Gender Bias in Intrahousehold Allocation: Evidence from an Unintentional Experiment Luis H. B. Braido

Pedro Olinto

Helena S. Perrone

November 09, 2007

Abstract We use a very special data set from a recent Brazilian social program to investigate the existence of gender bias in intrahousehold allocation of resources. The program was designed to make monetary transfers directly to mothers and pregnant women in the poorest households of Brazil. Bureaucratic mistakes, beyond the control of the applicants, have accidentally excluded many households who actually applied and were accepted to the program. These unintentional exclusions formed a control group in the molds of random experiments. This is used here to identify the impact of an exogenous variation in female nonlabor income on household decisions. Contrary to the existing literature, our results do not support the existence of gender-speci…c e¤ects on household decisions. Keywords: Bolsa Alimentação, Gender Bias, Collective Approach, Household Bargain, Unitary Model. JEL Classi…cation: D12, J16, O54, H31.

We are indebted to Marco Bonomo for insightful remarks that considerably improved the paper. Braido: Getulio Vargas Foundation, Rio de Janeiro, Brazil ([email protected]); Olinto: World Bank, Washington DC, USA ([email protected]); Perrone: University of Toulouse, Toulouse, France ([email protected]).

1

1

Introduction

Conditional Cash Transfer (CCT) programs have become pervasive in Latin America and the Caribbean. According to Lindert, Skou…as, and Shapiro (2005), di¤erent CCT programs currently reach 60 million people, which represents about 60% of the extremely poor in this region. These programs have a common design feature: they make monetary transfers directly to the highest ranking woman in the bene…ciary household, usually the mother of all children in the household. This design feature hinges on a growing belief that women present expenditure patterns that are more pro-child than men. Investigating the empirical validity of this belief is therefore of great relevance for the design and implementation of social policies. Studying the possibility of gender bias in the intrahousehold allocation of resources is also important from a theoretical point of view. It allows us to confront di¤erent theories of household behavior. On the one hand, the unitary model of household assumes that preferences of di¤erent members can be represented by a single well-behaved utility function. Therefore, redistributing resources across members should have no e¤ect on the household’s expenditure pattern. On the other hand, di¤erent models (such as the collective model) suggest that an exogenous variation in female nonlabor income a¤ects the bargaining power inside the household and, then, changes the household’s expenditure pattern toward goods that are typically preferred by women. In this paper, we use a unique data set to study the existence of gender-speci…c e¤ects on household income allocation. The data were collected in 2002 in the Northeast of Brazil with the objective of evaluating the impact of one of Brazil’s …rst CCT programs, the Bolsa Alimentação (henceforth, BA). The BA program aimed at reducing infant mortality and nutritional de…ciencies among children from very poor families. It started in 2001 and consisted of monetary transfers to pregnant women and mothers of children aging less than 7 years. As in most similar programs, the recipients of the BA transfers were always female. Therefore, given that the amounts transferred were not negligible (they averaged approximately 8% of the typical household total expenditure), participation in the program did represent a signi…cant exogenous variation in the income accruing to women. This exogenous variation is likely to be perceived as long termed, since the bene…ciary families knew that they were likely to continue to receive additional CCTs from the follow up program, the Bolsa Escola (BE), which was targeted to kids aging from 7 to 14 years and was conditioned on school attendance.1 Note however that even a transitory shock on female income could a¤ect their bargaining power, since it is reasonable to 1

These two programs, BA and BE, were uni…ed in 2003 under the name of Bolsa Família.

2

suppose that poor Brazilian households are credit constrained (which typically results in anxious consumption behavior). Therefore, if human preferences did present a gender-speci…c component and if intrahousehold decision-making power were affected by exogenous changes in the member’s income share, then participation in the BA program should a¤ect resource allocation in households with male and female members, net of pure income e¤ects. The uniqueness of the data lies on the fact that a group of eligible households, who actually applied and were accepted to the program, were randomly and unintentionally excluded from it. The exclusion of eligible households occurred due to three independent reasons, all of them beyond the control of the applicants and not correlated to household unobserved characteristics. First, some …les containing household-identifying information were lost during electronic transmission via the internet to the bank responsible for making the payments (Caixa Econômica Federal — henceforth, CEF). This type of exclusion was due to network problems and then, for each municipality, it is independent of household observed and non-observed characteristics. The second source of exclusion is somewhat exotic. While the software used by the municipal authorities in charge of program registration was adapted to the Portuguese language, the software used in the federal capital by the CEF bank for issuing the bene…ciary identi…cation number and processing payments was not. Thus, because the CEF software was not able to read special characters (such as ç, ~, ´, and ^), households in which at least one member had any special character in the name did not receive an identi…cation number and were not included in the program roster for some time. These characters are commonly found in Brazilian names (e.g., João, José, Ângela, Andréa, Tânia, Mônica) and surnames (e.g., Aragão, Gonçalves, Magalhães, Mendonça, Simões). Furthermore, these names and surnames are homogeneously distributed across the Brazilian population, that is, they are not linked to speci…c ethnic, gender, age, or income groups. Hence, this second exclusion criterion is also exogenous, conditional on the number of members in the household.2 Finally, the third source of exclusion is related to misspelling problems. Many households that applied and were eligible for the BA program had children at schooling age and were enrolled in the Bolsa Escola (BE) program. The federal bank responsible for the transfers of both programs, the CEF, decided that it would issue only one identi…cation number for each family, hoping that in the future this would 2

