthless, and determined combatants who
share its agenda and goals. Much like insurance companies seeking to screen
out individuals with serious pre-existing conditions, rebel groups will set up
mechanisms to restrict entry, and they will reject many volunteers, something
that we should not observe if the collective action problem holds. By the logic
of moral hazard, rebels should accept the security benefits of being in the

26 CIDE
 How Free is Free-Riding in Civil Wars ?…

organization, while trying to minimize their effort and risk. Furthermore, as

suggested by McNeill, rebel organizations do not always maximize
recruitment. Faced with limited logistic means, few weapons, and a limited
capacity to support thousands of troops, many organizations prefer to recruit
relatively few active full-time fighters.
We have argued in this paper that the assumption of a collective action
problem automatically emerging in situations of civil war is unsupported by
available systematic evidence. Rethinking the application of the collective
action problem to contexts of large scale violence is not merely a question of
getting the description right, though that is a benefit in and of itself. It also
suggests previously unexplored answers to key theoretical problems in the
study of violence. For instance, stalemated civil wars of long duration are
puzzling if we assume that combatant risk significantly exceeds civilian risk.
If, on the other hand, participation improves individuals’ short-run survival
prospects, the puzzle is easily resolved. Our approach is also a promising way
to account for the highly counterintuitive (yet common) phenomenon of
people who act as combatants for the “wrong” side in civil wars (e.g.
Chechens in pro-Russian militias, Kurds in the Turkish army, Mayan
Guatemalans in government militia, etc.). Last, we point to private goods
provided by rebel organizations, such as protection, which have been eclipsed
by the recent focus on looting; likewise, our analysis shows the importance of
conceptualizing civil wars as processes that generate incentives and
constraints, in contrast to the view, best summarized by Tullock, that payoffs
to participation are linked almost exclusively to expectations about outcomes.
In terms of comparative patterns, we would expect rebel recruitment to
display a convex function: On the one hand, recruitment should go up under
conditions of complete absence of the state; such absence would eliminate
the risk of state sanctions against rebels, consistent with Fearon and Laitin
(2003), while it would make possible the implementation of rebel sanctions
targeting free-riding. On the other hand, rebel recruitment should also
increase under conditions of extreme violence by the state, provided the
rebels are able to survive. Significant but lower levels of state violence
should, everything else, depress rebel recruitment, both because of the
disruption of rebel operations and the increase in participants’ risk. Likewise
our analysis suggests the importance of paying close attention to the temporal
and spatial variation of violence: not all civilians face similar risks and
incentives all the time. Additionally, the perception of risk may diverge from
real risk—and violence may be a factor in generating such misperception.
Furthermore, civilian preferences are heterogeneous, as is civilian behavior
during civil war: the behavioral range between the opposite stances of rebel
fighter and nonparticipant is very large and includes all types of collaboration.
Clearly, more complex models of participation and recruitment call for fine-
grained micro-data.

DIVISIÓN DE ESTUDIOS INTERNACIONALES 27

Stathis N. Kalyvas and Matthew Adam Kocher

In short, an enhanced understanding of civil wars requires a double

correction: an emphasis on systematic evidence from a wide cross-section of
individuals and an awareness of the dynamics of violence. Both require a
much more systematic micro-empirical approach than currently practiced.

28 CIDE
 How Free is Free-Riding in Civil Wars ?…

Appendix

To obtain the solution set D for the system of equations generated by our
assumptions about Phoenix, we created the set of all possible integer-valued
4–tuples D ∈ {(Iu , Ic , Vi , Vc )} such that:

(1) Vu + Iu = 50889
(2) Vc + I c = 8079

(i.e. such that Vietcong and innocents in the categories “confirmed” and
“unconfirmed” sum to the total for each category in the data).

Next, assuming that pin and pvn are proportions [see equations (3) and (4) in
the text], we dropped all 4-tuples for which these variables fell outside the
unit interval. We considered two cases, the first for killed + captured
[equations (3.1) and (4.1)], the second for killed only [equations (3.2) and
(4.2)]. In each case, we solved equation (3) for pvn, simplified [equations (3.1)
and (3.2)], substituted the result into equation (4), and simplified again,
resulting in equations (4.1) and (4.2).

