Conditional cash transfers and birth order effects : the Brazilian case

UNIVERSITEIT VAN AMSTERDAM Faculty of Economic and Business MSc Development Economics Thesis

Conditional Cash Transfers and Birth Order Effects: The Brazilian Case

Abstract

This paper examines the effect of conditional cash transfer program eligibility on birth-order effects. The paper relies on a sharp discontinuity in the eligibility for the Brazilian CCT program Bolsa Família to and focus on the relationship between birth order and school enrollment and child labor. The findings are that families eligible to receive the CCT present positive and mostly significant links between birth order and enrollment. Regarding child-labor outcome, although the estimates are ambiguous, the analyses of possible sources of heterogeneity shed some light on the mechanisms behind the family’s choice regarding their children’s time allocation and how a CCT can increase its

efficiency by incorporating differences in birth order on their payoff structure

Marcelo Vieira de Campos

Statement of Originality

This document is written by Marcelo Vieira de Campos who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Table&of&Contents&

&

1.&Introduction&

During the past decades the interest regarding the role that the birth-order has in the household decision of allocating their time has increased. Originally psychologists and sociologists gave the subject attention, driving economists to also address the issue, but mostly in developed scenarios. The birth order emerges in this context as one of the conceivable factors that helps the family decide which children will be sent to work and which will continue their formal education. A large part of the evidence of previous research points towards significant birth order effects in several outcomes that favor earlier born children.

The starting point is the assumption that earlier born children do not have to compete with their siblings for their parent’s time and other family’s resources. This would result in the first-born achieving better outcomes such as higher wages later in life.

A large part of the previous research, like Kessler (1991), Black, Devereux and Salvanes (2005) or De Haan (2010) however were conducted in the setting of a

developed country. This paper proposes a similar research in terms of analyzing the effects of birth order, but in Brazil that provides for a developing country setting.

The idea that birth order effects behave differently in a context where child labor is frequent was proposed by Basu and Van (1998). They argue that child labor is most commonly a consequence of poverty because parents are more prone to forfeit their child’s education in favor of the labor market if facing capital constraints. When this is the case, the likelihood of elder siblings participating in child labor is larger due to the fact that they can command a bigger wage in the market. This leads to a positive correlation between the order of birth and human capital achievements.

In one instance, if earlier born children have higher abilities it can mean that they would have higher returns to education than later born children, which might lead the decision to keep them in school. On the other hand, an earlier-born child with higher abilities may also earn a higher wage in the labor market than the later born, making them

more likely to be sent to the labor market. The decision is mostly dependent on the time value of money, once a higher education is an investment.

Emerson and Souza (2008) found, using Brazilian data from 1998, that first-borns are less likely to attend school than their later born sibling and male last-born children are less likely to work as child laborers that their earlier born siblings. De Haan, Plug, and Rosero (2014) relied on family-fixed effects model and found, using Ecuador data, that later-born children are ahead in their cognitive development, are more likely to attend school in their adolescence and to spend more time with their mothers on cognitive activities. The estimates point that birth order effects for poor families are persistently positive. Ejrnaes and Portner(2004), using Philippines data and an intra-household allocation model with endogenous fertility found that higher birth orders have an advantage over siblings with lower birth order. Most of the studies using data from a developing country found that the negative relationship between birth order and human capital attainment sometimes does not hold, specifically when families face tighter constraints.

Another influence to the time and resource allocation within the household in developing countries are conditional cash transfer programs, especially when the school enrollment and attendance are the conditionalities. In this context families must decide if it is in their best interest to keep their children in school and receive an additional income from the government or send them to the labor market.

The aim of this paper is to examine the impact of the Brazilian conditional cash transfer program Bolsa Família on the dynamics time allocation between children within a family that are evidenced by child-labor incidence and school enrollment.

In order to estimate the effect of the CCT program on the birth order effects, this paper will take advantage of the sharp discontinuity in the eligibility criteria for the CCT program, making it possible through the use of non-parametric methods to examine how the birth order effects behave around this threshold.

The paper will be structured as follows: Section 2 will discuss the link between birth order, child labor and time allocation within a family, section 3 will present the background of the CCT program Bolsa Família in Brazil, section 4 will address the data

set and sample selection, section 5 the identification strategy, section 6 the empirical results, followed by the conclusion.

2.&Theoretical&Framework&

A useful and simple way to begin approaching this problem the household faces when allocating their children’s time is to follow the approach of Emerson and Souza (2008), of the household utility function as the current consumption of the family, the human capital attainment of each child and the leisure time of each child. Conflicting incentives for the children’s time are inherently in the models assumption.

If children go to the labor market they can expand the domestic current

consumption, but they will be forfeiting their education and hence, their human capital will be smaller as well as their prospect wages in the labor market. If they go to school they can increase their human capital, which will increase their earnings in the future. If they choose to engage in leisure activities, they can contribute to the happiness level of the family. Although these activities are not exclusive, children that work, especially the ones coming from a poor family, may not be able to obtain the same level of education as they would if they did not had to work.

