UNITARY VERSUS COLLECTIVE

MODELS OF THE HOUSEHOLD:

IS IT TIME TO SHIFT THE BURDEN
OF PROOF?
Harold Alderman
Pierre-And ré
Chiap pori Lawrence
Had dad
]ohn Hod dinott
Ravi Kanbur

M ost development objectives focus on the well-being of individuals. Poli

cies are targeted to increase the percentage of individuals who avoid
poverty, who can read., who are free from hunger and illness, or who
can find gainful employment. Individual welfare, however, is based in
large part on a com plex set of interactions among family members.
Until recently most polic y analyses implicitly viewed the household as
having only one set of preferences. This assumption has been a powerful too/
for understanding household behavior, such as the distribution of tasks and
goods. But a growing body of evidence suggests that this view is an expe
dience that comes at considerable, and possibl y avoidable, cost. The article
argues that more effective polic y instruments will emerge from analyzing the
processes by which households balance the diverse interests of their
members.

E
xperience in severa! sectors shows that, when policymakers neglect pat terns
of distribution within households, they do so at their peril. Con
sider government attempts to target programs to individuals in certain
age groups, rather than to households: nutritionists, for example, recognizing
the vulnerability of preschool children, often target supplementary feeding to
this age group. lnternational experience, however, indicares that such interven
tions will not succeed unless the actions of other household members are taken
into account; households often reduce the amounr of food given to rhe target
child at home and distribute it among the child's siblings (Beaton and Ghassemi
1982; Kennedy and Alderman 1987). At the other end of the age spectrum, the

The World Bank Research Observer, vol. 10, no. 1 (February 1995), pp. 1-19
©.1995 Thc lntemational Bank for Reconstruction and Dcvclopmcnt/THE WORLD BANK 1
·full impact of targeting programs to the elderly can only be effectively assessed if
the responses of other family members are taken into consideration (Cox and
Jimenez 1992).
Similarly, many attempts to introduce new crops or agricultural technologies
have not fared as well as expected because policymakers did not give adequate
consideration to different household members' responsibilities for crops. For ex
ample, in her study of rice production in Cameroon, Jones (1986) found that rice
was considered to be a "male" crop, that is, men controlled any income generated
from rice, even if the crop was produced by women. Despite recommendations to
concentrate on rice cultivation, few women planted rice; instead, they grew sor
ghum, which they controlled, despite its lower returns.
As a result of these and similar experiences, there is increasing recognition
that distribution of tasks and goods within households is important in project
design. In this paper we provide an overview of the burgeoning literature on
this subject. Our principal goal is to demonstrate that understanding the
process by which resources :;ire distributed within households has important
implications for policy.
We call the prevailing model of distribution within households the unitary
model. This model implies that what matters for certain policy initiatives-such
as public works schemes or transfer programs-is the amount of income the
household re ceives, not the identity of the individual within the household who
is the target of the public program. Conversely, under sorne alternative models,
the efficacy of pro grams depends on the member of the household targeted.
The guide to policy making implied in the unitary model is simpler if it is correct,
inefficient if it is not.
Models that assume that households behave as if they had a single decision
maker can lead to a failure to understand the long reach of sorne public interven
tions. We provide examples in which understanding how resources are distrib
uted within a household can strengthen policy design. We also review the theory
and evidence accumulated by a range of studies that indicate weaknesses in the
unitary model.
Although this evidence still requires sorne bolstering, we suggest a shift in
emphasis; what we refer to as the collective models of household behavior
should be regarded as the standard approach, with the unitary model regarded
as an important but special case. We do not counsel abandonment of the unitary
model; it has proved to be a powerful and pliable tool for household studies. But
in many circumstances, using a unitary model of the household in inappropriate
situations has more serious policy consequences than using a collective model
when a unitary model would have been appropriate (see Chiappori 1992b;
Haddad, Hoddinott, and Alderman 1994; and McElroy 1992).