It will be necessary to control for the household population in the econometric estimations to be carried, because larger households are more likely to have at least one name with special characters, and household size might be correlated to the outcome variables of interest.

3

become a single social security number to be used by all federal programs. Therefore, the CEF blocked the registration of households whose data arriving from the BA registration showed any inconsistency with the information recorded during the earlier BE registration. For instance, if during the BA registration a household member’s name was spelled di¤erently from what was recorded in the BE registration, the entry in the BA program would be frozen until this inconsistency was clari…ed. Since the electronic records in both programs were …lled by sta¤ members, this third type of exclusion is related to administrative problems beyond the control of the applicants. It is thus exogenous, conditional on the number of household members, the municipality where registration was conducted, and previous enrollment status in the BE program. For our purposes, this experiment has two advantages over traditional randomizations in which the program implementation is delayed in some randomly chosen locations. First, it allows one to compare households living in the same location and facing the same relative prices. Second, the excluded households here are less likely to have anticipated future participation. Registration of candidate households was conducted in the …eld by municipal authorities. Both authorities and households did not communicate directly with the CEF bank, and they were aware that the registration did not guarantee participation. The federal government determined quotas for each municipality and announced that cuts were expected depending on the program budget. Therefore, it is likely that the excluded households have interpreted the exclusions as caused by lack of eligibility. The accidental exclusions were not immediately detected by program managers. Once an error was detected and …xed by the CEF bank, the household were reincluded into the program. However, when the survey team went to the …eld to conduct the interviews, less than 2% (19 observations) of the originally excluded households had been reincluded into the BA program. Since the reinclusions were conducted by the CEF sta¤ without any in‡uence from the municipality authorities and households themselves, they are also likely to be conditionally independent of household observed and unobserved characteristics. We measure the impact of BA participation on expenditure patterns in two di¤erent categories of households: (i) the mixed male/female households, formed by decision makers of di¤erent genders; and (ii) the female households, which contains no male adult and is headed by a woman (typically single mothers living with their children). Gender-speci…c bargaining e¤ects should be present only in the …rst category of households. The results indicate that, for households with male and female adults, the BA participants present expenditure patterns that are statistically identical to the nonparticipants. Moreover, the BA participants in the two household

4

categories present similar expenditure patterns. These results do not support the joint hypothesis that human preferences have a gender-speci…c component and that households are better modeled by a nonunitary model. The remaining of this paper is organized as follows. We review the empirical literature in Section 2 and describe the data in Section 3. The empirical strategy and results are discussed in Section 4. Section 5 presents a brief conclusion.

2

Literature Review

We review here some empirical tests for the income-pooling prediction of the unitary model. We …rst present papers that use reduced-form econometric models and conventional expenditure data. Next, we discuss the work of Browning, Bourguignon, Chiappori, and Lechene (1994), who derive (and test) predictions from a structural model of household behavior. Finally, we mention three papers which, similarly to ours, rely on natural experiments.

2.1

Reduced-Form Estimates

Most of the existing literature uses reduced-form models to estimate the e¤ect of women leadership on household-choice variables. Handa (1996) presents evidence that female-headed households in Jamaica dedicate a greater budget share to child clothes, health, and food goods, while male-headed households spend more with alcohol and tobacco. Thomas (1990) uses a household survey from Brazil (Estudo Nacional de Despesa Familiar ) to show that nonlabor income in the hands of women have a greater positive e¤ect on children’s anthropometrics than nonlabor income in the hands of men. Thomas (1994), using data from the United States, Brazil, and Ghana, shows that the mothers’education level tend to have a stronger e¤ect on the girls’ height (relatively to boys), while the fathers’ education have a larger impact on the boys’height. Thomas, Contreras, and Frankenberg (2002) use data from the Indonesia Family Survey to study how child morbidities (such as diarrhea, coughs, and fever) are a¤ected by the value of assets that wives and husbands bring to marriage (an indicator of economic independence). The results suggest that the unitary model performs well for most Indonesian regions, but not for Java and Sumatra. In a di¤erent perspective, Pezzin and Schone (1997) study resource allocation in households consisting of elderly parents and their adult children (instead of focusing on di¤erences in decision-making power of wives and husbands). Using the Survey of Assets and Health Dynamics (AHEAD), they …nd that the adult child’s nonlabor

