Extended abstract: The evidence on the effectiveness of universal child care programs on child development is far from unified, ranging from positive effects (especially for targeted programs) to negative effects (for universal child care programs). Such diverging results point to pervasive effect heterogeneity. We apply the unifying framework of marginal treatment effects (MTE) to study heterogeneous treatment effects of a universal child care program in Germany, exploiting exogenous variation caused by a reform that lead to a large expansion in available child care slots staggered across municipalities. We draw on unique administrative data from compulsory school entry examinations, which provide us with physician administered school readiness tests. We document that there is substantial heterogeneity in returns to early child care attendance with respect to unobservables (as well as observables). We find a positive but statistically insignificant average treatment effect (ATE), but returns to be positive and high for the average child not attending child care (TUT). The results reveal a pattern of ‘perverse selection on gains’: Children with lower (observed and unobserved) gains to child care are more likely to select into child care. We explain this by the fact that parents make the enrolment decision into child care based on multiple conflicting objectives, which include not only child well-being but also own labor market participation. Hence, children from more advantaged backgrounds are more likely to enter child care, but have smaller treatment effects because of their better counterfactual. We estimate that policies which would increase the probability of child care attendance for the children with a higher resistance to enter child care, e.g. through a quota for ethnic minority children or a further extension of child care slots, would yield positive and substantial returns. The seminar is a lunch seminar (lunch is on the house) - please sign up by e-mail to Centre Secretary Mette Vad Andersen