The Unitary Model of Household Behavior

Until fairly recently, most econornists viewed the household as a collection of
individuals who behave as if they agreed on how best to combine their time,

2 The World Bank Research Observer, vol. 10, no. 1 (February, 1995)
 goods purchased in the market, and goods produced at home. This approach
originates in standard demand analysis and has been extended to include house
hold decisions about child care, crop adoption, education, fertility, health, home
production, labor supply, land tenure, and migration. Indeed, this view even
offers a perspective on the formation and dissolution of the household, that is,
on marriage and divorce (Becker 1973).
This approach is appealing because it allows us to analyze the impact of
changes in policy and other relevant variables on individual behavior with rela
tive ease and it can address diverse issues. lt is sometimes called the common
preferences model, the altruism rnodel, or the benevolent dictator rnodel. We call
it the unitary model because this !abe! describes how the household is assumed to
act (as one). Other labels tend to reflect the means by which the household is hy
pothesized to act as one. Common preferences are only one way in which the
household can act as one; violence or the threat of violence is another. Altruism
has also been used to explain why households might behave as one individual,
but it is altruism under very restrictive conditions, as we shall show later.
The unitary model has sorne important limitations. lt can allow prices to
differ for various household members (the wife's and husband's wages, for ex
ample), but it assumes that all household resources (capital, labor, and land) are
pooled. This assumption requires that at least one member of the household is
able to monitor the other mernbers and to sanction those who fall foul of its
rules, an issue both of information flows and control.
We are critica! of the unitary rnodel because it fails to incorporate the process
by which resources are distributed within households. The model is able to
explain differences in individual welfare within a household, however, even
when these differences are exhibited systernatically by gender, age, or relation
to household head. 1 These distributional inequalities could be generated by
preferences for in equality shared by household members. Moreover, unequal
distribution of re sources may be considered efficient for households. For
example, resources may be distributed on the basis of differences in individuals'
ability to earn higher in comes, and the higher incorne would then be shared by
all members. In such a manner individual differences are treated as different
prices and wages.
Pitt, Rosenzweig, and Hassan (1990), who extend the agricultura! household
model of Singh, Squire, and Strauss (1986), illustrate the adaptability of unitary
models. They suggest that, if sorne individuals can earn or produce more for the
household when they are healthy and better fed, then it makes sense for the
house hold to provide extra calories to those individuals. They also find that, in
sorne seasons, individuals with the best health in a family do not receive enough
calories to compensate them fully for their effort. Thus, within households,
resources may be distributed so that consumption is more equitable than work
effort.
If, as it is likely, individual household members have different preferences,
then each individual's preferences would have to be considered in assessing the
total well-being of the household. A vast literature on social choice illustrates the
theoretical difficulties associated with aggregating individuals' preferences. lt is

Harold Alderman, Pierre-André Chiappori, Lawrence Haddad, John Hoddinott, and Ravi Kanbur 3
difficult to reflect the preferences of all household members and not just those of
a single member, even if a single member were to act as a so-called benevolent
dictator.
Various approaches to solving the problem of aggregation of preferences have
been offered. Samuelson (1956) suggested that the aggregation of preferences
and the pooling of household resources could be achieved by consensus among
household members, but he did not indicate how such a consensus is reached.
Other proposed solutions include the assumption that individuals tend to seek
spouses with similar preferences (assortative mating) and the treatment of
households as markets in which bartering or trade occurs (Becker 1973). These
solutions are not satisfactory because assortative mating does not resolve the
potential conflict in preferences across generations and a model of households as
markets fails to address the problems of monitoring and incentives. An alterna
tive approach is based on Sen's (1966) model of cooperatives. Here, family
welfare is the weighted sum of the net utility of all members, but the model begs
the question of how these weights are determined.
Another attempt to resolve the problems of aggregation and enforcement is
Becker's "rotten kid theorem" (1974, 1981). Becker considers the case of a
household with two members, a benefactor and a recipient. The benefactor,
who is an altruist, transfers consumption to the recipient, a selfish individual
who cares only about personal consumption. Now suppose the recipient under
takes sorne action that raises bis or her consumption but lowers that of the
benefactor (hence the "rotten kid" sobriquet). The benefactor could respond by
lowering transfers so that the recipient's new level of consumption is below the
original leve!. Consequently, the recipient is not likely to behave rottenly in the
fust place. Thus, the preferences of the altruist and the preferences of the house
hold converge.
Unfortunately, the rotten kid theorem only holds under restrictive circum
stances (Bergstrom 1989; Haddad, Hoddinott, and Alderman 1994). lt has
proved to be important, however, because it provides testable assumptions and
because the underlying altruism has strong policy implications regarding the
extent to which government policies are mitigated by prívate response.

