Estimating Marginal Effects for Zero-inflated Models: Introducing the R package mzim

• In Event: Innovative Statistical Methods and Computational Tools for Complex Research Designs

Abstract

Count data in psychological and medical research are often zero-inflated, complicating analysis. Traditional zero-inflated models (ZIP/ZINB) address this but provide complex, conditional parameter estimates that are difficult to interpret for population-level effects. To overcome this, marginalized zero-inflated models (mZIP/mZINB) directly estimate the population-averaged effect, offering more straightforward interpretations. However, a lack of accessible software has limited their adoption. This paper introduces mzim, the first comprehensive R package for estimating and interpreting these marginalized models. The paper serves as a tutorial, demonstrating mzim's workflow and its advantages over traditional approaches through an empirical analysis of abuse experiences, bridging a gap between statistical theory and applied research.

Authors

• Chendong Li, Texas A&M University

• Oiman Kwok, Texas A&M University