475 ⎛I ⎞
(3.1) 0 < + pin ⎜⎜ c ⎟⎟ < 1
Vc ⎝ Vc ⎠
26733Vc − 475Vu
(4.1) 0 < <1
IuVc − IcVu

366 ⎛I ⎞
(3.2) 0 < + pin ⎜⎜ c ⎟⎟ < 1
Vc ⎝ Vc ⎠

10341Vc − 366Vu
(4.2) 0 < <1
IuVc − IcVu

Finally, we calculated the odds ratio [equation (5)] for all 4-tuples, and
dropped those cases outside a 1-unit band around the odds ratio derived from
the data:

I V
(5.1) 25.64 < u c < 26.64
IcVu

DIVISIÓN DE ESTUDIOS INTERNACIONALES 29

Stathis N. Kalyvas and Matthew Adam Kocher

Taken together, D subject to the conditions (1), (2), (3.1), (4.1), (5.1) or (1),
(2), (3.2), (4.2), (5.1) is an exhaustive specification of our solutions.

We also subjected our results to a “robustness” check, by modifying condition

(5.1) for a 10-unit interval around the odds ratio taken from the data:

I uVc
(5.2) 21.14 < < 31.14
I cVu

Using this larger band for the odds ratio resulted in the solutions outlined in
Tables 5 and 6, below. We can see by comparing these figures to the ones
drawn from Tables 2 and 3 that the results do not differ greatly. This
alternative specification of our model permits solutions with somewhat
smaller proportions of innocents victimized. Nevertheless, the overall findings
do not change: under the new assumption, innocents were still far more likely
to suffer from the Phoenix Program than were Vietcong agents.

TABLE 6: HOW MANY INNOCENTS WERE VICTIMIZED? EXTREME AND INTERMEDIATE SOLUTIONS
(ROBUSTNESS CHECK WITH 21.14 < ODDS RATIO < 31.14)
MAXIMUM INTERMEDIATE MINIMUM
CONFIRMED INNOCENTS (IC) 666 398 214
CONFIRMED VIETCONG (VC) 7413 7681 7815
UNCONFIRMED INNOCENTS (IU) 37488 29202 26063
UNCONFIRMED VIETCONG (VU) 13401 21687 24826
PROPORTION OF INNOCENTS VICTIMIZED (PIN) 0.713 0.904 0.999
PROPORTION OF VIETCONG VICTIMIZED (PVN) 0.0000097 0.015 0.027
TOTAL INNOCENTS VICTIMIZED 27208 26768 26327
TOTAL VIETCONG VICTIMIZED 0 440 881
TOTAL VICTIMIZED 27208 27208 27208

TABLE 7: HOW MANY INNOCENTS WERE KILLED? EXTREME AND INTERMEDIATE SOLUTIONS
(ROBUSTNESS CHECK WITH 21.14 < ODDS RATIO < 31.14)
MAXIMUM INTERMEDIATE MINIMUM
CONFIRMED INNOCENTS (IC) 1592 350 54
CONFIRMED VIETCONG (VC) 6487 7729 8025
UNCONFIRMED INNOCENTS (IU) 45000 27864 8701
UNCONFIRMED VIETCONG (VU) 5889 23025 42188
PROPORTION OF INNOCENTS KILLED (PIN) 0.230 0.345 0.999
PROPORTION OF VIETCONG KILLED (PVN) 0.000025 0.032 0.039
TOTAL INNOCENTS KILLED 10707 9731 8755
TOTAL VIETCONG KILLED 0 976 1952
TOTAL KILLED 10707 10707 10707

30 CIDE
 How Free is Free-Riding in Civil Wars ?…

Bibliography

Anderson, Jon Lee (2004), Guerrillas. Journeys in the Insurgent World.