Collective Models of Household Behavior

Severa} formulations of the unitary model contain an assumption that inequi
table distribution of resources or leisure within a household represents a willing
act on the pan of all household members. Although models of the distribution of
resources within households are as much about sharing among generations as
between genders, this assumption is viewed as particularly restrictive when it is
applied to decisionmaking between spouses. As one of the most noted critics of
the unitary model, Folbre (1986, p. 251), comments:
The suggestion that women and female children "voluntarily" relin
quish leisure, education, and food would be somewhat more persua-

4 The World Bank Research Observer, vol. 10, no. 1(February, 1995)
 sive if they were in a position to dernand their fair share. lt is the
juxtaposition of wornen's lack of econornic power with the unequal
allocation of household resources that lends the bargaining power [col
lective rnodel] approach rnuch of its persuasive appeal.

Similar concerns have given irnpetus to severa! collective rnodels that focus on
the individuality of household rnernbers. These rnodels explicitly address the
question of how individual preferences lead to a collective choice. These are
sometirnes referred to as bargaining rnodels, but we prefer the more generic label
of "collective" rnodels, partly because sorne irnportant collective rnodels do not
explicitly address bargaining. Moreover, the phrase can be neatly juxtaposed
with the terrn "unitary" rnodels.
Cornrnon arnong the various collective rnodels is their interest in directly
addressing how individual household rnernbers reconcile different preferences.
These approaches can be subdivided into two broad categories: those that rely
on noncooperative relations; and those that rely on cooperative solutions.
In comrnon with the unitary rnodel, the cooperative approach begins by not
ing that individuals form a household when it is more beneficial to them than
rernaining alone. Higher benefits could occur because forming a household is a
more efficient way to produce household goods or because sorne goods can be
produced and shared by rnarried couples but not by single individuals. For
exarnple, household forrnation rnay generate benefits such as "love" or "corn
panionship." In any case, gains accrue frorn household formation, and these
need to be distributed across the rnernbers. Where collective rnodels depart frorn
unitary models is in the rule goveming this distribution.
The noncooperative approach ( Ulph 1988; Kanbur 1991; Carter and Katz
1992; Lundberg and Pollak 1993) relies on the assurnption that individuals
cannot enter into binding and enforceable contracts with each other. Instead,
individuals' actions are conditional on the actions of others. For exarnple, Carter
and Katz's fairly polar "reciprocal clairns" rnodel depicts the household as con
sisting of largely separate, gender-specific econornies linked by reciproca! clairns
on rnernbers' incorne, land, goods, and labor. A wife's budget is separate frorn
her husband's; she responds to changes in her husband's allocation of labor
solely according to her own needs. The transfer of incorne between thern estab
lishes the only link between the wife and husband. Sirnilarly, in Lundberg and
Pollak's rnodel, "each spouse rnakes decisions within his or her own sphere" (p.
994) and responds to the other's decisions by altering the level of voluntary
contribution to shared goods.
In efficient cooperative rnodels, it is assurned that household decisions are
always efficient in the sense that no one can be rnade better off without sorneone
being rnade worse off. The rnodels rnake no assurnptions about how resources
are distributed within households. A sirnplified version of this approach is as
follows. Suppose a household consists of two individuals. Once a decision has
been reached regarding expenditures on public goods, rernaining household

Harold Alderman, Pierre-André Chiappori, l.Awrence Haddad, ]ohn Hoddinott, and Ravi Kanbur 5
income is allocated among the prívate goods according to a sharing rule. The
sharing rule in tum is affected by the incomes of the two household members.
This model can illustrate how individual incomes affect the household's con
sumption of different goods (see Bourguignon, Browning, and Chiappori 1994).
Further, it is possible to identify the household's sharing rule even if no individ
ual consumption is observed.2
A key feature of the efficient cooperative approach is that the rules regarding
distribution within households come from the data and are not assumed. This is
especially convenient for assessing the relevance of the altemative models
(Chiappori 1992b).
Sorne cooperative models impose structure by representing household deci
sions as the outcome of sorne specific bargaining process and applying the tools
of game theory to this framework. Then the division of the gains from marriage
depend on the "fallback;' or "threat point;' position of each member. These
fallback positions are a function of extra-environmental parameters, that is,
demographic, legal, and other macroeconomic conditions external to the house
hold. These include sex ratios in marriage markets, laws conceming alimony
and child support, changes in tax status associated with different marital states,
and, in developing countries, the ability of women to retum to their natal homes
and prohibitions on women working outside the home (McElroy 1990, 1992).
Collective decisionmaking can be enforced in two ways. The first is through
the threat of household dissolution. McElroy ( 1992) notes, however, that in the
context of small daily decisions, it is not credible for either spouse to threaten
divorce. She suggests that a second way to analyze decisions regarding short-run
issues is to use differences in impatience to reach an agreement, with the nonco
operative solution acting as the threat point.

