Professional Documents
Culture Documents
CITATIONS READS
0 710
3 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Roberto Castro on 05 January 2015.
*Bobonis: Associate Professor, Department of Economics, University of Toronto, 150St. George Street,
Toronto, Ontario, M5S 3G7, Canada and BREAD and CIFAR (e-mail: gustavo.bobonis@utoronto.ca); Gonzlez-
Brenes: Cove Strategy, 15 Kenwood St., Somerville, MA 02144 (e-mail: melissa@covestrategy.com); Castro:
Professor, Regional Center for Multidisciplinary Research, National Autonomous University of Mexico, Apartado
postal 4-106, Cuernavaca, Morelos, 62431, Mxico (e-mail: rcastro@unam.mx). A preliminary version of the paper
was circulated as Womens Income, Female Status, and Spousal Violence: Effects of the Mexican Oportunidades
Program (March 2006). We are grateful to Michael Baker, Dwayne Benjamin, Morley Gunderson, Gillian
Hamilton, Robert Pollak, Deborah Reed, Heidi Shierholz, and Aloysius Siow, as well as two anonymous referees,
whose suggestions greatly improved the paper. We would also like to thank seminar participants at Toronto, the
PAA 2006 Conference, the ASSA 2007 Conference, and the CEA 2011 Conference for helpful comments, as
well as Caridad Araujo, Tania Barham, Marta Rubio, Iliana Yaschine, and the staff at Oportunidades for provid-
ing administrative data and their general support throughout. Research support from the Institute of Business and
Economics Research at UC Berkeley and the University of Toronto Connaught Fund is gratefully acknowledged.
We are responsible for any errors that may remain.
To comment on this article in the online discussion forum, or to view additional materials, visit the article page
at http://dx.doi.org/10.1257/pol.5.1.179.
1
Violence against women has been condemned at the highest international world population and women confer-
ences (i.e., Cairo 1994 and Beijing 1995) as an issue of human rights, public health, and womens personal security.
In the context of our study, rural Mexico, 7 percent of women in a marital union report being victims of physical
abuse inflicted by their male partners at some point during the previous 12 months (Castro, Riquer, and Medina
2006; Castro and Casique 2008).
179
180 American Economic Journal: economic policy february 2013
statuscould affect the incidence and severity of spousal abuse. A growing num-
ber of developing countries have introduced conditional cash transfer (CCT) pro-
grams, poverty alleviation programs that provide funds to households in exchange
for certain actions such as childrens school attendance, school performance, and
preventive health care visits. A common feature of these programs is that funds are
targeted to adult women, mothers of the target population of children. The basis for
this gender-specific targeting is a consensus among scholars and policymakers that
targeting resources to women is an effective way to channel resources to children,
and may also promote the empowerment of women within the household and in the
community.2 However, a potential unintended consequence of the program (and of
the gender-based targeting, in particular) may be an increased incidence of domestic
violence, since unexpected changes in womens economic conditions may increase
the incentives for male partners to use violence or threats of violence to (re)gain
control over household resources or decision making (McCloskey 1996; Kimerling
and Baumrind 2004).
The main objective of this paper is to provide evidence of the impacts of the
Mexican Oportunidades CCT program on the incidence of male-to-female spousal
violence. We use data from a new and recently available nationally representative
survey, the National Survey on Relationships within the Household (ENDIREH
2003), which includes detailed information on the prevalence of male-to-female
spousal abuse and threats of violence against women. We define violence from the
survey measures as including physical, sexual, and emotional abuse. These distinc-
tions allow us to better characterize the extent and nature of spousal abuse in this
population than previous work in the literature. Our research design is based on the
comparison of beneficiary and nonbeneficiary households within villages, as well as
a careful choice of households based on predetermined characteristics to minimize
potential omitted variable and selection biases.3
We find that women in beneficiary households are 5 to 7 percentage points
(approximately 40 percent) less likely to be victims of physical abuse than non-
beneficiary women, impacts that may result from an improvement in womens power
within the household. However, these women are 3 to 5 percentage points more
likely to be victims of emotional violence, including threats of physical violence,
with no associated physical abuse. These effects are concentrated among women in
households with low expected gains to marriage. The evidence gives a mixed view
of the effectiveness of these programs, as currently structured, to reduce the inci-
dence of domestic violence against women.
2
See a survey of this literature by Duflo (2005) and the seminal papers by Thomas (1990) and Schultz (1990).
There is a burgeoning literature on these programs impacts, and the gender-based targeting of these transfers in
particular, on the levels of female empowerment, the allocation of household resources towards investments in
children, child well-being, and marital transitions (Attanasio and Lechne 2002; Djebbari 2005; Rubalcava, Teruel,
and Thomas 2009; Bobonis 2009, 2011).
3
The estimates we obtain are using a nonexperimental estimator, because we do not use data from the short-
term evaluation of the program. The experimental design was only maintained over a short term (1.5 years in this
case, during the years 19981999), whereas our data is for a different sample of households and for the year 2003
(six years following the initial phase-in of the program). The nonexperimental estimates naturally require stronger
assumptions than the experimental estimates regarding the influence of unobservable characteristics on outcomes.
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 181
How should these findings be interpreted? They provide prima facie evidence
against traditional economic theories of the family, which generally predict that
these transfer programs, by improving womens bargaining position in the house-
hold, may decrease the incidence of both physical and emotional spousal abuse.4
One possible alternative explanation is that male partners may use threats of vio-
lence as instruments of coercion to appropriate or (re)gain control over household
resources or decision making (i.e., extract rents) from their female counterparts.
We reinterpret Bloch and Raos (2002) model of domestic violence, in which male
partners have private information regarding the gains to marriage, such as their
own private income or their status within the household based on traditional gender
roles. This private information enables husbands to use threats of abuse to coer-
cively demand transfers from their wives, and strategically use physical violence (as
a punishment) depending on whether their wife complies with the act of coercion.5
The model predicts that, if marginal increases in womens socioeconomic condi-
tions lead to an increase in the amount of rents that husbands are willing to extract,
this will lead to an increase in the threat of spousal abuse with no associated physical
violence, and a reduction in the actual incidence of physical abuse. These predic-
tions are consistent with the data.
That being said, the instrumental use of violence is not the only possible expla-
nation consistent with the data. The CCT program increases the level of household
resources, not just the socioeconomic opportunities available to (or more specifi-
cally, the income of) adult women. If this increase in household resources reduces
stress levels among partners and/or improves the emotional health of adults (or
children) in the family, it could lead to a reduction in the incidence or severity of
spousal abuse. Another possibility is that the program conditions generate conflict
and greater stress within the household. While our results on physical and emotional
violence are consistent with the instrumental violence hypothesis, in isolation they
do not rule out the possibility of these alternative hypotheses. To address this, we
assess whether there is some reduction in other measures of intrahousehold conflict
by examining whether there is a change in the level of violence against children
perpetrated by each parent. We find no significant evidence of increased violence
against children. Together with the results on physical and emotional violence, these
findings are most consistent with the instrumental violence hypothesis.
Our work relates to a small literature studying heterogeneous impacts of condi-
tional cash transfer programs on spousal violence. Angelucci (2008) finds that in the
context of the randomized evaluation of the rural PROGRESA program, alcohol-
related domestic abuse decreased among households receiving small transfers,
4
It is not obvious that the causal relationship between female socioeconomic opportunities and spousal violence
will be negativeas highlighted by the theories of male backlash. Yet evidence from a number of studies in devel-
oped countries suggests that this is the case; see Stevenson and Wolfers (2006) and Aizer (2010). The evidence for
couples in less developed countries is more mixed. For instance, Panda and Agarwal (2005) find that women who
own property are less likely to be victims of spousal violence in India, but Gonzlez-Brenes (2005) does not find
a relationship between female income shares and the probability of violence for women in several East African
countries.
