Minimum working age and child labor


An increase in the minimum working age in Brazil has no effect on child employment overall. It does, however, tend to reduce child labor in regions with high inspection rates, suggesting considerable heterogeneity in enforcement.

Global progress in eradicating child labor has stalled for the first time in 20 years. The latest estimates show a notable increase of over 8 million child workers since 2016 (UNICEF/ILO, 2021). In total, 160 million children worldwide, i.e. one in ten, are concerned with a form of labor. The situation has worsened recently with the COVID-19 crisis and its consequences in terms of school closures, economic hardships and health shocks for millions of families in the world.

One of the policies put high on policy agendas is the mere ban on child labor, taking the form of a minimum employment age. This type of regulation involves the ratification of the ILO Convention 138 accompanied by national measures. In principle, minimum working age laws, if applied uniformly to all sectors, should reduce child labor. They may be ineffective in practice if they simply lead to a reallocation of child labor from the regulated sector to an unregulated one, such as family businesses, where the law is not applied or difficult to enforce. This type of law may even have adverse effects if the lower wages associated with the additional risk to firms lead budget-constrained households to make their children work more (Basu, 2005; Bharadwaj et al., 2020).

To date, the empirical literature on the subject confirms that these laws are broadly ineffective (Edmonds and Shretha, 2012). Part of the explanation is contextual : these laws often tend to follow declining trends in child labor, rather than leading them. More generally, approaches based on a crude comparison of employment rates above and below the minimum legal age make it difficult to disentangle the effect of the law from nonlinear age-employment trends (Edmonds, 2014). Against this background, we suggest new evidence based on a change in the Brazilian law leading to a quasi-experimental setting.

The Brazilian minimum age reform: a quasi-experimental setting

On the 15th of December 1998, the minimum legal age for work was increased from 14 to 16 in Brazil. This country is an interesting case study as it met the conditions of a credible enforcement immediately after the announcement of the reform. This reform involved a change in the Constitution and was part of a broad set of measures intended to eradicate child labor in Brazil (such as conditional cash transfers programs). More importantly, enforcement could be supported by an operational labor inspection system that had been refocused to detect illegal labor in general and child labor in particular. In other words, if such a reform had any chance of working, it would be within the framework of the Brazilian reform of 1998

What we suggest in a recently published paper (Bargain and Boutin, 2021) is to exploit the features of the reform, which allow isolating a causal effect of raising the minimum working age on child labor. Precisely, the reform lends itself to discontinuity-based approaches since the new ban affected those who turned 14 years old after mid-December 1998 but not those who turned 14 before. This discontinuous eligibility based on children’s birthdate is used in a regression discontinuity design (RDD) to infer the potential effect of the reform from the employment differential between these half-cohorts at the cutoff. 

Combining the discontinuous treatment with the time change around the threshold is also possible using a difference-in-discontinuity approach (DDisc). We subtract from the difference in employment between treated and untreated half-cohorts the same difference the year before, i.e. before the implementation of the reform, either using the same pair of half-cohorts at a younger age (same-cohort approach) or the older cohort taken at the same age as our main groups of interest (same-age approach). This approach overcomes the difficulties pertaining to nonlinearities in age-employment trends that may invalidate difference-in-difference approaches (such as Piza and Portela Souza, 2016).

Findings: No general effect of the ban on child labor

RDD results point to the absence of a discontinuous change in child employment rate around the threshold, i.e. for those born after mid-December 1984 compared to those born before. DDisc estimations confirm this finding. DDisc results are illustrated in the graphs below. We plot the difference in employment rates by birth-date cells (in weeks) between September 1999 and September 1998, first for the same cohorts (upper graph), then at same age (lower graph). With the same-cohort approach, the employment difference is positive because the same children observed in 1999 are one year older and more likely to work. In the same-age approach, there is a one-year difference in employment between consecutive cohorts observed at the same age.


Importantly, in both graphs, all the birthdate cells to the right of the cutoff are relatively well aligned with those to the left: there is no sign of an abrupt drop in employment at the cutoff. Many sensitivity analyses on model specification, employment definition, power calculation, etc., confirm that there is no sign of an overall employment effect of the new ban on children around 14 years of age. We also find no evidence of work substitution within families (for instance, children withdrawing from the labor market and replaced by more work by the mother).

Heterogeneity: Inspectable Workplace and Regional Degree of Enforcement

Admittedly, great heterogeneity in enforcement and compliance across regions and types of employment was expected. We then explore potential heterogeneous effects of the reform. We question whether child activities in workplaces that are deemed more visible/accessible to labor inspectors (i.e., more “inspectable”) may induce better compliance with the new law. Inspectable work is defined as taking place in shops or factories while less-inspectable work is that done in private homes (e.g., domestic work), in vehicles, or outdoors (e.g., street vending). We do not find marked effects when focusing on inspectable workplaces. However, a significant diminishing effect of the new law on child labor concerns regions where the pre-existing rate of labor inspections is high. [1] This result can be interpreted as suggesting evidence of there being a law—and an enforceable discontinuous treatment.

1. We interact the treatment variable with high and low intensity of treatment defined as living in a state with above or below average rate of inspection by firm in the municipality.


Bargain, O., & Boutin, D. (2021). Minimum age regulation and child labor: New evidence from Brazil.  World Bank Economic Review35(1), 234-260.

Basu, K. 2005. “Child Labor and the Law: Notes on Possible Pathologies.” Economics Letters 87 (2): 169–74.

Bharadwaj, P., Lakdawala, L. K., & Li, N. (2020). Perverse consequences of well intentioned regulation: Evidence from India’s child labor ban. Journal of the European Economic Association18(3), 1158-1195.

Edmonds, E.V. (2014): "Does minimum age of employment regulation reduce child labor?", IZA World of Labour, 73

Edmonds, E. V., and M. Shrestha. 2012. “The Impact of Minimum Age of Employment Regulation on Child Labor and Schooling.” IZA Journal of Labor Policy 1 (1): 14.

Piza, C. and A. Portela Souza (2016): "The Causal Impacts of Child Labor Law in Brazil: Some Preliminary Findings", World Bank Economic Review, papers and proceedings, 30(1), S137-S144

UNICEF/ILO (2021), Child Labour: Global estimates 2020, trends and the road forward

Share this