Statistical Models for Count Data

Abstract

Statistical analyses involving count data may take several forms depending on the context of use, that is; simple counts such as the number of plants in a particular field and categorical data in which counts represent the number of items falling in each of the several categories. The mostly adapted model for analyzing count data is the Poisson model. Other models that can be considered for modeling count data are the negative binomial and the hurdle models. It is of great importance that these models are systematically considered and compared before choosing one at the expense of others to handle count data. In real world situations count data sets may have zero counts which have an importance attached to them. In this work, statistical simulation technique was used to compare the performance of these count data models. Count data sets with different proportions of zero were simulated. Akaike Information Criterion (AIC) was used in the simulation study to compare how well several count data models fit the simulated datasets. From the results of the study it was concluded that negative binomial model fits better to over-dispersed data which has below 0.3 proportion of zeros and that hurdle model performs better in data with 0.3 and above proportion of zero.