According to Emerson e Souza (2008) the decision of time allocation for each child needs to take into account the marginal effect that education has on human capital accumulation, how this additional human capital affects the household utility function and the returns of the children for their labor.

It is important to stress that a considerable part of the birth-order effect lies in the fact that it is well established in the literature that wages usually increase with the age of the child; consequently the child labor wage is positively correlated with the child birth-order.

In summary, if what the household would gain from the higher wage that can be commanded by the first-born compared to the later-born is larger than the potentially higher returns to schooling of the first-born, then it would be expected that earlier born children are more likely to work and less likely to attend school, which implies a positive birth-order effect.

Another circumstance where a positive relation between birth-order and

educational outcomes can emerge is situations where the household is too poor to send their first-born to school. The additional income brought by the first-born can make it possible for the second-born to attend school.

To comprehend how impactful the decision of sending a child to the labor market is, the future well being of his family and their next generations would have to be taken into account. Poverty typically shifts the attention of a family to the current consumption, putting pressure on the family to re-allocate their children to the labor market instead of school and leisure. Due to the positive link of labor income and schooling, it is easy to see that children who went to the labor market instead of to school will earn less as an adult, making them more likely to send their own children to work to complement their income. One common goal of Conditional Cash Transfers is to break this vicious cycle that builds through the inter-generational transmission of poverty.

Regarding CCTs policy implications, Janvry and Sadoulet (2006) argue, based on data from Mexico’s Progressa, that CCT programs need to take into account the birth-order effects to achieve a more efficient targeting. If the birth-birth-order is positively linked to education attainment, the CCT transfers should decrease its value with the birth-order to balance the positive effect of the birth order. Another implication regards differences in transfers for boys and girls. The study suggests that the optimal transfer for girls should be greater than for boys in order to mitigate the effects that arise from gender inequality.

Ferreira, Filmer and Schady (2009) use evidence from a CCT program in

Cambodia to argue that to be properly implemented and targeted, the CCT should look at the effect that the transfers have on other children within a family. This effect can be uncertain and depends the interaction between the positive income effect and the negative displacement effect.

De Haan, Plug, and Rosero (2014) argues that if the well being of the children is the main focus of CCT programs, than it should be designed with an emphasis in

improving the condition of the earlier-born children and the evaluation of the program success should also take into account this different responses of children within a family

3.&Bolsa&Família&Program&&

Numerous programs in developing countries are based on transfers to poor

households conditional on the school enrollment of school-aged children. These programs have been shown to increase school attendance in a variety of settings. The Brazilian Conditional-Cash transfer (CCT) program Bolsa Família, from now on referred to as BFP (Bolsa Família Program), was set-up by former president Luis Inácio Lula da Silva in 2004.

According to Reimers et al (2006) president Lula integrated several government cash transfer programs that aimed at increasing the school attendance of children, assisting families in need that lacked sufficient nutrition and provide universal access to gas and electricity through monthly grants to the families. This integration process of the previous CCT programs resulted in reduction of administrative costs and bureaucratic complexity, making it possible for the BFP to experience a quick expansion during the past decade.

According to Soares, Ribas and Osorio (2010), the PBF united the insertion and elimination criteria, the transfer amounts, the ministry responsible for the program implementation and the information system.

The eligibility criteria for the BFP must respect the following rules; Families with monthly per capita income (MFPCI) ranging from 70BRL and 154BRL, with children among 0 and 17 years old, conditional on a school attendance of 85% minimum and families with monthly per capita income of up to 70BRL, regardless of the family composition.

When looking only at children below 17 years old, the discontinuity in the eligibility criteria is turned into a sharp one, which will be helpful in the identification strategy that will be presented further.

The World Bank sees the BFP as an enormous success; BFP helped to reduce extreme poverty from 9.7% when the program was implemented to 4.3% of the population in 2013, the Gini inequality measure experienced a drop of more than 12% reaching the country’s lowest level of inequality since the measure began being applied. BFP is held by many as one of the main drivers of the recent fall in inequality, explaining 10 to 20% of the variation in inequality measured by the Gini coefficient depending on

the study method. Barros (2010) found!that!despite!representing!just!0.5%!of!

household!income,!BFP!contributed!10%!to!the!decline!of!inequality!between!2001! and!2007.Another positive impact that the BFP achieved is the break in the transmission of poverty from parents to their children. This was made possible by the education and health conditionalities of BFP.

Regarding the effect that BFP has on enrollment, Glewwe and Kassouf (2008) found that the program increase enrollment by 5.5% in the primary education and by 6.5% in the secondary education and lowered the dropout rates by 0.5%.

The objective of this paper is to see how the birth order effects behave across the eligibility threshold, and how Bolsa Família can incorporate this into their framework, to enhance their targeting and the marginal benefit of the transfer.