5
This argument is based on sociological theories of male backlash. These predict that as womens economic
opportunities improve, violence against them may increase if men feel their traditional gender role threatened (e.g.,
Macmillan and Gartner 1999).
182 American Economic Journal: economic policy february 2013
6
A large empirical literature assesses the extent of Pareto-efficiency in intrahousehold resource allocation deci-
sions. Evidence from a number of studies in developed countries suggests that the intrahousehold allocation of
resources is Pareto-efficient (Browning 1994; Browning and Chiappori 1998; Chiappori, Fortin, and Lacroix 2002).
However, Udry (1996), Dercon and Krishnan (2000), and Duflo and Udry (2004) find evidence inconsistent with
Pareto-efficiency among rural households in Africa, although evidence from other West African settings (Akresh
2006; Rangel and Thomas 2005) and from rural Mexico (Bobonis 2009) suggests that the results may not be as
generalizable as previously thought.
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 183
I.Context
7
However, these changes have been more prevalent within families living in urban areas (Schmukler 1998;
Garca and de Oliveira 2006).
184 American Economic Journal: economic policy february 2013
roles of women are changing, they may be doing so rather slowly. Evidence suggests
that these changes may be even slower among the poorest sectors of the country,
including indigenous households and the rural population (Vzquez Garca 1997).
As for the relationship of these transitions with the incidence of spousal violence,
evidence from the two recent national surveys on domestic violence shows that female
wage earners face a 30 percent higher risk of suffering any form of intimate partner
violence (physical, sexual, or emotional abuse), as compared to women not working
for pay. This association is consistent with sociological theories of male backlash,
showing that the fact that women are employed might constitute a threat to many men,
either because these women have an income of their own, or because having a job
requires diverting time and attention outside the household, or a combination of fac-
tors of this nature (Castro, Riquer, and Medina 2006; Castro and Casique 2008).8 This
is the background against which policies and programs intended to improve womens
opportunities and economic status are being considered.
8
Aizer (2010) has highlighted the concern of unobserved heterogeneity and endogeneity biases in studies of
this nature. She exploits variation in labor market conditions for women, in contrast with their actual employment
conditions, to estimate the extent to which these affect the incidence of physical spousal abuse in the household.
9
Receipt of the education-specific benefits is contingent on children attending school at least 85 percent of the
time, which is verified by school personnel.
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 185
The variables used to construct this marginality index were: (i) the localitys population, (ii) the number of
10
dwellings in the village, (iii) the proportion of the adult population who was illiterate, (iv) the proportion of adults
working in the agricultural sector (in 1990), the proportion of households (v) without potable water, (vi) without
drainage, (vii) without electricity, (viii) with a dirt floor (in 1990), and (ix) the average number of persons per room
in each household (in 1990).
11
Within a subsample of communities, a poverty indicator was constructed using household income data col-
lected from baseline surveys. A discriminant analysis was then separately applied in each region in order to identify
the household characteristics that maximized the correct classification of as poor and non-poor (minimizing Type
I and Type II targeting errors). Eligible households were identified on the basis of this welfare index (see Skoufias
2001 for a more detailed description of the targeting process).
186 American Economic Journal: economic policy february 2013
12
This follows closely the description provided in the documentation and results of the survey in Castro, Riquer,
and Medina (2006).
13
We restrict the questions to those least likely to involve the womans perceptions such as: a partner who stops
speaking to a woman, humiliates her, has turned her relatives against her, or has made her feel fear.
14
The exact survey questions and a more detailed description of the construction of these and other variables
employed in the analysis are available in online Appendix A. We also attempted to construct measures of the
severity of violence. These indices were constructed based on the responses to the questions described above (and
presented in detail in the online Appendix). However, we believe that the quality of the severity and intensity data
is quite limited, and chose not to use these in the main analysis.
15
An important caveat is that our data consists of self-reports of spousal abuse by female women in a marital
union. In this context, it has been impossible for us to collect data on more objective reports of violence, such as
incidents of female hospitalizations for assault (as in Aizer 2010). On the other hand, the survey data we use has
the advantage of being able to capture (imperfect) measures of threats of abuse and emotional abuse that would be
very difficult to capture with administrative data. We discuss the direction of potential biases in the results section.
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 187
Physical or sexual Indicator for any occurance of physical 0.16 0.13 0.18 0.049
violence or sexual abuse (0.037)
Physical Indicator for any occurance of physical abuse 0.11 0.09 0.13 0.039
violence (e.g., push, beating, attack with blade) (0.032)
Sexual Indicator for any occurance of sexual abuse 0.09 0.08 0.10 0.017
violence (e.g., use of force to have sexual relations) (0.027)
Threat of physical Indicator for any occurance of physical abuse 0.08 0.07 0.09 0.013
violence threat (e.g., threat of leaving, threat w/deadly (0.020)
weapon, threat to kill)
Emotional Indicator for any occurance of psychological, 0.11 0.10 0.12 0.016
violence abuse excluding perceptions questions (0.022)
(e.g., locked you in, threatened to leave you)
Notes: Sample means weighted by inverse sampling weights. Standard errors clustered at the village level. Significant
differences between beneficiary and nonbeneficiary households. N = 2,867. The sample includes women ages 25
and older in rural villages with children ages 11 and younger, in a marital union since the year 1997 or earlier.
***Significant at the 1 percent level.
**Significant at the 5 percent level.
*Significant at the 10 percent level.
of Oportunidades benefits. Although there is some noise in the data (since only
ten households per village are randomly selected to participate in the survey), the
data aggregated by percentiles of the marginality index distribution are incredibly
similar to the administrative data grouped in an analogous format. The correlation
of the proportion of beneficiary households using the ENDIREH survey data with
administrative data on the number of recipient households at the locality level in
2003 is 0.84 (not reported in the tables), which suggests that the information from
the household survey closely represents receipt of Oportunidades benefits.
The summary statistics based on the survey responses indicate that spousal vio-
lence is a pervasive phenomenon in rural Mexico. Sixteen percent of women in the
sample report having experienced some form of physical or sexual violence within
the last year (Table 1). The incidence rates of physical, sexual, and emotional vio-
lence are of similar magnitude in this population, with roughly 11 percent reporting
having experienced some form of physical violence, 9 percent reporting some act
of sexual violence, and 11 percent reporting emotional abuse in the previous year.
Threats of violent behavior and physical abuse are also quite common in this context.
Eight percent of women reported receiving threats by partners within the past year
of either leaving the household (with or without their children), of being physically
abused, or of being murderedrates in line with those reported for the incidence of
physical abuse. These measures compare favorably to reported incidence of abuse in
many other developing countries, but are high relative to the incidence rates reported
in developed countries. We also compare violence incidence rates between benefi-
ciary and nonbeneficiary women, and find that physical violence incidence rates are
3.9 percentage points (30 percent) lower among the former group; these patterns
preview the studys empirical findings.
188 American Economic Journal: economic policy february 2013
Notes: Sample means weighted by inverse sampling weights. Significant differences between beneficiary and non-
beneficiary households. (The standard errors of mean differences are clustered at the village level.) N = 2,867. The
sample includes women ages 25 and older in rural villages with children ages 11 and younger, in a marital union
since the year 1997 or earlier.
***Significant at the 1 percent level.
**Significant at the 5 percent level.
*Significant at the 10 percent level.
