Econometric methods for distributional policy effect


This project aims to develop new econometric methods that can measure distributional policy effects by accounting for heterogeneous policy impacts among observationally equivalent individuals. The project expects to develop quantile regression methods under a difference-in-differences framework and will further extend our analysis by accommodating issues of censoring and sample selection. The research outcomes of this project are expected to substantially broaden the scope of the standard mean difference-in-differences approach and have significant contributions to empirical studies in the future. The project intends to provide statistically valid inferential procedures and conduct simulation exercise and empirical studies.

People involved

Project background and aims

Causal analysis is one of fundamental objectives in economic research and has important implications for policies and programs evaluation. Despite its importance, identifying and estimating causal effects are challenging, because observational studies, rather than experimental ones, are more prevalent due to ethical concerns or logistical constraints.

A large body of the literature has shown that correlation information from observational data does not allow researchers to measure the causal effects of programs or policies. When researchers attempt to draw causal inferences about policy effects, they must use observational data to identify the counterfactual outcome that would have been realized in the absence of the policy.

If policy effects do not differ across individuals, the average of counterfactual outcomes serves as a basis for evaluating the policy effect. The econometrics and statistics literature, however, has highlighted the role of individual heterogeneity and documents the importance of accounting for the heterogeneity when policies are evaluated. Average effects are often used as a summary measure of policy effects especially in empirical studies, while there is relatively less work quantile effects, which allow researchers and policy makers to compare outcome distribution with and without a policy. This is surprising because the question of what kind of policies should be employed receives an enormous amount of attention in many sub-fields of economics, such as labour, development and health economics, and the effectiveness of policy should be evaluated based on welfare gain over the target population, rather than the average person. Quantile policy effects provide sufficient information on distribution impacts of the policy and the quantile effects can influence policy formation or business in industries. In addition, the recent improvement in data availability allows researchers to empirically document individual heterogeneity and the development in computational capabilities, due to advances in computational hardware and algorithmic advances, enables to implement methods previously considered impractical. Thus, causal analysis accounting for individual heterogeneity has become an increasingly important topic in econometric theory and can broaden the scope of empirical studies by uncovering heterogeneous impact of economic policies.

The overall aim of our research projects is to develop a set of new econometric methods that are designed to identify and estimate distributional policy effects in the present of the issues discussed above: individual heterogeneity, truncated outcomes due to censoring and sample selection, and the presence of unobservables correlated across regions. More precisely, we address those issues for estimating quantile treatment effects given panel or repeated cross-sectional data, and also we will apply our methods for empirical studies to demonstrate their usefulness. Our research projects consist of three inter-related projects that will extend the recent literature on quantile treatment effects by:

  1. accommodating the presence of censoring in QDID methods with conditioning variables,
  2. considering regional-level quantile regression with interactive fixed effects
  3. sample selection quantile model.