|
Abstract:
|
Count data are commonly used in biostatistics, demography, actuarial science, and economics (Cameron & Trivedi, 1998). They are often modeled using Poisson or negative binomial distributions. One of the issues that can appear in count data is having many zero values. That is when the observed data show a higher relative frequency of zeros than what is consistent with the Poisson model (Cameron & Trivedi, 1998). According to Miller (2007), analyzing this type of data as if they come from Poisson distribution has some unwelcome consequences. One of the assumptions of Poisson distribution is having equal values of mean and variance (McCullagh & Nelder, 1989). However when there is inflation in the number of zeros in the data, this assumption is violated and the variance becomes greater than the mean and that is what is called overdispersion (Zorn, 1996). One of the solutions to handle this type of data is to model them using zero-inflated Poisson (ZIP) model that has joint distribution Lambert (1992). In my paper, I used longitudinal type of data where there is inflation of the zeros. Therefore, I modeled the data using zero-inflated Poisson model. I compared two different zero-inflat
|