New Cork, Penguin.
Aussaresses, Paul (2001), Services Spéciaux / Algérie 1955-1957. Paris,
Perrin.
Binford, Leigh (1996), The El Mozote Massacre; Anthropology and Human
Rights, Tucson, Univresity of Arizona Press.
Brinkley, Douglas (2003), “Tour of Duty / John Kerry in Vietnam”, The
Atlantic Monthly, December, 47-60.
Brocket, Charles D. (2005), Political Movements and Violence in Central
America, New Cork, Cambridge University Press.
Browning, Christopher R. (1990), “Germans and Serbs: The Emergence of
Nazi Antipartisan Policies in 1941”, in Michael Berenbaum (ed.), A Mosaic of
Victims: Non-Jews Persecuted and Murdered by the Nazis, New York, New York
University Press, 64-73.
Collier, Paul (2001), “Economic Causes of Civil Conflict and their
Implications for Policy”, in Chester A. Crocker, Fen Osler Hampson, and Pamela
Aall (editors), Turbulent Peace: The Challenges of Managing International
Conflict. Washington, D.C.: United States Institute of Peace Press, 2001,
143-162.
Condit, D. M. (1961), Case Study in Guerrilla War: Greece During World
War II. Washington DC: Special Operations Research Office, The American
University.
Cooper, Matthew, (1979), The Nazi War Against Soviet Partisans, 1941-
1944. New Cork, Stein and Day.
Davis, Shelton H. (1988), “Introduction: Sowing the Seeds of Violence”, in
Robert M. Carmack (ed.), Harvest of Violence: The Maya Indians and the
Guatemalan Crisis, Norman, University of Oklahoma Press, 3-36.
Elliott, David W.P. (2003), The Vietnamese War: Revolution and Social
Change in the Mekong Delta, 1930 – 1975, Armonk, New York and London, M.E.
Sharpe.
Elster, Jon (1989), The Cement of Society: A Study of Social Order,
Cambridge, Cambridge University Press.
Fearon, James D. (2004), “Why Do Some Civil Wars Last So Much Longer
than Others?”, in Journal of Peace Research 41, 3:275 - 301.
Fearon, James D. and David D. Laitin (2003), “Ethnicity, Insurgency, and
Civil War”, in American Political Science Review 97, 1:75-86.
Gould, Roger V. (1995), Insurgent Identities: Class, Community, and
Protest in Paris from 1848 to the Commune, Chicago, University of Chicago
Press.
Gross, Jan T. (1979), Polish Society Under German Occupation: The
Generalgouvernement, 1939-1944, Princeton, NJ: Princeton University Press.
Gumz, Jonathan E. (2001), “Wehrmacht Perceptions of Mass Violence in
Croatia, 1941-1942”, in The Historical Journal 44, 4:1015-1038.

DIVISIÓN DE ESTUDIOS INTERNACIONALES 31

Stathis N. Kalyvas and Matthew Adam Kocher

Hayden, William (1999), “The Kosovo Conflict: The Strategic Use of

Displacement and the Obstacle to International Protection”, in Civil Wars 2,
1:35-68.
Herrington, Stuart A. (1997), Stalking the Vietcong / Inside Operation
Phoenix: A Personal Account, Novato, California, Presidio Press.
International Committee of the Red Cross (ICRC) (2004), Report of the
International Committee of the Red Cross (ICRC) on the Treatment by the
Coalition Forces of Prisoners of War and Other Protected Persons by the
Geneva Conventions in Iraq during Arrest, Internment and Interrogation.
Geneva, ICRC.
Jones, Adam (2000), “Gendercide and Genocida”, in Journal of Genocide
Research 2, 2:185– 211.
Kalyvas, Stathis N. (2006), The Logic of Violence in Civil War, New York,
Cambridge University Press.
Klinkhammer, Lutz (1997), Stragi Naziste in Italia: La guerra contro i civili
(1943-1944), Roma, Donzelli.
Kedward, H. R. (1993), In Search of the Maquis: Rural Resistance in
Southern France 1942-1944, Oxford, Oxford University Press.
Kheng, Cheah Boon (1983), Red Star Over Malaya: Resistance and Social
Conflict During and After the Japanese Occupation of Malaya, 1941-1946,
Singapore, Singapore University Press.
Lansdale, Edward G. (1964), “Viet Nam: Do We Understand Revolution?”,
in Foreign Affairs 43:1, 75-86.
Li, Lincoln (1975), The Japanese Army in North China, 1937-1941 /
Problems of Political and Economic Control, Tokio, Oxford University Press.
Lichbach, Mark Irving (1996), The Cooperator's Dilemma, Ann Arbor, The
University of Michigan Press.
_______, (1995), The Rebel’s Dilemma, Ann Arbor, The University of
Michigan Press.
_______, (1987), “Deterrence or Escalation? The Puzzle of Aggregate
Studies of Repression and Dissent”, in The Journal of Conflict Resolution 31, 2:
266-297.
Lichbach, Mark Irving and Ted Gurr (1981), “The Conflict Process: A Formal
Model”, in The Journal of Conflict Resolution 25, 1: 3-29.
Lotnik, Waldemar (1999), Nine Lives / Ethnic Conflict in the Polish-
Ukrainian Borderlands, London, Serif.
Maass, Peter (2005), “The Way of the Comandos”, in New York Times
Magazine, 1 May, 38–83.
Mason, T. David (1989), “Nonelite Response to State-Sanctioned Terror”, in
The Western Political Quaterly 42, 4:467-492.
Mason, T. David and Dale A. Krane (1989), “The Political Economy of Death
Squads: Toward a Theory of the Impact of State-Sanctionned Terror”, in
International Studies Quarterly 33, 175-198.
Mazower, Mark (1998), Dark Continent: Europe’s Twentieth Century,
London, Allen Lane.
McCoy, Alfred W. (1980), “‘Politics by Other Means’: World War II in the
Western Visayas, Philippines”, in Alfred W. McCoy (ed), Southeast Asia Under

