<<

Steps Backward? Lowering the Minimum Age for Child Labor in

Akito Kamei∗† University of Illinois at Urbana-Champaign

September 21, 2020

Abstract

Although many global policies seek to limit child labor, some countries argue that this additional labor can help households living in poverty. Despite the controversy, no previous empirical study documents how legalizing child labor impacts the total labor supply of children. In August 2014, Bolivia lowered its minimum working age from 14 to 10. This paper evaluates the effects of this policy. Using difference-in-differences, this paper shows that the policy increased participation in family work by 3.7 percentage points for boys aged 12-13 but had no effect on girls’ labor. The effect is explained by an increase in agricultural labor among farm households. Boys aged 12-13 from farm households are ten percentage points more likely to work (a 20 percent increase) due to this policy change. This paper finds no policy effect on schooling outcomes, at least in the short-term. Consistent with an increase in agricultural labor, families with children exposed to the policy increased household monthly food consumption (2.83 USD per capita), primarily due to the increased outputs from their farms.

Keywords: Child labor, Legal minimum age, Enforcement of , Bolivia

JEL Codes: J08, K42.

∗Department of Agricultural and Consumer Economics, University of Illinois at Urbana-Champaign, PhD Student in Agricultural and Applied Economics, 326 Mumford Hall, MC-710, 1301 West Gregory Drive, Urbana, IL 61801- 3605, E-mail: [email protected] †I would like to thank Mary Arends-Kuenning, Rebecca Thornton, Kath Baylis, Hope Michelson, Benjamin Crost for comments.

1 1 Introduction

As of 2020, 172 countries have ratified ILO Convention 138, which sets the general minimum working age at 15. However, child labor is still prevalent in many low- and middle- income countries. In

2016, 114 million children between the ages of 5 and 14 engaged in labor worldwide, almost one in nine of all children (ILO, 2017b).

Two main concerns justify child labor bans: the impact of labor on health and on human capital investment. Regarding health effects, working children can be exposed to extreme temperatures, soil contamination, or dangerous tools at working sites. Such hazardous factors could lead to injuries or deteriorate children’s long-term health (Beegle et al., 2009; Graitcer and Lerer, 1998;

O’Donnell et al., 2005). A second concern is the substitution between work and schooling; child workers have less time to attend to their studies. Increased labor supply could lead to decreased attendance or performance at school, which could increase dropout rates.

Although minimum working age are essential and indispensable for combating child labor

(ILO, 2017a; Th´evenon and Edmonds, 2019), it is not clear whether a legal ban alone serves as an effective means of reducing or eliminating child participation in labor markets. Basu and

Van(1998) theorize that child labor bans could potentially increase child labor by deteriorating household income given that preventing children from working can remove an important income source from poor households.1

Although bans on child labor have been well-studied, there is a gap in the literature with respect to policies that, instead, legalize child labor. In 2014, the Bolivian government lowered the minimum working age from 14 to 10 years old. The government considers this policy an essential part of a larger strategy to reduce the country’s poverty level by 2025.

Proponents of the reform do not view child work as inherently bad. Save the Children expressed the favorable response to the reform by recognizing that some cultures consider child labor as a part of an active economic family (BBC, 2014). labor is idealized in Bolivian culture, especially for labor-intensive tasks such as agricultural production (Hindman, 2009).2

However, many international organizations raised concerns about the policy reform. In addition

1Discussion and empirical evidence on child labor ban is provided in Section2. 2Angrist and Kugler(2008); Dammert(2008) study the change in coca production on child labor supply in Colombia and Peru.

2 to violating the international convention 138, the ILO was concerned that the new law might not adequately protect children from hazardous work (ILO, 2014). UNICEF was in agreement on this point and expressed additional concerns that the law would encourage young children to seek jobs early (Fontana and Grugel, 2015; Gaudin et al., 2014; The Wall Street Journal, 2014).

Despite the controversy, no previous empirical study documents how legalizing child labor im- pacts total labor supply of children. Whether legalization increases child labor is not a priori evident. The effect of lowering the legal working age on child labor depends on underlying assump- tions and legal enforcement of existing child labor law. This paper fills this gap in the literature by investigating the impact of lowering the minimum working age on child labor supply. This paper further investigates the effect of the policy on household welfare measured by focusing on changes in household food consumption.3

As the law was implemented nationwide, it is not straightforward to identify a counterfactual.

This study utilizes two cohorts of children from a nationally representative dataset to conduct a difference-in-differences (DID) analysis. The labor supply of children exposed to the policy is compared to the labor supply of children who were unexposed to the policy, controlling for trends in child labor of non-exposed age groups.

Results suggest that the policy increased boys’ labor supply for family work with no change in girls’ labor supply. The increase is driven by agricultural labor among boys in farm households.

Consistent with an increase in agricultural labor supply, households exposed to the policy consume more food, primarily due to the increased production of their farms. Households with policy- exposed children increased monthly consumption of food from their own farms by 2.01 USD per capita at constant 2010 prices. Finally, this study finds no policy impact on schooling outcomes, at least in the short term. This was measured in two ways: the number of children out of school, and years of schooling. This result is consistent with another study in Bolivia on schooling using conditional cash transfers (Canelas and Ni˜no-Zaraz´ua, 2019). The authors find that the conditional cash transfer in Bolivia (Bono Juancito Pinto) increased child schooling but had no effect on child labor. Schooling and child labor may not be a perfect substitute in the case of Bolivia, at least in the short term. 3The welfare of the policy should be evaluated from the multiple perspectives. However, the income information of the child labor is not analyzed given that the majority of the children work without cash payment.

3 The reform of lowering minimum working age is unprecedented. However, the discussion of child labor legalization is not exceptional to Bolivia. The president of defended the practice of child labor in 2019 ( 24, 2019). The results of the paper provide input on the future discussion of child labor policy.

While the data analyzed in this paper lacks information about the safety of the children’s working environment, a household survey collected in 2016 (two years after the reform) shows that approximately 20 percent of working children are exposed to hazardous factors at their working site. Although this study documents that households experience a small benefit from a child in the household being employed, we should be aware that the increase in labor participation may increase the number of children exposed to hazardous forms of work in Bolivia. Whether the legal minimum working age has an impact on a child’s working environment and his or her long-term outcomes remain as important research questions.4

The remainder of the paper is organized into five sections. Section 2 presents the previous study of child labor law and the background of the policy change. Section 3 presents data and descriptive statistics on child labor and household characteristics. Sections 4 discusses a framework to evaluate the policy and provides an estimation strategy. In Section 5, the paper presents estimation results, and Section 6 concludes with discussion.

2 Background

2.1 Child Labor Law and Labor Supply

Seminal papers by Basu and Van(1998) and Basu(2005) provide a theoretical framework for thinking through the effect of child labor bans. The papers argue that banning child labor could potentially increase child labor supply by reducing total household income. This seemingly contra- dictory result demonstrates the complexity of the child labor issue and has become the motivation for a number of empirical studies. Bharadwaj et al.(2013) documents an increase in child labor after ’s 1986 child labor ban. The results follow the theoretical prediction from Basu(2005).

On the other hand, other empirical studies conclude that child labor bans either decrease or

4I do not analyze the long-term impacts of the policy because the identification assumption is less likely to hold in the long-term.

4 have no impact on labor supply. Moehling(1999) exploits state variations in bans of under the age of 14 in the between 1880 and 1910. The study finds little impact on the child labor supply.5 Similarly, Piza and Souza(2016)’s study on the 1998 Brazilian policy reform that increased the minimum working age from 14 to 16 years old found a resulting decrease in child labor participation. Bargain and Boutin(2019) study the same Brazilian law. Although they find no impacts of the policy reform for the overall sample, the paper documents child labor reductions in sectors where child work is visible and in regions where labor inspections are more common and intensive.

In contrast with the literature on the effects of child labor bans, no empirical studies have been conducted on the effect of legalizing child work.

2.2 Legalization of Child Labor in Bolivia, 2014

In 2004, Bolivia’s Code for Children and Adolescents (Article 126) prohibited child labor under the age of 14 (Government of Bolivia, 1999). However, the country faced internal disagreement between international norms of banning child labor and national ideology of child work (Liebel,

2015). Child labor was still prevalent in Bolivia even after the implementation of Article 126, as

436,000 children between 6-13 years of age (22.8 percent) were estimated to be employed in 2008

(IPEC, 2012).

Given the high rate of child labor and the limited enforcement of the prohibition, the admin- istration of President Evo Morales decided to take a different approach; protecting child labor through legalization.

The Code for Children and Adolescents (Article129) entered into force on August 4th, 2014

(Ley 548 Codigo Ni˜no,Ni˜nay Adolescentes). According to 2014 policy, children can work legally in their family business with parental protection after the age of 10. At the age of 12, children can work for a third party. Children working for a third party under the age of 14 are required to obtain permission from the local Commissioner for Children’s Rights. The parents or guardians of children have to permit children to attend school as well as ensuring working conditions that children can relax and take part in leisure activities (Liebel, 2015). The commission is responsible for ensuring the safety of these children and making sure that labor hours do not exceed eight hours

5Manacorda(2006) exploits the state variation of minimum age to study the effect of labor supply among siblings.