Policy Imfclications of Altemative Models of Distribution within

Househo ds
Is the distinction between unitary and collective models merely an arcane
academic curiosity? Or do differences in how resources are distributed within
households, as the various theories imply, reflect appreciable differences in the
outcomes of policy measures?
Clearly, how resources are distributed within households affects the measure
ment of poverty and inequality. Consider a country in which the central govern
ment makes transfer payments to provincial or state authorities. The size of
these transfers is determined by estimated levels of poverty. Does it matter if
poverty is measured with reference to households or to individuals? If resources
are equally distributed among household members, either measure will yield the
same estímate of the degree of poverty. As Haddad and Kanbur (1990) demon
strate, however, this no longer holds if resources are unequally distributed
within the household. Drawing on individual- and household-level data on calo
ric availability in the Philippines, they estímate the incidence of poverty using the

6 The World Bank Research Observer, vol. 10, no. 1(Fcbruary, 1995)
income-gap poverty measures proposed by Foster, Greer, and Thorbecke (1984)
and find that ignoring unequal distribution within households understates pov
erty by 18 to 23 percent.
Haddad and Kanbur's illustration is based on a poverty measure that pays
particular attention to food consumption, but the general result also holds when
income is used instead of food. For example, Apps and Savage (1989) find that
the welfare rankings of households in Australia depend critically on transfers
between spouses. To the extent that there are empirical difficulties in identifying
how resources are distributed within households, however, determining the
changes in rankings caused by these spousal transfers is difficult.
Does the analytical complexity associated with collective models offer any
additional insights for policy interventions? We illustrate below four areas in
which the choice of model is important.

Public Transfers to Individual Household M embers

The daim that household decisions are not affected by the identity of the
individual eaming income has been refuted in a number of settings. This has
obvious implications for policy, as the following quotations illustrate.
Many participants in the public debate conceming actual govem
ment transfers take it for granted that intrafamily distribution will vary
systematically with the control of resources. When the British child
allowance system was changed in the mid-1970s to make child benefits
payable in cash to the mother, it was widely regarded as a redistribu
tion of family income from men to women and was expected to be
popular with women. (Lundberg and Pollak 1993, p. 989)

Indeed, so convinced did sorne Ministers become that a transfer of

income "from the wallet to the purse" at a time of wage restraint would
be resented by male workers, that they decided at one point in 1977 to
defer the whole child benefit scheme. (Brown 1984, quoted in
Lundberg and Pollak 1993, p. 989)
Most examples of income pooling revolve around the fact that women spend
more of their income on food and child care. Thomas (1990, 1992) finds that in
Brazil, for example, the identity of the household member controlling income
affects nutrient intake, fertility, child survival, and young children's weight for
height. The results for child survival are particularly dramatic; increases in the
mother's uneamed income raise child survival by twenty times that resulting
from a comparable increase in the father's uneamed income. We discuss such
evidence in greater detail in the next section.
The importance of potential policy failures arising from neglecting the identity
of the transfer recipient is likely to grow as social safety nets are implemented to
ameliorate the short-run negative effects of economic adjustment. Newman,

Harold Alderman, Pierre-André Chiappori, Lawrence Haddad, John Hoddinott, and Ravi Kanbur 7
Jorgensen, and Pradhan (1991) found that in Bolivia the urban infrastructure
construction project of the Social Emergency Fund bolstered the incomes of the
poor in a cost-effective manner. But only 2 percent of the participants in the fund
were women. The untested assumption seems to have been that fund income
would trickle clown to wives, mothers, and children or that they would be better
served through credit and other programs in which female participation was
substancial.

Public Transfers and Interactions between Household M embers

The nature of interactions among household members determines whether
changes in household behavior mitigare or enhance che effectiveness of public
transfers.
The potential for changes in household behavior to offset the effectiveness of
public transfers has been recognized as imponant since Barro's (1974) seminal
paper on family obligations and tax policy. Barro noted, for example, that
households may respond to the introduction of a social security system by
eliminating completely any private transfers from the young to the old.
If intergenerational sharing within a household is not purely altruistic, how
ever, households may respond differently. For example, Cox and Jimenez
(1990) consider a hypothetical family with young members residing in towns
and old members living in rural areas. They consider the case in which
transfers are made from the young to the old, and individual consumption
depends on the family's total income. Suppose a social security program is
introduced that taxes the young and subsidizes the old, leaving total family
income unchanged. If individuals in the household acted altruistically, this
might well lead to a reduc tion in urban-rural remittances (although
consumption is unchanged). But it may be that transfers from the young to
the old were undenaken in exchange for services (such as home production).
Once social security payments are provided, rural residents might be less
willing to provide services to urban residents. As a result, the urban
household members must transfer higher amounts to their elders to retain
the same services. This is the opposite result of that predicted by the altruistic
unitary model.

