School Attendance, Child Labour and Cash

Transcription

1 PEP-AusAid Policy Impact Evaluation Research Initiative 9th PEP General Meeting Cambodia December 2011 School Attendance, Child Labour and Cash Transfers: An Impact Evaluation of PANES Verónica Amarante Mery Ferrando Andrea Vigorito Instituto de Economía, Universidad de la República

2 Main objectives Estimate the effect of a temporary plan, PANES, on child labor and school attendance for children aged 6 to 17 Explore potential explanatory channels such as labour market outcomes, income and awareness of conditionalities Ch k th b t f lt i t diff t Check the robustness of our results using two different identification strategies (RD and difference in difference)

3 Outline of the presentation: 1. Background information 2. Program description 3. Cash transfers, child labour and school attendance 4. Methodology 5. Main findings 6. Final comments

4 1. Background information General Uruguay: 3.1 million individuals, 85% in urban areas 50 th HDI (3 rd in LAC) PPP per capita income US$ 10, /02 economic crisis (p.c. income -11.4%, unemployment 17%, poverty rate doubled) Existing social safety focused on transfers to the elderly Nearly 50% of children <5 in poverty whereas 8% for over 65 High literacy rates slow increase of average years of schooling High literacy rates, slow increase of average years of schooling among adults (9.8 in 2010)

5 1B 1. Background dif information Sh School attendance Graph 3. School attendance rates in Uruguay to 5 6 to to 17 Source: Based on household surveys High attendance rates at primary school (problems with repetition) Main failure at secondary school with high drop out rates concentrated Main failure at secondary school, with high drop out rates concentrated in lower income stata

6 1. Background information i School attendance and child labour Table 1. Socioeconomic characteristics of children (14-17) by income group. Uruguay Income group Attends Neither attends Attends school and Only works school nor school works works Total Indigent households 52% 5% 9% 34% 100% Poor households 62% 5% 9% 24% 100% All households 75% 3% 6% 16% 100% Child lb labor incidenceid is 1.5% for those aged 5-11 and 9.2% for those aged 12-17

7 2. Program description Objectives and main features PANES: Plan de Atención Nacional alaemergencia Social Temporary antipoverty program, April 05 December 07 Aims: -assist households hit by 2002 crisis -strengthen human + social capital of the poor Target population: HH from the bottom quintile below poverty line - 95% of HH at least one child ,000 HH (10% HH 14% population) Cost US$2,428 per beneficiary household, 0.41% GDP After the program ended, a permanent child allowances program was After the program ended, a permanent child allowances program was launched

8 2. Program description Enrolment and elegibility Means tested and only households with per capita income below a threshold (US$50 per month). The income condition disqualified around 10% of the initial applicants. 188,671 applicants, 102,000 beneficiary HHs Assignment based on proxy means test (including educational attainment, number of hh members, crowding, durable goods, sewage, etc.) Participation conditional on children's school attendance and health checks (monthly prenatal controls for pregnant women) Conditionalities not enforced due to limited coordination between involved institutions

9 2. Program description Components Monthly cash transfer (Ingreso ciudadano) US$56 (lump sum). Represents approximately 50% of the average self-reported pre-program p household income Food Card (Tarjeta alimentaria) for food and personal hygiene, HH with children and pregnant women. Between US$13 - US$30, depending di on household composition Other small components: workfare program job training adult educational Other small components: workfare program, job training, adult educational interventions and health care subsidies.

10 3. Cash transfers, child labour and school attendance Economic modeling dli has treatedt school attendance and child lb labor as alternative uses of time. Two main reasons for leaving school: lower net returns for human capital accumulation than for other assets, or capital market imperfections Empirical literature: lack of explanatory power from income, wealth and credit availability on child labor (Deb and Rosati, 2004). CCTs aim at alleviating short and long run poverty providing a monthly transfer and requiring compliance with conditions If the amount of a cash transfer is higher than a certain minimum, the household will modify its allocation of child s time in favor of schooling (Skoufias and Parker, 2001).

11 3. Cash transfers, child labour and school attendance Potential channels The transfer may cause an income effect on the beneficiaries; if leisure is a normal good, they end up consuming more leisure and working less. If this is the case, the hh would not necessary experience variations iti in income. Hh income variations depend on the manner in which intrahousehold decisions are modeled, this is particularly relevant in the case of teenage labor. Many studies have analyzed this aspect in LAC but they have not found evidence of disincentive effects on adult labor supply (see Alzúa et al, 2009; Fizbein and Schady, 2009) CCTs have been successful in increasing school attendance, in some cases at the expense of a reduction of child li leisure time but not of child work (Ravallion and Wodon, 2000)

12 4. Methodology Data Administrative records of successful and unsuccessful applicants ( hh): baseline socio-demographic characteristics (score), income and standard household survey variables Follow up panel surveys. Sample size: 3000 hh, 2500 eligible and noneligible applicants, in a neighborhood of 2% around the program eligibility threshold score. -1 st wave: December March Non-response rate was 30%: replacement households with approximately the same score -2 nd wave: February - April 2008 (after PANES ended) 92% of households were successfully re-surveyed. We mainly use administrative records and the second wave of the follow up survey. The first wave is only used to identify awareness of conditionalities

13 4. Methodology Model specification and identification (RD) Assignment to PANES was done on the basis of a predicted poverty score and an income threshold. The program was remarkably well-targeted. Graph. PANES elegibility and participation This design allows a sharp regression discontinuity approach.