4.&Data&and&Sample&Selection&

The Data set that will be used in this study comes from the annual household survey in Brazil, called Pesquisa Nacional por Amostra do Domicilios, from now on referred to as PNAD. The survey is conducted by the Instituto Brasileiro de Geografia e

Estatística (IBGE) and is a nationally representative survey, with the exclusion of rural

zones in the north of the country. In the survey questions regarding the geographic location, characteristics of the dwelling, the size of the household, the relationship between the individuals that live in the same household and their income from labor or other sources and other individual characteristics such as age, gender and education. In the year of 2014 over 150 thousand households were surveyed, accounting for over 350 thousand individuals.

The main sample of this study consists of 73,878 individuals between 7 and 16 years old. Families with less than 2 children or more than 4 were excluded from the sample. The exclusion of children younger than 7 years old is due to the fact that

compulsory schooling begins at this age in Brazil, and older than 16 due to labor market legislation that precludes children below the age of 15 from participating in the labor market. This selection of the sample is also useful to the identification strategy since it

provides a sharp discontinuity in the eligibility criteria for the CCT program Bolsa

Família.

One shortcoming of PNAD is that it does not present the information of whether a family actually receives the BFP transfers, hence the reliance in the assumption that all families that are eligible receive the transfers.

The education outcome variable of interest is the enrollment status of the children and the variable is binary, taking the value of 1 if the individual is enrolled in the current year in the educational system and 0 if the individual was not enrolled in the current year.

The labor market outcome variable of interest is whether the individual has worked in the previous week, taking the value of 1 if the individual realized any type of labor activity outside the household or 0 otherwise.

Table 1 provides some descriptive statistics for the treatment and control groups. The treatment group is composed by children that have a monthly family per capita income ranging between 124 BRL and 154 BRL, and the control group of children with monthly family per capita income ranging between 155 and 184. This particular sample selection will be helpful later on by supporting the identification strategy

(Table 1- Descriptive statistics)

! BFP!eligible!(Bandwidth!=30;!124<MFPCI<=154BRL)! BFP!ineligible!(Bandwidth!=30;!154<MFPCI<=184BRL)!

Outcome!Variables! Mean! SD! Within!family!

SD! Mean! SD! Within!family!SD! Enrollment! 0.782! 0.413! 0.323! 0.736! 0.440! 0.327! Child!Labor! 0.074! 0.261! 0.197! 0.097! 0.296! 0.216! Child!Characteristics! ! ! ! ! ! ! FirstSborn! 0.319! 0.466! 0.439! 0.342! 0.474! 0.448! SecondSBorn! 0.314! 0.464! 0.442! 0.337! 0.472! 0.453! ThirdSBorn! 0.198! 0.398! 0.372! 0.195! 0.396! 0.368! ForthSBorn! 0.101! 0.301! 0.276! 0.082! 0.274! 0.248! Gender(1=Female)! 0.477! 0.499! 0.394! 0.484! 0.499! 0.399! Age(months)! 409.720! 2620.800! 1649.400! 357.200! 2379.300! 1664.600! Family! Characteristics! ! ! ! ! ! ! HH!schooling! 6.230! 4.220! ! 6.850! 3.890! ! Mother!schooling! 4.880! 4.170! ! 5.350! 4.410! ! Age!mother!(years)! 29.010! 19.210! ! 29.160! 19.330! ! Teenage!Mother! 0.418! 0.493! ! 0.410! 0.491! ! Number!of!children! 3.700! 1.456! ! 3.380! 1.350! !

5.#Methodology#

5.1#Sharp#Discontinuity#Design#

!

Following Imbens and Lemieux (2008), the Sharp regression discontinuity design relies on the assignment variable (!_!) being a deterministic function of one of the

covariates, the forcing variable (!_!).

!! = 1 !! ≥ !

!_! 0 is the outcome that occurs in the absence of treatment and !_! 1 the outcome in the presence of the treatment.

!_! = ! !_! ∙ !_! 1 + 1 − !_! ∙ !_! 0 = !! 0 !!!!!"!!!_{! 1 !!!!"!!!!}! = 0

! = 1

Observations that lie below the threshold value c are the control group while the ones that are above the threshold are the treated group. Hence, it is possible to estimate the average causal effect at the discontinuity by looking at the conditional expectation of the outcome variable given the forcing variable.

!!"# = lim_!↓!!Ε !! !! = ! −!lim_!↑!Ε[!!| !! = !]

!!"# = Ε ! 1 − !(0) ! = ! = ! ! 1 ! = ! − ![!(0)|! = !]

Exploiting the program eligibility, children with family income below or equal to 154 BRL would be the treated group and having full probability of being eligible for the CCT and children with family income above this threshold have null probability of receiving the CCT.

Lee and Lemieux (2008) argues that as long as there is no sorting or manipulation around the eligibility threshold, one can look at families as being randomly assigned to treatment and control groups.

Given that, by design, no observations of the control group have income equal to 154BRL, there is the need to make a smoothness assumption. According to Imbens and Lemieux (2008), this assumption, formulated in terms of conditional expectation, is that the expectation of both Y(0) and Y(1), given that !_! = ! are continuous in !. This

continuity condition allows the average outcome of children right above the 154BRL eligibility threshold to be a valid counterfactual for children right below the threshold.

