A Conditional Cash Transfer and Women’s Empowerment: Does Bolsa Familia Influence Intimate Partner Violence?

Abstract

We apply mixed methods to explore how a conditional cash transfer (CCT) may influence intimate partner violence (IPV). Qualitative interviews with female beneficiaries of Bolsa Familia, Brazil’s CCT program, and service providers suggest positive, negative, and null associations between CCTs and IPV are all plausible. These associations result from a combination of economic and psychological motivations. We also use quantitative methods to examine if the expansion of Bolsa Familia between 2004–2009 affects the female homicide rate and marital separations. Using municipal level data in a difference-in-differences analysis, we find null associations between Bolsa Familia and female homicides. When we look at the impact on marriages and separations, indicative of a changing reservation utility of women within the household, we find no impact on marriages but Bolsa Familia is associated with an increase in separations, and, to a greater extent, separations of couples with children. These results suggest Bolsa Familia impacts women’s agency within the family, but not to the extent that it reduces the most extreme form of IPV.

INTRODUCTION

Conditional cash transfers (CCTs) are a common development program in which monetary stipends are often provided to impoverished families contingent on some action. Many programs provide mothers with a transfer if their children attend school or have health check-ups. In contexts where women have less earnings and women are the transfer recipients, these programs have the potential to be financially empowering, but this improved economic situation could be associated with increases or decreases in intimate partner violence (IPV). Low levels of financial autonomy is already a risk factor for becoming a victim of intimate partner violence (García-Moreno, Jansen, Ellsberg, Heise, & Watts, 2005). Economic theory indicates a conditional cash transfer can improve the financial ability to leave the union; the threat of exiting due to maltreatment is more credible (Manser & Brown, 1980). In support of this theory, Aizer (2010) found that more equal wages between genders results in fewer female hospital visits in California. Yet theory also suggest that men feel threatened when women are empowered—resulting in backlash against these women. Other studies do find increased economic opportunities are correlated with higher rates of violence. For example, in a study of violence in a garment district in Bangladesh among migrant families, Heath found a positive correlation between work and IPV (2012).

A literature review on impacts of CCTs on IPV, a specific measure of female empowerment, indicated mixed effects in developing countries (Buller et al., 2018). Few studies indicated an increase in IPV, while most showed a decline or null results. However, most of these studies rely on self-reports. Only in Uruguay are administrative data on reports of domestic violence used as an outcome variable rather than survey data (Borraz & Munyo, 2017). The bulk of the literature in Latin America relies on self-reports. In Mexico, Bobonis, Castro & Gonzales-Brenes found that women beneficiaries of the CCT Oportunidades reported less physical violence by spouses but more emotional violence two to six years following implementation (2011). However, five to nine years after implementation, Bobonis, Castro, and Morales found little difference in levels of physical and emotional violence between spouses; threats of physical violence were lower for women in more recently formed couples suggesting that recipient women may have more choice in selecting a non-violent partner (2015). Fernald and Hidrobo analyzed an unconditional cash transfer program in Ecuador and found that emotional violence decreases among those with education beyond primary school. For those with lower levels of education, emotional violence increases if they have an education level equal or greater than that of their husband (2012). They found no effect on physical violence. As far as we are aware, Peru is the only Latin American country where participation in the CCT Juntos has been associated with a reduction in physical as well as emotional violence, though the reduction is small (Perova, 2010).

Women’s empowerment more generally has also been studied as an impact of CCTs. For example, Bobonis found an increase in separations and unions as marriage turnover increased with the CCT program Progressa in Mexico (2011). This may be because women were able to move into better-quality unions. De Brauw et. al (2012), etc. found an increase in self-reported agency for recipients of Brazil’s CCT Bolsa Familia: women had increased decision making power regarding contraception, and women in urban areas and who are less educated than their spouses have a voice in their children’s health expenditures, children’s school attendance, children’s clothing, the woman’s clothing, the woman’s own labor supply, and durable good purchases. Qualitative literature on Bolsa Familia, reviewed by Bartholo, generally indicated positive outcomes for women’s empowerment from Bolsa Familia (2016). Bartholo summarized that six of the eight studies reviewed explore autonomy of women in the home and found that income security provided these women with more respect and decision-making power over purchasing choices. However, partners may still influence recipients by shaming them if they do not spend on items aligned with the social goals of Bolsa Familia.

Studying how CCTs may influence intimate partner violence is important for a number of reasons. First, if violence increases, then the design of the CCT may need to be altered so as not to put women in danger. Alternatively, additional policies could be implemented along with the existing CCTs to provide those protections. If IPV decreases with CCTs, a careful examination of the mechanisms may help to inform additional policies to more directly influence the reduction of IPV. Our study finds little correlation between the Brazilian CCT Bolsa Familia and extreme IPV (female homicides), though we do find an increase in separations, suggesting there may be some portion of women using Bolsa Familia to leave undesirable relationships. These less dramatic findings imply that small transfers may have some an impact on female empowerment, but the transfer is not sufficient to suppress extreme violence. Thus intimate partner violence needs to be addressed with more direct protections. Researchers could also test if programs with larger monetary transfers for women are more effective than Bolsa Familia in reducing violence against women.

We contribute to the literature on CCTs and female empowerment in two ways. First, our qualitative research corroborates pathways by which CCTs can influence IPV. Our interviewees include service professionals in addition to beneficiaries, whose expertise provides insights from a broad array of cases. We are unaware of qualitative research on CCTs and IPV that includes service providers, though some studies of the health impacts of CCTs include service providers (Adato & Roopnaraine, 2010). Secondly, our quantitative analysis assesses Bolsa Familia’s impact on women’s empowerment by examining two outcomes: female homicide as a proxy for IPV and union formation/dissolution as an additional proxy for women’s empowerment. As far as we are aware, this is one of the few studies within the CCT literature that uses non-self-reported data – both qualitative and quantitative–to examine if a CCT influences IPV and the larger realm of women’s empowerment. Women’s empowerment is not measurable through a single indicator – and it’s construct can even be questionable with multiple indicators. However, by including a variety of analyses, we hope to have contributed a broader understanding of how CCTs could plausibly be influencing female empowerment.