32 CIDE
 How Free is Free-Riding in Civil Wars ?…

Japanese Occupation. New Haven: Yale University Southeast Asia Studies, 191-
245.
McKenna, Thomas M. (1998), Muslim Rulers and Rebels: Everyday Politics
and Armed Separatism in the Southern Philippines, Berkeley, University of
California Press.
McNeill William H. (1947), The Greek Dilemma: War and Aftermath,
Philadelphia, J.B. Lippincott.
Moyar, Mark (1997), Phoenix and the Birds of Prey: The CIA’s Secret
Campaign to Destroy the Viet Cong. Annapolis, MA, Naval Institute Press.
National Archives and Records Administration (NARA). n.d. National Police
Infrastructure Analysis Subsystem – II. Records Group 330. Accession No. 3-349-
79-002-D.
Petersen, Roger D. (2001), Resistance and Rebellion: Lessons from Eastern
Europe, Cambridge, Cambridge University Press.
Simpson, John (2006), “No Surprises in the War on Terror”, BBC News, 13
February. http://news.bbc.co.uk/2/hi/middle_east/4708946.stm
Stoll, David (1993), Between Two Armies: In the Ixil Towns of Guatemala,
New Cork, Columbia University Press.
Tarrow, Sidney (1998), Power in Movement / Social Movements and
Contentious Politics, Cambridge, Cambridge University Press.
Thayer, Thomas C. (1985), War Without Fronts: The American Experience
in Vietnam, Boulder, CO, Westview Press.
Tilly, Charles, (1978), From Mobilization to Revolution, Reading: Addison-
Wesley.
Tone, John Lawrence, (1994), The Fatal Knot: The Guerrilla War in
Navarre and the Defeat of Napoleon in Spain. Chapel Hill, The University of
North Carolina Press.
Tucker, Shelby (2001), Among the Insurgents /Walking Through Burma.
London, Flamingo.
Tullock, Gordon (1971), The Paradox of Revolution. Public Choice, ***: 89-
99.
Vazeos, Emmanouil (1961), Ta agnosta paraskinia tis Ethnikis Antistaseos
eis tin Peloponnison (The unknown back-stage of the National Resistance in the
Peloponnese), Korinthos, Self-published.
Wickham-Crowley, Timothy P. (1991), Exploring Revolution: Essays on
Latin American Insurgency and Revolutionary Theory, Armonk, NY: M. E.
Sharpe.
Wood, Elisabeth Jean (2003), Insurgent Collective Action and Civil War in
El Salvador, New York, Cambridge University Press.
Wou, Odoric Y.K. (1994), Mobilizing the Masses: Building Revolution in
Henan, Stanford, Stanford University Press.