Women in this sample come from relatively poor socioeconomic status house-
holds, since Oportunidades is targeted to poor households in marginalized rural
communities (Table 2). Approximately 8 percent of women in the sample have no
schooling, although two thirds of them (65 percent) have completed some primary
school (Table 2, panel A). A significant share of women (14 percent) is indige-
nous (based on the linguistic definition of indigenous background), which is highly
correlated with low socioeconomic status in Mexico. The womens average age is
34.9years. The reported proportion of women exposed to spousal abuse between
their parents during childhood is quite large, at approximately 10 percent. There is
considerable evidence from psychology and sociology regarding the intergenera-
tional transmission of spousal behavior. Therefore, women in this context may have
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 189
not old enough to attend secondary school. Schultz (2004) and Behrman, Sengupta,
and Todd (2005) report that school enrollment rates were close to one hundred per-
cent for primary school children among both program and comparison village chil-
dren in a randomized evaluation of the program, and, thus, that the program had
no impacts on primary school enrollment. Based on this evidence, we assume that
conditionality constraints are not likely to be binding for households with primary
school children, and assume that the take-up of the program is complete for house-
holds with only primary school-aged (or younger) children.16 This reduces the con-
cerns of endogenous program take-up based on womens decision-making power
(and tolerance for violence).
Second, we restrict the sample to women ages 25 and older (women in this sam-
ple were 19 years old and older in 1997, at the start of the initial phase-in of the pro-
gram). We do this based on the reasoning that if at some point during their lifetimes
these women lived in program villages, they would have been too old to be eligible
for program benefits as children, which could have potentially improved their socio-
economic status before marriage and possibly affected their marital outcomes.
Finally, we restrict the sample to couples who have been in union since 1997
or earlierwho made their current marital choices preceding the start of the pro-
gramin order to minimize the likelihood that the program directly affected their
marital matching patterns. These restrictions minimize potential confounding due to
marital sorting based on the tolerance for spousal abuse and other potential determi-
nants of the incidence of violence, such as the gains to marriage (e.g., Pollak 2004;
Siow 2009). These restrictions result in a sample of 2,867 households.
These restrictions are insufficient to construct comparable groups of beneficiary
and nonbeneficiary households for the empirical analysis. A comparison of means of
these predetermined characteristics among these groups of households indicate that
beneficiary households in the sample are of disproportionately low socioeconomic
status, based on their indigenous status, educational attainment levels, and their own-
ership of durables and/or access to public infrastructure (Table 2, columns24). A
joint test of significance of these differences in means rejects the null of equality at the
1 percent significance level (2(25)=213.67; p-value<0.001). This is not surpris-
ing since Oportunidades is targeted to poor households in marginalized communities.
To address the targeting of the program to these poor communities, we make
comparisons of beneficiary and nonbeneficiary households within villages in order
to remove all selection based on the village-level targeting of the program. This
within-village comparison dramatically reduces the observed selection into the pro-
gram. A within-village means comparison of the same predetermined characteris-
tics among these groups of households shows a drastic reduction inalthough not
a complete elimination ofthe observed predetermined observable characteristics
differences (Table 2, column 5). A joint test of significance of these within-village
differences in means still rejects the null of equality at the 1 percent significance level
(2(25)=195.37; p-value<0.001). We thus employ additional statistical methods
16
Overall program take-up rates throughout this period are approximately 90 percent and are expected to be
even higher for this subsample of households whose conditionality constraints are less likely to be binding (see
Gertler, Martinez, and Rubio 2012).
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 191
B. Estimation
We mainly report ordinary least squares estimates of the average treatment effect
(ATE), conditioning flexibly on a set of predetermined individual and household
socioeconomic characteristics, as well as village fixed effects, in order to capture
any village-specific unobserved heterogeneity influencing spousal abuse patterns
(e.g., access to health clinics, community groups).17 The regression equation for
outcome Yivis the following:
Yiv= Ti v+
(1) Xiv+
v +iv ,
where the treatment indicator Ti v equals one for beneficiary household i in village
v and is zero otherwise; Xiv are the predetermined covariates that are possibly sig-
nificantly correlated with Ti v and Yiv; v are village fixed effects, and i v are unob-
served determinants of domestic violence. We cluster standard errors at the village
level. Under the usual conditional unconfoundedness condition, the estimates of
are those of the ATE.18
In order to address the potential concerns of unobserved heterogeneity in the
within-village household comparison, we pursue a set of tests and sensitivity analy-
ses inspired by the work on diagnostics of selection on observable and unobservable
variables (i.e., Imbens 2003; 2004; Altonji, Elder, and Taber 2005). Essentially, we
identify which observable characteristics (Xiv) that are correlated with treatment
assignment (Ti v) are also significant predictors of spousal abuse outcomes, as these
may plausibly be the covariates most correlated with the unobservable characteris-
tics that jointly determine program eligibility/take-up and abuse outcomes.19 For
those identified variablesthe womans age, partners age, partners schooling,
family size, and years in unionwe evaluate the robustness of the results to flex-
ible specifications that allow for high-order and interaction terms between these
variables, and also include interactions with the womans levels of education.20 The
results obtained from this sensitivity analysis are qualitatively and quantitatively
similar across specifications.
17
We cannot estimate village fixed-effects probit and logit models due to the low incidence of the dependent
variables and the relatively small within-village cell sizes. However, the estimates from probit and logit models
excluding village fixed effects are qualitatively similar to those from linear probability models, and are more pre-
cisely estimated in general. Estimates are available from authors upon request.
18
The consistency of the estimate also relies on the assumption of no spillovers across households within vil-
lages. To the extent that there are positive spillovers of the program in the same direction as the direct program
impacts, as shown for consumption and educational outcomes by Angelucci and De Giorgi (2009), Bobonis and
Finan (2009), and Lalive and Cattaneo (2009), the within-village comparison would lead to an underestimate of
average program effects.
19
In this context, we cannot effectively implement nonparametric or semiparametric methods such as
propensity-score matching or case-control matching, because our village-level cell sizes are too small to have a
sufficient sample for matching.
20
See online Appendix B for details.
192 American Economic Journal: economic policy february 2013
Finally, we also present estimates from empirical models that additionally con-
dition on households asset holding patterns and access to infrastructure. There is
a trade-off in doing so: on one hand, these controls may reduce concerns of unob-
served heterogeneity due to households varying wealth levels which may influence,
for instance, the opportunity costs of partners to engage in spousal abuse or the like-
lihood of separation. However, because these variables are measured at the time of
the survey, they are potentially endogenous regressors since households may choose
to make household improvements or purchase assets from the program-related cash
benefits (Gertler, Martinez, and Rubio-Codina 2012). The results are robust to the
inclusion and exclusion of these additional control variables.
Estimates of the average treatment effects of the program on physical and sexual
spousal violence, as well as emotional violence and threats outcomes, are displayed
in Table 3. Column 1 reports the estimated effects based on a specification where
control variables enter linearly. The following columns report specifications in
which we gradually include (i) interaction terms between the females educational
attainment and the partners age, the partners schooling attainment, household size,
and the couples years in union (column 2); (ii) polynomial terms for each partners
age, the partners education, household size, and years in union (column 3); and
(iii)the additional interaction of the higher-order terms (column 4). In column 5, we
present estimates with additional asset-holdings/infrastructure variables. Finally, in
column 6 we report OLS estimates excluding village fixed effects for comparison
purposes. Since the fully saturated models (column 4) are more likely to reduce or
eliminate potential biases, we consider these our preferred estimates.