BACKGROUND ON BOLSA FAMILIA

Brazil is home to the first CCT programs, implemented independently by two municipalities in 1995: Bolsa Escola (School Grant in Brasilia) and PGRFM (Guaranteed Minimum Family Income Program in Campinas, Sao Paulo). As the popularity of such programs expanded, by 2001 more than 200,000 families were covered by local or state programs, many of which were absorbed in 2001 into the federal adoption of Bolsa Escola and Bolsa Alimentação. These precedents paved the way for Bolsa Familia to be established as a national project in October 2003, combining these and a few other programs into a centralized system. The program grew from 3.8 million families in 2004 to 12 million families by 2009.

Under Bolsa Familia legislation, mothers in poor families receive a stipend if they comply with health and education conditions. (Preferentially the stipend is given to women; men can be recipients, although only if they are single fathers or in households that are extremely poor without female membership.) Children must adhere to vaccine schedules, attend school 85% of the time, and pregnant women must complete prenatal appointments. The value of the monthly stipend was about US$5 per child with a maximum of US$19 in 2006, about 30% of the per capita poverty line. Money is deposited into mothers’ accounts and withdrawn from ATMs using Bolsa Familia cards; only 2.2% of recipients do not use electronic withdrawal (Lindert et. al. 2007). Extremely poor families receive an additional monthly stipend regardless of family composition and compliance with conditionalities.

BACKGROUND ON IPV IN BRAZIL

In 2006, Brazil passed legislation to legally define five types of IPV violence: physical, sexual, patrimonial (financial), psychological (emotional manipulation), and moral (public humiliation). Our qualitative research touches on patrimonial violence, however, most of our interviews and analysis focus on physical violence. Seven years after this legislation, 91% of respondents on a nationally representative survey agreed to some degree that husbands who beat their wives belong in jail and 89% disagreed to some degree that a husband can yell and shout at his wife (Guerreiro Osorio & Fontoura, 2014). Yet underlying biases against women are still strong, with 58% of respondents agreeing at least partially that if women knew how to behave themselves there would be less rape and 82% agreeing, at least partially, that outsiders should not get involved in marital disputes.

In 2003, the prevalence of physical abuse within couples in capital cities was estimated to range from 13.2–34.8% as measured by the the Conflict Tactics Scales, with higher rates in the poorer regions (Reichenheim et al., 2006). However, quantitative measures of IPV are challenging based on the private nature of the event. Much happens within the home, unobservable to the data collector. Measurement issues related to self-reported data are a challenge because of selection issues regarding the participants’ willingness to report incidents. Although overall violence in households may be falling, women may report more IPV as their outside options improve or as it is perceived as more socially acceptable to renounce abusers. Women who are abused but live in a world where violence against them is a cultural norm may not perceive the injustice nor report it: less than twenty percent of Brazilian women surveyed in the 2005 WHO study had reported IPV to the police (García-Moreno et al., 2005). In addition, another twenty percent of women had not told anyone about the abuse inflicted on them. Thus, we have turned to more objective measures: counts of female homicides of 15–49 year olds, since it is impossible to self-report a homicide. Although this measure limits us to the most extreme form of IPV and is still imperfect as not all homicides are due to IPV, it is superior to the alternatives due to its reliability and independence from self-reported bias. See Perova and Reynolds (2017).

Brazil has one of the highest female homicide rates in Latin America, only behind El Salvador, Guatemala and Honduras – all countries with lower levels of economic development. Moreover, much of the violence against women in Brazil is perpetrated by family members: Movimento Nacional de Direitos Humanos (National Movement of Human Rights) estimates it to be close to 70 percent. Furthermore, 66.3 percent of these homicides against women involve intimate partner violence (AGENDE 2007). The Brazilian national death statistics indicate that of all female homicides, 40% of these incidents occurred inside the family home or residence, where IPV is most likely to occur. By comparison, the portion of male murders occurring inside the family residence stood at 14.7% (Waiselfisz, 2012). In 2011, of the almost 50,000 women in the health system who responded to a questionnaire for victims of violence, 70% of those who reported the relationship to the attacker indicated it to be family (Waiselfisz, 2012). All of these statistics indicate that IPV makes up a large portion of the homicides against women within Brazil. This is also the case internationally: in South Africa, Israel, Canada or the US, 70% of female homicides are attributed to IPV (United Nations Office on Drugs and Crime, 2011). In California alone, 70% of hospital visits are due to aggression from an intimate partner (Aizer, 2010). A dip in female homicides in 2007—without a corresponding dip for men—aligns with the Brazilian government passing a law with tougher penalties for IPV, further validating the link between homicides and IPV (Perova & Reynolds, 2017).

QUALITATIVE METHODS

Location

In January of 2016 we conducted field research in a city in northeastern Brazil. We selected this location for data collection because there is a high rate of participation in Bolsa Familia. We partnered with administrators of a local NGO who helped us recruit participants from a low-income neighborhood on the periphery of the city; one of the authors had worked with this NGO on field work for a previous study. The names of the municipality and the NGO are omitted to protect study participants. Some service providers were contacted prior to traveling to Brazil and our Brazilian contacts assisted us in scheduling interviews. Others we met when requesting meetings at service centers to which we were referred.

Qualifications

The translating author had established trust with the community on a prior research project; she had led focus groups to inform quantitative data collection in the same community. The interviewing author is certified by the state of California to work with survivors of intimate partner violence and had previously conducted interviews with administrators of Bolsa Familia under the supervision of a Brazilian political scientist. We designed the interviews in consultation with a Brazilian researcher in the health profession and a Brazilian expert on Bolsa Familia and gender issues. Since data collection, the interviewer has been more formally trained in qualitative research methods (drafting data collection instruments, conducting interviews and focus groups, and systematically analyzing qualitative data) both through coursework and employment at a research firm. These new experiences indicate some limitations of our qualitative data collection, such as a more representative sampling technique could have been employed and a more structured interview could have been used. However, for the most part, the formal training confirmed that the data collection and analyses were generally aligned with rigorous qualitative research practice.

Sampling

The local NGO introduced us to female community members, some of whom they had helped escape IPV, and we subsequently relied on snowball sampling to contact additional community members, interviewing eight women in total. Seven of these eight women had been beneficiaries of Bolsa Familia (Table 1, Panel B). We also interviewed five local professionals from the same municipality who either specialized in supporting victims of IPV or who had expert knowledge of Bolsa Família. These professionals were: a social worker and a psychologist who worked at a Centro de Referencia de Asistencia para la Mujer (Referral Center for Women’s Support - CRAM), a professor of public health, a legal administrator who worked at the family court, and the director of a female police station. Our sample size was limited by our time constraints in Brazil.