5 per day and/or 40 hours in a week.

3 Data and Descriptive Statistics

3.1 Data

To evaluate the impact of the policy, this study uses annual nationally representative household surveys (Encuesta de Hogares) from 2012-2014. The survey is conducted by the Bolivian national statistical department (Instituto Nacional de Estadistica).6 The data were collected in November and December (four to five months after July 2014, when the law was passed).7

The survey collects data on the labor supply for each household member, including the employ- ment type, working sector, and labor hours. In this study, children working as an employee are considered as children working for a third party. The other forms of employment considered are related to children working for a family business (self-employed, family work with or without pay, and household employed).

The analysis sample is restricted to children aged 7-16 years old. This includes children 1-3 years younger and 1-3 years older than the age group that was exposed to the law change.

3.2 Descriptive Statistics

Table1 summarizes children’s characteristics by gender. 8 Nearly 70 percent of interviewed house- holds reside in urban areas, and the average household size is 5.6 persons. Only five percent of children are out of school, where out of school is defined as children not enrolled in school or not attending school. Years of schooling is 5.5-5.6 years old on average. The majority of children have fathers with primary or secondary school level education. The high rate of missing information about fathers reflects the prevalence of male migration in Bolivia (Coon, 2016).

A small number of children work for a third party (Panel B). Almost no children aged 10-11 work for a third party, and the number is less than 1.8 percent for children aged 12-13 years old

6The data is obtained from Instituto Nacional de Estadistica. 7The age of the children is set as age in July, the month that legislation has passed. I use age in July, rather than recorded age at the time of the survey, because whether children are affected by this policy is determined by the age at the time of the policy implementation. 8This paper estimates the effect of the policy separately by gender. Data collected in 2016 focusing on child labor reveals that girls aged 10-13 years old spend 1.9 more hours per week on domestic chores. Boys spend 5.39 and girls spend 7.26 hours on average. See AppendixA for more details.

6 —instead, most child laborers work in family businesses. Overall, 14-15 percent of all children aged

7-16 work for a family business.

Together with differential effects by gender, this study conducts sub-sample analysis by family employment status. Households that own family businesses have a higher labor demand and more working opportunities within the household than those without businesses. If any of the household members older than 15 years old work in a family business, this is considered as the household owning a family business. For boys, 67 percent are from households with a family business (Panel

C). Among those 67 percent, 21 percent are from farm households, and 44 percent have a family member working in the non-farm family businesses.

Figure1 shows that households owning a family business have higher child labor participation.

The rate is especially high for children from farm households. Around 45-48 percent of children whose households have a farm business work as agricultural laborers. Approximately ten percent of children work in non-farm sectors if their households own a non-farm family business.

Table2 presents statistics for working children. For both boys and girls, around 64 percent of children who work are employed in agriculture, and 13 percent work for sales and retail. Almost all children in agricultural work for family businesses without pay, and only 12 percent of working children report earnings information.9

Working children, on average, engage in 22.9 hours of labor per week. This is little more than four hours a day if we assume five working days in a week.10 Following the standard set by the

Commissioner for Children’s Rights of Bolivia, this paper considers work over eight hours a day or

40 hours a week as overwork.11 Approximately 20 percent of children work over 40 hours a week; rates are especially high for children working in manufacturing and construction, with 36 percent of children workers in these sectors considered overworked.

Work Environment

Although information about the working environment is not available in the data, the national survey conducted in 2016 focused on child labor reveals that almost twenty percent of working

9Due to this unobservability of child labor earnings, I do not investigate the effect of the policy on child wages. 10The survey collects data on the primary and secondary work if the respondents engage in more than one occupation. The labor hour is calculated as the sum of the two. 11Even thought this restriction is mostly applied for child labor working for a third party, the paper applies the standard to define overwork.

7 children aged 10-13 face some hazardous factors in their working site (Table3). 12

Agricultural labor is prevalent and some households consider it as a safe form of employment

(ILO, 2014). However, children working in agriculture are more likely to be exposed to hazardous factors than children working in sales or retails. Approximately 12 percent of children in agriculture report exposure to extreme cold or heat. The second most common hazardous factor for agricultural labor is exposure to contaminated soil or dust.

4 Estimation Approach

Intuitively, lowering the minimum working age would result in more children entering the work force. However, whether legalization increases child labor is not a priori evident. The effect of lowering the legal working age on child labor depends on underlying assumptions.

The assumption that households send their children to work only out of economic desperation, is called the luxury axiom in existing theoretical frameworks (Basu and Tzannatos, 2003; Basu and

Van, 1998). If the luxury axiom holds (i.e. children work because it is necessary for the family to survive), then the legalization of child labor should have little bearing on the number of children working. Households with no illegal child labor before the policy would not change their child’s labor supply with the introduction of a law legalizing the labor, because the labor was not occurring in the first place. Therefore, under this assumption, lowering the minimum working age does not increase child labor, at least on the extensive margin.13,14

This section discusses an estimation framework for the legalization of child labor and child labor supply following Edmonds and Shrestha(2012).

12The statistics are based on ENNA 2016 (Encuesta de ninos, ninas y adolescentes que realizan actividad laboral o trabajan 2016). 13If the legalization increases children’s wage, the change in labor supply essentially depends on how child labor responds to wages. Previous empirical studies document an increase in child labor in response to a higher wage (Angrist and Kugler, 2008; Boozer and Suri, 2001; Duryea and Arends-Kuenning, 2003; Kruger, 2007; Li and Sekhri, 2020). This study does not study the outcome of children’s wage as only 12 percent of the working children report earning information in the Bolivian data. 14Similarly, the social norm of child labor plays a crucial role for parents’ decisions (Edmonds and Shrestha, 2012). Given that the previous minimum age of 14 years old has been enforced for a long time, parents may not increase child labor just because its legal status has changed.

8 4.1 Estimation Framework: Impact of Lowering Minimum Working Age

Figure2 depicts child labor supply by age. The y-axis is the number of children working in a society, and the x-axis represents child age. Assuming that each household has random preferences for child time allocation, child labor participation is a smooth increasing function of age (higher opportunity cost for older children).15 The yellow area (below the red line) represents the child labor supply in the society.

Figure2 (a) depicts a society where there is a legal minimum working age (14 years old). Under perfect , the labor supply would be zero up to the legal minimum age (described with the red line). Edmonds and Shrestha(2012) study the labor supply discontinuity at the legal minimum working age using MICS data from 59 low-income countries. Although many countries set their minimum legal working age at 14, the data show a large number of working children under the age of 14. In a society where the law is not perfectly enforced, the discontinuity is less apparent

(Figure2 (b)). The authors conclude that child labor laws have weak enforcement power given that the regression discontinuity shows small magnitude of coefficients.16

Rather than a regression discontinuity, which evaluates the impact around the cut-off, this paper uses a DID estimate utilizing the incidence of the policy change.17 The DID estimate is described as areas: A for children aged 10-11 and B for children aged 12-13 in Figure2 (c).

4.2 Estimation Equation

Figure3 presents the structure of the analysis sample with respect to exposure to the policy.

Children aged 7-9 and 14-16 (sample with blue) serve as control groups. Children in those age brackets are not directly affected by the policy change.

15The function is not smooth when the schooling system creates a discontinuity in return from education. However, for the simplicity of the argument, we will dismisses this aspect. 16AppendixB presents child labor participation rate before and after 2014 reform. The data from Bolivia also shows no perfect compliance of child labor law, as some children under the are observed to be working. 17Piza and Souza(2016) and Bargain and Boutin(2019) study a policy reform that increased the minimum working age from 14 to 16 years old that happened in 1998 in Brazil. Both papers focus on the sample of children who turned 14 years old with respect to the day of law implementation. Compared to the data from Brazil (PNAD), the Bolivian data used in this paper is smaller in sample size and do not allow for estimation using regression discontinuity.

9 To quantify the change in labor supply, I estimate the following equation:

yit = α + β1Afterit + β2Exposed(10-11)it + β3Exposed(12-13)it (1) + [β4Exposed(10-11)it + β5Exposed(12-13)] × Afterit + βX Xit + θt + it

where yit is a binary variable of child labor. The parameter of interests are β4 and β5, which are the interaction of Exposed age group and After (Year 2014). The β4 coefficient corresponds to the area of A and β5 corresponds to the area of B in Figure2 (c). The estimation model includes household and individual controls such as rural/urban residence and categorical variables including the father’s/mother’s education level and household size. Age dummies are included to control for each age-specific factors. The estimation includes year fixed effects to control for factors such as overall macroeconomics in each year. The nine regional fixed effects are included to control for regional differences such as labor market conditions. My preferred specification includes controls for precision, but the main findings are robust in models without controls.

