A Generalized Conway–Maxwell–Poisson Model for Handling Inflation, Truncation, and Dispersion

Abstract:

Count data in social science studies exhibit various issues, including truncation, inflation, and under- or over-dispersion issues, making them unsuitable for standard Poisson and Negative Binomial models. Conway-Maxwell-Poisson (CMP) model has been recommended for its flexibility and interpretability of regression coefficients. However, empirical implementations remain limited, particularly in widely used commercial software familiar to social scientists.

To address this limitation, this study implemented the CMP model and extended it to a general modelling strategy capable of incorporating truncation, inflation, and potential clustering. A series of simulations were conducted to evaluate the implementation under various truncation and inflation scenarios.

Our results indicated that standard models exhibited substantial bias and poor coverage, deteriorating performance as inflation increased, and our CMP implementation yielding unbiased estimates, low RMSEs, and satisfactory coverage. Additionally, the two methods of approximating the normalizing constant Z in the CMP model were compared.

Speaker:

Prof. Tianji CAI

Professor
Department of Sociology
University of Macau

Webinar Links:

Related Posts