Policy Initiatives Directed to Individual Household M embers

The unitary model implies that it does not matter how policy initiatives are
directed; the household will respond to that policy independentl y of the recipient
of information or services. This assumption gives rise to two potencial policy
failures: a resistance to panicular policies that appear beneficia!, and the unin
tended costs of policies that are adopted. Consider the consequences of these
two policy failures in terms of the adoption of new technology in developing
countries.3
The first example (mentioned in the introduction) described women's opposi
tion to recommendations to plant rice in Cameroon. In another instance, rural

8 The World Bank Research Observer, vol. 10, no. 1 (February, 1995)
 households in Zambia were encouraged to intercrop maize, a male-controlled
crop, with beans, a female-controlled crop (Poats 1991). Diets would have
improved as a result of the well-known complementary benefits from consuming
these two crops, and less work would have been required because the interplant
ing scheme reduced weeding time. Women refused to adopt this change, how
ever, because if they planted beans on land normally allocated to maize, they
would lose ownership of the beans.
By contrast, a project in Togo to encourage soybean production to supplement
the household diet with much-needed protein succeeded precisely because it
took into account the collective nature of household behavior ( Dankelman and
Davidson 1988). At the outset, the project was targeted toward women,
through exchange visits and workshops organized in women's homes. Also,
soybeans were not introduced as a cash crop, which would have changed
women's status within the househ9ld. Instead, they were promoted as legumes
that could be used to make sauces. The result was that the crop remained in the
hands of women, who in sorne cases were allocated small plots of land for
cultivation.
Even when the neediest group is correctly identified, however, the policy
might have adverse unintended impacts. Von Braun and Webb (1989) report
that in the early 1980s in the Gambia , rice irrigation was introduced to an area
of swamp rice production to raise yields and commercialize the product. Al
though this initiative was designed to raise women's share of household income,
it reduced that income because yield increases transformed the status of rice
from a prívate crop under women's control into a communal crop under men's
control. Prior knowledge of the relative bargaining positions of men and women
might have helped predict the outcome and enabled policymakers to redesign the
program to meet the original goals.4

Long Reach of Policy M easures

The unitary model depicts as impotent several policy initiatives that neither
directly affect the technology of production nor affect household preferences but
that may in fact have a major impact on household allocation decisions. An
example particularly relevant to developing countries is that of managing com
mon property resources, such as access to common grazing land.
Haddad and Kanbur (1992) oudine the following model. A household has
two individuals, each of whom produces output as the result of two tasks. Each
individual is better at one or the other task, so it pays to specialize and cooperate
in tasks. But how should the two individuals divide the benefits of cooperation?
Suppose that the fallback option for each individual is to work alone. Now
suppose that the govemment introduces a scheme that guarantees better access
for ali to common property resources. How will the govemment scheme affect
inequality within the household? The income generated from improved access to
common property resources might be higher than the income from working

Harold Alderman, Pierre-André Chiappori, Lawrence Haddad, ]ohn Hoddinott, and Ravi Kanbur 9
alone, but still less than the income from cooperation. Then, even if it is not
actually used, more equitable access to common property resources actually
improves equality within the household. The scheme has a remarkably long
reach-it equalizes distribution within the household by altering outside
options.
Several other features of this example are worth noting. The credibility of the
scheme is at the heart of sorne of the policy debates on the extent of access to
common property resources. Access might be rationed in a manner that imposes
limits and does not effect distribution within households. Moreover, the results
of the model hold for several other policy interventions, such as Maharastra' s
employment guarantee program in India, if the guarantee of employment acts as
an inalienable property right.
Most important, Haddad and Kanbur's work illustrates the importance of
distinguishing among different classes of collective models. For example, in a
cooperative model based on bargaining, if improved access to common property
resources is guaranteed only to married women, distribution within households
will be unaffected because women's position outside of marriage is unaffected.
Here, only changes in access to common property resources for women outside
as well as inside marriage would alter the distribution of resources within house
holds. By contrast, if households are operating in a noncooperative setup,
changing married women's access to common property resources would be suffi
cient to affect the position of women within the household.
In a similar manner, a policy might aim to change the distribution of transfers
within a household without influencing the distribution in the event the house
hold dissolves. Such a policy would be ineffective in the context of a cooperative
bargaining model, but it might lead to a redistribution within the household if a
noncooperative model holds. Lundberg and Pollak (1993) modeled a shift in the
distribution of public child support supplements from fathers to mothers but left
intact the distribution of support payments to mothers in the event of a divorce.
In this example, the entitlement influenced the woman's position within mar
riage in a manner similar to the increased access to common property resources.
Because the shift did not affect the situation in the event of household breakup
(by assumption), it had no influence in the cooperative model.