4.3 Testing Identification Assumptions

The identification assumption is that in the absence of the law change, the child labor trends for children aged 10-13 and children aged 7-9/14-16 in 2014 would be the same (parallel) as before.

This assumption is likely to hold, given that there were no other policy changes in Bolivia that affected specifically children aged 10-13 in 2014. Having a control group sandwiching the 10-13 cohort on both sides reduces the concern that the time trend in the younger/older age group drives the estimation results. Similarly, as the treatment status is defined by the age of children, concern of self-selection to the treatment group is small.

AppendixC examines the validity of the assumption. Appendix C.1 tests whether the labor trend during the study period drives the result. The section presents (1) the result of placebo implementation (law implementation one year before the actual law implementation), and (2) controls for pre-trend labor supply for each age group in linear form. The null results from false specification provide the validation that the main result is driven by the 2014 policy change and not by a general trend around the study period. The control of pre-trend in linear form provides

10 another supportive evidence that the result is not driven by general trends in labor supply.

One threat to the identification assumption is a spillover of labor among siblings; children in the control age groups may be affected by the policy if they have siblings exposed to the policy.

Appendix C.2 provides the estimation results with consideration of possible spillover among siblings.

Appendix C.3 provides the estimated results for asset index, household size and parents’ ed- ucation as outcomes. Household size, parents’ education and asset holdings are relevant living standard measures (Carter and Barrett, 2006). However, as the 2014 data were collected a few months after the implementation of the law, changes in those outcomes are not expected to be observed. Non-significant results on those outcomes affirm that the main results are not driven by some biased characteristic of the selected sample, such as only capturing the poor households.

5 Empirical Results

5.1 Main Effects: Child Labor Supply

Table4 presents the effect of the policy on children work for a third party and for those employed by a family business (including agriculture). I present three specifications with different controls.

I describe the estimation result with all control variables as preferred specification (columns 3 and

6) though the magnitude of the coefficient barely changes in any specifications.

Overall, the policy has no impact on employment for a third party. Regarding the child labor supplied to family enterprises, the policy increased the probability of boys aged 12-13 working by

3.7 percentage points. Given that the average boys’ labor participation is 14.5 percent for this age group before the reform, the effect is a 25 percent increase from the average. I do not find any change in labor participation for girls.18

An increase in boys’ labor is mainly driven by their increase in agricultural labor supply. Table

5 provides a sub-sample analysis based on household employment status. The increase in child labor in agriculture is 11.7 percentage points for boys from farm family businesses (column 3).

Although the estimate is not statistically significant, the coefficient for agricultural labor for boys aged 10-11 is high with 6.5 percentage points (t=1.42). There is no increase in agricultural labor

18Robustness checks in AppendixC show that the main result is not driven by a trend in labor supply around the time of the study nor the random chance that the main estimate capture the poor households. Estimates considering possible sibling spillover shows 4.2 percentage points, only slightly hire than 3.7, estimates presented in Table4.

11 from the households that own non-farm family business or do not own a family business (columns

1 and 2).19

The results suggest that the policy had no impact on labor hours nor the incidence of overwork

(AppendixD). The mean prevalence of overwork for farm households was five to seven percent for children aged 10 to 11, and eleven to twelve percent for children aged 12-13 (Table D.1).20 Results show no clear evidence of the policy impacting labor hours.21 Although the legalization of the child labor intended to reduce the overwork as one protection measure, the estimates do not suggest strong evidence of change in labor hours. Results suggest that the policy increased child labor in the extensive margin, but not the intensive margin of labor supply.

5.2 Household Impacts: Monthly Food Consumption Per Capita

Previous studies have documented a positive impact of child labor on household production (Andre et al., 2017; Jacoby, 1993; Rosenzweig, 1980; Skoufias, 1994). If child labor contributes to household income or agricultural production, we expect to see an increase in food consumption.

This section provides estimated results for food consumption analyzed at the household level.22

The estimation considers households as treated if they have some children exposed to the policy

(children aged 10-13 years old). Food consumption is represented as per capita by dividing the household consumption by the number of household members (Appendix E.1 provides descriptive statistics).

Table6 shows estimates for monthly food consumption per capita. Consistent with the increase in agricultural labor supply, there is an increase in food consumption from the farm households, but not from other types of households (columns (7)-(9)). Farm households with policy-exposed children increase monthly food consumption per capita by 2.83 USD (2010 level). The increase in consumption is driven by increased consumption from the families’ own production. Having children exposed to the policy increases per capita monthly food consumption from self-provision by 2.01 USD.

19The study did not find any change in child labor supply for non-agriculture sector (results upon request). 20Overwork is defined as children who work more than eight hours a day, or 40 hours a week. 21There is statistically significant decrease in labor hours for girls aged 10-11 for households with no family business, but the number of the observation is small (N=52). 22Unfortunately, over the study period, the survey changed the consumption items asked. The only section with no change in the questionnaire is the section on food consumption.

12 The amount of a government cash transfer is comparable to the gain in household consumption estimated in this paper. An increase in monthly food consumption of 2.01 USD means that the gain from child labor is 24 USD annually. The Bolivian conditional cash transfer (Bon Juancito

Pinto) initiated in 2006, provides 200 Bolivianos (approximately 22-25 USD) annually to every child registered in the public education system with a condition on school attendance.23

Appendix E.2 investigates the robustness of the results. In the IHS form, the increase in food consumption from self-provision is 0.28 in log USD (Table E.1). The increase is only observed when the estimate considers having policy-exposed boys as treated. Table E.2 shows that having policy-exposed girls in the household does not affect food consumption. This is consistent with labor results, as the policy did not affect girls’ labor supply.

5.3 Schooling Outcomes

Neither the theory nor the literature are clear on the effect of child labor on schooling. Some argue that child labor can serve as a means to support household expenditures on education including school supplies, school fees, or transportation. Whether there is a trade-off between child labor and human capital investment remains an empirical question.

AppendixF provides estimated results for schooling. I find no reduction in the likelihood of farm households boys being out of school. Schooling measured as years of education shows no change either. The result is consistent with Canelas and Ni˜no-Zaraz´ua(2019), a study of conditional cash transfer program in Bolivia (Bono Juancito Pinto) conducted using data from 2005-2013. The authors find that the conditional cash transfer increased child schooling, but had no effect on child labor. They conclude that substitution between schooling and labor is small for Bolivian children.

Although the analysis in this paper is limited to short-term effects, the policy effects on edu- cation performance (e.g., study time, test scores) and long-term schooling remains as important topics for the future research.24

23Whether the amount of 200 Bolivianos is large or small depends on household living standard. 24This study does not analyze the long-term impact as the identification assumptions are less likely to hold for the cohort difference-in-difference analysis.

13 6 Conclusion and Discussion

This paper evaluates the policy effect of lowering the legal minimum working age. The policy increased labor participation in family work by 3.7 percentage points for policy-exposed boys aged

12-13. The increase is explained by an increase in agricultural labor from farm households. Boys aged 12-13 from farm households are ten percentage points more likely to work. Following an increase in agricultural labor, households experience a small increase in household food consumption from a child in the household being employed. The interpretation of the results needs caution as the external validity of findings in this study is limited to children aged 10-14, the ages that this policy targeted.

One puzzle in the empirical result is that the effect of the policy change is only observed for boys. A study from Brazil shows a similar results that the policy reform only affects the labor supply for boys (Piza and Souza, 2016). The authors interpret the result as labor policy was only binding for boy’s labor supply. Higher domestic labor burden for girls may be another reason why the policy effect was not observed for girls.

The limitations of this paper include the lack of information on several outcomes. As the central question around child labor has shifted from “whether children are working or not” to

“whether children are working in an environment that is harmful to their physical and mental development,” we should be aware that the increase in labor participation may increase children exposed to hazardous forms of labor. A household survey collected two years after the reform shows that approximately 20 percent of working children are exposed to hazardous factors at their working site.

The policy impact on children’s wages, their working environments, and long-term outcomes are noteworthy and should be explored in the future.25

25Children’s wages are not analyzed in this paper as most children do not report earning information. This study does not analyze the long-term impact as the identification assumptions are less likely to hold for the cohort difference-in-difference analysis.

14 References

Andre, P., Delesalle, E., and Dumas, C. (2017). Returns to farm child labor in Tanzania. Available

at SSRN 2900121.

Angrist, J. D. and Kugler, A. D. (2008). Rural windfall or a new resource curse? Coca, income,

and civil conflict in Colombia. The Review of Economics and Statistics, 90(2):191–215.

Bargain, O. and Boutin, D. (2019). Minimum Age Regulation and Child Labor: New Evidence

from Brazil. The World Bank Economic Review.

Basu, K. (2005). Child labor and the law: Notes on possible pathologies. Economics Letters,

87(2):169–174.

Basu, K. and Tzannatos, Z. (2003). The global child labor problem: What do we know and what

can we do? The World Bank Economic Review, 17(2):147–173.

Basu, K. and Van, P. H. (1998). The economics of child labor. American Economic Review, pages

412–427.

