Exploring Propensity Score Methods With Semi-Continuous Exposure

• In Event: Application of Longitudinal Models and Statistical Theory

Abstract

This study aims to investigate the appropriate definitions for the propensity score of semi-continuous treatment and the corresponding weighting equations when using Hurdle and Zero-Inflated models. To achieve these aims, the study conducted a comprehensive simulation that examined the impact of various factors, including sample size, number of covariates, zero proportions of semi-continuous exposure, dispersion parameter values, and nonlinear term status, on the estimation of treatment effects. The study's findings indicate that neither of the two Semi-GPS definitions works well with the Hurdle model, and no weighting equations are effective for ATE estimation for the Hurdle and Zero-Inflated models. Based on these results, the study recommends using the conditional mean definition with the Zero-Inflated model to estimate the ATE.

Authors

• Huibin Zhang, University of Tennessee

• Walter L. Leite, University of Florida