Panel Count Data Models: Examining Dengue Incidence in Malaysia

Understanding the discrete non-negative nature of count data is important for econometricians to model panel count data type. Panel count data is one of the nonlinear data type where the underlying assumption of a linear model to follow normal distribution is not appropriate. The Poisson distribution and its extensions are more suitable to provide the basis for modeling count and panel count data. The generalized linear model (GLM) is widely used for count data as these models fulfilled the statistical properties of such data that is it follows the exponential family distribution. However, if the count data type is repeated over individuals and nested in time period and known as panel data, the fixed effects Poisson and Negative Binomial Generalized Linear Model (GLM) is more appropriate choice to model panel count data. When there exists unobserved heterogeneity caused by spatial and temporal variability within the individuals, the fixed effects are replaced by the random effects in the GLM model. The random effects model allows the model to vary across individuals and time within the same group. Specifically, the negative binomial random effects model can effectively manage the unobserved heterogeneity and over-dispersion in the panel data. Thus, econometricians may consider several options of the fixed effects or the random effects of generalized linear model (GLM) to model panel count data. This chapter also discusses different types of panel count data models involving fixed effects and random effects of GLM models with an application to dengue disease panel count data. Other variants of dynamic panel count model and multivariate panel count model for multiple counts are also discussed in this recent development.