TABLE 1:

DESCRIPTIVE STATISTICS FROM INTERVIEWS

PANEL A: Demographic Information	Community Members	Service Providers
Bolsa Familia Recipient	88%	0%
Female	100%	60%
Black or mixed race	100%	20%
Married	67%*	NA
Employed	33%*	NA
PANEL B: Reporting Frequencies
Increase in IPV (Overall)	50%	60%
Economic Mechanism: Men desire the  money for their consumption	25%	20%
Psychological Mechanism: Male  backlash from threat of female  empowerment	25%	60%
No Change in IPV (Overall)	38%	40%
Economic Mechanism: The stipend is  too small to change partner dynamics  is for the children / not for women so is	13%	40%
	25%	0%
Reduction in IPV (Overall)	13%	40%
Economic Mechanism: Fewer conflicts  from less financial stress  more self-esteem and respect from  fulfilling conditionalities and receiving	13%	0%
	13%	0%
Other Mechanism: More trust in public  policy	0%	40%
N	8	5

^*We did not ask this question explicitly, but information emerged in the interviews of 6 community members Reporting frequencies are based on coding done in MAXQDA12; percentages sum to greater than 100 in some

Interviews

Our interviews were semi-structured, including questions on the following themes: the role of women in Brazilian culture and society, the prevalence of violence in their communities, female empowerment, relationship and household roles and processes, and the differential impacts of paid female employment versus social assistance. The interviews were conducted in the service providers’ offices, the community member’s home, or in a private room of the community center. At the beginning of each interview, the purpose of the study was explained and informed consent secured. Interviews were translated between English and Portuguese in real time, with one author translating and the other making note of body language, key ideas, interruptions, etc. The research protocol was approved by the Institutional Review Board at the University of California, Berkeley.

Analysis

The interviews were recorded and the English translation was transcribed. Data analysis was completed using qualitative data analysis software, MAXQDA 12. The data was organized around three hypotheses based on the existing literature: Bolsa Familia reduces IPV, Bolsa Familia has no impact on IPV, and Bolsa Familia increases IPV. Within each perspective, we developed a coding system around the thematic mechanisms that emerged_.

QUALITATIVE RESULTS

For each hypothesis, we uncovered both an economic and a psychological mechanism which are summarized in Table 1, Panel B.

CCT Increases IPV

Seven out of thirteen respondents, including four community members and three service providers, reported that the CCT increases IPV. Two explanations were given for this perceived outcome. The first is that some men want control of the CCT for their own consumption, which is regarded as patrimonial or economic violence. Subsequently, physical violence erupts. This may happen even if women feel more empowered. Community Member 2 noted that although she felt more empowered after receiving benefits from Bolsa Familia, her participation in the program also resulted in increased levels of violence. She attributed this to more conflict between her and her husband.

In other cases, men gain control over the transfer without physical violence, through threats only. Community Member 1, a cook at the local NGO, formerly received Bolsa Família benefits, but no longer qualified for the program after she gained employment.

Community Member 1: It [the CCT] causes fights. Sometimes the men will steal the cards and use all the money...Many have told me this experience. Sometimes when the man is dependent, he threatens, and the woman goes and gives him the money, because who wants to lose their lives? Not always, but sometimes it happens that men will threaten women. We don’t feel very protected.

Other community members similarly believed that participation in Bolsa Família caused more conflict between partners, but there were slight variations in their explanations of why this occurred. In the interview above, Community Member 1 believed that husbands acted violently as a means of acquiring the cash transfer, whereas Community Member 9 insisted that physical violence between spouses occurred only after the husband had successfully retrieved the transfer from the wife and spent the money. The cause of violence in these cases was framed as marital unrest and mutual fighting between spouses, rather than intimate partner violence.

Interviewer: Do husbands and wives fight over who controls the money from Bolsa Família? Community Member 9: No. The men don’t fight because they just grab it. After they get it and spend it, then they fight.

Community Member 1, acknowledged the prevalence of physical violence inflicted on women receiving the cash transfer by friends or other relatives, namely their own children.

Community Member 1: Their husbands fool them. These are the uneducated women. And other people fool them too. Other people take out some money from their card but they don’t know what amount is supposed to be taken out and then other people fool them. Sometimes people use the money for drugs. Sometimes the children take the money. They [the women] try to be alert but they are fooled. It happens.

This power dynamic was reported by women beneficiaries as well as by Professional 1, a psychologist working at a local women’s resource center. She explained that it is common for her clients to experience escalated levels of economic abuse, or patrimonial violence, as a result of the CCT.

Service provider 1: He can do this [engage in patrimonial violence] with Bolsa Família. When he takes your card, and keeps you away from your money, your rights, this is patrimonial violence. So this happens a lot. And this is also subjective because he doesn’t need to get your card and go to the bank and withdraw your money. He will make you go and then you will give him all the money. That’s also patrimonial.

Some respondents also reported perceived increases in physical IPV associated with the CCT due to the men’s desire to maintain control and dominance over their female partners. Interpretations focused around the idea of “male backlash”, which posits that when there is an inconsistency in traditional power relations (e.g. a woman has greater resources at her disposal than her husband), violence will increase because the man’s dominance is threatened. Themes of machismo, patriarchy, and jealousy were common in women’s narratives explaining why the CCT triggers IPV. Community Member 5, who joined Bolsa Família a year before the interview explained the desire of men to have dependent female partners.

Community Member 5: There are cases where the husband doesn’t like that the woman receives money. Husbands are like that...they always want the woman to be dependent.

Community Member 3, a prior recipient of Bolsa Família, hid the fact that she was participating in the program from her husband out of fear that he would become angry and violent if he found out.

Interviewer: When you received Bolsa Família, why did you choose to keep it a secret from your husband?

Community member 3: Because I had a lot of fear of him. He threatened me...he came with a knife, sharp blades...he was a crazy man. If he discovered he would be mad because he couldn’t dominate me anymore. He taught me to say I can’t do it without him. He said, “You’re mine, you’re not of anybody else, and if you run away, I will kill you.”

Three service providers gave similar reports, based on the experiences of their clients. They explained that the male desire to control female partners, an attitude they described as “machista,” is common in the Northeast.

