O IMPACTO DO PROGRAMA BENEFÍCIO VARIÁVEL

JOVEM SOBRE A ALOCAÇÃO DO TEMPO DOS JOVENS

E A OFERTA DE TRABALHO DOS ADULTOS

(THE IMPACT OF THE EXPANSION OF THE BOLSA FAMÍLIA PROGRAM ON THE TIME ALLOCATION OF YOUTHS AND

LABOR SUPPLY OF ADULTS)

Lia Chitolina

Universidade de São Paulo

Miguel Nathan Foguel

IPEA

Naercio Menezes Filho

Insper e Universidade de São Paulo

MDS – Setembro de 2016

Introduction • Conditional Cash Transfer (CCT) Programs have been extensively used

by many governments worldwide: 2 countries in 1997, 28 countries in 2008

(Fizbein and Schady, 2009), and 70 countries in 2014 (Lindert,presentation)

• Dual purpose: alleviating poverty in the short term and incrementing

investment in human capital in children from poor families.

• Success of such programs in reducing poverty depends on how and to what

extent transfers and programs’ conditions impact the allocation of

family time, particularly school attendance and labor supply decisions.

• Theory: the combination of the income and the “price” (associated with the

education condition) effects does not tell the direction of the final effect

on school attendance and family labor supply. It becomes then an empirical

question.

Introduction

• Here, we empirically assess the school and labor supply

effects of extending the coverage of a CCT program to

youths.

• Specifically, we evaluate the impacts of expanding the

Programa Bolsa Família - PBF: in 2007 the Benefício

Variável Jovem – BVJ was introduced giving cash transfers to

poor families conditional on school attendance of members

between 16 and 17 years of age.

Bolsa Variável Jovem • Before the creation of the BVJ: (1) extremely poor families received the Basic Benefit

and the Variable Benefit for each child up to 15 y-o; (2) poor families received only the Variable Benefit for each child up to 15 y-o;

• After the launching of the BVJ in 2007, both extremely poor and poor families with adolescents aged 16 to 17 became eligible to receive the Variable Benefit for Youngsters on the condition that each adolescent is regularly enrolled in school with at least 75% attendance. At present, each family can receive up to five Variable Benefits and up to two BVJs.

2004 2005 2006 2007 2008 2009

Eligibility criteria (income):

Extremely Poor 50 50 60 60 60 70

Poor 100 100 120 120 120 140

Value of benefits:

Basic Benefit 50 50 50 58 62 68

Variable Benefit 15 15 15 18 20 22

Variable Benefit for Youngsters - - - - 33 33

Bolsa Variável Jovem • The value of the BVJ is relevant in marginal terms, especially for

extremely poor families and those with only one 16 y-o adolescent

0%

20%

40%

60%

80%

100%

120%

10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110 115 120

family per capita income

Percentage gain in family income from receiving one BVJ: families with different age compositions of siblings

age composition 1 age composition 2 age composition 3

AC1:only one 16 y-o AC2: one younger than 15 y-o and one 16 y-o AC3: : two younger than 15 y-o and one 16 y-o

Related Literature • Effect on school enrollment: Overall it is positive

• Skoufias and Parker (2001): increases of 5.4 p.p. for boys aged 16 and 17 y-o in Mexico (Progresa);

• Attanasio et al. (2005): positive impact of 14- to 17-y-o of 6 p.p. both in urban and rural areas in Colombia (Familias en Accion);

• Schady and Araujo (2008): impact of 10 p.p. for children aged 6 to 17 y-o in Ecuador (Bono de Desarrollo Humano);

• Khandker et al. (2003): impact of 12 p.p. on secondary school enrollment of girls aged 11 to 18 years for each year of exposure to the program in Bangladesh (Female Secondary School Assistance Project);

Related Literature • Effect on labor supply: Overall it is statistically nil or very small in magnitude

• Skoufias and Parker (2001): reduced the labor force participation of children aged 12 to 17, for both boys and for girls. Focusing on the 16-17 year-old subgroup: effect on the probability of working was negative but not statistically significant. Mexico (Progresa);

• Skoufias and Maro (2008): no significant effect on adults’ labor supply. Mexico (Progresa);

• Pedrozo (2010): a negative impact on adults’ labor supply, especially that of single or divorced mothers. Brazil (PBF);

• Tavares (2008): mothers experienced a 5.6 p.p. increase in the probability of participating in the labor market and extended their weekly working hours by 1.6 percent. Brazil (PBF);

• Ferro and Nicolella (2007): program did not affect the probability that parents participate in the labor force. Brazil (PBF);

• Foguel and Barros (2010): impact on female participation is not significant; impact on males is positive, but very small. Brazil (PBF).

Econometric methodology • We do not observe whether families receive the BVJ & this might

be endogenous.

• Focus on families in the poorest 20%, which are more likely to receive the PBF/BVJ (50% of beneficiaries in 2006): Intention to Treat – ITT (effect of being entitled).

