Seminar SERIES

 

 

THURSDAY, JANUARY 27, 2022 | Zoom 3:30 p.m.

Zhehui Luo, PhD

Professor of Epidemiology and Biostatistics
Michigan State University

"Correlated random effects models for zero-inflated outcomes: An estimation of the impact of a state sponsored Medicaid prenatal home visiting program for black pregnant women"

Motivated by the need to evaluate a non-randomized program with zero-inflated count outcomes, we propose a correlated random effects model with unbalanced panels and use the simulated maximum likelihood estimation method to estimate the conditional effect of a multi-valued treatment. We then explore the potential to extend the Mundlak approach to this setting. We make three points in this paper (1) when the average treatment effects are the target parameters, others are all nuisance (from conditional models with correlated random slope to marginal between-within models); (2) the average treatment effect on the treated is sensitive to the proportion treated; and (3) no single method dominates the others for a causal effect of interest in observational studies.