14 4. Methodology Model specification and identification (RD) Following Lee and Card (2008), we run the following regression: where: y i =bb 0 + b 1 1(N i >0) + f(n i ) + 1(N i >0) g(n i ) + X g + u i y i: variable ibl of fit interest tfor household hldi, N i =S i -E be the normalized income score (E: eligibility threshold). N i >0 indicator for households above the threshold 1() and two parametric polynomials in the normalized score (f(n i ) and g(n i )), on each side of the threshold, such that f(0)=g(0)=0: X : additional covariates This identification strateg onl pro ides local treatment effects Valid if This identification strategy only provides local treatment effects. Valid if variables other than the score are continuous at the discontinuity point

15 4. Methodology Model specification and identification (DD) Y it Ti 1t Ti 1 t X i it where -T=1 reflects the presence of the program at time t=1, whereas T=0 denotes lack of treatment at time t=1 -the coefficient B gives the average DD effect of the program -X are control variables We run models using individual fixed effects We lack of another wave of pre-treatment data to check whether trends among treatment and control groups were similar

16 5. Main findings

17 Table 4. Effects on school attendance by age group. Marginal effects coefficient and standard deviation of the treatment variable. RDD DD 3-5 years old Linear specification Quadratic specification Quadratic specification with control Individual fixed effects DD with RD polynomial (1) (2) variables (3) (4) (5) Total years old (0.0616) (0.1028) (0.0853) (0.0487)** (0.102) Total years old (0.0292) (0.0404) (0.0123) (0.0172)* (0.0386) Total years old (0.0364) (0.0350) (0.0182) (0.0139) (0.0295) Total (0.0883) (0.0802) (0.0696) (0.0440) (0.110) Standard errors in brackets; * significant at 10%; ** significant at 5%; *** significant at 1% Source: Based on PANES administrative record and 2 nd follow-up survey

18 Table 5 Effects on child labour. Marginal effects coefficient and standard deviation of the treatment variable. RDD estimation. Population Linear specification Quadratic specification Quadratic specification with control variables 6-17 years old 6-13 years old years old (0.0277) 0277) (0.0442) 0442) (0.0291) 0291) (0.0206) (0.0395) (0.0228) (0.0513) (0.0840) (0.0775) Standard errors in brackets; * significant at 10%; ** significant at 5%; *** significant at 1%. Source: Based on 2 nd follow-up survey

19 Table 6. Effects on child labour. Children Marginal effects coefficient and standard deviation of the treatment variable. DD estimation Individual fixed effects (1) DD including RD polynomial (2) Total (0.0426) 0426) (0.1021) Boys (0.0648) (0.1591)* Girls (0.0588) (0.1432) Standard errors in brackets; * significant at 10%; ** significant at 5%; *** significant at 1%. Source: Based on PANES administrative record and 2 nd follow up-survey.

20 Table 7. Effects on labour market. Marginal effects coefficient and standard deviation of the treatment variable. RDD DD Population and Linear Quadratic Quadratic specification variable specification specification with control Activity Unemployment Employment Hours of work variables Individual fixed effects DD with RD polynomial (0.0289)* (0.0449) (0.0289)* (0.0160) (0.0373) (0.0157) (0.0271) (0.0312) (0.0133) (0.0230) (0.0267)* (0.0410)** (0.0457)*** (0.0167) (0.0362) (1.888) (2.796)* (2.943) (1.497) (2.348)

21 Table 8. Effects on personal labour income and total household income. Marginal effects coefficient and standard deviation of the treatment variable (people older than 20). RDD DD Population Personal labour income Hh total income (per capita) Linear specification Quadratic specification Quadratic specification with control variables Individual fixed effects DD with RD polynomial (0.0823) (0.133) (0.112) (0.0784) (0.203) (0.0647) (0.0999) (0.0952) (0.0557) (0.107)* Standard errors in brackets; * significant at 10%; ** significant at 5%; *** significant at 1%. Source: Based on PANES administrative record and 2 nd follow-up survey.

22 Role of conditionalities: only 20% of respondents were aware of the school enrolment requirement for children aged We estimated the effect of the beneficiaries awareness of conditions on the probability of school attendance among children aged The positive correlation between children s school attendance and respondents awareness of conditionalities i i disappears once control variables are included Table 9. Effects of conditionality on school enrolment. Marginal effects coefficient and standard deviation of the treatment variable years. 2 nd follow-up survey DD (fixed effects) Without control With control Without control With control variables variables variables variables (0.0140)*** ( ) (0.0192)** (0.0163) ( ) ( ) (0.0165)** (0.0107) (0.0489) (0.0524) (0.0524) (0.0523) Robust standard errors in brackets; * significant at 10%; ** significant at 5%; *** significant at 1% Robust standard errors in brackets; significant at 10%; significant at 5%; significant at 1%. Source: Based on PANES administrative record and 2 nd follow-up survey.

23 6. Final comments Our results indicate that the program did not affect school attendance or child labour, for children as a whole or when disaggregating by age group or sex. Results on child school attendance are not surprising for children at primary school age (attendance is almost universal) For children at secondary school: either the amount of the transfer was not enough as an incentive or other variables rather than income are related to this decision Our results suggest that the transfer amount should probably be considerable higher to foster schooling and not a lump sum to influence schooling decisions at the household level. The fact that money was paid to adults needs to be better analyzed

24 Awareness of conditionalities does not affect school attendance. There is a positive association between both variables but it is explained by other observable variables The specific features of this intervention and the fact that it was launched in amiddle-income i country like Uruguay may provide useful flinsightsi into how the effects of cash transfer programs vary across contexts. It highlights that interventions differ from one country to the next and that it is risky to make generalizations in relation to successful policies. At present, however, the policy debate tends to emphasize general policy impacts on the basis of evidence from well known successful experiences