Domestic violence incidence rates are lower among beneficiary households than
among nonbeneficiary households; these effects are concentrated in reductions in
partners use of physical abuse. The estimated reduction in the incidence of either
physical or sexual violence varies among specifications, ranging between 8.0 and 10.5
percentage points (44 and 57 percent) (Table 3, row 1). The preferred specification
(Table 3, column 4) implies a reduction in the incidence of physical or sexual violence
by 8.2 percentage points (45 percent; significant at the 90 percent confidence level).
We find large and statistically significant reductions of 5.17.0 percentage points
(4055 percent) in the incidence of physical abuse (Table 3, row 2). Our preferred
estimate shows a reduction of 5.5 percentage points, or 43 percent (significant at the
90 percent confidence). The estimated reductions in sexual abuse are of similar mag-
nitude to those of physical abuse, although the estimated effects are less precisely
estimated. The point estimate from our preferred specification implies a reduction
in sexual abuse of 5.0 percentage points (51 percent), though imprecisely estimated.
To highlight a point on the direction of potential bias, note that the estimated
effects on both physical and sexual violence are of larger magnitude in absolute
terms than both the across-village OLS estimates (column 6) and the comparison
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 193
Mean of
dependent
Coefficient estimate on beneficiary status [1/0] (s.e.) variable
Dependent variables (1) (2) (3) (4) (5) (6) (7)
Physical or sexual violence 0.096** 0.105** 0.088* 0.082* 0.080* 0.052 0.184
(0.047) (0.050) (0.050) (0.044) (0.041) (0.033)
Physical violence 0.062* 0.070** 0.051 0.055* 0.052* 0.051* 0.127
(0.032) (0.033) (0.033) (0.031) (0.030) (0.030)
Sexual violence 0.066* 0.073* 0.056 0.050 0.046 0.004 0.098
(0.039) (0.041) (0.037) (0.035) (0.033) (0.020)
Threat of violence 0.025 0.018 0.035 0.018 0.016 0.000 0.086
(0.037) (0.024) (0.037) (0.023) (0.022) (0.019)
Emotional violence 0.041 0.031 0.045 0.027 0.017 0.005 0.121
(0.040) (0.030) (0.042) (0.032) (0.029) (0.021)
Village fixed effects Yes Yes Yes Yes Yes No
Individual and household Yes Yes Yes Yes Yes Yes
controls (linear)
Individual control interaction No Yes No Yes Yes Yes
terms
Individual control polynomial No No Yes Yes Yes Yes
terms
Asset controls No No No No Yes No
Observations 2,867 2,867 2,867 2,867 2,867 2,867
Notes: Each reported coefficient is from a different regression. Robust standard errors in parentheses, clustered at
the village level. Coefficient estimates from village fixed effects and/or OLS regressions weighted by survey sam-
pling weights. Controls include indicator variables for woman and partners age, woman and partners indigenous
status, womans schooling-level indicators, the partners schooling attainment level, household size, cohabiting
couple indicator, years in union, and variables measuring reported histories of spousal abuse in parental household
during childhood. For details on polynomial and interaction terms, see Section IV.
***Significant at the 1 percent level.
**Significant at the 5 percent level.
*Significant at the 10 percent level.
of raw differences (see Table 1). This suggests that, under our identifying assump-
tions, cross-village unobserved heterogeneity leads to an underestimate of the true
program effects.21
In contrast to the results for physical abuse, we do not find evidence of program
impacts on the incidence of emotional violence or threats of violence. The point
estimates suggest moderate increases of 1.6 to 3.5 percentage points in the incidence
of threats of abuse and of 1.7 to 4.5 percentage points in the incidence of emotional
abuse, but none are statistically distinguishable from zero (Table3, rows 4 and 5).
Our preferred point estimates suggest a moderate increase of 1.8 percentage points
21
We can also characterize the bias in our program impact estimates if there is systematic underreporting of
physical abuse, under certain conditions. Suppose the only measurement error problem is one of underreport-
ing (i.e., there are no cases of reports of violence when in reality these do not occur) and the treatment induces
less underreporting of any form of domestic violence, or: Pr(reportic =1|Tic =1, Xic , ic , violenceic =1)>
Pr(reportic =1|Tic =0, Xic , ic , violenceic =1), then since our estimated effects are negative, the lesser under-
reporting should lead to an underestimate (in absolute terms) of the estimated impact on physical/sexual violence.
194 American Economic Journal: economic policy february 2013
and 2.7percentage points, respectively. Note again that the within-village com-
parison estimated effects are greater, suggesting that the cross-village unobserved
heterogeneity leads to an underestimate of the true program impacts.22
The results indicate substantial reductions in physical abusebut not in other
forms of male-to-female spousal abuseamong beneficiary households. Moreover,
the estimates of emotional violence (weakly) suggest that the incidence of the pat-
tern of verbal non-physical abuse in the household may be increasing. We would
like to further examine whether there is some degree of substitution in the male part-
ners use of physical and emotional abuse, as predicted by theories of instrumental
violence. This is examined in more detail in the following section.
The findings provide prima facie evidence against traditional cooperative and
noncooperative theories of the family, which generally predict that these transfer pro-
grams, by improving womens bargaining position in the household, will decrease
the incidence of spousal abuse, both physical and emotional.23 One possible alter-
native explanation is that male partners may use threats of violence as instruments
of coercion to appropriate or (re)gain control over household resources or decision-
making (i.e., extract rents) from their female counterparts. In this model, a rein-
terpretation of Bloch and Raos (2002) theory of domestic violence, male partners
have private information regarding the gains to marriage, such as their own private
income or their status within the household based on traditional gender roles.24 Due
to this private information, they can use threats of abuse to coercively demand trans-
fers from their wives, and use violence strategically depending on whether their wife
abides to the act of coercion. The model has a main testable implication:
22
A caveat to this result is that if the treatment induces less underreporting of any form of domestic violence,
this should lead to a positive bias in the estimated impacts.
23
Economic theories of spousal violence focus on noncooperative bargaining models in which female partners
incomes, and financial resources for women outside the marriage more generally, influence the womans threat point
and thus reduce the level of violence in equilibrium (Tauchen, Witte, and Long 1991; Farmer and Tiefenthaler 1996;
1997).
24
Bloch and Raos (2002) model is one of spousal violence as a bargaining instrument used by male partners
to extract larger dowry payments from the brides family. The bargaining takes place between the families of the
bride and groom, rather than within the couple itself.
25
This prediction is consistent with sociological theories of male backlash. These predict that as womens
economic opportunities improve, violence against them may increase because men feel their traditional gender role
threatened (e.g., Macmillan and Gartner 1999).
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 195
The model predicts opposing effects for the incidence of threats of spousal abuse
without actual physical abuse, and the incidence of physical abuse. This result is
driven by the fact that an increase in the transfer demanded by male partners will
lead to a change in the threshold value of the husbands cost of using threats, increas-
ing the proportion of individuals who strictly prefer to exercise threats of violence
and extract resources from their partners, as opposed to using physical violence. A
concise exposition of the model and of the general prediction is available in online
Appendix C.
Since the theory predicts that the transfers may lead to an increase in the inci-
dence of threats of spousal abuse with no associated physical violence, in order to
implement these tests we need to construct measures that capture this phenomenon.