The figure below is a plot of the probability of being eligible to the BFP, which depicts the sharp discontinuity in the 154 BRL mark.

FIG (1) – Bolsa Família eligibility and Monthly Household Income per capita

Assuming that all the BFP eligible households actually gets the grant (according to IBGE, the BFP penetration is above 95%) there is the expectation that the outcome variables that will be examined in this papers of whether a child is enrolled or has worked in the past week, will also have a discontinuity around the same threshold.

It is useful to proceed with graphical analysis of the variables of interest and examine how they behave across this threshold. If they are significantly different across the threshold

For robustness, different degrees of local polynomial and different bandwidths were used and the different estimates can be found below.

The following figures represent the regression discontinuity plot for the preferred estimation method, which can be found in column (2) and consists of a local polynomial of order 1 and a rectangular kernel. The bandwidth selection procedure is based on a! common!minimum!squared!error!optimal!bandwidth!selector!for!the!RD!treatment! effect!estimator.

Table&2.A:&Enrollment&vs.&MFPCI&

#Variable# (1)# (2)# (3)# (4)# (5)#

Enrolled#

RD&Estimate& 0.075& 0.096& 0.179& 0.053& 0.061& Robust&95%&CI& [0.152;&0.052]& [0.189;&0.058]& [0.249;&0.109]& [0.348;&0.161]& [0.148;&0.067]& Kernel&Type& Uniform& Uniform& Triangular& Triangular& Uniform& Std.&Error& 0.01681& 0.0216& 0.0357& 0.0109& 0.0153&

PQValue& 0& 0& 0& 0& 0&

Order&loc.&Poly& 1& 1& 2& 1& 2&

Bandwidth&Loc.&Poly&

N& 10.693&50& 6.779&30& 6.779&30& 28.316&150& 28.316&150&

FIG (3) (Worked in the past week vs MFIPC)

Table&2.B&Worked&past&week&vs.&MFPCI&

Variable# (1)# (2)# (3)# (4)# (5)#

# Worked#in#the#Past#Week#

RD&Estimate& Q0.044& Q0.0365& 0.0095& Q0.0252& Q0.049& &95%&CI& [Q0.022;&Q0.065]& [Q0.009;&&Q0.064]& [Q0.056;&0.036]& [Q0.006;&Q0.044]& [Q0.023;&Q0.075]& Kernel&Type& Uniform& Uniform& Triangular& Triangular& Uniform& Std.&Error& 0.0108& 0.0161& 0.026& 0.0096& 0.0131&

PQValue& 0.000& 0.010& 0.686& 0.009& 0.000&

Order&loc.&Poly& 1& 1& 2& 1& 2&

Bandwidth&Loc.&Poly& N& 50& 10.693& 30& 6.779& 30& 6.779& 150& 28.316& 150& 28.316&

Gelman and Imbens (2014) argue against the use of high-order polynomials

approximations for the conditional expectation of the outcome variables of interest given the forcing variable. Their argument lies in the fact that the estimates are sensitive to the order of the polynomial and there are no good methods for deciding which is the optimal order of the

polynomial.

5.2#Internal#Validity#

There are some concerns that must be addressed for implementing a regression discontinuity design.

To further validate the identification strategy, Imbens and Lemieux (2008) advocate in favor looking at possible jumps in the value of other covariates at the cutoff point, testing for discontinuity in the conditional density of the forcing variable and using various values of the bandwidth.

To check whether there is a discontinuity in the conditional density of the forcing variable this paper will follow the test proposed by McCrary (2007). This test is based on an estimator for the discontinuity at the cutoff in the density function of the running variable.

(Fig 4. MFIPC histogram)

McCrary suggests testing the null hypothesis of continuity of the density of the covariate that triggers the assignment at the discontinuity point, against the alternative of a jump in the density function. McCrary test results in a p-value of 0.2535, indicating that the null hypothesis that there is no manipulation at cutoff holds.

Another test to reassure the identification strategy is to check for discontinuities in the conditional expectation of potential covariates. Below there are plots of potential covariates and their respective RD estimate.

(Fig 5. Computer ownership vs. MFPCI) (Fig 6. Refrigerator ownership vs. MFPCI) Rd estimate = 0.016; p-value:!0.365 Rd estimate = -0.006; p-value:0.714

(Fig 7. Home ownership vs. MFPCI) (Fig 8. Car ownership vs. MFPCI) Rd estimate = -0.016; p-value:0.714 Rd estimate = .0086; p-value:!0.644

As it is possible to see rom the graphs above, the presented covariates does not show significant difference in their conditional expectation across the eligibility threshold. This fact goes in the direction to support the assumption that the only thing causing significant differences in enrollment and child labor across the eligibility threshold is in fact the BFP.

5.3#Birth#Order#Effects##

!

In order to check whether the BFP eligibility affects children within a family in different ways regarding the educational and labor market outcomes, the analysis in section 5.1 will be replicated using 4 different subsamples, consisting of the first, second, third and fourth-born.

