|
Activity Number:
|
127
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Social Statistics Section
|
| Abstract - #304807 |
|
Title:
|
Identification, Inference, and Sensitivity Analysis for Causal Mediation Effects
|
|
Author(s):
|
Kosuke Imai and Luke Keele and Teppei Yamamoto*+
|
|
Companies:
|
Princeton University and The Ohio State University and Princeton University
|
|
Address:
|
Department of Politics, Princeton, NJ, 08544,
|
|
Keywords:
|
causal inference ; causal mediation analysis ; direct and indirect effects ; linear structural equation models ; sequential ignorability ; unmeasured confounders
|
|
Abstract:
|
Causal mediation analysis is routinely conducted by applied researchers in various disciplines. In this paper, we first prove that the average causal mediation effect (ACME) is nonparametrically identified under a sequential ignorability assumption. We also show that the popular estimator based on the linear structural equation model (LSEM) can be interpreted as an ACME estimator if an additional assumption is satisfied. Second, we consider a simple nonparametric estimator of ACME to relax distributional and functional form assumptions. Third, we propose a new sensitivity analysis which can be easily implemented by applied researchers within the standard LSEM framework. Such an analysis is essential in many settings to examine the robustness of empirical findings to unmeasured confounding. Finally, we apply the proposed methods to a randomized experiment in political psychology.
|