Therefore, we use two sets of additional violence measures: (i) indicator variables
for the incidence of threats of spousal abuse and no incidence of physical (or sexual)
violence; and (ii) indicator variables for the incidence of emotional violence and no
incidence of physical (or sexual) violence.26 Moreover, to examine heterogeneous
responses, we employ a measure of the female partners decision-making power
within the household and another of the households expected gains to marriage. The
former is an index measure created by Casique (2006) and reported in Castro, Riquer,
and Medina (2006) that aggregates the female partners decision-making power based
on responses to 13 questions regarding which partner(s) contribute to the households
decisions in areas such as: the female spouses labor force p articipation decision; the
households expenditure decisions; decisions regarding when to have sexual rela-
tions; the use of contraceptives; and fertility decisions.27 The latter is the product of
a secondary school completion indicator for the female partner and the educational
attainment levels of the male partner, as partners educational attainment are comple-
mentary in households marital output (e.g., Becker 1991; Siow 2009).
To test Prediction 1, we estimate empirical models analogous to those in equa-
tion(1), using our theory-driven threats and emotional violence outcomes as depen-
dent variables. We report estimates from our preferred specifications with fully
saturated controls (Table 4). According to our estimates, the program led to substan-
tial increases in the incidence of violent threats or acts of emotional violence with no
associated acts of physical or sexual abuse, in some cases more than doubling the inci-
dence of these acts. Threats of violent behavior with no physical abuse increased by
approximately 3.0 to 3.1 percentage points, depending on the specification (Table 4,
row 1). The results are robust to expanding the definition of physical abuse to include
26
The assumption that the level of spousal abuse that triggers the noncooperative outcome is physical abuse
rather than forms of emotional abuse such as forcible confinementgenerates the result. It is not clear that some
dimensions of emotional violence (e.g., forcible confinement, threats of murder) are at a lower bar than acts of
physical violence. We address this empirically by modifying the definition that we use to construct the measure
of emotional violence, including the responses to the questions on whether (i) the male partner stops speaking to
a woman; (ii) humiliates her; (iii) has turned her relatives against her; or (iv) has made her feel fear. Using these
alternative measures of low intensity emotional violence suggest the same result.
27
For details, see online Appendix A.
196 American Economic Journal: economic policy february 2013
Notes: Each reported coefficient is from a different estimator. Robust standard errors in parentheses, clustered at
the village level. Coefficient estimates from village fixed effects regressions weighted by survey sampling weights.
Controls include indicator variables for woman and partners age, woman and partners indigenous status, womans
schooling-level indicators, the partners schooling attainment level, household size, cohabiting couple indicator,
years in union, and variables measuring reported histories of spousal abuse in parental household during childhood.
For details on polynomial and interaction terms, see Section IV. The sample sizes in specifications conditioning on
no episode of physical violence or no episode of physical or sexual abuse are 2,611 and 2,479, respectively.
***Significant at the 1 percent level.
**Significant at the 5 percent level.
*Significant at the 10 percent level.
s exual violence, with estimates ranging from 2.8 to 2.9 percentage points (statistically
significant at 95 percent confidence; Table 4, row 2). The similarity in the estimated
impacts is expected, since the violent threat measure includes threats of sexual abuse.
We also find evidence of a substantial increase in the incidence of emotional violence
associated with a lack of physical violence (Table 4, row 3). We estimate increases
of 3.94.4percentage points (8192 percent) (significant at conventional confidence
levels). These results are also robust to taking into account the lack of incidence of
sexual abuse (Table 4, row 4). The point estimates are in the 5.05.2 percentage points
range and are very precisely estimated. The estimated effects across dependent vari-
ables are jointly significantly different from zero at conventional confidence levels
(p-values: 0.025 and 0.031). That said, we get somewhat smaller point estimates in
specifications where we use as dependent variables the incidence of threats or of emo-
tional abuse conditional on the lack of physical/sexual abuse (reducing the sample
size by 9 to 14 percent)and lose some precision (columns 34).28
28
It is not possible to characterize the direction of bias of these estimates as a result of measurement error
due to underreporting (without making additional assumptions). To see this, decompose the dependent variable
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 197
Notes: Robust standard errors in parentheses, clustered at the village level. Coefficient estimates from village fixed
effects regressions weighted by survey sampling weights. Controls include indicator variables for woman and part-
ners age, woman and partners indigenous status, womans schooling-level indicators, the partners schooling
attainment level, household size, cohabiting couple indicator, years in union, and variables measuring reported his-
tories of spousal abuse in parental household during childhood. For details on polynomial and interaction terms,
see Section IV.
***Significant at the 1 percent level.
**Significant at the 5 percent level.
*Significant at the 10 percent level.
We also estimate subgroup average program impacts for the main variables of inter-
est for the couples in which female partners are categorized as having low or moderate/
high decision-making power, respectively (Table 5). Although we have large differ-
ences according to our point estimates in both physical and sexual violence incidence
rates, we do not have sufficient precision to say that these are lower among beneficiary
households than among nonbeneficiary households in both groups of households
(Table 5, rows 12). In contrast, our theory-based measures suggest that the sub-
stitution of physical/sexual violence for threats of abuse and emotional violence is
concentrated among households in which women have a moderate degree of decision-
making power (Table 5, rows 36, columns 45). There is no evidence of substitution
in the following manner: Pr(report threat=1 and report violenceic=0|Tc )=Pr(reportic =1|Tic , threatic =1)
Pr(threatic =1|Tic )Pr(reportic =0|Tic , report threatic =1, violenceic=0)Pr(violenceic =0|Tic , report threatic =1).
The direction of bias additionally depends on the last two terms on the right-hand side of the equation (which
conditions on the incidence of reporting the threat of abuse, one of the dependent variables measured with error).
198 American Economic Journal: economic policy february 2013
among the subsample of couples where women have very low d ecision-making
power (Table 5, rows 36, columns 12), although we cannot statistically distin-
guish the effects for this group relative to the moderate decision-making power sub-
group. Again, tests based on the conditional sample give similar point estimates with
a lower degree of precision (Table 5, columns 3 and 6).
To examine whether the effects vary across households with varying expected
gains to marriage, we estimate models similar to (1) where we allow the treatment
to vary (linearly) with the expected gains to marriage (i.e., with the product of part-
ners educational attainment levels), of the following form:
Yiv =1Ti v+
(2) 2Ti v GMi v+
1 GMi v+
Xiv2+
v +iv ,
where GMivdenotes our measure of the expected gains to marriage (the product of
partners educational outcomes). The Xiv term includes flexible controls for each
partners educational attainment, thus capturing some of the potential heterogeneity
due to individuals endowments and preferences.29 The heterogeneous effects esti-
mates from these models are reported in Table 6, columns 1 and 2.
We find large reductions in the incidence of physical or sexual violence among
couples with low expected gains to marriage, in the order of 5.6 to 8.3 percentage
points, and no statistically significant evidence of a reduction in the magnitude of the
effect among couples with higher expected gains to marriage (Table 6, rows1 and 2).
In contrast, our models on the substitution effects provide strong evidence consistent
with the theoretical predictions. We find substantial increases in the incidence of
threats of violence associated with no physical (or sexual) abuse of 3.3to 3.6per-
centage points among couples with very low gains to marriage, effects that decrease
in magnitude for couples with greater expected gains to marriage (Table6, rows3
and 4). The estimates for the substitution between emotional and physical/sexual
violence provide analogous results: a substantial increase of 4.8 to 5.6 percentage
points among couples with low expected gains to marriage, and effects that decrease
in magnitude for couples with greater expected gains (Table6, rows5 and 6).30
Finally, we examine heterogeneous responses by couples type of marital
union/arrangement. We estimate subgroup average program impacts for the main
variables of interest for the couples reporting being married or cohabiting separately
(including an interaction term of the cohabitation indicator with the couples pro-
gram beneficiary status) (Table 6, columns 34). The point estimates suggest that
the effects vary significantly across marital and consensual unions, with some of
the point estimates being significantly larger (in absolute value) for the latter group,
although not consistently across all measures. The estimates are also not statistically
significant across subgroups.31
29
A note of caution is warranted: the models exploiting heterogeneous effects could be subject to concerns of
unobserved heterogeneity in the preferences or endowments of these households, which could place some doubt on
our interpretation of the findings.