5

income is negatively correlated with (child’s) full-time work and provision of parental care, and positively correlated with the demand for prescription drugs for the parent. In principle, none of the results above support the unitary model of household behavior. However, endogeneity of the leadership variables (e.g., headship, income share, education, and asset ownership) potentially bias the results. For instance, it could be possible that more altruistic women tended to marry men with weak leadership characteristics in order to divert resources towards children-assignable goods. Du‡o (2003) attempts to overcome the endogeneity problem by using instrumental variables. She studies the di¤erential impact of South African old age program on nutritional status of children by comparing anthropometrical measures of children that live in households where there is a male pension recipient (typically the grandfather) and where there is a female recipient (typically the grandmother). Information about whether the child has a grandparent alive is used to instrument the variable describing the gender of the pension recipient. The results show that pensions received by women have a positive e¤ect on the nutritional status of girls but no e¤ect on boys, while pensions received by men have no impact on the nutritional status of either girls or boys.

2.2

Structural Estimates

Browning, Bourguignon, Chiappori, and Lechene (1994) model household decisions under the assumption that the interactions among household members with di¤erent preferences would lead to Pareto e¢ cient outcomes. The parameters of the structural model are estimated using Canadian data on couples with no children. The main goal of the analysis is to investigate how …nal outcomes depend on the income each member brings into the household. They also compare expenditure behavior in single-person households with that of couples. They …nd that older and higher-income partners are able to divert a higher share of total household expenditure towards their own consumption. Besides, holding age di¤erences and relative income shares constant, women are able to divert more income towards her own consumption in wealthier households.

2.3

Natural Experiments

We are aware of three works which, similarly to ours, use natural experiments to test the incoming-pooling hypothesis. Lundberg, Pollak, and Wales (1997) use the 1979 reform in the UK family allowance policy, which transferred a substantial amount of income to the hands of the mothers. Their data include the average con-

6

sumption of clothes for di¤erent family categories (de…ned according to income and number of children). They use the period 1973-1976 to represent the consumption regime before the policy change and the period 1980-1990 to represent the regime after the reform. Comparing weighted averages, they …nd that the ratio of child/male and female/male clothing expenditures increased in the households bene…ted by the program. This analysis was limited to clothing expenditures and do not take into account eventual changes in relative prices over the years studied. In a recent working paper, Ward-Batts (2003) uses this same data source, but disaggregated to the household level, to test the e¤ect of the 1979 reform in the UK family allowance policy on household expenditure shares. The periods 19731976 and 1980-1983 represent the regimes before and after the reform. She uses data on households with: (i) one man; (ii) one woman aged less than 60 years and recorded as the wife of the household head; and (iii) one to three children (aged less than 18 years). She ends up with a sample consisting of 18,810 households over the 8 years studied in the paper. A demand system is estimated using data on consumption expenditures, a price index for each category of goods, and household characteristics. The demand curves of bene…ciary households shifted up for childrenassignable goods (such as clothes and toys) and shifted down for men’s clothing and tobacco. Attanasio and Lechene (2002) examine the e¤ects of participation in a cash transfer program, the PROGRESSA, which started in 1998 in rural Mexico. The transfer is conditional on school enrollment and the mother is always the recipient. The program only started one and a half years later for a number of randomly selected villages that form the control group. The authors show that self-reported decision-making power varies across recipients, non-recipients, and future recipients in the control villages. They also perform a more conventional test using data on expenditure shares. Eight types of non-durable expenditures are considered: food, alcohol and tobacco, transportation, services, and clothing for women, men, girls, and boys. The results indicate a positive impact of female income on girls’and boys’ clothing and a negative impact on alcohol. Our work is directly related to these three last papers. The main di¤erence from the before-after exercises in Lundberg, Pollak, and Wales (1997) and Ward-Batts (2003) is that we compare treated and non-treated households in the same period. When compared to Attanasio and Lechene (2002), our exercise uses treated and non-treated households living in the same municipality. These two features allow us to avoid concerns about eventual changes in relative prices over time or across municipalities. In the case studied here, the results do not reject the income pooling hypothesis.