By continuing the analysis with the five specifications proposed earlier in section 5.1, the aim is to observe if the discontinuity found earlier when accounting for all birth orders, emerges for each specific order of birth.

If the same discontinuity is observed for all birth orders, it is reasonable to think that the eligibility towards the BFP affects all children within a family in a similar way. On the other hand, if significant differences on the behavior of children within a family are found, it would be an indication that the BFP has different impact and effectiveness within children belonging to the same family.

6.#Empirical#results#

We begin this section by presenting, as in section 5.1, the regression discontinuity figures of the preferred specification (2) accompanied by a table presenting the estimates for the other specifications.

The most relevant information to take into account from the following regression discontinuity plots and estimates is whether they behave in a similar manner. The signal of the effect and its significance will provide important insight on the behavior of the family

regarding the decision to send each child to the labor market or to the school.

! ! ! ! !

! (Fig&9.&Enrollment&vs.&MFPCI:&FirstQBorn&Only)& ! Table&3.A:&Enrollment&vs.&MFPCI:&FirstQBorn&Only& Variable# (1)# (2)# (3)# (4)# (5)# # Enrollment#FirstJBorn#Only#

RD&Estimate& 0.0069& Q0.0235& 0.0471& Q0.0061& Q0.0034& &95%&CI& [Q0.0513&;&0.0651]& [Q0.0984&;&&0.0512]& [Q0.0831&;&0.1774]& [Q0.0434;&0.0312]& [Q0.0554&;.04851]& Kernel&Type& Uniform& Uniform& Triangular& Triangular& Uniform&

Std.&Error& 0.0297& 0.0382& 0.06648& 0.01903& 0.02653& PQValue& 0.814& 0.537& 0.479& 0.749& 0.895&

Order&loc.&Poly& 1& 1& 2& 1& 2&

Bandwidth& N& 50& 3.554& 30& 2.254& 30& 2.254& 150& 9.848& 150& 9.848& ! (Fig&10.&Worked&past&week&vs.&MFPCI:&FirstQBorn&Only)& ! Table&3.B&Worked&past&week&vs.&MFPCI:&FirstQBorn&Only& Variable# (1)# (2)# (3)# (4)# (5)# Worked#in#the#past#week#FirstJBorn#Only#

RD&Estimate& 0.03576& 0.07575& 0.17999& 0.04863& 0.04967& &95%&CI& [Q0.0089&;&0.0804]& [0.0178&;&&0.1336]& [0.0734&;&0.2865]& [0.0192;&0.0780]& [0.0091&;&0.0901]& Kernel&Type& Uniform& Uniform& Triangular& Triangular& Uniform&

Std.&Error& 0.0228& 0.02955& 0.0543& 0.01499& 0.02067& PQValue& 0.117& 0.01& 0.001& 0.001& 0.016&

Order&loc.&Poly& 1& 1& 2& 1& 2&

Bandwidth& N& 50& 3.554& 30& 2.254& 30& 2.254& 150& 9.848& 150& 9.848&

(Fig&11.&Enrolled&vs.&MFPCI:&SecondQBorn&Only)&

! Table&4.A:&Enrollment&vs.&MFPCI:&SecondQBorn&Only&

Variable# (1)# (2)# (3)# (4)# (5)#

## Enrollment#SecondJBorn#Only#

RD&Estimate& 0.06405& 0.1207& 0.2545& 0.0588& 0.0702& &95%&CI& [0.0082&;&0.1198]& [0.0482&;&&0.1931]& [0.1347&;&0.3742]& [0.0229&;&0.0948]& [0.0198&;&0.1205]& Kernel&Type& Uniform& Uniform& Triangular& Triangular& Uniform&

Std.&Error& 0.0284& 0.0369& 0.0611& 0.0183& 0.02567&

PQValue& 0.024& 0.001& 0& 0.001& 0.006&

Order&loc.&Poly& 1& 1& 2& 1& 2&

Bandwidth&&

N& 3.509&50& 1.221&30& 1.221&30& 9.771&150& 9.771&150& (Fig&12.&Worked&past&week&&vs.&MFPCI&:&SecondQBorn&Only)&

Table&4.B:&Worked&in&the&past&Week&vs.&MFPCI:&SecondQBorn&Only&

Variable# (1)# (2)# (3)# (4)# (5)#

# Worked#in#the#past#week##SecondJBorn#Only#

RD&Estimate& Q0.07959& Q0.0959& Q0.05988& Q0.02612& Q0.05471& 95%&CI& [Q0.1161&;&Q0.0430]& _{[Q0.1430&;&&Q0.0487]& [Q0.1346&;&0.0148]& [Q0.0496&;&Q0.0025]& [Q0.0866&;&Q0.0227]&}

Kernel&Type& Uniform& Uniform& Triangular& Triangular& Uniform& Std.&Error& 0.0186& 0.02404& 0.03814& 0.012& 0.0163&