30
The effects are quantitatively similar in specifications including additional asset controls (available from
authors).
31
Estimates with subgroup-specific regression models suggest even larger differences in program impacts esti-
mates across subgroup, although (again) these are imprecisely estimated (available from authors).
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 199
Notes: Robust standard errors in parentheses, clustered at the village level. Coefficient estimates from village fixed
effects regressions weighted by survey sampling weights. Controls include indicator variables for woman and part-
ners age, woman and partners indigenous status, womans schooling-level indicators, the partners schooling
attainment level, household size, cohabiting couple indicator, years in union, and variables measuring reported his-
tories of spousal abuse in parental household during childhood. For details on polynomial and interaction terms,
see Section IV.
***Significant at the 1 percent level.
**Significant at the 5 percent level.
*Significant at the 10 percent level.
This is not surprising given two important methodological issues with the esti-
mate of these subgroup-specific effects. First, the sample of cohabiting couples is
quite smallthere are 524 cohabiting coupleslimiting our ability to get sufficient
precision for these subgroups effects estimates. Also, a comparison of predetermined
characteristics across beneficiary and nonbeneficiary households among cohabiting
couples suggests that these are quite dissimilar, even within villages.32
In summary, although we ultimately cannot rule out some degree of selection
and omitted variable biases in our estimates, we take comfort in the fact that the
theoretical predictions and our set of results suggest a clear symmetry between the
incidence of physical and sexual abuse on one hand and the incidence of emotional
32
See Table A2 in online Appendix A. Although the differences among cohabiting couples are not precisely esti-
mated (partly due to the small sample sizes), beneficiary households seem substantially more negatively selected
than nonbeneficiary ones. As a result, it is not surprising that the point estimates for cohabiting couples are possibly
overestimated.
200 American Economic Journal: economic policy february 2013
abuse and threats of violence on the othera symmetry that is difficult to reconcile
with monotonic selection bias arguments. Furthermore, we address other potential
concerns of unobserved heterogeneity in Section VC.
The asymmetric information story for the instrumental use of violence is not
the only possible explanation for these results. For instance, the program increases
the level of household resources, not just the socioeconomic opportunities available
to (or more specifically, the income of) adult women. If this increase in house-
hold resources reduced stress levels among partners and/or improved the emotional
health of adults (or children) in the family, it could lead to a reduction in the inci-
dence or severity of spousal abuse. Another alternative explanation is that the pro-
gram conditions generate conflict and greater stress within the household. While
our results on physical and emotional violence are consistent with the instrumental
violence hypothesis, in isolation they do not rule out the possibility of these alterna-
tive hypotheses.
Although we do not have measures of the emotional well-being or mental health
of adult household members, we can assess whether there is some reduction in other
measures of intrahousehold conflict, by examining whether there is a reduction in
the level of violence against children perpetrated by each parent. The measures of
incidence of violence against children consist of various dichotomous variables
indicating whether either the male or the female partner (or each exclusively) beat
or insult the child as a result of him or her behaving badly.33
The resulting estimates suggest no impact of the program on the incidence of
physical violence (i.e., beating) of any parent against children (Table 7, panel A).
The point estimates suggest a small 1.6 percentage points increase in the incidence of
abuse, but the results are statistically indistinguishable from zero (Table 7, panelA,
column 1). In a specification where we allow for the effects to vary by couples
expected gains to marriage, we do not observe heterogeneous responses based on
this dimension (Table 7, panel A, column 2).34 Additionally, although we observe
very moderate increases in violence against children of approximately 2.6 points by
female partners (statistically insignificant; Table 7, panel A, column 3), the effects
are estimated to be essentially zero for male partners (Table 7, column 5). These
estimates do not vary significantly by couples expected gains to marriage (Table 7,
columns 4 and 6).35
The analogous estimates for the incidence of insults towards children suggest
no distinct evidence of reductions in violence, overall, and for female partners
(Table 7, panelB, columns 1 and 3). The point estimates do suggest that the male
partners insults towards children decrease by 4.2 percentage points (statistically
The exact questions are: Do you (does your husband/partner) beat your children when they behave badly?
33
(Yes, No).
34
We thank a referee for suggesting this test.
The effects are quantitatively similar in specifications including additional asset controls (available from
35
authors).
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 201
Notes: Robust standard errors in parentheses, clustered at the village level. Coefficient estimates from village fixed
effects regressions weighted by survey sampling weights. Controls include indicator variables for woman and part-
ners age, woman and partners indigenous status, womans schooling-level indicators, the partners schooling
attainment level, household size, cohabiting couple indicator, years in union, and variables measuring reported his-
tories of spousal abuse in parental household during childhood. For details on polynomial and interaction terms,
see Section IV.
***Significant at the 1 percent level.
**Significant at the 5 percent level.
*Significant at the 10 percent level.
insignificant; Table 7, panelB, column 5). Again, the estimates do not vary sig-
nificantly by couples expected gains to marriage. Most relevant for our argument,
among male partners in relationships with minimal gains to marriage, we find a sta-
tistically insignificant reduction in the incidence of insults by 3.7 percentage points
(Table 7, column 6). That said, the precision of all these estimates is not sufficiently
strong to reject moderately important increases or decreases in the incidence of vio-
lence against children.
The program impact estimates could also suffer from bias if Oportunidades caused
higher marital dissolution rates, and beneficiary couples who remained in a marital
union had differing unobserved determinants of spousal abuse than nonbeneficiary
couples. As a result of this attrition bias, being a beneficiary could potentially be corre-
lated with unobservable characteristics of couples who chose to remain in a relation-
ship. To the extent that unions in which women would be more likely to suffer from
abuse are the least [most] likely to dissolve as a result of the program, it would lead
to a downward [upward] bias (in absolute terms) in the estimates of physical abuse
and an upward [downward] bias in the emotional abuse effects. Unfortunately, we
do not have the necessary longitudinal or recall data among these women to estimate
these divorce-driven attrition rates for this sample. However, Bobonis (2009, 2011)
202 American Economic Journal: economic policy february 2013
VI.Conclusion
The main objective of this paper is to provide evidence of the effect of the
Oportunidades conditional cash transfer program on the prevalence of male-to-
female spousal violence in rural Mexico. The evidence suggests that women in
beneficiary households are approximately 40 percent less likely to be victims of
physical abuse than nonbeneficiary women, impacts that may come as a conse-
quence of an increase in womens empowerment within the household. However,
women in beneficiary households are as likely as nonbeneficiary women to receive
threats of violent behavior and be victims of emotional abuse, and substantially more
likely to receive threats of abuse with no associated physical abuse, than women in
nonbeneficiary households. The results may have important implications for policy,
since they provide a mixed view of conditional cash transfer programs effective-
ness in improving womens empowerment within the household. The program may
increase the likelihood of violent threats, which may in turn compromise womens
emotional health and other aspects of their well-being.
As a second contribution, we show how these results are consistent with a model
of asymmetric information in household bargaining, enriching the existing work
studying the relationship between womens income and economic opportunities,
the extent of spousal abuse, and the degree of female empowerment. Specifically,
the study contributes to the empirical work opening the black box of intrahousehold
decision making. For instance, as presented in our bargaining framework, infor-
mation asymmetries may lead to distinct equilibria across households: nonviolent
households achieve Pareto optimal allocation decisions whereas violent ones suffer
from the destruction of a share of the gains to marriage. Considering this hetero-
geneity may help inform future empirical analyses regarding the consequences of
CCTs and other social policies on intrahousehold resource allocation decisions (het-
erogeneity which is generally ignored in this empirical literaturesee for instance
Attanasio and Lechne 2002; Rubalcava, Teruel, and Thomas 2009; and Bobonis
2009) as well as empirical work examining the Pareto efficiency of intrahousehold
allocations (e.g., Browning 1994 and Udry 1996).