BBC, M. (2014). ¿Llegal o parte de la cultura? El trabajo infantil divide a Bolivia. www.bbc.com/

mundo/noticias/2014/01/140110 bolivia trabajo infantil vs. Beegle, K., Dehejia, R., and Gatti, R. (2009). Why should we care about child labor? The

education, labor market, and health consequences of child labor. Journal of Human Resources,

44(4):871–889.

Bharadwaj, P., Lakdawala, L. K., and Li, N. (2013). Perverse consequences of well intentioned

regulation: Evidence from India’s child labor ban. Journal of the European Economic Asso-

ciation.

Boozer, M. and Suri, T. (2001). Child labor and schooling decisions in Ghana. Mimeograph, Yale

University.

Canelas, C. and Ni˜no-Zaraz´ua,M. (2019). Schooling and labor market impacts of Bolivia’s Bono

Juancito Pinto program. Population and Development Review.

Carter, M. R. and Barrett, C. B. (2006). The economics of poverty traps and persistent poverty:

An asset-based approach. The Journal of Development Studies, 42(2):178–199.

Coon, M. (2016). Remittances and child labor in Bolivia. IZA Journal of Migration, 5(1):1.

Dammert, A. C. (2008). Child labor and schooling response to changes in coca production in rural

15 Peru. Journal of Development Economics, 86(1):164–180.

Duryea, S. and Arends-Kuenning, M. (2003). School attendance, child labor and local labor market

fluctuations in urban Brazil. World Development, 31(7):1165–1178.

Edmonds, E. V. and Shrestha, M. (2012). The impact of minimum age of employment regulation

on child labor and schooling. IZA Journal of Labor Policy, 1(1):14.

Fontana, L. B. and Grugel, J. (2015). To eradicate or to legalize? Child labor debates and ILO

Convention 182 in Bolivia. Global Governance: A Review of Multilateralism and International

Organizations, 21(1):61–78.

France 24 (2019). Bolsonaro, who worked from age 8, defends child labor in Brazil.

https://www.france24.com/en/20190706-bolsonaro-who-worked-age-8-defends

-child-labor-brazil. Gaudin, A. et al. (2014). Bolivia defends decision to legalize child labor from age 10. Latin America

Digital Beat (LADB), The University of New .

Government of Bolivia (1999). Ley del codigo ni˜na,ni˜noy adolescente, ley no. 2026. http://

www.ilo.org/dyn/natlex/docs/WEBTEXT/55837/68387/S99BOL01.htm. Graitcer, P. L. and Lerer, L. B. (1998). Child labor and health: Quantifying the global health

impacts of child labor.

Hindman, H. D. (2009). The world of child labor: An historical and regional survey. ME Sharpe.

ILO (2014). ILO’s concerns regarding new law in bolivia dealing with . https://

www.ilo.org/ipec/news/WCMS 250366/lang--en/index.htm. ILO (2017a). Ending child labour by 2025: A review of policies and programmes. Technical report,

Geneva.

ILO (2017b). Global estimates of child labour: Results and trends, 2012-2016. Technical report,

International Labour Office, Geneva.

IPEC (2012). Child labour and education in Bolivia. Technical report, International Programme

on the Elimination of Child Labour, Geneva.

Jacoby, H. G. (1993). Shadow wages and peasant family labour supply: An econometric application

to the Peruvian Sierra. The Review of Economic Studies, 60(4):903–921.

Kruger, D. I. (2007). Coffee production effects on child labor and schooling in rural Brazil. Journal

of Development Economics, 82(2):448–463.

16 Li, T. and Sekhri, S. (2020). The spillovers of employment guarantee programs on child labor and

education. The World Bank Economic Review, 34(1):164–178.

Liebel, M. (2015). Protecting the rights of working children instead of banning child labour: Bolivia

tries a new legislative approach. The International Journal of Children’s Rights, 23(3):529–

547.

Manacorda, M. (2006). Child labor and the labor supply of other household members: Evidence

from 1920 America. American Economic Review, 96(5):1788–1801.

Moehling, C. M. (1999). State child labor laws and the decline of child labor. Explorations in

Economic History, 36(1):72–106.

O’Donnell, O., Rosati, F. C., and Van Doorslaer, E. (2005). Health effects of child work: Evidence

from rural . Journal of Population Economics, 18(3):437–467.

Piza, C. and Souza, A. P. (2016). The causal impacts of child labor law in Brazil: Some preliminary

findings. The World Bank Economic Review, 30(Supplement 1):S137–S144.

Rosenzweig, M. R. (1980). Neoclassical theory and the optimizing peasant: An econometric analysis

of market family labor supply in a developing country. The Quarterly Journal of Economics,

94(1):31–55.

Skoufias, E. (1994). Using shadow wages to estimate labor supply of agricultural households.

American Journal of Agricultural Economics, 76(2):215–227.

The Wall Street Journal (2014). Newest legal laborers in Bolivia: Children. https://www.wsj.com/

articles/newest-legal-laborers-in-bolivia-kids-1414627368. Th´evenon, O. and Edmonds, E. (2019). Child labour: Causes, consequences and policies to tackle

it. OECD Social, Employment and Migration Working Papers, No. 235, OECD Publishing,

Paris, (235).

17 Figure 1: Child Labor Participation by Family Employment Status

Girls Boys

47.9 50 43.9 40

32.8

28.5 30 20

11.1 10.1 9.8

% of children working 8.8 10 4.4 2.6 1.6 1.4 0.7 0.7 0.2 0.5 0 No Farm Non farm Both No Farm Non farm Both family family family farm and family family family farm and business business business non farm business business business non farm

Agriculture Child Labor Non−agriculture Child Labor

Graphs by gender of children

Notes: The figure shows percent prevalence of the child labor by household characteristics (Aged 7-16: N=21,669). There are 7,321 households (34 percent) with no family business, 4,397 family farm business households (20 percent), 9,370 non-farm family business households (43 percent), and 584 households with both farm and non-farm business households (3 percent).

18 Figure 2: Framework of Minimum Working Age Change

(a) Perfect Enforcement (b) Imperfect Enforcement

(c) Regression Discontinuity and Difference-in-differences Estimate

Notes: The figure presents child labor supply under three scenarios of law enforcement. Figure 2 (a) depicts the situation where there is a legal minimum working age (14 years old). Under the perfect law enforcement, the labor supply drops at the legal minimum age (described with the red line). Figure 2 (b) shows the case where law enforcement is not perfect. Figure 2 (c) shows the change in labor supply due to the change in the minimum working age. Although law enforcement is not perfect, there are children not working given its illegality. However, change in the minimum working age let those children work freely. The increase of child labor after the law is described as the area of A (for children aged 10-11) and B (for children aged 12-13).

19 Figure 3: Exposure to the Policy: Age and Year of Data Collection

Notes: The figure visualizes the sample by exposure to the policy. In the Difference-in-differences setting, the sample with blue serves as a control group. Children age 7, 8, 9, 14, 15, and 16 are never affected by the policy change. Therefore, those children are the control group (not exposed to the policy). The second difference comes from time difference depending on the year of the data collection. The data collected before 2014 are called before data set and in 2014 is considered as after data set.

20 Table 1: Characteristics of 7-16 Year-Old Children in Bolivia, 2012-2014

Boys Girls Min Max Panel A: Child/Household Characteristics Age 11.55 11.60 7 16 Urban 0.68 0.69 0 1 Household size 5.53 5.58 1 16 Household size is 1-4 0.33 0.32 0 1 Household size is 5 0.23 0.23 0 1 Household size is 6-7 0.29 0.30 0 1 Household size is 8+ 0.15 0.15 0 1 Out of school 0.05 0.05 0 1 Years of schooling 5.53 5.64 0 15 Father’s education No education 0.02 0.02 0 1 Primary 0.27 0.26 0 1 Secondary 0.31 0.32 0 1 Higher 0.15 0.15 0 1 Missing 0.25 0.26 0 1 Mother’s education No education 0.07 0.08 0 1 Primary 0.33 0.32 0 1 Secondary 0.31 0.30 0 1 Higher 0.14 0.14 0 1 Missing 0.15 0.15 0 1 Panel B: Labor Outcomes Child labor for a third party 0.03 0.02 0 1 Among age 7-9 0 0 0 1 Among age 10-11 0.01 0 0 1 Among age 12-13 0.03 0.01 0 1 Among age 14-16 0.08 0.04 0 1 Child labor for family 0.15 0.14 0 1 Among age 7-9 0.09 0.09 0 1 Among age 10-11 0.14 0.13 0 1 Among age 12-13 0.17 0.17 0 1 Among age 14-16 0.18 0.17 0 1 Panel C: Household Work Environment No family business 0.33 0.34 0 1 Family business 0.67 0.66 0 1 Family farm business 0.21 0.20 0 1 Non-farm family business 0.44 0.43 0 1 Both farm and non-farm family business 0.03 0.03 0 1 Observations 10916 10753 10916 10753 Notes: The observation is at the individual level. The sample pools sample from year 2012, 2013, and 2014. Child labor working for a third party is a dummy variable of one if the child engages in labor as employee. Child work for family is a binary variable of one if the child engages in child labor for family work or self-employment (this includes patron and household employment).