7

3

Data

We use data from a survey conducted by the International Food Policy Research Institute (IFPRI). This research was contracted by the Inter-American Development Bank (IDB) and the Brazilian Ministry of Health in order to evaluate the impact of the BA program on several nutritional and health outcomes. The BA program consisted of cash transfers to low-income families with pregnant women or mothers of children aging less than 7 years. It started in 2001 and eligible households were those with estimated monthly per capita income of 90 reais (US$ 37.50) or lower. The mother or the highest ranking woman in the household was the sole recipient of this cash transfer, which value ranges from 15 to 45 reais per month (i.e., $6.25-$18.75), depending on the number of qualifying children in the household. The transfers were conditional on women committing to a ‘Charter of Responsibilities’ that commanded regular attendance at prenatal care and child growth monitoring, and compliance with vaccination schedules and health education classes. Eligibility for the BA expired when children completed 7 years of age. Poor families with children aging 7 to 14 years would then become eligible for the Bolsa Escola (BE) program, which ensured continued cash transfer of the same amount as the BA program, but conditional on school attendance. As in the BA program, the transfer recipient was always the mother or the highest ranking woman in the household. The special feature of the data is the existence of a control group formed by households who had actually applied for the program and were eligible to bene…t from it, but were unintentionally excluded. Three types of accidental exclusion were detected: (i) some household …les were lost during the electronic transmission to the CEF— the national bank responsible for issuing identi…cation numbers and transferring the payments; (ii) the computer program used by the CEF was not adapted for the Portuguese language and then excluded households in which one or more members had special characters in their names (e.g., José, Ângela, and Assunção); and (iii) some households previously enrolled in the BE program were excluded from the BA program because the CEF blocked the registration of those whose data coming from the BA registration showed any inconsistency with the data previously recorded for the BE registration. Note that, for each municipality, the …rst type of exclusion is completely exogenous. The second type of random exclusion, however, is only conditionally exogenous. That is, it is exogenous conditional on the number of household members, since larger households were more likely to have at least one name with a special character. Likewise, the third type of exclusion is also exogenous conditional on the

8

number of household members, the municipality where registration was conducted, and previous enrollment status in the BE program. It is also worth stressing that the accidentally excluded households were aware that registration did not guarantee participation, since cuts were expected to be made depending on the program budget allocated to each municipality. Nevertheless, the municipality authorities were asked to register considerably more than the speci…ed quotas because many of the registered households were likely to be found not eligible if their per capita incomes were estimated to be above the cuto¤ level. In addition, households were not able to communicate directly with the CEF. Therefore, it is plausible that households typically interpreted these accidental exclusions as being caused by the lack of eligibility and did not act on it. When the surveys were …nally …elded, around 2% of the households in the sample that had been accidentally excluded had already been reincluded into the program and have reported receiving transfers. Nevertheless, it is likely that these few reinclusions are independent of household characteristics. They were conducted by the CEF sta¤ who did not communicate directly with the households.

3.1

Survey Design

Excluded households were found in 67 Brazilian municipalities. Two criteria were established for the selection of the municipalities to be included in the evaluation study: (i) only municipalities from the Northeastern region were to be included in the study, given that 60% of the bene…ciaries resided in this region; (ii) for cost saving reasons, only municipalities with at least 40 excluded families participating in the program for at least six months were surveyed. In April of 2002, when the survey team went to the …eld, there were four municipalities that …t those two criteria: Teotônio Villela, in the state of Alagoas; Mossoró, in the state of Rio Grande do Norte; Itabuna and Teixeira de Freitas in the state of Bahia. In terms of the number of randomly excluded households, Teixeira de Freitas, with 240 exclusions, was the municipality with the largest sample, followed by Mossoró with 116 excluded applicants, Itabuna with 87, and Teotônio Vilela with 63. To ensure the comparability of the two groups, a sample of matching pregnant women and children was selected from the roster of receiving bene…ciaries. To increase the power of the statistical tests, two bene…ciaries were matched to each excluded woman and child. The matching criteria used were the following: (i) residence in the same municipality; (ii) same gender and age; (iii) similar socioeconomic characteristics. After a pool of included and excluded households was matched by criteria (i) and (ii), the data collected during the registration process were used to 9

further improving the matching in terms of socioeconomic characteristics (criterion iii). The variables used were: declared per-capita income, number of household members, rent paid, and the value of water, electricity, and gas expenses (all variables collected before the BA program had started). Naturally, the households’ observed characteristics are balanced across BA participants and nonparticipants, since they have been incorporated into the sample design. Morris et al. (2004) and IFPRI (2006) present further details on the matching mechanism employed.