The cases where Bolsa Familia increases violence may be limited to situations of extreme poverty. Community Member 5 indicated that if husbands were already employed, and thus still had control of resources outside of the transfer, they would not seek to take control of the resources. Because of this resource control outside of the home, these men felt less threatened by their wife’s access to the CCT and consequently did not behave aggressively.

CCT has no effect on IPV

Five respondents reported no change in violence associated with the CCT. Community members and service professionals reported that the value of Bolsa Família is too small to have any real impact on violence. Community Member 1, the cook from the NGO, emphasizes this point.

Community Member 1: It’s [Bolsa Família] more of an income coming in to help, but we continue living in the same situation. To be able to feel more power, they need to work and earn more money not just on Bolsa Família. They have to go out and fight for power. Bolsa Família is not enough. The husbands are cruel. There is no respect whether you have Bolsa Família or not. Husbands are the same.

Service provider 2, a social worker at the resource center for IPV victims in Salvador, laughed audibly when asked whether she had clients who had been able to leave their abusive partners using the income from the CCT: “[laughter] No. It’s too little money.”

The second reason identified for why Bolsa Familia may have no impact on violence is that not everyone was aware of the important role women are required to play in Bolsa Família. If women are perceived as key players in an established federal program, they may be respected for that role and subsequently experience less violence. However, if their role is perceived as insignificant or even nonexistent, it likely explains the absence of change in the respect shown to them (and the violence experienced by them) in their households and communities.

Community Member 4 is a young female college student who does not participate in the CCT program. Despite being educated and living in close proximity to numerous CCT participants, she was unaware that the stipend is given preferentially to women.

Community Member 4: Here in Brazil...there is a false image of what a woman is, what a woman can do—that a woman is respected. That’s not true. Those who live know that it’s not true. I didn’t even know that Bolsa Família was only for women. I’m only learning this now.

Similarly, when Community Member 8, a Bolsa Família recipient, was asked whether her husband had any problem with the fact that he did not control the transfer as she did, the woman replied that the transfer was not “her transfer” but was instead a resource for her children. This also signals a way in which perceptions of the program may influence ideas about household resource control and determine levels of IPV.

Community Member 8: No, there’s no difference [in violence]. He knows that it’s not for me or for him, but that it’s for the children. The food and school for the children.

CCT decreases IPV

Only Community Member 9 reported that participation in Bolsa Família decreased IPV for her friends (she personally had never experienced violence by her husband, before or after joining the program). She provided two distinct explanations for the observed reduction. First, she explained that husbands who had behaved violently prior to their wives receiving the CCT modified their behavior after the fact because of the sense of relief they felt knowing that they were no longer responsible for acting as the sole household provider. In other words, in these contexts the CCT reduced household financial stress and subsequently reduced IPV.

Community Member 9: The husbands of my friends feel better because they don’t have to be the sole provider...I think it [IPV] went down because there were more constant fights earlier and the women receiving this money...there were fewer problems and so there’s fewer fights. Less financial stress.

Additionally, the respondent reported that her friends who had experienced less IPV after joining Bolsa Família felt more independent because they no longer had to rely solely on their husbands and felt that they had more power and respect in the household because they were responsible for receiving the transfer and fulfilling the conditionalities. This, coupled with reduced financial stress, reportedly lessened the incentive of the already employed husband to use violence to maintain control or to extract the transfer.

Two service providers similarly indicated that Bolsa Família could reduce violence, although this was uncommon. These respondents highlighted the role of Bolsa Família in creating trust in public policy more generally among beneficiary women, which reportedly gives them more faith that the system will aid them in seeking justice for the violence they experience. Additionally, one service provider in our case study explained how the CCT reduced violence indirectly: women believe that the government has their best interest in mind as a result of the program and assume that perpetrators of violence will be held accountable; therefore, the beneficiaries tolerate less violence. Subsequently, they will seek justice and be less likely to tolerate abuse. Service provider 2, the social worker at the resource center for IPV victims in Salvador, elaborated.

Service provider 2: What I have seen is that the woman comes here without Bolsa Família. We take her to get the Bolsa Família and have this right and then she begins to receive the money and then she begins to see that she has rights to things. She finally sees that this exists and she has rights to this and she receives it. She starts to believe in public policy and this helps her believe that she can begin to complain about husbands and justice. This is the first point for starting to believe in public policy. This effect is easier to achieve than actually sustaining the separation financially. And then the belief in public policy leads to the separation instead of the money itself.

USING QUALITATIVE RESULTS TO INFORM THE QUANTITATIVE STUDY

The qualitative study informs our quantitative study in two ways. First, it supports the validity of any of the hypotheses at the aggregate level. The variety of experiences from our qualitative findings reveal that a positive, negative, or null relationship between a CCT and IPV would not be surprising. The conclusion that any hypothesis is plausible does not make our quantitative analysis moot – given that we interviewed only 13 respondents, our qualitative investigation does not provide any information on which of the uncovered mechanisms dominates, and consequently, what the average impact of Bolsa Familia on IPV is. Our quantitative analysis has a potential to provide this information, and consequently, inform national level policy on mitigating IPV.

Specifically, we evaluate the impact of Bolsa Familia using a proxy for one of the harshest forms of IPV: female homicide (we discuss advantages and limitations of this measure in greater detail in the next section). We also examine the impact of Bolsa Familia on marriages and separations, as an intermediate outcome which can be related to IPV. Specifically, separation from an abusive partner may decrease exposure to less dangerous forms of violence, which we are not capturing with our variable of interest.

Secondly, since there are different experiences reported, we also perform heterogeneity analyses to see if different groups have distinct patterns. Our data does not allow us to divide along specific economic and cultural/psychological differences, but we do stratify by women’s age, city size, and geographical regions. Younger women have different economic opportunities and cultural values than older women. The same can be said regarding city size, with capitals and large metropolitan areas often having more diverse economic opportunities than smaller cities. Additionally, some regions of Brazil are more impoverished and have distinct cultural values than other regions. For example, in contrast with the Northeast where our qualitative study occurred, the South is much wealthier and has a much larger population of European decent in contrast to African descent in the Northeast.

QUANTITATIVE DATA

Even though our qualitative data are from 2016, we analyze data from the years from 2004 to 2009 because this was the period of Bolsa Familia expansion and thus provides the most variation in Bolsa Familia coverage. We chose not to analyze 2010 because some of our control variables are missing for 2010, a census year which interrupted some data collection.