21 Table 2: Working Characteristics of 7-16 Year-Old Children by Sector, 2012-2014

Sector Manufacture Sales Food Across Agriculture Construction Retails Restaurant Others Panel A: Employment Type Worker 0.06 0.02 0.37 0.02 0 0.06 Office worker 0.08 0 0.12 0.24 0.29 0.28 Self-employed 0.03 0.01 0.05 0.06 0.04 0.16 Patron social without pay 0 0 0 0 0 0 Family work (no pay) 0.81 0.96 0.45 0.68 0.66 0.25 Household employed 0.01 0 0 0 0 0.26 Panel B: Labor Outcomes Weekly labor hours 22.92 20.45 29.85 26.73 23.82 27.79 Overwork 0.20 0.14 0.36 0.28 0.23 0.33 Earnings Reported (0/1) 0.15 0.03 0.49 0.25 0.29 0.58 Panel C: Working Site Number of people working 4.38 4.60 4.87 3.31 3.88 3.80 Home 0.04 0 0.20 0.05 0.05 0.12 Land (Local) 0.87 1 0.67 0.64 0.52 0.50 Mobile 0.02 0 0.01 0.06 0.18 0.01 Small shop 0.04 0 0.02 0.18 0.20 0.01 Other 0.04 0 0.10 0.06 0.05 0.36 Observations 3610 2314 397 493 218 187 Notes: Top one percent of the weekly labor hours is trimmed. Overwork is defined as children who work more than eight hours a day, or 40 hours a week. The number of people working includes the child him/herself.

Table 3: Hazardous Factors in Working Environments by Sector (Age 10-13, Bolivia in 2016)

Sector Manufacture Sales Food Across Agriculture Construction Retails Restaurant Others Any hazard reported 0.19 0.19 0.37 0.12 0.23 0.23 Hazardous factors Contaminated soil/dust 0.07 0.07 0.12 0.02 0 0.12 Fire, gas, flames 0.01 0 0.05 0 0.06 0 Loud noise or vibration 0.01 0 0.16 0.03 0 0.08 Extreme cold or heat 0.11 0.12 0.09 0.05 0.03 0.16 Dangerous tool 0.05 0.05 0.04 0.02 0.18 0 Work at height 0.01 0 0.07 0 0 0 Work in the water 0 0 0 0 0 0 Chemical products 0 0 0 0.01 0 0.01 Others 0 0 0 0.02 0 0.03 Observations 713 457 48 107 55 46 Source: Encuesta de ni˜nos,ni˜nasy adolescentes que realizan actividad laboral o trabajan 2016. Notes: Enumerators list up to three hazardous facotor for each children.

22 Table 4: Effect of the Policy on Probability of Engaging in Child Labor

Boys Girls (1) (2) (3) (4) (5) (6) Outcome A: Child labor for a third party After × Exposed (Aged 10-11) -0.007 -0.006 -0.006 -0.007 -0.007 -0.007 (0.009) (0.009) (0.009) (0.007) (0.007) (0.006)

After × Exposed (Aged 12-13) -0.010 -0.010 -0.009 -0.001 -0.001 -0.001 (0.009) (0.009) (0.009) (0.006) (0.006) (0.006) Mean (Age 10-11 before policy) 0.00 0.00 0.00 0.00 0.00 0.00 Mean (Age 12-13 before policy) 0.03 0.03 0.03 0.01 0.01 0.01

Outcome B: Child labor for family After × Exposed (Aged 10-11) -0.002 0.006 0.007 0.020 0.007 0.008 (0.018) (0.016) (0.016) (0.017) (0.016) (0.016)

After × Exposed (Aged 12-13) 0.037∗∗ 0.036∗∗ 0.037∗∗ -0.005 -0.006 -0.006 (0.017) (0.016) (0.016) (0.017) (0.016) (0.016) Mean (Age 10-11 before policy) 0.13 0.13 0.13 0.12 0.12 0.12 Mean (Age 12-13 before policy) 0.14 0.14 0.14 0.16 0.16 0.16 After/Exposed Yes Yes Yes Yes Yes Yes Year/Region control Yes Yes Yes Yes Yes Yes Household control No Yes Yes No Yes Yes Age dummy control No No Yes No No Yes Observation 10,916 10,913 10,913 10,753 10,752 10,752 Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The Year/Region controls for year of data collection and nine fixed effects of regions (departamiento). The Household control includes dummy variables of household size, residence of urban/rural, and father’s/mother’s education. Age dummy is the dummy variables of each age.

23 Table 5: Effect of the Policy on Probability of Engaging in Agriculture Child Labor

Dependent Variable: Engaging in Agriculture Child Labor (0/1) Boys Girls (1) (2) (3) (4) (5) (6) No Non-farm Farm No Non-farm Farm family family family family family family Sample business business business business business business After × Exposed (Aged 10-11) -0.003 0.004 0.065 -0.012 0.003 0.006 (0.011) (0.011) (0.046) (0.008) (0.010) (0.045)

After × Exposed (Aged 12-13) -0.010 0.008 0.117∗∗∗ 0.001 0.002 -0.046 (0.011) (0.010) (0.045) (0.007) (0.010) (0.045)

After/Exposed Yes Yes Yes Yes Yes Yes Year/Region control Yes Yes Yes Yes Yes Yes Household control Yes Yes Yes Yes Yes Yes Age dummy control Yes Yes Yes Yes Yes Yes Mean (Age 10-11 before policy) 0.01 0.02 0.37 0.01 0.02 0.39 Mean (Age 12-13 before policy) 0.03 0.03 0.43 0.01 0.02 0.46 Adjusted R2 0.08 0.13 0.27 0.05 0.13 0.28 Observation 3,617 5,054 2,538 3,703 4,897 2,440 Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The Year/Region controls for year of data collection and nine fixed effects of regions (departamiento). The Household control includes dummy variables of household size, residence of urban/rural, and father’s/mother’s education. Age dummy is the dummy variables of each age.

Table 6: Effect of the Policy on Monthy Consumption Per Capita (2010 USD)

No family business Non-farm family business Farm family business (1) (2) (3) (4) (5) (6) (7) (8) (9) Purchase Pur- Purchase Pur- Purchase Pur- provision chase Provision provision chase Provision provision chase Provision combined only only combined only only combined only only After × Exposed -0.26 0.17 -0.43 -0.07 -0.31 0.24 2.83∗ 0.82 2.01∗∗ (1.23) (1.20) (0.31) (0.98) (0.98) (0.40) (1.49) (1.22) (0.91) Control Yes Yes Yes Yes Yes Yes Yes Yes Yes Mean 43.35 42.39 0.97 41.11 38.40 2.72 36.04 24.23 11.81 Adjusted R2 0.14 0.17 0.13 0.16 0.17 0.06 0.22 0.30 0.09 Observation 4,444 4,444 4,444 5,746 5,746 5,746 2,443 2,443 2,443 Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The Control includes year of data collection, nine regions (departamiento), dummy variables of household size, residence of urban/rural, and father’s/mother’s education.

24 Appendices

A Gender Gap in Domestic Labor

Although the domestic labor information is not available in the data used in the analysis, the nationally representative survey conducted in 2016 (ENNA 2016: Encuesta de ninos, ninas y ado- lescentes que realizan actividad laboral o trabajan) specializing child labor reveals that Bolivian girls engage in domestic labor more than boys.

Figure A.1 (left) shows participation rate for each domestic labor activity for boys and girls aged 10-13. Boys are more likely to engage in shopping and fixing home, and girls are more likely to engage in tasks such as cooking, cleaning house or washing clothes. More than ninety percent of children aged 10-13 years old engage in at least one hour of domestic labor.

Figure A.1 (right) shows the distribution of the domestic labor hours. Boys, on average, spend

5.4 hours per week on domestic labor, while the same statistics 7.3 hours for girls. Although the economic activity labor supply seems similar by gender, Bolivian girls face a higher domestic chores labor burden. Therefore, the policy reform for economic activity may affect differently by gender.

Figure A.1: Domestic Labor for Children Aged 10-13 in Bolivia

Percent of children engage in activity Density distribution of domestic labor hours Sample: Children aged 10−13 Sample: Children aged 10−13 .15 100

91

80 80 73 .1 65 60 62 60 58

Percent 49 40 40 39 .05 25

20 25 14 16 8 0 4 Boys Girls 0

Shopping Fixing home stuff 0 10 20 30 40 Cooking Cleaning house and place Hours for domestic labor per week Washing clothes Taking care of children Boys Girls Fetching water Other tasks Notes: Four sample is dropped as the domestic labor hour was more than 40 hours.

25 B Compliance of Child Labor Law by Age

Table B.1 highlights the low compliance of legal minimum working age in Bolivia both before and after the policy. For children aged 10-13, around 12 to 16 percent were working for family business even before the policy change. For children 7-9 years old, 8-11 percent worked for family even thought their work has never been legalized. The low compliance rate for labor law even before the policy indicates the weak enforcement power of child labor law in Bolivia.