DIVISIÓN DE ESTUDIOS INTERNACIONALES 33

 Novedades

DIVISIÓN DE ADMINISTRACIÓN PÚBLICA

Mariscal, Judith, Rivera Eugenio, Organización industrial y competencia en las

telecomunicaciones en América Latina: estrategias empresariales, AP-173
Rivera Urrutia, Eugenio, La construcción del gobierno electrónico como
problema de innovación institucional: la experiencia mexicana, AP-174
Bravo Pérez, Héctor Manuel, et al., Evaluación económica del convenio de
distribución de aguas superficiales del lago de Chapala y su efecto en el
bienestar social, AP-175
Bravo Pérez, Héctor Manuel, Juan Carlos Castro Ramírez, Construcción de una
matriz de contabilidad social con agua para el estado de Guanajuato, AP-
176
Bracho González, Teresa y Giulianna Mendieta, El proceso de focalización
y las estrategias estatales de selección de beneficiarios: El caso del
Programa Escuelas de Calidad, AP-177
Arellano, David y Walter Lepore, Publicness y Nueva Gestión Pública: hacia una
recuperación de los valores de lo público, AP-178
López Ayllón, Sergio y Alí Bernardo Haddou Ruiz, Rendición de cuentas en los
órganos reguladores autónomos: algunas consideraciones sobre el diseño
institucional de las autoridades reguladoras en México, AP-179
Sour, Laura, Pluralidad y transparencia en el proceso de aprobación
presupuestal al interior de la Cámara de Diputados, AP-180
Cabrero, Enrique, Los retos institucionales de la descentralización fiscal en
América Latina, AP-181
Merino, Mauricio, La profesionalización municipal en México, AP-182

DIVISIÓN DE ECONOMÍA

Brito Dagobert L. y Juan Rosellón, The Political Economy of Solar Energy, E-357
Torres Rojo, Juan M., et al., Índice de peligro de incendios forestales de largo
Plazo, E-358
J. M. Torres et al., Bequest Motive for Conservation in Forest Production
Communities, E-359
Hernández, Fausto y Brenda Jarillo Rabling, Is Local Beautiful? Decentralization
in Mexico in the Presence of Elite Capture, E-360
Unger, Kurt, El desarrollo económico y la migración mexicana: el TLCAN e
Impactos en las regiones, E-361
Gómez Galvarriato, Aurora, Rafael Dobado and Jeffrey Williamson,
Globalization, De-Industrialization and Mexican Exceptionalism, 1750-1879,
E-362
Gómez Galvarriato, Aurora and Aldo Musacchio, Larger Menus and
Entrepreneurial Appetite: An Empirical Investigation of Organization Choice
 in Mexico, E-363
Mayer, David y Carlos Bazdresch, Hacia un consenso para el crecimiento
económico en México: puntos mínimos para una estrategia coherente de
desarrollo, E-364
Mayer, David y Ma. Fernanda López Olivo, Transmisión intergeneracional de
habilidades cognitivas por niveles socioeconómicos…, E-365
Mayer, David, Dinámica geográfica de productividad e innovación en la
manufactura mexicana, E-366

DIVISIÓN DE ESTUDIOS INTERNACIONALES

González González, Guadalupe, México ante América Latina: Mirando de reojo

a Estados Unidos, EI-132
Ortiz Mena L.N., Antonio Ortiz y Ricardo Sennes, Brasil y México en la economía
política internacional, EI-133
Minushkin, Susan y Matthew Adam Kocher, Trade and Investment Policy
Preferences and Public Opinion in Mexico, EI-134
Ortiz Mena L.N., Antonio, México ante el sistema monetario y comercial
internacional: lecciones de Bretton Woods a la actualidad, EI-135
Meseguer Covadonga et al., The Diffusion of Regulatory Reforms in Pension
Systems: Latin America in Comparative Perspective, EI-136
Schiavon, Jorge A., La relación especial México-Estados Unidos: Cambios y
continuidades en la Guerra y Pos-Guerra Fría, EI-137
Ortiz Mena, Antonio, The Domestic Determinants of Mexico’s Trade Strategy,
EI-138
Kocher, Matthew Adam and Stathis N. Kalyvas, How free is “Free Riding” in
Civil Wars? Violence, Insurgency, and the Collective Action Problem, EI-139
Chabat, Jorge, Mexico: The Security Challenge, EI-140
Kydd, Andrew, The Ball is in your Court: Mediation and Blamecasting, EI-141