• Specifically, within the first two deciles, we compare families with 16 y-o (treated) with families with 15 y-o (control) before and after the program (PNADs 2006 and 2009).

• Placebo: compare same groups between 2003 and 2006.

• The effect on school enrollment and labor supply was estimated via DD:

Yit = 𝜷0 + 𝜷1Treati + 𝜷2Aftert + 𝜷3(Treati*Aftert) + 𝜷4’Xit + eit

Raw differences in school enrollment

Deciles of Family Income Distribution

Distribution

Across the Deciles Proportion Within the

Deciles

1 (poorest) 26.0 43.2

2 24.4 41.5

3 19.5 29.5

4 12.9 25.4

5 8.5 15.1

6 4.4 6.7

7 2.5 4.2

8 1.1 1.8

9 0.5 0.7

10 (richest) 0.3 0.4

Total 100.0 -

Table 2 - Share of Beneficiary Families by Decile (2004)

Descriptive statistics: PNAD 2006 15 years-old 16-years-old

Control Group Treatment Group Difference

Household:

Household size 5.70 5.57 0.13*

Number of children 3.76 3.61 0.14**

Age of the head of household 43.68 44.82 -1.14***

Age of the youngest child 9.13 9.85 -0.72***

Age of the oldest child 17.32 18.34 -1.02***

Urban 0.65 0.62 0.03

Other income 87.82 88.40 -0.58

Individuals:

Mother:

Age 40.26 41.58 -1.31***

Educational level 3.68 3.48 0.20

Employment 0.65 0.62 0.02

Weekly working hours 27.13 27.49 -0.35

Wage from main job 187.77 189.07 -1.28

Father:

Age 44.26 45.76 -1.49***

Educational level 3.09 2.87 0.22

Employment 0.93 0.91 0.02

Weekly working hours 44.42 43.72 0.70

Wage from main job 286.19 293.54 -7.35

Children:

Educational level 5.64 6.14 -0.50***

Employment 0.33 0.41 -0.08***

Weekly working hours 23.73 26.18 -2.45***

Wage from main job 96.23 111.12 -14.90*

Impact on school attendance: all and by area

Notes: Dependent variable is a binary indicator of school attendance. Sample includes households among the

poorest 20 per cent with children aged 15 and 16 years only. Robust standard errors in parentheses. Column (2)

includes controls for number of children, education, age and race of head, household composition, urban areas

and state dummies. Starred coefficients are significant at the 5% level (*).

Variable (1) (2) (3) (4)

All All Urban Rural

Treated -0.080*** -0.073*** -0.065*** -0.090***

(0.015) (0.015) (0.018) (0.026)

2009 0.031*** 0.036*** 0.042*** 0.015

(0.012) (0.015) (0.018) (0.029)

Treated*2009 0.049*** 0.047*** 0.029 0.090***

(0.019) (0.019) (0.022) (0.033)

Market Wages - -0.041** -0.051** 0.008

(0.021) (0.024) (0.045)

Constant 0.890*** 1.035*** 1.069*** 0.884***

(0.009) (0.066) (0.077) (0.140)

Observations 4781 4781 3193 1588

R² 0.015 0.052 0.058 0.058

Variables Boys Girls Youngest Boys and Youngest

All

Treated*2009 0.0637*** 0.032 0.101** 0.150***

(0.027) (0.025) (0.044) (0.065)

N 2,577 2,204 1,035 569

Urban

Treated*2009 0.035 0.025 0.036 0.066

(0.033) (0.030) (0.053) (0.085)

N 1,673 1,520 704 370

Rural

Treated*2009 0.124*** 0.057 0.231*** 0.300***

(0.049)

(0.043) (0.084) (0.106)

N 904 684 331 199

Impact on school attendance by gender/age and area

Notes: Dependent variable is school attendance. Sample includes households among the poorest 20 per cent

with children aged 15 and 16 years only. Robust standard errors in parentheses. All columns include controls

for number of children, education, age and race of head, household composition, urban areas and state

dummies. Each column reports the results a different regression. Starred coefficients are significant at the 5%

level (*).

Variables Midwest Northeast North Southeast South

All

Treated*2009 -0.017 0.076*** -0.004 0.093** -0.032

(0.076) (0.025) (0.047) (0.045) (0.078)

Observations 306 2523 770 820 362

Urban

Treated*2009 -0.036 0.063** -0.010 0.093* -0.137

(0.088) (0.032) (0.058) (0.048) (0.093)

Observations 240 1551 491 647 264

Rural

Treated*2009 -0.028 0.100*** -0.012 0.071 0.259*

(0.148) (0.040) (0.085) (0.114) (0.142)

Observations 66 972 279 173 98

Controls Yes Yes Yes Yes Yes

Impact on school attendance by region

Notes: Dependent variable is a binary indicator of school attendance. Sample includes households among the

poorest 20 per cent with children aged 15 and 16 years only. Robust standard errors in parentheses. Column (2)

includes controls for number of children, education, age and race of head, household composition, urban areas

and state dummies. Starred coefficients are significant at the 5% level (*).

