Script for: Zhou et al. (2023). A Simulation Study of the Performance of Statistical Models for Count Outcomes with Excessive Zeros
Description
This repository contains the R code for the data generation and analysis for the paper: Zhou, Z., Li, D., Huh, D., Xie, M., & Mun, E. Y. (2023). A Simulation Study of the Performance of Statistical Models for Count Outcomes with Excessive Zeros. arXiv preprint arXiv:2301.12674. Abstract Background: Outcome measures that are count variables with excessive zeros are common in health behaviors research. Examples include the number of standard drinks consumed or alcohol-related problems experienced over time. There is a lack of empirical data about the relative performance of prevailing statistical models for assessing the efficacy of interventions when outcomes are zero-inflated, particularly compared with recently developed marginalized count regression approaches for such data. Methods: The current simulation study examined five commonly used approaches for analyzing count outcomes, including two linear models (with outcomes on raw and log-transformed scales, respectively) and three prevailing count distribution-based models (i.e., Poisson, negative binomial, and zero-inflated Poisson (ZIP) models). We also considered the marginalized zero-inflated Poisson (MZIP) model, a novel alternative that estimates the overall effects on the population mean while adjusting for zero-inflation. Motivated by alcohol misuse prevention trials, extensive simulations were conducted to evaluate and compare the statistical power and Type I error rate of candidate statistical models and approaches across data conditions that varied in sample size (N = 100 to 500), zero rate (0.2 to 0.8), and intervention effect sizes conditions. Results: Under zero-inflation, the Poisson model failed to control the Type I error rate, resulting in higher than expected false positive results. When the intervention effects on the zero (vs. non-zero) and count parts were in the same direction, the MZIP model had the highest statistical power, followed by the linear model with outcomes on the raw scale, negative binomial model, and ZIP model. The performance of linear model with a log-transformed outcome variable was unsatisfactory. When only one of the effects on the zero (vs. non-zero) part and the count part existed, the ZIP model had the highest statistical power. Conclusions: The MZIP model demonstrated better statistical properties in detecting true intervention effects and controlling false positive results for zero-inflated count outcomes. This MZIP model may serve as an appealing analytical approach to evaluating overall intervention effects in studies with count outcomes marked by excessive zeros.
Files
Steps to reproduce
Steps to follow: 1) Download the following files: "script_source_final.R" (source R code) and “script_simulation_analysis.R" (R script to run simulation and analysis). 2) Run “script_simulation_analysis.R" in R to replicate the simulation and analysis. All necessary instructions are included.