Casting Doubts on the Unitary Model-Evidence

We have argued that the unitary household model faces serious theoretical
challenges and that using this approach entails important policy implications. In
this section, we review evidence that supports these challenges. We begin with
sorne informal evidence. This material is not necessarily nested within a formal
test procedure; nevertheless, it casts doubt upon certain aspects of the unitary
model.
We then present sorne more formal evidence although we recognize that not
all the studies are ideal. Particularly problematic is the fact that income reflects

10 The World Bank Research Observer, vol. 10, no. 1 (February, 1995)
past and current household choices and that measurement is complicated by the
assumption-also held by many applications of the unitary model as well-that
household formation and sometimes composition can be regarded as predeter
mined. Other studies pertain only to specific regions or are based on relatively
small samples. In the examples referred to here, aspects of the unitary model are
seriously questioned or rejected using data drawn from twelve different coun
tries.5 lt is this steady accretion of results that legitimates concerns about the
unitary model.
Few researchers defend the unitary model on the basis of the validity of its
assumptions; these "do violence to reality" ( Rosenzweig 1986, p. 233). Echoing
a previous debate in economics, however, one could argue that realism is unim
portant. As Samuelson (1963, p. 232) put it: "a theory is vindicable if its
consequences are empirically valid to a useful degree of approximation; the
(empirical) unrealism of the theory 'itself; or its 'assumptions; is quite irrelevant
to its validity and worth:' Ultimately, then, the accumulation of the type of
evidence discussed here shifts the starting point in household studies. Again,
echoing the earlier debate: "if the abstract models contain empirical falsities, we
must jettison the models, not gloss over their inadequacies" (Samuelson 1963,
p. 236).

Informal Evidence
Many studies-from severa! disciplines and from both industrial and develop
ing countries-indicate that income is not pooled within a household. Other
arrangements that households adopt indude systems where one person manages
ali finances and expenditures except for personal spending money; a "spheres of
responsibility" system where, for example, a husband gives his wife a set amount
for purchasing specified commodities; and an "independent management" sys
tem, whereby each individual has income and is responsible for certain expendi
tures and no one has access to ali household funds (Pahl 1983). Not surprisingly,
the different ways in which households control income translate into different
patterns of expenditures. Case study material from anthropological and socio
logical studies indicates that men spend more of the income they control for their
own consumption than do women. Alcohol, cigarettes, status consumer goods,
even "female companionship" are noted in these studies. By contrast, women are
more likely than men to purchase goods for children and for general household
consumption.
There is considerable evidence that domestic violence is prevalent in both
industrial and developing countries and that it affects income distribution within
the household. Domestic violence dearly refutes justification for the unitary
household model based on altruism. Violence may underlie a dictatorial version
of a unitary household model. Jones (1986), for example, relates that respon
dents to a survey said that the threat of a beating influenced their decision to
work. Rao (1994) finds that food purchases in India are influenced by domestic

Harold Alderman, Pierre-André Chiappori, Lawrence Haddad, ]ohn Hoddinott, and Ravi Kanbur 11
violence. The study also indicates that extra-environmental parameters may
affect domestic violence and attempts to incorporate domestic violence into a
collective model of household behavior. Tauchen, Witte, and Long (1991) pre
sent further evidence that community factors, such as access to public assistance,
affect the probability of domestic violence.
Extra-environmental parameters have an explicit role in sorne collective
models and thereby provide additional indirect support for such models
(Lundberg and Pollak 993). A few recent studies use these conditions to sup
port bargaining models. An illustration is Rao and Greene's (1993) detailed
analysis of the impact of bargaining on fertility in Brazil. They find that fertility
is lower than average when the ratio of males aged 25 to 29 to females aged 15
to 19 in the region is higher than average. lt is reasonable to interpret this ratio
as a measure of the availability of altemative spouses. As the ratio increases,
women have a greater chance of remarrying if they leave their current house
hold, hence a greater ability to bargain for the smaller families they prefer. 6