DIVISIÓN DE ESTUDIOS JURÍDICOS

Posadas, Alejandro, Canada Trade Law & Policy after NAFTA and the WTO, EJ-8
Hernández, Roberto, Alcances del “juicio oral” frente a la Reforma Integral a
la Justicia Penal propuesta por presidencia, EJ-9
Magaloni, Ana Laura, El impacto en el debate sobre la reforma judicial de los
estudios empíricos del sistema de justicia: el caso del estudio del Banco
Mundial sobre le Juicio Ejecutivo Mercantil, EJ-10
Bergman, Marcelo, Do Audits Enhance Compliance? An Empirical Assessment of
VAT Enforcement, EJ-11
Pazos, María Inés, Sobre la semántica de la derrotabilidad de conceptos
jurídicos, EJ-12
Elizondo Carlos, Luis Manuel Pérez de Acha, Separación de poderes y garantías
individuales: La Suprema Corte y los derechos de los contribuyentes, EJ-13
Fondevila Gustavo, Estudio de percepción de usuarios del servicio de
 administración de justicia familiar en el Distrito Federal, EJ-14
Pazos, Ma. Inés, Consecuencia lógica derrotable: análisis de un concepto de
consecuencia falible, EJ-15
Posadas, Alejandro y Hugo E. Flores, Análisis del derecho de contar con un
juicio justo en México, EJ-16
Posadas, Alejandro, La Responsabilidad Civil del Estado /Análisis de un caso
hipotético, EJ-17
López, Sergio y Posadas Alejandro, Las pruebas de daño e interés público en
materia de acceso a la información. Una perspectiva comparada, EJ-18

DIVISIÓN DE ESTUDIOS POLÍTICOS

Murillo, María Victoria y Martínez Gallardo Cecilia, Policymaking Patterns:

Privatization of Latin American Public Utilities, EP-178
Cermeño Rodolfo, Sirenia Vázquez, What is Vote Buying? The Limits of the Market
Model, EP-179
Schedler Andreas, Electoral Authoritarianism Concept, Measurement, and Theory,
EP-180
Negretto L. Gabriel, Confronting Pluralism: Constitutional Reform in Mexico After
Fox, EP-181
Beltrán Ulises, Contextual Effects on the Individual Rationality: Economic
Conditions and retrospective Vote, EP-182
Nacif Benito, ¿Qué hay de malo con la parálisis? Democracia y gobierno dividido en
México, EP-183
Langston Joy, Congressional Campaigning in Mexico, EP-184
Nacif Benito, The Fall of the Dominant Presidency: Lawmaking Under Divided
Government in Mexico, EP-185
Lehoucq, Fabrice E., Constitutional Design and Democratic Performance in
Latin America, EP-186
Martínez Gallardo, Cecilia and John D. Huber, Cabinet Turnover and Talent
Searches, EP-187

DIVISIÓN DE HISTORIA

Pani, Erika, Saving the Nation through Exclusion: The Alien and Sedition Acts
and Mexico´s Expulsion of Spaniards, H-32
Pipitone, Ugo, El ambiente amenazado (Tercer capítulo de El Temblor…), H-33
Pipitone, Ugo, Aperturas chinas (1889, 1919, 1978), H-34
Meyer, Jean, El conflicto religioso en Oaxaca, H-35
García Ayluardo Clara, El privilegio de pertenecer. Las comunidades de fieles y
la crisis de la monarquía católica, H-36
Meyer, Jean, El cirujano de hierro (2000-2005) H-37
Sauter, Michael, Clock Watchers and Stargazers: On Time Discipline in
Early-Modern Berlin, H-38
Sauter, Michael, The Enlightenment on Trial…, H-39
Pipitone, Ugo, Oaxaca prehispánica, H-40
Medina Peña, Luis, Los años de Salinas: crisis electoral y reformas, H-41
 Ventas

DIRECTAS:

57-27-98-00 ext. 6094

Fax: 5727 9800 ext. 6314

INTERNET:

Librería virtual: www.e-cide.com

Página web: www.cide.edu
e-mail: [email protected]

LIBRERÍAS DONDE SE ENCUENTRAN DOCUMENTOS DE TRABAJO:

• LIBRERÍAS GANDHI Tel. 56-6110-41

• LIBRERÍA CIDE/F.C.E. Tel. 57-27-98-00 EXT. 2906
• SIGLO XXI EDITORES S.A. DE C.V. Tel. 56-58-75-55
• UAM AZCAPOTZALCO Tel. 53-18-92-81
• ARCHIVO GENERAL DE LA NACIÓN Tel. 51-33-99-00