Modelling Clustered and Hierarchical Count Data: Poisson-Gamma Regression
64th ISI World Statistics Congress - Ottawa, Canada
Format: CPS Abstract
Keywords: clustered-data, count, generalized linear models, mixed-mode, mixed-models
Session: CPS 06 - Clustering
Monday 17 July 8:30 a.m. - 9:40 a.m. (Canada/Eastern)
Count data are very common in many practical fields such as medical studies, biology, epidemiology, and actuarial sciences. To analyze such data, Poisson regression is considered in the context of a generalized linear model. However, in a real data set, count responses may be clustered and hierarchical indicating the clustering effects. A hierarchical Poisson mixed model is a further improvement for analyzing such data by incorporating clustering effects, where random effects distribution is usually assumed to be normal. However, random effects distribution may be a member of the conjugate family i.e. a gamma distribution for count responses. In this study, we explored the performance of conjugate random effects distribution in the generalized linear mixed model framework by using both simulation studies and a real-life data set extracted from the latest Bangladesh Demographic and Health Survey (BDHS) 2017–18.