36
These are available from the authors upon request.
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 203
References
Aizer, Anna. 2010. The Gender Wage Gap and Domestic Violence. American Economic Review 100
(4): 184759.
Akresh, Richard. 2006. (In)Efficiency in Intra-Household Allocations. Unpublished.
Anderson, Kristin L. 1997. Gender, Status, and Domestic Violence: An Integration of Feminist and
Family Violence Approaches. Journal of Marriage and the Family 59 (3): 65569.
Angelucci, Manuela. 2008. Love on the Rocks: Domestic Violence and Alcohol Abuse in Rural Mex-
ico. B. E. Journal of Economic Analysis and Policy: Contributions to Economic Analysis and Pol-
icy 8 (1): Article 43.
Angelucci, Manuela, and Giacomo De Giorgi. 2009. Indirect Effects of an Aid Program: How Do
Cash Transfers Affect Ineligibles Consumption? American Economic Review 99 (1): 486508.
Ariza, Marina, and Orlandina de Oliveira. 2001. Familias en Transicin y Marcos Conceptuales en
Redefinicin. Papeles de Poblacin 28 (46): 939.
Ashraf, Nava. 2009. Spousal Control and Intra-Household Decision Making: An Experimental Study
in the Philippines. American Economic Review 99 (4): 124577.
Attanasio, Orazio, and Valrie Lechene. 2002. Tests of Income Pooling in Household Decisions.
Review of Economic Dynamics 5 (4): 72048.
Basu, Kaushik. 2006. Gender and Say: A Model of Household Behavior with Endogenously Deter-
mined Balance of Power. Economic Journal 116 (511): 55880.
Becker, Gary S. 1991. A Treatise on the Family. Cambridge, MA: Harvard University Press.
Behrman, Jere R., Piyali Sengupta, and Petra Todd. 2005. Progressing through PROGRESA: An
Impact Assessment of a School Subsidy Experiment in Rural Mexico. Economic Development and
Cultural Change 54 (1): 23775.
Benera, Lourdes, and Martha Roldn. 1987. The Crossroads of Class and Gender: Industrial Home-
work, Subcontracting, and Household Dynamics in Mexico City. Chicago: University of Chicago Press.
Bloch, Francis, and Viyajendra Rao. 2002. Terror as a Bargaining Instrument: A Case Study of Dowry
Violence in Rural India. American Economic Review 92 (4): 102943.
Bobonis, Gustavo J. 2009. Is the Allocation of Resources within the Household Efficient? New Evi-
dence from a Randomized Experiment. Journal of Political Economy 117 (3): 453503.
Bobonis, Gustavo J. 2011. The Impact of Conditional Cash Transfers on Marriage and Divorce. Eco-
nomic Development and Cultural Change 59 (2): 281312.
Bobonis, Gustavo J., and Frederico Finan. 2009. Neighborhood Peer Effects in Secondary School
Enrollment Decisions. Review of Economics and Statistics 91 (4): 695716.
Browning, Martin, Franois Bourguignon, Pierre-Andr Chiappori and Valrie Lechene. 1994.
Income and Outcomes: A Structural Model of Intrahousehold Allocation. Journal of Political
Economy 102 (6): 106796.
Browning, Martin, and Pierre-Andr Chiappori. 1998. Efficient Intrahousehold Allocation: A Gen-
eral Characterization and Empirical Tests. Econometrica 66 (6): 124178.
Casique, Irene. 2001. Power, Autonomy, and Division of Labor in Mexican Dual-Earners Families.
Lanham, MD: University Press of America.
Casique, Irene. 2006. Indices de empoderamiento femenino y su relacin con la violencia de gnero.
In Violencia de gnero en las parejas mexicanas. Resultados de la Encuesta Nacional sobre la
Dinmica de las Relaciones en los Hogares 2003, edited by Castro Roberto, Florinda Riquer, and
Mara Eugenia Medina, 75107. Cuernavaca: CRIM-UNAM, Inmujeres.
Castro, Roberto. 2004. Violencia Contra Mujeres Embarazadas: Tres Estudios Sociolgicos. Cuer-
navaca: Centro Regional de Investigaciones Multidisciplinarias.
Castro, Roberto, and Irene Casique, eds. 2008. Violencia de Gnero en las Parejas Mexicanas. Anli-
sis de Resultados de la Encuesta Nacional Sobre la Dinmica de las Relaciones en los Hogares
2006. Cuernavaca: CRIM-UNAM, Inmujeres.
Castro, Roberto, Florinda Riquer, and Mara Eugenia Medina, eds. 2006. Violencia de Gnero en las
Parejas Mexicanas: Resultados de la Encuesta Nacional sobre la Dinmica de las Relaciones en
los Hogares 2003. Cuernavaca: CRIM-UNAM, Inmujeres.
Chiappori, Pierre-Andr, Bernard Fortin, and Guy Lacroix. 2002. Marriage Market, Divorce Legis-
lation, and Household Labor Supply. Journal of Political Economy 110 (1): 3772.
Chias, Beverly N. 1992. The Isthmus Zapotecs: A Matrifocal Culture of Mexico. Fort Worth: Harcourt
Brace Jovanovich.
Dercon, Stefan, and Pramila Krishnan. 2000. In Sickness and in Health: Risk Sharing within House-
holds in Rural Ethiopia. Journal of Political Economy 108 (4): 688727.
204 American Economic Journal: economic policy february 2013
Djebbari, Habiba. 2005. The Impact on Nutrition of the Intra-Household Distribution of Power.
Institute for the Study of Labor (IZA) Discussion Paper 1701.
Duflo, Esther. 2005. Gender Equality in Development. http://economics.mit.edu/files/799.
Duflo, Esther, and Christopher Udry. 2004. Intra-Household Resource Allocation in Cte DIvoire:
Social Norms, Separate Accounts, and Consumption Choices. Unpublished.
Elmendorf, Mary. 1972. The Mayan Woman and Change. PhD diss. Union Graduate School.
Farmer, Amy, and Jill Tiefenthaler. 1996. Domestic Violence: The Value of Services as Signals.
American Economic Review 86 (2): 27479.
Farmer, Amy, and Jill Tiefenthaler. 1997. An Economic Analysis of Domestic Violence. Review of
Social Economy 55 (3): 33758.
Felson, Richard B., and Steven F. Messner. 2000. The Control Motive in Intimate Partner Violence.
Social Psychology Quarterly 63 (1): 8694.
Follingstad, Diane R., and Dana D. Dehart. 2000. Defining Psychological Abuse of Husband Toward
Wives: Contexts, Behaviors, and Typologies. Journal of Interpersonal Violence 15 (9): 891920.
Garca, Brgida, and Orlandina de Oliveira. 2006. Las familias en el Mxico metropolitano: visiones
femeninas y masculinas. Mexico City: El Colegio de Mxico.
Gertler, Paul J., Sebastian W. Martinez, and Marta Rubio-Codina. 2012. Investing Cash Transfers to
Raise Long-Term Living Standards. American Economic Journal: Applied Economics 4 (1): 16492.
Gonzlez-Brenes, Melissa. 2005. Domestic Violence and Household Decision-Making: Evidence
from East Africa. Unpublished.