Table B.1: Labor Participation Before and After the Policy in Bolivia, 2012-2014

Boys Girls Before After Difference Before After Difference Child labor for a third party 0.03 0.04 0.01 0.02 0.02 0 Among age 7-9 0 0 0 0 0 0 Among age 10-11 0 0.01 0 0 0 0 Among age 12-13 0.03 0.03 0 0.01 0.01 0 Among age 14-16 0.08 0.09 0.02 0.04 0.05 0.01 Child labor for family 0.13 0.18 0.05 0.13 0.16 0.03 Among age 7-9 0.08 0.11 0.03 0.08 0.11 0.02 Among age 10-11 0.13 0.17 0.04 0.12 0.16 0.05 Among age 12-13 0.14 0.22 0.08 0.16 0.18 0.02 Among age 14-16 0.17 0.21 0.05 0.15 0.19 0.04 Observations 7144 3772 10916 7051 3702 10753 Notes: The observation is at the individual level. The Before columns represent the data collected in the year 2012 and 2013. After represents the data collected in the year 2014. Child labor working for a third party is a binary variable of one if the child engages in labor as employee. Child work for family is a binary variable of one if the child engages in child labor for family work or self-employment (this includes patron and household employment).

26 C Robustness Check

This section tests the robustness of the main results. The robustness tests shows that the main result is not driven by a trend in labor supply around the time of the study nor the random chance that the main estimate is correlated with the poor households. The estimate considering possible sibling spillover shows a similar magnitude with estimated results presented in Table4.

C.1 Trend in Labor Supply

This subsection provides the robustness of the main results considering trend in labor supply in two ways. First, I run the same estimate with main results but changed the law implementation one year before the actual law implementation. The null result from the false specification would provides a validation that the main result is not driven by general trend in child labor around the period of study. The second estimation controls for labor supply pre-trend for each age group in linear form.

The results for false policy enforcement show no effect on child labor (columns 1 and 2, Table

C.1). The data used in this section are survey collected in the year 2011, 2012, and 2013.26 I consider the year 2013 as the year of false policy implementation (one year before the actual law enforcement). The results show non-significant impact of the false policy suggesting that the main result in this paper not likely to be driven by trends in labor supply around the time of the study period.

The results with linear control of pre-trend labor supply for each group is shown in columns

4 and 6.27 The coefficient of the boy’s labor supply aged 12-13 is 4.3, and remains similar in magnitude with the main estimate (3.7 percentage points, columns 3).

26The data is not collected in 2010, and the data prior to 2011 was 2007. 27The before data used in this section is extended to 2011 to control for pre-trend in a longer period (the results do not change even if the sample is restricted to 2012, 2013, and 2014).

27 Table C.1: Effect of the Policy on Probability of Engaging in Child Labor for Family Business Robustness Check: Trend in Labor Supply

1 Year Before Controlling for a Linear Time Trend (1) (2) (3) (4) (5) (6) Boys Girls Boys Boys Girls Girls False After × 10-11 0.005 -0.020 (0.017) (0.017) False After × 12-13 0.024 0.013 (0.017) (0.017) After × 10-11 0.001 -0.010 0.002 -0.021 (0.016) (0.024) (0.016) (0.025) After × 12-13 0.037∗∗ 0.043∗ 0.001 0.016 (0.015) (0.025) (0.016) (0.025)

After/Exposed Yes Yes Yes Yes Yes Yes Year/Region control Yes Yes Yes Yes Yes Yes Household control Yes Yes Yes Yes Yes Yes Linear Pre-Trend No No No Yes No Yes Age dummy control Yes Yes Yes Yes Yes Yes Mean (Age 10-11 before policy) 0.20 0.18 0.18 0.18 0.16 0.16 Mean (Age 12-13 before policy) 0.19 0.20 0.18 0.18 0.20 0.20 Adjusted R2 0.27 0.24 0.29 0.29 0.24 0.24 Observation 10,936 10,779 14,839 14,839 14,631 14,631 Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The Year/Region controls for year of data collection and nine fixed effects of regions (departamiento). The Household control includes dummy variables of household size, residence of urban/rural, and father’s/mother’s education. Age dummy is the dummy variables of each age.

28 C.2 Spill-over to Younger/Older Siblings

Children in the control age group may be affected by the policy if they have siblings exposed to the policy. If it exists, the spillover of labor to siblings becomes a threat to the identification assumption. Given that the study documents an increase in boys’ labor supply, the estimate in this section includes the term of sibling spillover: Effect of having a male sibling in the exposed age group.

yit = α + β1Afterit

+ δ1MaleSibExposed(7-9)it + β2Exposed(10-11)it + β3Exposed(12-13)it + δ2MaleSibExposed(14-16)it

+ [δ3MaleSibExposed(7-9) + β4Exposed(10-11) + β5Exposed(12-13) + δ4MaleSibExposed(14-16)it]

× Afterit + βX Xit + θt + it (2)

The equation (2) includes MaleSibExposed(7-9) and MaleSibExposed(14-16) and its interaction to Afterit. δ3 captures the change in labor supply for children whose age is 7-9 years old, and have an older male sibling who is exposed to the policy. Similarly, δ4 capture the labor supply change for children aged 14-16 years old whose younger male sibling is exposed to the policy.

Table C.2 shows the estimation results from equation (2). No statistically significant spillover is detected. In the preferred specification (column 3), the coefficient of the boy’s labor supply aged

12-13 is 4.2. This is slightly higher than 3.7, the estimation result from equation (1).

29 Table C.2: Effect of the Policy on Probability of Engaging in Child Labor for Family Business (Spillover)

Boys Girls (1) (2) (3) (4) (5) (6) After × Siblings Exposed (Aged 7-9) 0.014 0.029 0.028 -0.043 -0.014 -0.014 (0.036) (0.032) (0.032) (0.036) (0.034) (0.033)

After × Exposed (Aged 10-11) -0.000 0.011 0.011 0.017 0.010 0.010 (0.018) (0.017) (0.017) (0.018) (0.016) (0.016)

After × Exposed (Aged 12-13) 0.039∗∗ 0.041∗∗ 0.042∗∗∗ -0.007 -0.004 -0.004 (0.018) (0.016) (0.016) (0.018) (0.016) (0.016)

After × Siblings Exposed (Aged 14-16) 0.028 0.043 0.045 0.015 0.043 0.043 (0.038) (0.035) (0.034) (0.036) (0.033) (0.033)

After/Exposed/Sib Exposed Yes Yes Yes Yes Yes Yes Year/Region control Yes Yes Yes Yes Yes Yes Household control No Yes Yes No Yes Yes Age dummy control No No Yes No No Yes Mean (Age 7-9 before policy) 0.08 0.08 0.08 0.08 0.08 0.08 Mean (Age 10-11 before policy) 0.13 0.13 0.13 0.12 0.12 0.12 Mean (Age 12-13 before policy) 0.14 0.14 0.14 0.16 0.16 0.16 Mean (Age 14-16 before policy) 0.17 0.17 0.17 0.15 0.15 0.15 Adjusted R2 0.07 0.24 0.25 0.08 0.21 0.22 Observation 10,916 10,913 10,913 10,753 10,752 10,752 Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The outcome variable is the binary variable of child labor for a family business. The Year/Region controls for year of data collection and nine fixed effects of regions (departamiento). The Household control includes dummy variables of household size, residence of urban/rural, and father’s/mother’s education. Age dummy is the dummy variables of each age.

30 C.3 False Specification: False Outcomes

As the 2014 data were collected a few months after the implementation of the law, we do not expect any policy effect on permanent wealth outcomes. If the estimate from the main analysis happens to capture the relatively poor households (rather than the actual effect of the policy), we would expect outcomes correlated to permanent income to be statistically significantly different from zero.

Table C.3 presents the estimation results on the three outcomes that reflect the household wealth level, but not likely to be affected by the policy change in the short-term. The first one is the household asset index.28 The second is the household size.29 The third is the dummy variable whether the biological father has no education.30

The results do not show any significant results for those three outcomes.

Table C.3: Effect of the Policy on Household Wealth Measure Outcomes

Asset Household Father has index size no education (1) (2) (3) (4) (5) (6) Boys Girls Boys Girls Boys Girls After × Exposed (Aged 10-11) 0.042 -0.018 0.119 0.073 -0.005 -0.010 (0.033) (0.033) (0.098) (0.098) (0.007) (0.007)

After × Exposed (Aged 12-13) -0.030 -0.026 0.133 0.103 -0.006 0.005 (0.032) (0.033) (0.093) (0.097) (0.007) (0.007)

After/Exposed Yes Yes Yes Yes Yes Yes Year/Region control Yes Yes Yes Yes Yes Yes Household control Yes Yes Yes Yes Yes Yes Age dummy control Yes Yes Yes Yes Yes Yes Mean (Age 10-11) -0.21 -0.21 5.64 5.62 0.02 0.02 Mean (Age 12-13) -0.19 -0.19 5.50 5.63 0.02 0.02 Adjusted R2 0.65 0.64 0.12 0.11 0.04 0.04 Observation 10,913 10,752 10,913 10,752 10,915 10,753 Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The year/region control for year of data collection and fixed effects of regions (departamiento). The household control includes urban, parents education and household size. However, household size and parents education are omitted out from controls when they are estimated as outcomes.