Formal Evidence
Several of the assumptions (or restrictions) of the unitary model do not hold
when tested empirically. Three challenges to restrictions of the unitary model are
considered here: nonpooling of labor income; strategic behavior in the context
of intergenerational relations; and the impact of one family member's labor
choices on those of another family member. Tests of the nonpooling of labor
income challenge an underlying assumption of the unitary model. Such tests also
show that policy measures might differ depending on the various methods by
which household members control resources. The other tests of formal restric
tions are not phrased in terms of policy. The tests might seem to state obvious
points, but they contribute evidence that challenges the unitary model.

THE INCOME-POOLING RESTRICTION. A key assumption of the unitary model

is the pooling of household income. Income pooling implies that the identity of
the individual earning the income has no effect on the household demand for
goods and leisure, except through the eaming individual's choice between
leisure and work. Direct tests of the pooling of labor income have
econometric problems. Sorne studies therefore focus on unearned income. In his
study, which concludes that not all households pool income, Schultz (1990, pp.
601-02) notes,
If non-eamed income (or ownership of the underlying asset) influences
family demand behavior differently, depending on who in the family
controls the income (or owns the asset), then the preferences for that
demand must differ across individuals and such families must not com
pletely pool unearned income.
Similarly, Thomas (1990, 1992) finds that increased (nonlabor) income held by
women leads to a greater share of the household budget devoted to expenditures
on education, health care, and food.7

12 Tbe World Bank Researcb Observer, vol. 10, no. 1(February, 1995)
 The use of unearned income to test the income-pooling hypothesis is subject
to econometric criticisms because it likely reflects past choices (Haddad, Hod
dinott, and Alderman 1994). Although none of the existing tests are definitive
(the strongest proof may require an experimental design that randomly assigns
transfers to males and females), many of the more recent studies have addressed
the possibility that the results are an econometric anifact of an unmeasured
factor or reverse causality. Thomas observed (1992, fonhcoming) that fathers
and mothers behave differently toward daughters and sons. One explana tion-
albeit somewhat strained-may be that mothers with daughters choose to work
or invest differently from those ·with sons. In his household fixed-effects
model, however, Thomas also indudes several controls for unobserved individ
ual factors by taking differences across children and across parents. His results
do not disappear with these controls.
Taking a different tack, Hoddinott and Haddad (fonhcoming) use traditional
cropping patterns in Cote d'lvoire to model household behavior based on in
come sources. They also find that income is not pooled. When the share of cash
income received by wives in Cote d'Ivoire is increased, expenditures for food
increase and expenditures on alcohol and cigarettes decrease. In their study of
informal credit programs in Bangladesh, Pitt and Khandker (1994) also find that
credit affects household education and consumption choices differently if it is
given to women rather than to meo. (This study also employs a methodology
that treats the availability of credit much as an experiment and uses that to
control for the fact that credit choices reflect household preferences.)
Collective models provide additional tests of income pooling. Moreover, the
efficiency assumption within collective models strongly restricts the way in
which different income sources may influence consumption patterns. Thus, the
efficiency assumption provides additional tests of how changes in household and
individual income affect household consumption (Bourguignon, Browning, and
Chiappori 1994). Bourguignon and others (1992, 1993) construct a general
model that encompasses both the unitary and the collective frameworks as
special cases. Using data from France and Canada, they find that household
income is not pooled in either country. The restrictions of the collective models
do hold up; that is, the efficiency assumption is valid. Even more interesting is
the comparison, in the second paper, between a sample of couples and two
subsamples of singles. The unitary model fails for the couples, but not for the
singles. This would be the case if the unitary model failed because of a sharing
process negotiated between family members.

INTERGEN ERATIONAL TRANSFERS. The unitary model implies that

benefactors have no incentive to behave strategically. Children, even rotten
ones, do not attempt to raise their consumption at the expense of others
because if they did, an altruistic benefactor would automatically reduce the
size of the transfers made to the children. Correspondingly, it is possible to
test the hypothesis that altruistic benefactors will not intentionally manipulate
the behavior of the recip-