Time Frame

Our data forms a municipal level panel spanning 2004 through 2009. The panel begins in 2004, after a law was passed in 2003 which required health professionals to report abuse to the police; this law could affect trends in IPV. Additionally, this is also the first year for which some control variables become available. National data on Bolsa Familia are available since 2004, the year after Bolsa Familia was established as a national program. We end our panel in 2009, when 11 million families received the stipend. Since then, only an additional 2.8 million families were added by 2015. Thus, we focus on earlier years.

Outcome Variables: Female Homicides

Our primary outcome variable of interest is female homicides, the most extreme form of physical IPV. We choose this variable for two reasons: death is the ultimate limitation of freedom and thus there is a strong imperative to examine this outcome first, and this data is the most reliable. Our data come from Brazil’s Sistema de Informações de Mortalidade (National mortality data base – SIM), a registry of every death in the nation. We limit the data to the 15–49 age group, which parallels that used in the WHO study on IPV as this demographic has the highest concentration of victims from IPV. We consider a death to be a homicide if it is classified as one, or if, in a separate question with more specific delineations of the cause of death as defined by the WHO international classification of diseases, the cause of death is coded to be aggression (the discrepancy is 5%.) We do not include homicides from traffic accidents.

Data on homicides is also collected by individual states’ departments of public security, but these are more difficult to access. We were able to perform a small data validation exercise, comparing SIM data with homicides from Sao Paulo. Concordance has improved over time, from a thirty percent difference in 1998 to a four percent difference in 2005, the most recent year for which Sao Paulo data is available. The discrepancy between these numbers and the SIM database has been recognized as resulting from slight differences in definitions: deaths resulting from legal interventions and some other situations reclassified using health data (such as finding corpses) are not included in the police definition (Poverty Reduction and Economic Management Sector Unit Latin America and the Caribbean Region, June 2006).

To ensure the most accurate analysis possible, we eliminate municipalities from our sample that fail to meet international standards in quality control (more than 10% unexplained deaths occurring from external causes). In any one year about 1,200 of Brazil’s 5,564 municipalities have more than 10% unexplained deaths, but only 2,128 municipalities have always maintained this international standard during 2004–2009. We also eliminate municipalities that had ever had female homicide rates in the 100^th percentile, calculated over the entire time period. These are typically smaller municipalities, where several homicides in a year result in a very high homicide rate due to the small denominator. An additional 11 municipalities did not have complete control variables in every year. The remaining 2199 municipalities in the sample encompass 34% of the nation’s population of women ages 15–49 years.

We present main results for two variants of this outcome: we examine the log of the count of the number of homicides and the homicide rate for women ages 15–49 years. We account for zeros by adding 0.001 to our count values before taking the natural log.

Outcome Variables: Marriages & Separations

Prior to 2010, Brazilian law requires that before divorce, which allows for legal remarriage, a separation must be registered for one year. Thus we focus on separations, which is a more immediate measure of union dissolution. Using data from Brazil’s civil registry provided by Brazil’s National Statistics Institute (Instituto Brasileiro Geografia e Economia – IBGE) we implement the same analyses with these secondary outcome variables: number of marriages, number of separations to couples with children, and number of separations to couples without children. We limit the municipalities to capital cities and large metropolitan areas since filing is not available in all municipalities. We postulate that within the largest cities, most of the separations registered will be for citizens of that municipality. We maintain the requirement that causes of death be less than 10% unknown as an indication that the municipality maintains well-documented records. We remove an additional four municipalities that started with several years without any marriage in the panel, suggesting that courts (or improved reporting) were implemented later in the time frame. Though cohabitation and dissolution without legal registration is common in Brazil, this is not the majority of cases. In 2000, 28.58% of couples cohabited instead of being legally married, and the likelihood that an illiterate woman was cohabiting was 35% (Covre-Sussai & Matthijs, 2010). Marriage rates increase with education and childbearing.

Treatment Variable: Bolsa Familia Coverage

Allocation of Bolsa Familia funds to municipalities is based on quotas originally determined by poverty analyses from 2000 and 2001, and the quotas were again updated in 2006 (Lindert, Linder, Hobbs, & de la Briere, 2007). Since these allocations were established by the national ministry of social development, the municipalities could not influence this division. Although poverty and violence are correlated in Brazil (The World Bank, 2006) so more violent municipalities will receive more Bolsa Familia resources, small, yearly fluctuations in violence should not impact the change in funds. Given that our preferred identification strategy relies on using variation not only over municipalities, but also over time, we consider the Bolsa Familia variable exogenous. From the Ministry of Social Development, we have the total sum of the transfer from Bolsa Familia to each municipality in each year. We divide this by the census projections of the number of women age 15–49 as a measure of the per capita yearly transfer to women age 15–49 (IBGE). Since poor Brazilians tend to have children at young ages, it is likely that the 15–49 age range captures the majority of recipients. Though women technically cannot become recipients of the Bolsa Familia stipend until age 18, we still include those between the ages of 15–17 in the denominator since they are included in our outcome variable group.

Control Variables

Since homicides due to IPV are a subset of our outcome variable, we include controls that are likely to impact IPV as well as violence more generally, such as street crime. We use data from a variety of sources provided by Brazil’s publicly available national statistics: IBGE, the Tribunal Superior Eleitoral (Electoral Superior Tribunal, TSE), the Ministerio de Desenvolvimento Social (Ministry for Social Development, MDS), Finanzas Brasil (a data base from the National Secretary of the Treasury, FINBRA), and other variables from the MUNIC survey (municipality survey by Brazil’s statistical office IBGE).

We include female population in our age group of interest and population density, as the number of homicides depends on the size of the population and the proximity of people to one another. We calculate the percent of population that is female, should scarcity of women influence how they are treated; a significant negative correlation between this variable and the female homicide rate supports this hypothesis. To capture municipal wealth, a variable generally correlated with crime, we include GDP per capita and GDP per capita from agriculture and industry. These variables we accessed from IBGE.

Gender gap in education may affect intra-household bargaining, so we include the ratio of the percent of female voters who have finished primary schooling to the percent of male voters who have finished primary schooling, data from the TSE. Given that voting is obligatory in Brazil for people aged 18 to 70, this is a valid proxy for the gender education gap.