3.2

Summary Statistics

The resulting database contains detailed information on household consumption of a great variety of food and non-food items. Information about expenditure patterns was aggregated in nine expenditure groups:3 Food (grains,vegetables and fruits, meat, dairy, salt, sugar, oil, spices, soft drinks, etc.); Health & Education (school fees, schooling material, uniforms, and expenses with hospital, doctors, health plans); Child Clothes; Adult Clothes; Alcohol, Tobacco & Gambling; Bars & Restaurants (food and beverages consumed in bars and restaurants); Utilities (water, telephone, gas, and electricity); Durables & Services (articles of personal hygiene, house-cleaning products, maids, lawyers, cooking articles, home maintenance, furniture, home insurance, mobile phone, weddings, donations, and funerals); Transport & Obligations (gasoline, auto services, taxes, and pensions). The Household Monthly Expenditure, de…ned as the sum of these expenditures, is used to construct the expenditure shares. Other variables such as the number of household members, a dummy variable describing whether the household were already enrolled on the BE program, and dummies for each municipality are also used in our analysis. Their summary statistics are presented in Table 1. 3

Rent was not included in the expenditures since this variable is poorly measured.

10

[Tables 1] An important feature of the data, used later in our identi…cation strategy, is the presence of two categories of families: (i) the mixed male/female households, formed by male and female adults (older than 18 years) and, in 5 observations, by teenager couples (aging between 14 and 18 years old); and (ii) the female households, which contains no male adult and is headed by a woman. Around 69.3% of each household category are BA participants; and about 88.8% of the sample refers to mixed male/female households. About 73.6% of the female households are composed of single-mothers and their children.4 The next table presents the sample means of each variable for included and excluded households. It is divided in two blocks according to BE enrollment status. Households previously enrolled in the BE program must have at least one child at schooling age; they present higher total expenditure and are slightly more populated. [Tables 2]

4

Expenditure Patterns

The theory of household expenditure behavior can be divided in two broad groups: one predicting that households behave as if they maximized a well de…ned objective function that only depends on the amount of goods consumed (the unitary approach); and another predicting that the distribution of nonlabor earnings across household members would a¤ect the …nal allocation (the collective approach). We brie‡y describe them here. The Unitary Approach Consider a household composed of a male and a female member earning (respectively) ym 0 and yf 0, and consuming L di¤erent goods represented by L L xm 2 R+ and xf 2 R+ . Consumption goods are traded at prices p 2 RL ++ and household preferences are represented by a function V (xm ; xf ). Hence, the optimal consumption bundle must solve: max V (xm ; xf ) s:t: (xm ; xf ) 2 R2L + : p (xm + xf )

ym + y f :

(1)

In this model, the demand for goods depend only on prices and household total income (ym + yf ). If preferences were homothetic, changes in (ym + yf ) would not 4

The sample presents 1 household that is BA participant and has no female member. This is likely due to misreporting, since all families should have one eligible woman. This observation was not included in the analysis.

11

a¤ect the share of income expended on each good consumed by this household. If V were not homothetic, then the income e¤ects associated to changes in (ym + yf ) would a¤ect the household expenditure shares. The Collective Approach Suppose now that household members have distinct preferences and allocate the household income through a bargaining process, where each member bargaining weight depends on the household income composition (ym ; yf ). In this case, the household choices should solve: max um (xm ) + (ym ; yf ) uf (xf ) s:t: (xm ; xf ) 2 R2L ym + y f ; + : p (xm + xf )

(2)

where the female bargaining power (ym ; yf ) is increasing in the female income participation, yf =ym . The demands will now depend on the income composition, since (ym ; yf ) a¤ect the objective function. Changes in (ym ; yf ) would a¤ect the share of income expended in each good due to income e¤ects and bargaining e¤ects resulted from the fact that members weights depend on the income vector, (ym ; yf ).

4.1

Econometric Model

Ideally, we would like to have an experiment where the cash bene…ts were randomly assigned to males and females in di¤erent households. Unfortunately, such an experiment does not exist. Our data present some households with a female bene…ciary and others with no bene…ciary. When studying families with male and female adults, BA participation can impact household consumption patterns through two di¤erent channels: income e¤ects and a potential increase of female bargaining power. Fortunately, the data also contain bene…ciary households with no male adult, the vast majority of which composed of single mothers and their children.5 Gender-speci…c bargaining e¤ects must be absent in the female households. Hence, the e¤ect of the BA program on female households can proxy the income e¤ects. Naturally, this strategy is imperfect since income e¤ects are not expected to be the same across the two household categories.6 Nevertheless, this comparative exercise is still informative. De…ne S as the share of the total expenditures that is allocated by each household to some general expenditure group (say food or child clothing); and let M IX and 5

The results remain identical when we use single-mother households instead of female households. One could also argue that participation into the program a¤ected the divorce rates and, then, the compostion of each category. 6