Harold Alderman, Pierre-André Chiappori, Lawrence Haddad, )ohn Hoddinott, and Ravi Kanbur 13
ient. If Becker's (1981) model holds, we should not find evidence of benefactors
behaving strategically by using bequests to obtain attention or monetary trans
fers from their offspring. Lucas and Stark (1985) and Hoddinott (1992), how
ever, find that parents do behave strategically; in Botswana increased holdings of
inheritable assets lead to higher monetary transfers from nonresident family
members (Lucas and Stark). The same pattem occurs in western Kenya in the
case of sons who anticípate receiving an inheritance (Hoddinott). Similarly, the
results in Cox and Jimenez's (1992, p. 167) study of Peru were found to be
"inconsistent with the strict Barro-Becker altruism motive:'
In a related vein, Altonji, Hayashi, and Kotlikoff ( 1992) formally test
altruism by modeling the movement of children's expenditures when parents'
income changes. They interpret their rejection of altruism as supporting the
presence of bargaining within households and as a direct challenge to the central
assumption in Barro's ( 1974) model. 8

LABOR SUPPLY DECISIONS. Restrictions on decisions about labor supply (time

spent eaming income) provide sorne very specific tests of the effect of wages on
the labor supply of spouses. The unitary model implies that, for a given increase
in total income, an increase in the husband's wage will affect the amount of time
his wife works to eam income exactly the same as an increase in the wife's wage
will affect the husband's time spent eaming income.9 Evidence from the United
States (Ashenfelter and Heckman 1974) rejects the equality of these effects (see
also Killingsworth 1983). Further, using panel data to control for unobserved
fixed effects, Lundberg (1988) rejects the hypothesis that the labor supply of the
husband and wife is jointly determined, as predicted by the unitary model.
Similarly, recent work has focused on empirical tests characterizing the "col
lective approach. Chiappori (1988, 1992a) derives restrictions on labor supply
in a model where consumption and leisure are prívate goods and extends the
analysis to collective models. Fortín and Lacroix (1993) estímate a general
model of labor supply in which both the unitary and the collective frameworks
can be tested as special cases. Using data from Canada, they find that the
restrictions of the unitary model are strongly rejected but that the analogous
ones from collective models are not. Browning and Chiappori (1994) analyze
data on consumption and find that the restrictions are rejected for couples but
not for singles (who do not have to share consumption). Moreover, the collective
generalization is not rejected for couples.

Conclusions
Despite the accumulated evidence that income is not pooled, the unitary
model, bolstered by ad hoc assumptions, retains an impressive ability to explain
the new body of evidence on inequality within the household. Moreover, in
many cases the choice of models will not affect either policy or research; Oc
cam's razor argues that in these cases the simplest approach be taken.

14 The World Bank Research Observer, vol. 10, no. 1(February, 1995)
 That said, we maintain that the burden of proof should be shifted onto those
who would claim that the unitary model should be the rule and the collective
models the exception. Our intention is neither to discard the unitary model in its
entirety nor to ignore legitimate concerns about interpreting the evidence critica}
of it. Although the rejection of the restrictions in sorne of the models discussed
has few direct qualitative implications for policymakers, collective models
themselves do have policy ramifications. Under many circumstances, acceptance
of a unitary model of the household, when it is inappropriate, has more serious
consequences for policy than does rejecting a unitary model when it is
appropriate.
To reiterate an example, rejection of the collective model implies
·(erroneously) that targeting transfers to women is pointless; if the model is
rejected when, in fact, it is valid, an efficient means of directing resources to
women and children may be forgone. If the unitary model is rejected when it
is sound, additional costs of targeting may be incurred. But most collective
models imply equal or greater investment in children from income using
resources controlled by women than the unitary model implies. Thus, unless the
costs of targeting programs to women in poor households are significantly
higher than targeting programs to poor households as a unit, the available
evidence may be considered adequate to indicate that false rejection of the
collective model is the more serious error.10
Equally important, a shift of theoretical focus will emphasize the need to
question how resources and activities are allocated and monitored in program
design and implication. In so doing, we surmise that more implications of house
hold allocation processes will emerge.
Implicit in our arguments is a view that household economics has not taken
Becker (1965) seriously enough. "A household;' he wrote, "is truly a 'small
factory': it combines capital goods, raw materials, and labor to clean, feed,
procreate, and otherwise produce useful commodities." (p. 496) We, too, per
ceive the household as a factory, but like ali factories, it consists of individuals
who-motivated at times by altruism, at times by self-interest, and often by
both-cajole, cooperate, threaten, help, argue, support, and, indeed, occa
sionally walk out on each other. In fields such as labor economics, policy has
been informed by research that goes inside the "black box" of the factory and
discusses individual incentives within a corporation. Those interested in the
welfare of individuals, especially in developing countries, may benefit from a
similar approach to the household.