We include dummies for whether the municipality has a local militia and a Public Safety Council, as these may affect overall crime (from MUNIC). We also include overall per capita spending within the municipality on civil defense, intelligence, and public security (from FINBRA). Similarly, per capita spending on hospital and health (from FINBRA) and a dummy variable indicating a municipal health council (from MUNIC) reflects capacity that may prevent attempted homicides from becoming deaths. In an effort to take public attitudes into account, we include dummy variables for whether the municipality has a human rights council (from MUNIC) and per capita spending on human rights and public communications (from FINBRA).

EMPIRICAL STRATEGY

We apply a difference-in-difference approach to identify effects of Bolsa Familia on female homicide outcome variables, both numbers and rates, as well as marriage and separation outcomes, controlling for municipality and year fixed effects. Specifically, we estimate:

$$Y_{mt}=\beta T_{mt}+\rho x_{mt}+{\alpha }_m+{\gamma }_t+{\epsilon }_{mt}$$

where Y_mt is the outcome variable in municipality m at time t and T_mt is per capita transfer from Bolsa Familia, where the numerator is the yearly transfer amount within the municipality and the denominator is the number of women age 15–49 in the municipality m at time t. Vector X_mt controls for time-variant municipal level characteristics, α_m is municipality fixed effect, and γ_t is year fixed effects, which control for national time trends, such as the 2007 dip in female homicides following the implementation of the Maria da Penha law.

In the computation, each municipality is weighted by the female population in our age group. We run each regression without controls and with controls. We also include a specification controlling for male homicide. A drop in violence against women associated with Bolsa Familia could be an income effect if higher incomes in community due to Bolsa Familia reduce crime. If poverty reduction is reducing violence for all rather than for women in particular, controlling for male homicides will capture this trend. However, if these economic gains resulting in an overall drop in violence indeed only benefit women, controlling for male homicides would not change the estimates. Another reason to include male homicides as a control is if the Bolsa Familia variable is correlated with other unobservables that impact homicides in general, not specific to female homicides. We check for robustness by examining results using an alternate measure of Bolsa Familia intensity: percentage of women receiving Bolsa Familia, as calculated by number of recipients divided by women ages 15–49. We also examine unweighted results, results that control for municipal linear time trends instead of year and municipal fixed effects.

For the homicide analysis, we stratified the sample by a variety of demographic heterogeneities (homicides of women younger than age 25 compared to homicides of older women, homicides of residents of capital cities and large metropolitan areas compared to homicides in smaller cities, and homicides in the Northern, poorer geographic area of the country compared to the Southern, wealthier area). The marriage and separations analysis is already limited to large cities because that’s where the legal processes occur. Similarly, we do not stratify by region due to the smaller sample size and we have less information on age.

QUANTITATIVE RESULTS

Figure 1 illustrates the base levels of violence and reais from Bolsa Familia in 2004 to show how these have changed over time. Although there has been a large expansion of Bolsa Familia, the eye is not able to draw out any particular pattern link between Bolsa Familia and changes in homicide rates. The average female homicide rate among women 15–49 in 2004 was 3.4 per 100,000 women in our sample, which is similar to the national average around 4 (Table 2). The male homicide rate was 43.8. On average, per woman municipal transfer of Bolsa Familia amounted to R$142.60 per year, or about US$50. (We divide total municipal transfer by the number of all women, not only eligible women. Consequently, this number is much lower than the average amount of transfer to Bolsa Familia recipients.) Our data also suggest that education gap favors women, and while close to 90 percent of municipalities have a health council, only a tiny fraction of them has a human rights council. Comparisons to the municipalities out of our sample are found in the Appendix Table.

Figure 1.

Female Homicides and Bolsa Familia Over Time

TABLE 2:

2004 SUMMARY STATISTICS

	N	mean	s.d.	min	max
Female Homicide Count Ages 15–49	2076	0.5	2.0	0.0	32.0
Male Homicide Count Ages 15–49	2076	6.6	29.3	0.0	482.0
Female Homicide Rate 15–49	2076	3.4	9.5	0.0	74.7
Male Homicide Rate 15–49	2076	43.8	64.0	0.0	627.0
Reais per woman per year from Bolsa Familia	2076	142.6	124.1	0.0	722.7
% Population Female Age 15–49	2076	0.5	0.0	0.4	0.6
Population Density	1932	0.1	0.4	0.0	6.9
Ratio of % of women voters who finished primary to % of men voters who finished primary	2076	1.3	0.2	0.8	3.1
Per Capita GDP	1932	7.0	7.0	1.2	122.6
Per Capita GDP from Agriculture	1932	2.1	3.4	0.0	101.7
Per Capita GDP from industry	1932	1.4	3.4	0.1	64.2
Municipality has a human rights council	2076	0.0	0.1	0.0	1.0
Municipality has a health council	2076	0.9	0.3	0.0	1.0
Municipality has a Public Safety Council	2076	0.1	0.3	0.0	1.0
Municipality has a Municipal Police Corps	2076	0.1	0.3	0.0	1.0
Per Capita Expenditure on Civil Defense	1932	0.3	1.6	0.0	35.7
Per Capita Expenditure on Public Security, Information and Intelligence	1932	51.5	87.4	0.0	1351.3
Per Capita Expenditure on other Defense Functions	1932	0.0	0.3	0.0	13.5
Per Capita Expenditure on Basic Health	1932	0.1	0.9	0.0	21.7
Per Capita Expenditure on General Health	1932	112.9	107.4	0.0	1269.3
Per Capita Expenditure on Hospitals	1932	348.5	218.6	0.0	3029.6
Per Capita Expenditure on Other Health	1932	5.5	23.5	0.0	298.3
Per Capita Expenditure on Social Communication	1932	0.8	3.7	0.0	72.8
Per Capita Expenditure on Human Rights	1932	0.1	1.3	0.0	32.0
Per Capita Municial Federal Transfer	1932	570.7	309.7	124.0	4586.9
population of women ages 15–49 (used for weighting)	2076	8131.8	28221	206	517496
population of men ages 15–49 (used for weighting)	2076	7871.4	25770	228	477421

Sample limitied to municipalities with accurate death statistics (less than 10% from unknown causes)

Also eliminates municipalities which ever had female homicides in the top 100th percentile.

Number of municipalities are lower than in the regression because more data is avialable in later years

We find no association between Bolsa Familia and female homicides whether using log count or homicide rates (Table 3). This finding is robust and includes male homicides as an additional control. Similarly, our null results persisted with all of our other robustness checks (using the percent of women receiving Bolsa Familia as the variable of interest, removing population weights, and controlling for municipal linear time trends instead of year and municipal fixed effects).