12

BA be binary indicator variables, where M IX = 1 indicates that the household has male and female decision makers, and BA = 1 denotes participation into the BA program. Consider then the following version of the di¤-in-di¤ model: S=

0

+

1 M IX

BA +

2 BA

+

3 M IX

+ ";

(3)

where j is a real-valued parameter, j 2 f0; 1; 2; 3g, and indicates the iteration of the two dummy variables. The parameter 2 measures the average impact of the BA transfer on S (i.e., the income e¤ect), while 1 captures the di¤erentiated impact of that transfer for mixed male/female households relatively to female households (i.e., the bargaining e¤ect). If participation in the BA program was exogenously assigned, the OLS estimator would consistently identify the parameters of equation (3). However, since BA participation was accidental, identi…cation cannot be taken for granted. There were three sources of accidental exclusions: (i) data loss during electronic transmission; (ii) special characters in the name of some household member; and (iii) inconsistency with previously recorded data, for households already registered in another federal program (the BE program). The …rst type of exclusion is only related to network problems, thus independent of household characteristics (for each municipality). The second type of exclusion is a¤ected by the number of members in the household— the larger the household, the more likely to have a name with some special character. Nevertheless, conditional on the number of household members (which is observed), this type of exclusion is exogenous. Similarly, the third type of exclusion is only exogenous once we control for the number of household members, municipality, and enrollment status in the BE program.7 Let Z be a vector containing variables describing the number of household members, municipality, and participation in the BE program. According to the description of the accidental exclusions, the BA dummy is exogenous conditional on Z. Then, de…ne F as the distribution function describing the exclusion process, namely: Pr (BA = 1 j Z) = F (Z ) ;

(4)

and assume for parsimony that F is represented by the linear probability model: BA = Z + u:

(5)

Given the features of the exclusion process, the term u is fully exogenous, and then orthogonal to Z and all other household characteristics embedded in ". 7

The municipality dummies also account for di¤erent relative prices faced by the families in each municipality.

13

Now, consider the linear projection of " on Z: L (" j Z) = Z ;

(6)

and rewrite equation (3) as: S=

0

+

1 M IX

BA +

2 BA

+

3 M IX

+ Z + v;

(7)

where v = " L (" j Z). By de…nition of linear projection, v is orthogonal to Z. Since u is exogenous, we have that, conditional on Z, the independent variables in (7) are orthogonal to v. Hence, the OLS estimator consistently identi…es the parameters of equation (7).

4.2

Estimation Results

We estimate equation (7) for nine di¤erent categories of goods. The results are presented in Table 3. As mentioned before, the coe¢ cient associated with the dummy BA M IX measures the bargaining e¤ect. It is not statistically di¤erent from zero in any regression. Moreover, the estimated coe¢ cients are negative for expenditure categories typically associated with children (such as Health & Education and Child Clothes). This suggests that the impact of the exogenous participation of mixed male/female households into the BA program did not distorted the expenditures towards good typically preferred by women, as one would expect from the previous results in the literature. Note also that the coe¢ cient associated with the BA dummy (which measures the income e¤ect) is also statistically non-signi…cant at the 5% level in all regressions. This is consistent with homothetic preferences. [Table 3] 4.2.1

Excluding BE Participants

One could expect that participation in the BA program had di¤erentiated impact on households that were already enrolled in the BE program. This could happen either because female decision-making power had already occurred when the BE transfers started or because these households suspected they were truly eligible and would be included in the BA program in the future. Moreover, although the electronic …les were …lled by the program’s sta¤, some individual characteristics (such as illiteracy) could have increased the probability of spelling problems. For these reasons, we re-estimated the regressions excluding BE participants from the sample. The results are qualitatively identical to the previous ones, as shown in Table 4. The coe¢ cients associated with the BA and BA M IX dummies are not statistically signi…cant in all regressions. 14

[Table 4]

5

Conclusion

We study gender bias in the intrahousehold allocation of resources using data from a recent social program called Bolsa Alimentação (BA). The BA program was designed to reduce nutritional de…ciencies and infant mortality among the poorest households in Brazil. It relies on demand-side incentives by means of money transfers to pregnant women and mothers of young children in very low-income families. Due to bureaucratic mistakes, many eligible applicants did not receive the cash bene…t. These unintentional exclusions formed a control group in the molds of random experiments. Our results do not support the idea that household consumption decisions are a¤ected by exogenous changes in the income share of male and female members.