28The household asset index is calculated using principal component analysis of various household characteristics; Home tenancy, material of wall , interior, roof, floor, bathroom, energy access, and garbage collection. 29In this specification, the household size is omitted from the control variables. 30In this specification, parents’ education is omitted from the control variables.

31 D Labor Hours and Overwork

Table D.1: Effect of the Policy on Child Labor Hours and Overwork

Boys Girls (1) (2) (3) (4) (5) (6) No Non-farm Farm No Non-farm Farm family family family family family family Sample business business business business business business Outcome A: Weekly Labor Hours After × Exposed (Aged 10-11) -16.990 -1.619 0.586 -48.767∗ 5.443 2.083 (12.137) (3.857) (1.979) (25.681) (3.796) (1.996)

After × Exposed (Aged 12-13) -11.383 -1.184 -0.446 -9.407 0.931 0.842 (8.415) (3.201) (1.830) (14.947) (3.205) (1.921) Control Yes Yes Yes Yes Yes Yes Mean (Age 10-11 before policy) 16.00 21.14 16.84 16.25 19.32 17.24 Mean (Age 12-13 before policy) 21.43 20.04 19.62 26.06 22.80 20.67 Observation 62 457 1,185 52 503 1,060

Outcome B: Overwork After × Exposed (Aged 10-11) 0.053 -0.070 0.005 -0.269 0.043 0.049 (0.384) (0.105) (0.051) (0.766) (0.108) (0.050)

After × Exposed (Aged 12-13) 0.047 -0.033 -0.035 0.016 -0.109 0.025 (0.266) (0.087) (0.047) (0.446) (0.091) (0.048) Control Yes Yes Yes Yes Yes Yes Mean (Age 10-11 before policy) 0.00 0.10 0.07 0.00 0.17 0.05 Mean (Age 12-13 before policy) 0.27 0.10 0.12 0.12 0.23 0.11 Observation 62 457 1,185 52 503 1,060 Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The Year/Region controls for year of data collection and nine fixed effects of regions (departamiento). The Household control includes dummy variables of household size, residence of urban/rural, and father’s/mother’s education. Age dummy is the dummy variables of each age.

32 E Household Consumption

This section provides descriptive statistics and robustness for the analysis of monthly household food consumption. The amount of consumption is represented as per capita by 2010 USD constant.31

E.1 Descriptive Statistics of the Household Food Consumption

Figure E.1 shows the distribution of total food consumption and food consumption from the self- provision by household types.32 The average food consumption per capita is 30.1 USD for house- holds who engage in the farm business. In comparison, the same statistic is 41.2 USD for farm business households.

E.2 Robustness Checks

Figure E.1 (b) shows the distribution of food consumption from self-provision. Although farm households have lower overall food consumption, self-provision of food is larger for farm households.

The distribution has a long right tail as a few households produce a large amount of self-provision of food. To estimate the outcome with less skewed distribution, Table E.1 shows estimation results using IHS transformation. The increase in self-provisioned food due to the policy is 0.28 in log term.

The increase in food consumption is only observed when the estimate considers having exposed boys as treatment households. Table E.2 shows that having a policy-exposed girls in the household does not increase food consumption. This is consistent with the findings from labor supply, as girls’ labor supply is not affected by the policy reform.

31I divide the household consumption by the number of household members. 32Consumption from the self-provision is food consumption from own production. This excludes consumption from purchase. The top one percent of the distribution is trimmed as extreme value for the estimation. The consumption was adjusted to the inflation level in 2010.

33 Figure E.1: Density of Per capita Monthly Food Consumption by Household Type .03 .15 .1 .02 .05 .01 0

0 0 50 100 150 0 50 100 150 Monthly Production Consumption Percapita (2010 Boliviano) Monthly Consumption Percapita (2010 Boliviano) No family business Non−farm family business No family business Non−farm family business Farm family business Farm family business Notes: Exclude the case of zero. The number of household with zero production value is N=3570 (80.4%) for households with no family business, N=3990 (69%) for households with non−farm Notes: 143 USD = 1000 BOB family business, and N=217 (8.8%) for households with farm family business. 143 USD = 1000 BOB

(a) Monthly food consumption per capita (b) Monthly food self-provision per capita .5 .8 .4 .6 .3 .4 .2 .1 .2 0

0 0 2 4 6 8 0 2 4 6 8 Monthly Production Consumption Percapita (IHS) Monthly Consumption Percapita (IHS) No family business Non−farm family business No family business Non−farm family business Farm family business Farm family business Notes: Exclude the case of zero. The number of household with zero production value is N=3570 (80.4%) for households with no family business, N=3990 (69%) for households with non−farm Notes: family business, and N=217 (8.8%) for households with farm family business.

(c) Monthly food consumption per capita (IHS) (d) Monthly food self-provision per capita (IHS)

Table E.1: Effect of the Policy on Monthy Consumption Per Capita (IHS)

No family business Non-farm family business Farm family business (1) (2) (3) (4) (5) (6) (7) (8) (9) Purchase Pur- Purchase Pur- Purchase Pur- provision chase Provision provision chase Provision provision chase Provision combined only only combined only only combined only only After × Exposed 0.00 0.00 -0.06 0.01 -0.02 0.01 0.10∗∗ 0.02 0.28∗∗∗ (0.04) (0.04) (0.05) (0.03) (0.03) (0.06) (0.05) (0.07) (0.10) Control Yes Yes Yes Yes Yes Yes Yes Yes Yes Mean 4.35 4.32 0.28 4.33 4.22 0.63 4.12 3.56 2.61 Adjusted R2 0.12 0.16 0.22 0.13 0.13 0.12 0.24 0.36 0.17 Observation 4,521 4,521 4,521 5,844 5,844 5,844 2,463 2,463 2,463 Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The Control includes year of data collection, nine regions (departamiento), dummy variables of household size, residence of urban/rural, and father’s/mother’s education.

34 Table E.2: Effect of the Policy on Monthly Consumption per capita (Exposed Girls as Treatment)

No family business Non-farm family business Farm family business (1) (2) (3) (4) (5) (6) (7) (8) (9) Purchase Pur- Purchase Pur- Purchase Pur- provision chase Provision provision chase Provision provision chase Provision combined only only combined only only combined only only USD (2010) After × Exposed -2.51∗ -2.66∗∗ 0.15 -0.77 -0.47 -0.30 0.62 0.83 -0.21 (1.34) (1.32) (0.34) (1.06) (1.05) (0.43) (1.48) (1.21) (0.91) Control Yes Yes Yes Yes Yes Yes Yes Yes Yes Mean 42.52 41.18 1.35 40.40 37.76 2.64 33.22 22.52 10.70 Obs 4,444 4,444 4,444 5,746 5,746 5,746 2,443 2,443 2,443

IHS Transformation After × Exposed -0.06 -0.08∗ -0.03 0.01 0.02 -0.04 0.01 0.00 0.02 (0.04) (0.04) (0.06) (0.03) (0.04) (0.07) (0.05) (0.07) (0.10) Control Yes Yes Yes Yes Yes Yes Yes Yes Yes Mean 4.32 4.27 0.33 4.29 4.20 0.65 4.02 3.46 2.49 Obs 4,521 4,521 4,521 5,844 5,844 5,844 2,463 2,463 2,463 Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The year/region controls for year of data collection and regions (departamiento). The household control includes household size, urban, parents education.

35 F Schooling Outcomes

Table F.1 shows schooling outcome for children by age. The prevalence of children out of school is law for children under age 13.33 The rate of out of school is higher for children age 14-16, especially after the policy implementation.

Table F.1: Schooling for Children Age 7-16 in Bolivia, 2012-2014

Boys Girls Before After Difference Before After Difference Out of school 0.04 0.06 0.01 0.05 0.06 0.01 Among age 7-9 0.02 0.02 0 0.02 0.02 0 Among age 10-11 0.02 0.02 0.01 0.02 0.02 0 Among age 12-13 0.04 0.04 0 0.04 0.06 0.03 Among age 14-16 0.09 0.12 0.04 0.10 0.12 0.01 Years of schooling 5.45 5.69 0.24 5.58 5.77 0.19 Among age 7-9 2.30 2.54 0.23 2.32 2.57 0.25 Among age 10-11 4.57 4.76 0.19 4.66 4.83 0.17 Among age 12-13 6.27 6.57 0.30 6.46 6.60 0.15 Among age 14-16 8.52 8.59 0.07 8.56 8.74 0.18 Observations 7144 3772 10916 7051 3702 10753 Notes: Observation at the child level. The observation is at the individual level. The Before columns represent the data collected in the year 2012 and 2013. After represents the data collected in the year 2014.