TABLE 3:

BOLSA FAMILIA & HOMICIDE OLS

Panel A: Log Number of Female Homicides				Panel B: Female Homicide Rate
Year & Municipality FE	Y	Y	Y	Y	Y	Y
Controls*	N	Y	Y	N	Y	Y
Men’s Homicide Control	N	N	Y	N	N	Y
Bolsa Familia	0.0007	0.0007	0.0002	0.0015	0.0012	−0.0004
SE	(0.0006)	(0.0008)	(0.0003)	(0.0020)	(0.0023)	(0.0017)
Adj R2	0.705	0.703	0.535	0.278	0.277	0.155
N	12543	11936	11936	12543	11936	11936
Municipalities	2119	2119	2119	2119	2119	2119

Bolsa Familia is reais per year per woman in the municipality

Significance levels: +.1

^*.05

^**.01

All regressions with standared errors clustered by municipality

OLS regresssions weighted by population of men or women 15–49

^*See table 1 for controls

Comparing our results stratified by a variety of demographic heterogeneities (homicides of women younger than age 25 compared to homicides of older women, homicides of residents of capital cities and large metropolitan areas compared to homicides in smaller cities, and homicides in the Northern, poorer geographic area of the country compared to the Southern, wealthier area) does not result in any sub-population for which a significant decline in female homicides is robust to the inclusion of the male controls and the homicide rate outcome.

We repeat our regressions on number of marriages and separations. This sample is smaller with fewer municipalities and years available; summary statistics are in Table 4. Using the same empirical structure and controls (but not male homicide rate), our regressions are weighted by overall population as we do not have information on the ages of the parties in the unions. The number of separations for couples with and without children under age 18 years have risen as Bolsa Familia has expanded (Table 5). In our robustness check without population weights, magnitudes are only about 1/6^th as large and tend toward marginal significance, but the coefficients on separations with children remain larger than the coefficients on separations without children. Results using percentage of women receiving Bolsa Familia also support the main regressions; coefficients on marriages and separations of couples without children are not statistically significant. The coefficients on separations with children suggests that a one percentage point increase in women receiving Bolsa Familia results in 2–3 more separations per municipality (statistically significant at 5% for the lagged independent variable specification, marginally significant at 10% otherwise). We find no impact on the number of marriages, a finding robust to our additional specifications.

TABLE 4:

2004 SUMMARY STATISTICS

	N	mean	s.d.	min	max
Marriages	115	1022.2	1857.0	16	12113
Separations (couples with children)	117	83.8	175.9	0	1595
Separations (couples without children)	117	28.1	65.5	0	573
Reais per woman from Bolsa Familia	117	71.0	65.1	7.8	339.4
% Population Female Age 15–49	117	0.5	0.0	0.5	0.5
Population Density	108	0.0	0.0	0.0	0.2
Ratio of % of women voters who finished primary to % of men voters who finished primary	117	1.1	0.1	1.0	1.7
Per Capita GDP	108	10.4	9.5	1.4	62.1
Per Capita GDP from Agriculture	108	0.4	0.7	0.0	4.0
Per Capita GDP from industry	108	3.3	4.1	0.2	27.0
Municipality has a human rights council	117	0.0	0.2	0.0	1.0
Municipality has a health council	117	0.9	0.3	0.0	1.0
Municipality has a Public Safety Council	117	0.2	0.4	0.0	1.0
Municipality has a Municipal Police Corps	117	0.3	0.5	0.0	1.0
Per Capita Expenditure on Civil Defense	108	1.9	5.0	0.0	35.7
Per Capita Expenditure on Public Security, Information and Intelligence	108	42.8	60.0	0.0	259.9
Per Capita Expenditure on other Defense Functions	108	0.2	1.3	0.0	13.5
Per Capita Expenditure on Basic Health	108	0.2	1.5	0.0	14.7
Per Capita Expenditure on General Health	108	80.6	60.7	0.0	316.5
Per Capita Expenditure on Hospitals	108	271.0	150.5	17.0	638.2
Per Capita Expenditure on Other Health Expenditures	108	8.1	29.4	0.0	207.4
Per Capita Expenditure on Social Communication	108	2.1	4.2	0.0	23.7
Per Capita Expenditure on Human Rights	108	0.2	0.6	0.0	4.5
Per Capita Municial Federal Transfer	108	278.2	108.2	124.0	731.6
Population (used for weighting)	108	221206.9	334559	7272	1727010

Sample limitied to municipalities that are capitals or large metropolitan cities with accurate death statistics (less than 10% from unknown causes)

TABLE 5:

BOLSA FAMILIA, MARRIAGES & SEPARATIONS

	Marriages			Separations (couples with children)			Separations (couples without children)
Year & Municipality FE	Y	Y	Y	Y	Y	Y	Y	Y	Y
Controls	N	Y	Y	N	Y	Y	N	Y	Y
Lagged outcome variable	N	N	Y	N	N	Y	N	N	Y
Bolsa Familia	2.5710	2.6523	2.786	0.4280*	0.4719*	0.3271*	0.2423+	0.2589*	0.1375*
SE	(2.6059)	(2.5915)	(2.3826)	(0.2008)	(0.1848)	(0.1304)	(0.1461)	(0.1267)	(0.0685)
Adj R2	0.967	0.971	0.971	0.920	0.940	0.948	0.85	0.88	0.913
N	663	663	663	674	674	674	674	674	674
Municipalities	113	113	113	115	115	115	115	115	115

Sample limitied to municipalities that are capitals or large metropolitan cities with accurate death statistics (less than 10% from unknown causes)

Significance levels:

⁺ .1

^*.05

^**.01

All regressions weighted by population, with standared errors clustered by muncipality

DISCUSSION

Our qualitative research with service professionals and community members reveals a variety of experiences and perceptions of how a CCT may influence IPV: it can increase, decrease, or have no influence on IPV. It is notable that all of these perspectives emerged in a small geographic location, which suggests a great deal of heterogeneity in experiences. The themes that emerged suggest that a combination of economic and psychological reasons is behind any of these associations. Our quantitative results suggest that at a national level as well as for subpopulations, Bolsa Familia has not had an impact on female homicides, the most extreme form of violence, although we find some evidence that Bolsa Familia is associated with an increased number of separations among couples with children. This increase suggests Bolsa Familia allows for a few women to exit unions, although this mechanism is not strong enough to reduce the most extreme form of IPV, female homicides.