PQValue& 0& 0& 0.116& 0.03& 0.001&

Order&loc.&Poly& 1& 1& 2& 1& 2&

(Fig&13.&Enrolled&vs.&MFPCI:&ThirdQBorn&Only)#

Table&5.A:&Enrollment&vs.&MFPCI:&ThirdQBorn&Only#

Variable# (1)# (2)# (3)# (4)# (5)#

# Enrollment#ThirdJBorn#Only#

RD&Estimate& 0.1389& 0.1851& 0.2268& 0.11422& 0.12344& &95%&CI& [0.0670&;&0.2107]& [0.0919&;&&0.2784]& [0.0791&;&0.3743]& [0.0667&;&0.1617]& [0.0579&;&0.1889]& Kernel&Type& Uniform& Uniform& Triangular& Triangular& Uniform&

Std.&Error& 0.0366& 0.0476& 0.0753& 0.02424& 0.03343&

PQValue& 0& 0& 0.003& 0& 0&

Order&loc.&Poly& 1& 1& 2& 1& 2&

Bandwidth&& N& 50& 2.083& 30& 1.333& 30& 1.333& 150& 5.088& 150& 5.088& (Fig&14:&Worked&in&the&past&Week&vs.&MFPCI:&ThirdQBorn&Only)# Table&5.B:&Worked&in&the&past&week&&&vs.&MFPCI:&ThirdQBorn&Only& Variable# (1)# (2)# (3)# (4)# (5)# # Worked#in#the#past#week ThirdJBorn#Only #

RD&Estimate& Q0.11003& Q0.11703& Q0.0962& Q0.06486& Q0.09211& &95%&CI& [Q0.1497&;&Q0.0703]& [Q0.1696&;&&Q0.0644]& [Q0.1806&;&Q0.0117]& [Q0.0914&;&Q0.0382]& [Q0.1284&;&Q0.0558]& Kernel&Type& Uniform& Uniform& Triangular& Triangular& Uniform&

Std.&Error& 0.02026& 0.02685& 0.04308& .01356& 0.01854&

PQValue& 0& 0& 0.026& 0& 0&

Order&loc.&Poly& 1& 1& 2& 1& 2&

Bandwidth&&

N& 2.083&50& 1.333&30& 1.333&30& 5.088&150& 5.088&150& &

(Fig&15:&Worked&in&the&past&Week&vs.&MFPCI:&FourthQBorn&Only)

Table&6.A:&Enrollment&&vs.&MFPCI&:&FourthQBorn&Only#

Variable# (1)# (2)# (3)# (4)# (5)#

# Enrollment#FourthJBorn#Only#

RD&Estimate& 0.22001& 0.23543& 0.2652& 0.14864& 0.1549& &95%&CI& [0.1163&;&0.3236]& [0.1025&;&&0.3682]& [0.0653&;&0.4651]& [0.0784&;&0.2188]& [0.0598&;&0.2499]& Kernel&Type& Uniform& Uniform& Triangular& Triangular& Uniform&

Std.&Error& 0.05289& 0.06778& 0.10198& 0.03581& 0.04852&

PQValue& 0& 0.001& 0.009& 0& 0.001&

Order&loc.&Poly& 1& 1& 2& 1& 2&

Bandwidth&& N& 50& 960& 30& 611& 30& 611& 150& 2.136& 150& 2.136& (Fig&16:&Worked&in&the&past&Week&vs.&MFPCI:&FourthQBorn&Only)# Table&6.B:&Worked&in&the&past&week&&&vs.&MFPCI:&FourthQBorn&Only& Variable# (1)# (2)# (3)# (4)# (5)# # Worked#in#the#past#week##FourthJBorn#Only#

RD&Estimate& Q0.06779& Q0.0769& Q0.04439& Q0.0305& Q0.04713& &95%&CI&

[Q0.1065&;&Q0.0289]& [Q0.1680&;&&0.0141]& [Q0.1026&;&0.0138]& [Q0.0599&;&Q0.0011]& [Q0.0848&;&Q0.0094]&

Kernel&Type& Uniform& Uniform& Triangular& Triangular& Uniform& Std.&Error& 0.0198& .0464& 0.02973& 0.015& 0.01923& PQValue& 0.001& 0.098& 0.135& 0.042& 0.014&

# #

In order to summarize the most relevant findings of the previous tables and graphs, the table 7 presents the preferred regression discontinuity estimates (2) and their respective standard errors for each order of birth regarding the outcomes variables of interest.

#

Table#7:#RD#estimates#preferred#specification

*** significant at 1 percent, ** significant at 5 percent, * significant at 10 percent

Regarding the enrollment variable, a positive pattern emerges from the regression discontinuity estimates. Although the estimates for the first-borns are not significant, it has a different sign than the later-born. Concerning the estimates of the second-born children, those eligible to the BFP have slightly more than 12% higher probability of being enrolled than those ineligible. When looking at later-born children the effect becomes even bigger, with 18,5% for third-born and 23,5% for fourth-born children. It is important to remember that this can only be interpreted as local average treatment effects.

This fact, together with a larger and significant effect for the higher orders of births, is a strong indicative that the BFP has highly heterogeneous effects among siblings when

looking at the probability of being enrolled.