References [1] Attanasio, Orazio and Valérie Lechene (2002). “Tests of Income Pooling in Household Decisions.” Review of Economic Dynamics, 5 (4), 720-748. [2] Browning, Martin (1992). “Children and Household Economic Behavior.”Journal of Economic Literature, 30 (3), 1434-1475. [3] Browning, Martin, François Bourguignon, Pierre-André Chiappori, and Valérie Lechene (1994). “Income and Outcomes: A Structural Model of Intrahousehold Allocation.” Journal of Political Economy, 102 (6), 1067-1096. [4] Du‡o, Esther (2003). “Grandmothers and Granddaughters: Old Pension and Intra-Household Allocation in South Africa.” World Bank Economic Review, 17 (1), 1-25. [5] Foster, Andrew (1998). “Marriage-Market Selection and Human Capital Allocation in Rural Bangladesh.” Mimeo: University of Pennsylvania. [6] Handa, Sudhanshu (1996). “Expenditure Behavior and Children’s Welfare: An Analysis of Female Headed Households in Jamaica.” Journal of Development Economics, 50 (1), 165-187. [7] IFPRI (2006). “Estudo de Avaliação de Impacto para a Bolsa Alimentação.” International Food Policy Research Institute, Food Consumption and Nutrition Division, Washington, D.C. 15

[8] Lindert, Kathy, Emmanuel Skou…as, and Joseph Shapiro (2005). “Redistributing Income to the Poor and the Rich: Public Transfers in Latin America and the Caribbean.” Mimeo: World Bank. [9] Lundberg, Shelly J., Robert A. Pollak, and Terence J. Wales (1997). “Do Husbands and Wives Pool Resources?: Evidence from the UK Child Bene…t.”Journal of Human Resources, 32 (3), 463-480. [10] Morris, Saul S., Pedro Olinto, Rafael Flores, Eduardo A.F. Nilson, and Ana C. Figueiró (2004). “Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil.” Journal of Nutrition, 114, 2336-2341. [11] Pezzin, Liliana E. and Barbara S. Schone (1997). “The Allocation of Resources in Intragenerational Households: Adult Children and Their Elderly Parents.” American Economic Review Papers and Proceedings, 87 (2), 460-464. [12] Thomas, Duncan (1990). “Intra-Household Resource Allocation: An Inferential Approach.” Journal of Human Resources, 25 (4), 635-664. [13] Thomas, Duncan (1994). “Like Father, Like Son; Like Mother, Like Daughter: Parental Resources and Child Height.” Journal of Human Resources, 29 (4), 950-988. [14] Thomas, Duncan, Dante Contreras, and Elizabeth Frankenberg (2002). “Distribution of Power within the Household and Child Health.” Mimeo: UCLA. [15] Ward-Batts, Jennifer (2003). “Out of the Wallet and into the Purse: Using Micro Data to Test Income Pooling.” Mimeo: Claremont McKenna College.

16

Table 1. Summary Statistics – Full Sample Full Sample Mean

(S. Dev.)

Min.

Max.

Freq. of Zeros

Food

0.654

(0.136)

0

0.944

0.001

Health & Education

0.042

(0.055)

0

0.379

0.345

Child Clothes

0.023

(0.042)

0

0.519

0.583

Adult Clothes

0.014

(0.038)

0

0.289

0.789

Alcohol, Tobacco & Gambling

0.016

(0.032)

0

0.328

0.522

Bars & Restaurants

0.006

(0.017)

0

0.189

0.753

Utilities

0.122

(0.078)

0

0.529

0.026

Durables & Services

0.091

(0.066)

0

0.517

0.011

Transport & Obligations

0.031

(0.048)

0

0.373

0.381

Monthly Expenditure (in R$)

373.5

(166.6)

29

1,116.7

0

Number of Household Members

5.50

(2.160)

2

16.0

0

Participation in the BE Program

0.346

(0.476)

0

1.000

--

Mossoró

0.268

(0.443)

0

1.000

--

Teixeira de Freitas

0.428

(0.495)

0

1.000

--

Teotônio Villela

0.133

(0.340)

0

1.000

--

Itabuna

0.171

(0.377)

0

1.000

--

Expenditure Shares:

Municipality Dummies:

Sample Size Note: Data from the IFPRI.

1,005

Table 2. Summary Statistics – Subsamples Households Not Enrolled in the BE Program

Households Previously Enrolled in the BE Program

Treated Group (BA=1)

Control Group (BA=0)

Treated Group (BA=1)

Control Group (BA=0)

Mean

(S. Dev.)

Mean

(S. Dev.)

Mean (S. Dev.)

Mean (S. Dev.)

