Propensity Score Estimation Models: Use of Complementary Log-Log Link Function

• In Event: Poster Session 2
  In Poster Session: Division D Section 2 Poster Session 1

Abstract

This study examined four different propensity score estimation models with the use of different link functions under single level and multilevel modeling frameworks through a simulation study. When the proportion of units within control and treatment groups was extremely imbalanced, a propensity score estimation model with complementary log-log link performed better than those with logit link. A single level propensity score estimation models performed better than multilevel propensity score estimation models in terms of the accuracy in the estimation of causal effects on an outcome. The findings suggested the use of complementary log-log link as an alternative to logit link when there is a relatively small treatment group and the vast majority of units are in a control group.

Authors

• HyeSun Lee, University of Nebraska - Lincoln

• Kurt F. Geisinger, University of Nebraska