The contrast between the negative estimate for the first-born and the positive signal for the second, third and fourth-born, along with the fact that effect of BFP eligibility becomes larger for higher birth order is in accordance with the ideas proposed by Basu and Van (1998) of a positive correlation between order of birth and human capital achievements.

Order&of&birth& & Enrollment&_(1)& Worked&in&the&Past&Week&_(2)& 1st& _Std.&Error&Estimate& Q0.0235&_(0.0382)& 0.07575**&_(0.02955)&

2nd& _Std.&Error&Estimate& 0.1207***&_(0.0369)& Q0.09591***&_(0.02404)&

3rd& _Std.&Error&Estimate& 0.1851***&_(0.0476)& Q0.11703***&_(0.02685)&

The considerable increase in the effect of the fourth-born in respect to the second-born might be an indicative that the family re-use and recycle old resources that improve the chance of their later-born children attending school, such as books, and school uniform

Turning the attention towards the labor outcome variable, it is clear to see that the estimates reconcile with what was proposed during the theoretical framework section. The effects of BFP eligibility regarding first-borns and labor market have a positive sign and are significant. This means that first-borns in the treatment group, or who are eligible to the BFP, have 7,5% higher probability of being sent to the labor market than the first-borns in the control groups, that are not eligible to the BFP program.

This effect is not observed for higher order of births, in fact, as intended by the program, the BFP eligibility makes it less likely for later-born to be sent to the labor market. Second and Third-born that are eligible to the BFP are 9,5% and 11% less likely of being having worked in the last week in activities out-side the household. These estimates are significant at 99% level of confidence

After analyzing the whole set of regression discontinuity estimates and plots, it is somewhat clear that the eligibility towards the BFP have different effects in children within the same family depending on their order of birth.

The underlying story that emerges from the estimates and plots for birth orders is that the BFP transfers are not enough to provide enough stimulus for the family to enroll their first-born. This is a indicative that, from the family’s perspective, the amount of income they would gain from sending their first-born to work is larger than the BFP transfer.

The first-born have a negative effect on enrollment and a positive effect on working in the past week, indicating the choice of the family to send this first-born to the labor market so that the earlier-born child enables the family making it possible for the later born ones to be enrolled and out of the labor market.

Another thing that emerges from the analysis above is that with exception of the first-born estimates regarding enrollment, all of the rest has the five specifications being mostly significant and pointing towards the same direction. This robustness strongly indicates that there is in fact a difference in the way the BFP eligibility affects children within a family

7.#Conclusion#

In this paper, the objective was to identify changes in birth order effects regarding enrollment and child labor for children that are eligible for receiving the Brazilian CCT program Bolsa Família.

Regarding the effect that the eligibility to BFP has on the correlation between birth order and enrollment in primary or secondary school, the estimates point towards a change in the direction of the birth order effects for children eligible for receiving social transfers. What was found is that the BFP have a higher impact on enrollment for higher birth order.

The program appears to be much more effective for later born children as is evidenced in column (2) and (3) of table 7, while the estimates which consider only the first-born show a slightly negative and insignificant impact of the BFP eligibility, the fourth born is large and significant, with a a gain in the probability of being enrolled close to 25%. These findings are similar to what was found in the existing literature regarding birth order effects in a developing country.

The estimates for the child labor variable of whether the child has worked in some activity outside the household during the previous week also show a interesting pattern, with a positive and significant effect for the first-born sample, and a negative and mostly significant for the later-born. This is a strong indication that the payment structure of the BFP should be adjusted to focus on the earlier born, who seems to still be suffering from the effects of poverty that are still not alleviated by the program.

The possible implications of the finding regarding the implementation of the BFP is that the government should take into account this positive pattern regarding birth order and school enrollment in the payoff structure of the BFP, providing enough incentives for every child within the family to attend and complete their schooling.

The potential differences regarding the gender of the child, which are not covered by this study but are well defined and explained in Janvry and Sadoulet (2006) and in De Haan, Plug, and Rosero (2014), should also be included for an optimal targeting and an increase in the efficiency of the program in order to attack gender vulnerabilities and

inequalities. Higher transfers for school-aged girls could address this, especially for the later-born ones, who have a higher chance of dropping-out of school. BFP tries to tackle this issue through gender specific conditionality, particularly for pregnant woman, but the result on this paper points towards an improve in efficiency if earlier action were to be taken, focusing on young girls that have not yet started their sexual life.

This paper gives rise to some interesting questions concerning the incentives that CCT programs have on the intra-household allocation of their children’s time and adds to the results found by Emerson and Souza (2008) that it is reasonable to expect that birth order have an opposite effect on children in the developed countries than those in

developing ones. What the results of this paper suggests is that the signal of the effects of birth order can vary within a country, with children from rich families having a opposite birth order effect than children coming from poor families or households.

This study adds to the existing literature by relying on a strong identification strategy, the sharp discontinuity design, and on few assumptions regarding the BFP penetration to estimates the heterogeneity in the program impact across children within a family concerning school enrollment and work outside the household.