The mixed-methods nature of this study provides complementary insights. The qualitative findings help validate our null quantitative findings: no effect is plausible, but so are positive and negative effects, which may have canceling impacts. Additionally, even though the association between Bolsa Familia and separations for couples with children was positive, the small magnitude of the coefficient may explain why service providers did not perceive Bolsa Familia to be empowering.

In addition to economic considerations, two themes emerged from the qualitative research. First, the psychological mechanisms suggest a need for training in conflict resolution as part of the conditionality if female empowerment is one of the intentions of a CCT. Particularly helpful may be training in conflict resolution regarding finances, a repeated theme in our interviews. Previous research suggests that conflict resolution training may have positive outcomes on female empowerment. Ashraf et al. (2018) found that negotiation training can improve girls’ education outcomes as women may negotiate with their parents for a future.

The sense of impotence in the relationship expressed by some of the service providers suggest that changes need to come from men. Group programs for men may also help improve female empowerment and reduce IPV (Ellsberg et al., 2015). Furthermore, by incorporating men into the design of a CCT with a financial transfer to women may reduce the reinforcement of gender roles that it is women who are to care for the children. Though there is little evidence from developing countries, a literature review concludes economic empowerment programs plus gender equality training are more promising than economic empowerment programs alone (Puerto Gómez, 2016). Currently, ProMUNDO Brasil is implementing a complementary program with Bolsa Familia in two states (Martinez et al., n.d.).

Finally, the importance of a woman’s perception of government and more specifically, the justice system, remains largely unexplored in the literature on CCTs and IPV. For example, though laws are part of the context of the pathways diagram in Buller et al.’s mixed-methods review of CCTs impact on IPV, these are not discussed, nor is the political commitment of the state to impoverished women (2018). An example of the protective nature of the psychology behind public policy is documented by Perova and Reynolds (2017): female homicides dip significantly following the passing of the Maria da Penha law against domestic violence. Unfortunately, though, the levels of female homicides returned to previous rates after a short time.

Though we attempt to address the question “Does Bolsa Familia Influence IPV?” the answer is not conclusive. We examined different facets of empowerment, but our work has a number of limitations. Our qualitative study was done in a single city with a small sample; we do not have complementary perspectives from distinct regions. The study relies on female homicides, which is the most extreme and therefore not as common form of IPV. Furthermore, it is a noisy measure, as not all homicides are from IPV, and not all homicides from IPV are reported as a homicide. Thus we cannot make conclusions regarding if Bolsa Familia influences other forms of IPV. Similarly, our legal measure of separations do not represent all separations in the country. Many Brazilian cohabitating couples are not legally married; 36.4% of all unions were consensual in 2010 (Covre-Sussai & Matthijs, 2010). This is higher among the poor, so Bolsa Familia recipients are even less likely to be legally wed. Our estimates for separations are thus not representative of the general status of the population and could be under-estimating the number of women who are able to leave an abusive union.

Nevertheless, our study has a number of strengths. Our qualitative study includes service professionals as well as beneficiaries. Because the service providers have worked with many victims of IPV, the small number of expert informants actually includes information distilled from a broad number of cases. Our quantitative analysis considers separations in addition to considering homicides and include a number of robustness checks. Our quantitative results are not self-reports, and cover a broad portion of a large country.

Future research could examine these associations at an individual level. Our quantitative study examines aggregate data and thus could be subject to the ecological fallacy. Data on individual recipients’ experience of violence with participation in the program could provide more precise estimates. New mandatory reporting laws for health care workers have generated a registry of violence data (SINAN) that could be linked to individual recipients from the Ministry of Social Development. Furthermore, the 2015 Lei do Feminiciodo offers an opportunity to examine which homicides are tried as femicides, with domestic violence being included as one criteria.

Regarding future policy developments, in addition to considering economic aspects of empowerment, designers of CCTs need to comprehensively examine trust in public policy and government as well as psychological factors. The effectiveness of these programs will also be improved with additional research determining risk and protective factors relating to the different patterns (increase, decrease, null) reported by the qualitative interviews. Studies of distinct subpopulations can improve targeting and design of interventions that could help prevent cases where CCTs are associated with an increase IPV.

Qualitative data & longitudinal administrative data used to see if BF empowers women.

Service providers think the BF conditional cash transfer is too small to influence IPV.

BF recipients think BF could increase, decrease, or have no change on IPV.

No association between BF and female homicides found.

BF associated with increase in separations of couples with children, but not without.

APPENDIX

APPENDIX TABLE:

SUMMARY STATISTICS (2004)

	Not in Sample	In Sample	difference in means, p-value
Female Homicide + Death for Agressions Rate per 100,000 Women 15–49 (SIM & Census)	5.336	3.372	0.000
Male Homicide + Death for Agressions Rate per 100,000 men 15–49 (SIM & Census)	36.268	43.815	0.000
% Population Female Age 15–49	0.489	0.489	0.533
Population Density	0.106	0.145	0.000
Ratio of % of women voters who finished primary to % of men voters who finished primary	1.275	1.287	0.057
Per Capita GDP	7.429	7.025	0.101
Per Capita GDP from Agriculture	1.733	2.111	0.000
Per Capita GDP from industry	1.796	1.372	0.001
Municipality has a human rights council	0.008	0.008	0.855
Municipality has a health council	0.854	0.859	0.602
Municipality has a Public Safety Council	0.075	0.084	0.259
Municipality has a Municipal Police Corps	0.140	0.090	0.000
Per Capita Expenditure on Civil Defense	0.484	0.291	0.036
Per Capita Expenditure on Public Security, Information and Intelligence	48.600	51.533	0.201
Per Capita Expenditure on other Defense Functions	0.021	0.012	0.502
Per Capita Expenditure on Basic Health	0.096	0.067	0.492
Per Capita Expenditure on General Health	98.789	112.868	0.000
Per Capita Expenditure on Hospitals	316.090	348.516	0.000
Per Capita Expenditure on Other Health	6.514	5.483	0.207
Expenditures
Per Capita Expenditure on Social Communication	0.535	0.754	0.024
Per Capita Expenditure on Human Rights	0.135	0.136	0.970
Per Capita Municial Federal Transfer	462.543	570.720	0.000
N	3488	2076