Testing Rank Preservation in Quantile Treatment Effects Evaluation
2015/10/16
研討會日期 : 2015-10-16
時間 : 14:30
主講人 : Professor Ping Yu
主持人 : Professor Jen-Che Liao
地點 : Conference Room B110
演講者簡介 : Professor Yu received his Ph.D. in Economics from University of Wisconsin–Madison at 2009. He is currently an Assistant Professor at The University of Hong Kong. His research fields are Theoretical and Applied Microeconometrics.
演講摘要 : It is commonly believed that rank preservation cannot be tested in quantile treatment effects evaluation due to the fundamental problem of causal inference. In this paper, we propose a Hausman-type test to test unconditional rank preservation under unconfoundedness when covariates are available. The basic idea is that unconditional rank preservation implies conditional rank preservation but the converse is not true, so significant difference between two statistics with one preserving conditional rank and the other preserving unconditional rank is an indicator of rank nonpreservation. In other words, we are testing rank preservation across covariates values rather than within a covariate value. We develop both parametric and nonparametric tests for both the overall quantile treatment effect and the quantile treatment effect on the treated. Under the null, the parametric test statistics follow mixed chi-square distributions asymptotically, while the nonparametric test statistics follow normal distributions. We suggest to use the exchangeable bootstrap to obtain critical values for the former and the simulation method for the latter. We also extend the rank preservation tests on the treated to a special confounded case with one-sided noncompliance. We illustrate the tests under unconfoundedness on data from the National Supported Work Program and the tests with confoundedness on data from training programs administered under the Job Training Partnership Act in the United States.