To summarize, this paper found there is some evidence that birth order effects behave in a different manner in situations of poverty and capital constraints as suggested by Bazu and Van (1998), regarding both enrollment and child labor, and that CCTs programs can and should incorporate this distinct birth-order pattern on their transfer and pay-out scheme to optimally target and increase the marginal benefit of each transfer mainly by increasing the transfers made for earlier-born children and including gender differentiation.

#

8.#Bibliography#

Barros, R., De Carvalho, M., Franco, S., & Mendonça, R. (2010). Markets, the State, and the Dynamics of Inequality in Brazil. Declining inequality in Latin America: A decade of progress, 134-74.

Basu, K., & Van, P. H. (1998). The economics of child labor. American economic review, 412-427.

Black, S. E., Devereux, P. J., & Salvanes, K. G. (2005). The more the merrier? The effect of family size and birth order on children's education. The Quarterly Journal of Economics, 669-700. Booth, A. L., & Kee, H. J. (2009). Birth order matters: the effect of family size and birth order on educational attainment. Journal of Population Economics,22(2), 367-397.

Bourguignon, F., Ferreira, F. H., & Leite, P. G. (2003). Conditional cash transfers, schooling, and child labor: micro-simulating Brazil's Bolsa Escola program. The World Bank Economic

Review, 17(2), 229-254.

Calonico, S., Cattaneo, M. D., & Titiunik, R. (2015). Optimal data-driven regression discontinuity plots. Journal of the American Statistical Association,110(512), 1753-1769.

Calonico, S., Cattaneo, M. D., & Titiunik, R. (2015). rdrobust: An R Package for Robust Nonparametric Inference in Regression-Discontinuity Designs. R Journal, 7(1), 38-51.

De Haan, M. (2010). Birth order, family size and educational attainment.Economics of Education Review, 29(4), 576-588.

De Haan, M., Plug, E., & Rosero, J. (2014). Birth Order and Human Capital Development Evidence from Ecuador. Journal of Human Resources, 49(2), 359-392.

De Janvry, A., & Sadoulet, E. (2006). Making conditional cash transfer programs more efficient: designing for maximum effect of the conditionality.The World Bank Economic Review, 20(1), 1-29.

Ejrnæs, M., & Pörtner, C. C. (2004). Birth order and the intrahousehold allocation of time and education. Review of Economics and Statistics, 86(4), 1008-1019.

Emerson, P. M., & Souza, A. P. (2003). Is there a child labor trap? Intergenerational persistence of child labor in Brazil. Economic development and cultural change, 51(2), 375-398.!

Emerson, P. M., & Souza, A. P. (2002). Birth order, child labor and school attendance in Brazil. Nashville: Vanderbilt University, Department of Economics.

Ferreira, F. H., Filmer, D., & Schady, N. (2009). Own and sibling effects of conditional cash transfer programs: Theory and evidence from Cambodia.World Bank Policy Research Working Paper Series, Vol.

Gelman, A., & Imbens, G. (2014). Why high-order polynomials should not be used in regression discontinuity designs (No. w20405). National Bureau of Economic Research.!

Glewwe P., & Kassouf A. (2008). The impacts of the bolsa escola/familia conditional cash transfer programa on enrollment, grade promotion and drop out rates in Brazil. Encontro Nacional de Economia (ANPEC). Baia.!

Imbens, G., & Kalyanaraman, K. (2011). Optimal bandwidth choice for the regression discontinuity estimator. The Review of economic studies, rdr043.

Imbens, G. W., & Lemieux, T. (2008). Regression discontinuity designs: A guide to practice. Journal of econometrics, 142(2), 615-635.

Kessler, D. (1991). Birth order, family size, and achievement: Family structure and wage determination. Journal of Labor Economics, 413-426.!

Lee, D. S. (2008). Randomized experiments from non-random selection in US House elections. Journal of Econometrics, 142(2), 675-697.

Lee, D. S., & Lemieuxa, T. (2010). Regression discontinuity designs in economics. Journal of economic literature, 48(2), 281-355.

McCrary, J. (2008). Manipulation of the running variable in the regression discontinuity design: A density test. Journal of Econometrics, 142(2), 698-714.

Oosterbeek, H., Ponce, J., & Schady, N. (2008). The impact of cash transfers on school enrollment: Evidence from Ecuador. Available at SSRN 1118249.

Reimers, Fernando, Carol DeShano Da Silva, and Ernesto Trevino. Where is the" Education" in Conditional Cash Transfers in Education?. Montreal: UNESCO Institute for Statistics, 2006.!

Soares, F. V., Ribas, R. P., & Osório, R. G. (2010). Evaluating the impact of Brazil's Bolsa Familia: Cash transfer programs in comparative perspective.Latin American Research Review, 45(2), 173-190.

Tenikue, M., & Verheyden, B. (2010). Birth order and schooling: Theory and evidence from twelve sub-saharan countries. Journal of African Economies,19(4), 459-495.

Zajonc, R. B., & Markus, G. B. (1975). Birth order and intellectual development. Psychological review, 82(1), 74.

! ! ! !