🤖 AI Summary
This paper addresses the fundamental identification challenge in causal inference concerning heterogeneous treatment effects: individual treatment effects (ITEs) are inherently unobserved, and counterfactual outcomes are missing. To tackle this, we propose a novel nonparametric sharp bounding method that relies solely on the marginal distribution of the observed outcome variable—without imposing distributional assumptions or conditional independence restrictions on covariates. For the first time, this approach rigorously characterizes the maximal information about heterogeneity that the data can provide. By integrating marginal distribution constraints with counterfactual reasoning, it enables nonparametric identification of heterogeneity structures beyond the average treatment effect (ATE). Empirically, applying the method to microfinance interventions reveals significant effect heterogeneity—even when the ATE is statistically insignificant; in a welfare reform evaluation, it uncovers bidirectional changes in working hours for a substantial share of workers, exposing structural divergence in policy impacts.
📝 Abstract
Treatment effect heterogeneity is of major interest in economics, but its assessment is often hindered by the fundamental lack of identification of the individual treatment effects. For example, we may want to assess the effect of a poverty reduction measure at different levels of poverty, but the causal effects on wealth at different wealth levels are not identified. Or, we may be interested in the proportion of workers who benefit from the minimum wage increase, but the proportion is not identified in the absence of counterfactuals. This paper derives bounds useful in such situations, which only depend on the marginal distributions of the outcomes. The bounds are nonparametrically sharp, making clear the maximum extent to which the data can speak about the heterogeneity of the treatment effects. An application to microfinance shows that the bounds can be informative even when the average treatment effects are not significant. Another application to the welfare reform identifies a nonnegligible portion of workers who increased and decreased working hours due to the reform.