Table F.2 provides estimation results for children out of school (either not enrolled or not attending) and years of schooling.34 Children in farm household shows a higher rate of out of school children, especially for children aged 12-13. The rate is seven percent for boys and eight percent for girls from farm households.

Although, the estimated results show an increase in child labor supply for boys from farming households, there is no change in schooling for this population (result shows higher girls out of school for aged 12-13). Change in schooling measured as years of schooling is not observed neither.

33Out-of-school is defined as children not enrolled into school or not attending school, excluding the case of school vacation. 34If the reason for not attending school is because of school vacation, the study does not count children as out of school.

36 Table F.2: Effect of the Policy on Education (Out of School and Years of Schooling)

Boys Girls (1) (2) (3) (4) (5) (6) No Non-farm Farm No Non-farm Farm family family family family family family Sample business business business business business business Outcome A: Out of School After × Exposed (Aged 10-11) -0.016 -0.021 0.006 0.000 0.002 -0.047 (0.018) (0.014) (0.029) (0.017) (0.014) (0.030) After × Exposed (Aged 12-13) -0.041∗∗ -0.003 -0.008 -0.009 0.009 0.059∗ (0.017) (0.013) (0.028) (0.017) (0.014) (0.030) Mean (Age 10-11 before policy) 0.02 0.02 0.02 0.01 0.01 0.04 Mean (Age 12-13 before policy) 0.04 0.02 0.07 0.04 0.01 0.08 Observation 3,616 5,052 2,538 3,703 4,896 2,440

Outcome B: Years of Education After × Exposed (Aged 10-11) -0.121 -0.005 0.110 -0.046 -0.122 0.115 (0.112) (0.096) (0.149) (0.110) (0.095) (0.149) After × Exposed (Aged 12-13) 0.159 0.055 0.216 0.028 -0.066 -0.149 (0.106) (0.090) (0.145) (0.107) (0.093) (0.148) Mean (Age 10-11 before policy) 4.71 4.68 4.18 4.77 4.78 4.25 Mean (Age 12-13 before policy) 6.40 6.37 5.88 6.61 6.60 5.98 Observation 3,609 5,049 2,537 3,698 4,889 2,435 After/Exposed Yes Yes Yes Yes Yes Yes Year/Region control Yes Yes Yes Yes Yes Yes Household control Yes Yes Yes Yes Yes Yes Age dummy control Yes Yes Yes Yes Yes Yes Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The Year/Region controls for year of data collection and nine regions (departamiento). The Household control includes dummy variables of household size, residence of urban/rural, father’s/mother’s education. Age dummy is the dummy variables of each age.

37 G Using Data After 2014

The reason that the main analysis employed 2014 data was that treated children surveyed in 2014 does not have any anticipation effect of the law.

Figure G.1 shows the treatment exposure by year of data collection. In 2015, children age 14 years old are also affected by policy change given that they were 13 years old at the time of the policy change (year 2014).

• Age 10: Children in this age knows that they can work once they turn to 10 years old.

• Age 11-13: Children in this age range are the ones that has been treated with no anticipation

effect.

• Age 14: Children in this age are exposed to the policy (at the time of policy change) .

However, they reached to the working age before the policy reform (I do not know I should

not expect any treatment effect or some treatment effect, given that children age 14 in control

group were able to work).

38 Figure G.1: Exposure to the Policy: Age and Year of Data Collection

Notes: The figure visualizes the sample by exposure to the policy. In the DID setting, the sample with blue serves as a control group. Children age 7, 8, 9, 15, and 16 are never affected by the policy change. Therefore, those children are the control group (not exposed to the policy). The second difference comes from time difference depending on the year of the data collection. The data collected before 2014 are called before data set and in 2014 is considered as after data set.

Table G.1: Effect of the Policy on Probability of Engaging in Child Labor for Family Business

Boys Girls (1) (2) (3) (4) (5) (6) Treat=1 -0.001 -0.012 0.045∗∗∗ 0.001 -0.013 0.037∗∗ (0.014) (0.013) (0.014) (0.014) (0.013) (0.015) Treat=3 0.013 0.015 0.072∗∗∗ 0.051∗∗∗ 0.049∗∗∗ 0.100∗∗∗ (0.014) (0.012) (0.014) (0.013) (0.012) (0.014) After × Treat=1 -0.005 -0.007 -0.004 0.027 0.030∗ 0.031∗ (0.019) (0.018) (0.018) (0.019) (0.018) (0.018) After × Treat=2 0.021 0.026∗∗ 0.029∗∗ -0.008 -0.008 -0.007 (0.013) (0.012) (0.012) (0.012) (0.012) (0.012) After × Treat=3 0.016 0.017 0.021 0.030 0.021 0.022 (0.022) (0.021) (0.021) (0.022) (0.021) (0.021)

After/Exposed/Sib Exposed Yes Yes Yes Yes Yes Yes Year/Region control Yes Yes Yes Yes Yes Yes Household control No Yes Yes No Yes Yes Age dummy control No No Yes No No Yes Observations 14459 14035 14035 14196 13840 13840 Adjusted R2 0.053 0.212 0.227 0.062 0.194 0.204 Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The outcome variable is the binary variable of child labor for a family business. The Year/Region controls for year of data collection and nine fixed effects of regions (departamiento). The Household control includes dummy variables of household size, residence of urban/rural, and father’s/mother’s education. Age dummy is the dummy variables of each age.

39 Table G.2: Effect of the Policy on Probability of Engaging in Child Labor for a Third Party

Boys Girls (1) (2) (3) (4) (5) (6) Treat=1 -0.037∗∗∗ -0.036∗∗∗ 0.007 -0.015∗∗∗ -0.014∗∗∗ 0.008 (0.008) (0.007) (0.008) (0.005) (0.006) (0.006) Treat=3 0.014∗ 0.012∗ 0.055∗∗∗ 0.012∗∗ 0.010∗ 0.033∗∗∗ (0.007) (0.007) (0.008) (0.005) (0.005) (0.006) After × Treat=1 -0.010 -0.011 -0.009 -0.009 -0.010 -0.010 (0.011) (0.011) (0.010) (0.008) (0.008) (0.008) After × Treat=2 -0.012∗ -0.011 -0.008 -0.003 -0.004 -0.003 (0.007) (0.007) (0.007) (0.005) (0.005) (0.005) After × Treat=3 -0.017 -0.010 -0.007 -0.016∗ -0.014 -0.013 (0.012) (0.012) (0.012) (0.009) (0.009) (0.009)

After/Exposed/Sib Exposed Yes Yes Yes Yes Yes Yes Year/Region control Yes Yes Yes Yes Yes Yes Household control No Yes Yes No Yes Yes Age dummy control No No Yes No No Yes Observations 14459 14035 14035 14196 13840 13840 Adjusted R2 0.010 0.022 0.061 0.004 0.014 0.033 Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The outcome variable is the binary variable of child labor for a family business. The Year/Region controls for year of data collection and nine fixed effects of regions (departamiento). The Household control includes dummy variables of household size, residence of urban/rural, and father’s/mother’s education. Age dummy is the dummy variables of each age.

Table G.3

Boys Girls (1) (2) (3) (4) (5) (6) Treat=1 -0.840∗∗∗ -0.814∗∗∗ 2.683∗∗∗ -0.936∗∗∗ -0.882∗∗∗ 2.753∗∗∗ (0.111) (0.111) (0.060) (0.114) (0.113) (0.061) Treat=3 2.707∗∗∗ 2.707∗∗∗ 6.255∗∗∗ 2.534∗∗∗ 2.559∗∗∗ 6.244∗∗∗ (0.109) (0.109) (0.059) (0.111) (0.110) (0.059) After × Treat=1 0.132 0.036 0.200∗∗∗ 0.131 0.033 0.111 (0.156) (0.158) (0.074) (0.159) (0.160) (0.074) After × Treat=2 0.106 0.003 0.170∗∗∗ 0.014 -0.050 0.053 (0.101) (0.103) (0.048) (0.103) (0.104) (0.048) After × Treat=3 0.216 0.037 0.268∗∗∗ 0.236 0.078 0.177∗∗ (0.179) (0.184) (0.087) (0.186) (0.190) (0.088)

After/Exposed/Sib Exposed Yes Yes Yes Yes Yes Yes Year/Region control Yes Yes Yes Yes Yes Yes Household control No Yes Yes No Yes Yes Age dummy control No No Yes No No Yes Observations 14440 14016 14016 14177 13821 13821 Adjusted R2 0.117 0.128 0.806 0.102 0.116 0.811 Notes: Standard errors in parentheses, ∗p < .10,∗∗ p < .05,∗∗∗ p < .01. The outcome variable is the binary variable of child labor for a family business. The Year/Region controls for year of data collection and nine fixed effects of regions (departamiento). The Household control includes dummy variables of household size, residence of urban/rural, and father’s/mother’s education. Age dummy is the dummy variables of each age.

40