Food

0.655

(0.134)

0.630

(0.157)

0.678 (0.122)

0.647 (0.130)

Health & Education

0.039

(0.054)

0.048

(0.061)

0.043 (0.052)

0.041 (0.056)

Child Clothes

0.027

(0.047)

0.020

(0.035)

0.021 (0.037)

0.019 (0.038)

Adult Clothes

0.015

(0.040)

0.014

(0.039)

0.011 (0.032)

0.016 (0.037)

Alcohol, Tobacco & Gambling

0.015

(0.032)

0.021

(0.039)

0.015 (0.029)

0.015 (0.028)

Bars & Restaurants

0.007

(0.017)

0.008

(0.024)

0.005 (0.013)

0.006 (0.017)

Utilities

0.121

(0.076)

0.132

(0.085)

0.111 (0.068)

0.129 (0.087)

Durables & Services

0.092

(0.067)

0.096

(0.074)

0.080 (0.057)

0.093 (0.062)

Transport & Obligations

0.029

(0.042)

0.030

(0.046)

0.035 (0.055)

0.034 (0.057)

372.1

(162.6)

363.4

(169.6)

384.2 (149.1)

374.7 (194.2)

(2.21)

6.6 (2.2)

6.5 (2.1)

Expenditure Shares:

Monthly Expenditure Numb. Household Members

4.9

(1.89)

5.0

Municipality Dummies: Mossoró

0.262

(0.440)

0.268

(0.445)

0.324 (0.469)

0.219 (0.415)

Teixeira de Freitas

0.468

(0.499)

0.416

(0.495)

0.340 (0.475)

0.438 (0.498)

Teotônio Villela

0.142

(0.349)

0.128

(0.335)

0.138 (0.346)

0.106 (0.309)

Itabuna

0.136

(0.343)

0.188

(0.392)

0.197 (0.399)

0.234 (0.427)

Sample Size Note: Data from the IFPRI.

508

149

188

160

Table 3. OLS Regressions of Expenditure Shares

Food

Health & Educat.

Child Clothes

Adult Clothes

Tobacco, Alcohol & Gamb.

Bars & Restaur.

Utilities

Durab. & Services

Transport & Obligat.

BA*MIX (Bargaining Effect)

0.016 (0.038)

-0.022 (0.014)

-0.007 (0.008)

-0.004 (0.01)

0.003 (0.005)

-0.007 (0.005)

0.029 (0.022)

0.002 (0.021)

-0.011 (0.008)

BA Dummy (Income Effect)

0.01 (0.036)

0.015 (0.013)

0.011 (0.007)

0.002 (0.009)

-0.006 (0.005)

0.005 (0.005)

-0.037* (0.021)

-0.009 (0.020)

0.009 (0.007)

MIX Dummy

0.004 (0.033)

0.009 (0.011)

0.006 (0.006)

0.002 (0.009)

0.006 (0.005)

0.003 (0.004)

-0.034* (0.019)

-0.012 (0.018)

0.017*** (0.006)

0.006*** (0.002)

-0.001* (0.001)

-0.001 (0.001)

0.001 (0.001)

0.000 (0.000)

-0.001*** (0.000)

-0.003*** (0.001)

-0.001 (0.001)

0.001 (0.001)

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

1,005

1,005

1,005

1,005

1,005

1,005

1,005

1,005

1,005

Numb. Household Members

Constant, BE Participation and Municipality Dummies Sample Size

Note: Robust standard deviation in parenthesis; + significant at 10%; * significant at 5%; ** significant at 1%.

Table 4. OLS Regressions of Expenditure Shares for Households Not Enrolled in the BE Program

Food

Health & Educat.

Child Clothes

Adult Clothes

Tobacco, Alcohol & Gamb.

Bars & Restaur.

Utilities

Durab. & Services

Transport & Obligat.

BA*MIX (Bargaining Effect)

0.022 (0.049)

-0.01 (0.018)

0.002 (0.011)

0.003 (0.015)

0.002 (0.008)

-0.006 (0.007)

0.004 (0.026)

-0.01 (0.025)

-0.006 (0.011)

BA Dummy (Income Effect)

0.004 (0.046)

-0.001 (0.018)

0.004 (0.010)

-0.002 (0.015)

-0.007 (0.007)

0.004 (0.007)

-0.014 (0.025)

0.006 (0.024)

0.005 (0.01)

MIX Dummy

0.003 (0.042)

0.002 (0.015)

-0.004 (0.009)

-0.008 (0.014)

0.008 (0.007)

0 (0.006)

-0.001 (0.023)

-0.01 (0.020)

0.01 (0.009)

Household Members

0.005* (0.003)

-0.001 (0.001)

-0.001 (0.002)

0.000 (0.001)

0.001 (0.001)

-0.001*** (0.000)

-0.002* (0.001)

-0.001 (0.001)

0.001 (0.001)

Constant, BE Participation and Municipality Dummies

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Sample Size

657

657

657

657

657

657

657

657

657

Note: Robust standard deviation in parenthesis; + significant at 10%; * significant at 5%; ** significant at 1%.