We study the effects of enrichment activities such as reading, homework, and extracurricular lessons on children’s cognitive and non-cognitive skills. We take into consideration the opportunity cost of spending time on enrichment, as it may replace activities such as sleep and socializing.

Our study controls for selection on unobservables using a control function approach that leverages the fact that many children spend zero hours per week on enrichment activities. At zero enrichment, confounders vary but enrichment does not, giving us direct information about the effect of confounders on skills. Using time diary data available in the Panel Study of Income Dynamics (PSID), we find that the net effect of the last hour of enrichment is close to zero for cognitive skills and negative for non-cognitive skills.

The negative effects for non-cognitive skills are concentrated in high school, consistent with elevated academic pressure related to college admissions.

…We thus adopt an alternative approach to control for confounders that exploits the fact that children in our data bunch at zero hours of enrichment activities per week (see Figure 1). We argue that many of these children are at a corner solution: time spent on an activity cannot be negative, so children with low propensities to spend time on enrichment all choose the lowest feasible amount of enrichment, zero.¹ This yields useful variation in confounders: at zero enrichment, all children chose the same amount of the “treatment”, namely zero. However, these children have different propensities towards enrichment: while some of them are nearly indifferent between their choice of zero enrichment and some other activity, others are far from indifferent; even a large shift in the costs/benefits of enrichment time would not induce them to move away from zero hours. Thus, at zero enrichment, confounders vary but treatment does not, so the variation in the outcome among observations at zero enrichment gives us direct information about the effect of confounders on skills. We use this idea to create a control function approach that corrects for the effect of confounders on skills.

This approach is developed formally in Caetano et al 2023 and has been also used in other contexts, such as the study of the effect of maternal labor supply on the skills of the child (Caetano et al 2021). A related approach has also been used to test a selection-on-observables assumption in the time use literature discussed earlier, when other time inputs are included as controls (eg. Caetano et al 2019 and Jürges & Khanam2021).

Using time diary data from the Child Development Supplement (CDS) of the Panel Study of Income Dynamics (PSID), we find that the net effect of enrichment on cognitive skills is small and indistinguishable from zero and that the net effect of enrichment on non-cognitive skills is quite negative and statistically-significant. This negative effect on non-cognitive skills is concentrated in high school, which is when enrichment activities become more oriented around homework and less oriented around social activities. Our measure of non-cognitive skills combines both externalizing behaviors related to outward aggression/antisocial behavior and internalizing behaviors related to anxiety/depression/self esteem. Despite the apparent differences in these behaviors, we nonetheless find similar negative effects considering internalizing and externalizing behaviors separately. We likewise find null effects for each of the 3 constituent achievement tests that combine to form our cognitive skills measure.

…Our findings highlight the pitfalls and trade-offs associated with intensive investment in enrichment activities, especially around high school, when enrichment activities become more oriented towards academic activities. Many youth seem to be spending so much time on enrichment that, on average, their last hour on these activities is actively harming their non-cognitive skills with no offsetting gain to their cognitive skills.

…In addition to these substantive empirical results, our paper highlights the potential for using bunching in the treatment to correct for selection on unobservables. Bunching of a treatment variable at one extreme due to a constraint is very common in many applied settings, including settings of interest to education and child development researchers.

For example, many other activities measured in terms of time use display bunching, and indeed we have examined in other work the effects of maternal working hours and television time on childhood skills (Caetano et al 2023, Caetano et al 2021). The method could be applied to study the effects on childhood development of many other types of activities including social media usage, homework, active time with parents, time with friends, playing sports, etc. (Caetano et al 2019, Jürges & Khanam2021). More broadly, consumption variables often display bunching, so for instance the effect on fetal and childhood development of maternal consumption of goods such as cigarettes (Almond et al 2005_19ya, Caetano2015, Oken et al 2008_16ya) and vitamins (Fawzi et al 1993_31ya) could also be studied with the method. Finally, it is possible to study the effect of a school resource on student outcomes, as long as that school resource is bunched at the extreme of the distribution. For instance, the distribution of specific school resources across schools might be bunched at zero, such as the amount of dollars allocated to a given subcategory of school spending (eg. funding for school police as in Weisburst2019), or the proportion of students attending the school who are eligible for subsidized school meals (Millimet & Tchernis2013).

See Also: