Abstract: When analyzing longitudinal data, conventional estimators can produce biased inference if cluster sizes are informative, meaning the frequency of repeated measures per subject is associated with the outcome. This presentation examines the performance and utility of modified Poisson regression paired with cluster-weighted generalized estimating equations (MP-CWGEE) to estimate individual-averaged marginal risk ratios in the presence of informative cluster sizes. Using results from a 1,000-iteration Monte Carlo simulation spanning 324 scenarios (evaluating empirical bias, convergence, and Type I error against five competing methods) alongside an applied cohort study of 423 people who used illicit opioids in New York City, we highlight MP-CWGEE’s robust performance and numerical stability relative to log-binomial and unweighted models. To conclude, the talk will touch on recent methodological extensions and showcase ongoing applied and theoretical work developed in collaboration with master’s and doctoral trainees in biostatistics.
About the Speaker: Jemar R. Bather, PhD, is an Assistant Professor of Biostatistics at the NYU School of Global Public Health and Co-Director of the Biosocial Approaches to Stress and Embodiment (BASE) Lab. His research intersects biostatistics, epidemiology, and social genomics, focusing on the biological embedding of psychosocial stress across the life course and complex survey methodology. He earned his PhD in Biostatistics from Harvard University.
