The next Workshop on Quantitative Methods in Education, Health and the Social Sciences (QMeHSS) is this Friday, May 15th, from 11:00-12:30pm. The workshop will be led by Dr. Guanglei Hong from the University of Chicago. She will be joined by Cheng Yang, Xu Qin, and Edward Bein. As with previous workshops, it will be held in NORC conference room 344. NORC is located at 1155 E. 60th Street. We look forward to seeing you there.
Two-Step Estimation in Causal Mediation Analysis
Guanglei Hong, PhD
(Association Professor, Department of Comparative Human Development and the Committee on Education)
Cheng Yang (research scientist at NORC)
Xu Qin (Doctoral Student in the Department of Comparative Human Development)
Edward Bein (Statistician at Abt Associates)
Causal mediation analysis through ratio-of-mediator-probability weighting (RMPW), similar to some other strategies for mediation analysis, involves two steps. Step 1 fits a logistic regression model to each treatment group and estimates the coefficients from the sample data. A weight is computed as a ratio of two estimated counterfactual probabilities. Applying the estimated weight to the sample data, step 2 then estimates the causal effects of interest. This two-step estimation procedure is potentially problematic if the standard error of a causal effect estimate does not reflect the sampling uncertainty in the estimation of the weight. This study extends to RMPW analysis an m-estimation solution to the two-step estimation problem by stacking the score functions from both steps. We derive the asymptotic variance-covariance matrix for the two-step estimators. Simulation results show satisfactory performance of the two-step estimation procedure in comparison with bootstrapping. We further extend this solution to the two-step maximum likelihood estimation of direct and indirect effects when RMPW is applied to data from multisite randomized trials.