TREATMENT EFFECT AND SENSITIVITY ANALYSIS FOR SELF-SELECTED TREATMENT AND SELECTIVELY OBSERVED RESPONSE
2005/07/26
研討會日期 : 2005-07-26
時間 : 15:00
主講人 : Prof. Myoung-jae Lee
地點 : C棟103室
演講者簡介 : Prof. Myoung-jae Lee 為Ph.D.，University of Wisconsin-Madison (1989)。
現為香港中文大學客座教授。
其主要研究領域為Econometrics 及Statistics。
演講摘要 : In finding effects of a treatment d on a response (or performance) variable y, sometimes the performance variable is observed only if the subject participates first (δ = 1). For instance, if the treatment is a childhood education intervention and the response variable is college grades (performance), then the subject should enter a college first (participation) for the response variable to be observed. An intriguing question arises: should the counter-factual performance of non-participants be included in the treatment effects? Depending on the answer, in addition to “participation effect”, we get either “invisible” or “visible” performance effect. Two selection issues occur: d may be self-selected instead of being randomized, and y is observed only when δ = 1. Nonparametric estimators for participation and visible effects are proposed with little restriction on δ but under a ‘selection-on-observable’ assumption for d; then a sensitivity analysis is conducted to relax the assumption. Additionally assuming independence between δ and y, ‘weighting’ estimators for invisible effect are explored as well. An empirical example of job-training effects on female work-hours is provided and the empirical finding suggests no or little invisible effect: the visible effect is mostly owing to the participation effect.