Horowitz, Joel L., and Charles F. Manski. 2000. Nonparametric Analysis of Randomized Experi-
ments with Missing Covariate and Outcome Data. Journal of the American Statistical Association
95 (449): 7784.
Imbens, Guido W. 2003. Sensitivity to Exogeneity Assumptions in Program Evaluation. American
Economic Review 93 (2): 12632.
Imbens, Guido W. 2004. Nonparametric Estimation of Average Treatment Effects under Exogeneity:
A Review. Review of Economics and Statistics 86 (1): 429.
Kimerling, Rachel, and Nikki Baumrind. 2004. Intimate Partner Violence and Use of Welfare Ser-
vices among Californian Women. Journal of Sociology and Social Welfare 31 (4): 16176.
Kishor, Sunita, and Kiersten Johnson. 2004. Profiling Domestic Violence: A Multi-Country Study. Cal-
verton, MD: Opinion Research Corporation Macro.
Lalive, Rafael, and M. Alejandra Cattaneo. 2009. Social Interactions and Schooling Decisions.
Review of Economics and Statistics 91 (3): 45777.
Ligon, Ethan. 2002. Dynamic Bargaining in Households (With an Application to Bangladesh).
http://escholarship.org/uc/item/1t52k4c5#page-1.
Lundberg, Shelly, and Robert A. Pollak. 1993. Separate Spheres Bargaining and the Marriage Mar-
ket. Journal of Political Economy 101 (6): 9881010.
Lundberg, Shelly, and Robert A. Pollak. 2003. Efficiency in Marriage. Review of Economics of the
Household 1 (3): 15367.
Macmillan, Ross, and Rosemary Gartner. 1999. When She Brings Home the Bacon: Labor Force
Participation and the Risk of Spousal Violence Against Women. Journal of Marriage and the
Family 61 (4): 94758.
Maluccio, John A., and Rafael Flores. 2004. Impact Evaluation of a Conditional Cash Transfer Pro-
gram: The Nicaraguan Red de Proteccin Social. International Food Policy Research Institute
Food Consumption and Nutrition Division Discussion Paper 184.
McCloskey, Laura Ann. 1996. Socioeconomic and Coercive Power within the Family. Gender and
Society 10 (4): 44963.
Oliveira, Orlandina de, and Marina Ariza. 1997. Divisin sexual del trabajo y exclusin social.
Revista Latinoamericana de Estudios del Trabajo 3 (5): 183202.
Panda, Pradeep, and Bina Agarwal. 2005. Marital Violence, Human Development, and Womens
Property Status in India. World Development 33 (5): 82350.
Pollak, Robert A. 2004. An Intergenerational Model of Domestic Violence. Journal of Population
Economics 17 (2): 31129.
Rangel, Marcos, and Duncan Thomas. 2005. Out of West Africa: Evidence on the Efficient Alloca-
tion of Resources within Farm Households. University of Chicago Harris School of Public Policy
Working Paper 05.15.
Rasul, Imran. 2008. Household Bargaining over Fertility: Theory and Evidence from Malaysia.
Journal of Development Economics 86 (2): 21541.
Rawlings, Laura B., and Gloria M. Rubio. 2003. Evaluating the Impact of Conditional Cash Trans-
fer Programs: Lessons from Latin America. World Bank Policy Research Working Paper 3119.
Vol. 5 No. 1 bobonis et al.: public transfers and domestic violence 205
Rivera, Leonor, Bernardo Hernndez, and Roberto Castro. 2006. Asociacin entre la violencia de
pareja contra las mujeres de las zonas urbanas en pobreza extrema y la incorporacin al Programa
Oportunidades. In El Programa Oportunidades Examinado Desde el Gnero, edited by Maria de
la Paz Lpez and Vania Salles, 6995. Mexico City: Colegio de Mxico.
Roldn, Martha. 1987. Class, Gender, and Asymmetrical Exchanges within Households. In The
Crossroads of Class and Gender: Industrial Homework, Subcontracting, and Household Dynam-
ics in Mexico City, edited by Lourdes Benera and Martha Roldn, 75103. Chicago: University of
Chicago Press.
Rosenbaum, Paul, and Ronald D. Rubin. 1983. The Central Role of the Propensity Score in Observa-
tional Studies for Causal Effects. Biometrika 70 (1): 4155.
Rubalcava, Luis, Graciela Teruel, and Duncan Thomas. 2009. Investments, Time Preferences, and
Public Transfers Paid to Women. Economic Development and Cultural Change 57 (3): 50738.
Rubin, Donald B. 1974. Estimating Causal Effects of Treatments in Randomized and Non-Random-
ized Studies. Journal of Educational Psychology 66 (5): 688701.
Rubin, Donald B. 1978. Bayesian Inference for Causal Effects: The Role of Randomization. Annals
of Statistics 6 (1): 3458.
Schmukler, Beatriz, ed. 1998. Familias y Relaciones de Gnero en Transformacin. Mxico: Edamex.
Schultz, T. Paul. 2004. School Subsidies for the Poor: Evaluating the Mexican Progresa Poverty Pro-
gram. Journal of Development Economics 74 (1): 199250.
Sen, Amartya. 1999. Development as Freedom. Oxford, UK: Oxford University Press.
Siow, Aloysius. 2009. Testing Beckers Theory of Positive Assortative Matching. University of
Toronto Working Paper 356.
Skoufias, Emmanuel. 2001. PROGRESA and Its Impacts on the Human Capital and Welfare of
Households in Rural Mexico: A Synthesis of the Results of an Evaluation by IFPRI. International
Food Policy Research Institute. Washington, DC, December.
Skoufias, Emmanuel, Benjamin Davis, and Sergio de la Vega. 2001. Targeting the Poor in Mexico:
An Evaluation of the Selection of Households into PROGRESA. World Development 29 (10):
176984.
Stevenson, Betsey, and Justin Wolfers. 2006. Bargaining in the Shadow of the Law: Divorce Laws and
Family Distress. Quarterly Journal of Economics 121 (1): 26788.
Strauss, Murray A., and Richard J. Gelles. 1990. Physical Violence in American Families: Risk Fac-
tors and Adaptations to Violence in 8,145 Families. New Brunswick: Transaction Publishers.
Tauchen, Helen V., Ann Dryden Witte, and Sharon K. Long. 1991. Domestic Violence: A Nonrandom
Affair. International Economic Review 32 (2): 491511.
Thomas, Duncan. 1990. Intra-Household Resource Allocation: An Inferential Approach. Journal of
Human Resources 25 (4): 63564.
Udry, Christopher. 1996. Gender, Agricultural Production, and the Theory of the Household. Jour-
nal of Political Economy 104 (5): 101046.
Vzquez Garca, Vernica. 1997. Mujeres que respetan su casa: Estatus marital de las mujeres
y economa domstica en una comunidad nahua del sur de Veracruz. In Familias y mujeres en
Mxico, edited by Soledad Gonzlez Montes and Julia Tun, 16394. Mexico City: El Colegio
de Mxico.
Whitehead, Ann. 1981. Im Hungry Mum: The Politics of Domestic Budgeting. In Of Marriage
and the Market, edited by Kate Young, Carol Wolkowitz, and Roslyn McCullagh, 88111. Lon-
don: CSE Books.
Wolf, Eric. 1959. Sons of the Shaking Earth. Chicago: University of Chicago Press.
World Bank. 2003. Gender Equality and the Millennium Development Goals. World Bank. Washing-
ton, DC, April.
World Bank. 2010. Gender: Working Towards Greater Equality. World Bank. Washington, DC.