學術研究

Quasi-Difference in Differences with Panel Data: Complete Pairing and Linear-Model LSE

日期 2012-01-10 (週二)
時間 02:30 PM
地點 Conference Room C103
演講者 Professor Myoung-Jae Lee
演講者簡介 Myoung-Jae Lee got his Ph.D. in Economics from University of Wisconsin-Madison in 1989. He is currently serving as Professor of Economics at Korea University. His research field is in econometrics and statistics, and labor and health economics.
摘要 With one treated and one untreated periods, difference in differences (DD) requires a ‘zero selection-effect (or the same time-effect) condition' that the untreated response changes be the same across the treatment and control groups, if the treatment were withheld contrary to the fact. A natural way to check out the plausibility of the condition is to backtrack one period and examine the response changes in two pre-treatment periods. If the condition holds in the pre-treatment periods, then one would go ahead and apply DD; otherwise, the pre-treatment discrepancy/gap should be accounted for in modifying DD. Under the assumption that the pre-treatment gap is persistent, the modified DD takes the form of ‘quasi-difference in differences (QD)' that may be also called ‘triple differences'. QD generalizes DD with a weaker identification condition: a time-constant, but not necessarily zero, selection effect. This paper addresses how to test whether the pre-treatment gap is persistent or not, and how to modify DD with QD accordingly. Nonparametric estimators/tests and panel linear-model-based LSE/tests are explored. A simulation study is presented to show that the proposed estimators/tests work as they are supposed to.
檔案下載226.81 KB