Variables All Urban Rural

Not Studying nor Working -0.038

(0.025)

-0.028

(0.029)

-0.086

(0.059)

Studying Only

0.004

(0.013)

0.012

(0.010)

-0.020

(0.040)

Working Only -0.010 -0.009 0.008

(0.029) (0.034) (0.055)

Studying and Working 0.044**

(0.022)

0.026

(0.025)

0.097**

(0.048)

Observations 4781 3193 1588

Impact on time allocation

Notes: The dependent variable is time allocation (four options). Sample includes households among the poorest

20 per cent with children aged 15 and 16 years only. Robust standard errors in parentheses. Entries are

marginal effects of each variable on the predicted probability of each option. Columns report results of a single

(multinomial logit) regression and include controls for number of children, education, age and race of head,

household composition, urban areas and state dummies. Starred coefficients are significant at the 5% level (*).

Alternative Boys Girls Youngest Boys and Youngest

Not Studying nor Working -0.052 -0.021 -0.030 -0.020

(0.039) (0.029) (0.055) (0.078)

Studying Only -0.003 0.003 -0.108 -0.231

(0.024) (0.012) (0.095) (0.151)

Working Only 0.025 -0.033 0.077 0.126

(0.040) (0.036) (0.078) (0.100)

Studying and Working 0.030 0.050** 0.061 0.125***

(0.035) (0.023) (0.039) (0.052)

N 2,577 2,204 1,035 569

Impact on time allocation by gender/age

Notes: Dependent variable is time allocation (four options). Sample includes households among the poorest 20

per cent with children aged 15 and 16 years only. Robust standard errors in parentheses. Entries are marginal

effects of treatment interacted with time on the predicted probability of each option. All columns include controls

for number of children, education, age and race of head, household composition, urban areas and state

dummies. Each column reports the results a different regression (multinomial logit). Starred coefficients are

significant at the 5% level (*).

Variables Mothers Fathers

Probability of Work Working Hours Probability of Work Working Hours

All

Treated*2009 0.036 1.852 0.016 -0.0005

(0.027) (1.276) (0.018) (0.930)

Observations 4620 2405 3622 3015

Urban

Treated*2009 0.019 0.685 0.017 -0.304

(0.035) (1.860) (0.027) (1.271)

Observations 3090 1341 2194 1674

Rural

Treated*2009 0.092** 4.034*** 0.017 0.444

(0.044) (1.683) (0.018) (1.377)

Observations 1530 1064 1428 1341

Impact on labor supply of parents

Notes: Dependent variable is a binary indicator for work (columns 1 and 3) and a continuous variable for hours

of work (columns 2 and 4). Sample includes households among the poorest 20 per cent with children aged 15

and 16 years only. Robust standard errors in parentheses. All columns include controls for number of children,

education, age and race of head, household composition, urban areas and state dummies. Each column reports

the results of a different regression. Starred coefficients are significant at the 5% level (*).

Pre-Program trends in school attendance

70%

75%

80%

85%

90%

95%

100%

2001 2002 2003 2004 2005 2006 2007 2008 2009

15 years-old 16 years-old

Variables Without Controls With Controls

Treated -0.063*** -0.061***

(0.015) (0.015)

2006 0.001 -0.013

(0.013) (0.015)

Treated*2006 -0.014 -0.008

(0.021) (0.021)

Constant 0.885*** 0.829***

(0.010) (0.067)

Observations 4,601 4,601

R² 0.010 0.043

Placebo: “impact” on school attendance

Notes: Dependent variable is school attendance. Sample includes households among the poorest 20 per cent

with children aged 15 and 16 years in 2003 and 2006 only. Robust standard errors in parentheses. Column (2)

includes controls for number of children, education, age and race of head, household composition, urban areas

and state dummies. Starred coefficients are significant at the 5% level (*).

Conclusions

• Positive effect of 4 p.p. on school attendance of 16-year-olds from the

poorest 20% families. Difference in school enrollment between the top and

bottom quintiles is 16 p.p., so impact bridges the gap by 25%.

Conclusions

• Positive effects were particularly high (10 p.p.) in rural areas and zero on

statistical grounds in urban areas.

• High impact on school attendance if the sibling is the youngest in the family

(10 p.p.) and even higher if the sibling is the youngest and a boy (16 p.p.).

These were even higher in rural areas (23p.p. and 30 p.p. respectively).

Threat to lose the benefit seems to be important.

• Positive effects were also found on the decision to 'Study and Work’.

Maybe good if studying and working are complementary activities.

• Did not find evidence of impact on the labor supply of parents.