Modeling Mite Counts Using Poisson and Negative Binomial Regressions

Abstract

Poisson regression is frequently used for analyzing dependent variable based on count data. The main feature of the Poisson distribution is the assumption that the mean and variance are equal. However, this equal the mean and the variance relationship rarely occurs in application. In many case, the variance is larger than the mean, which is called overdispersion. When overdispersion occurs in a data set, negative binomial regression, in which parameter estimations are obtained by considering the effect that stems from overdispersion, is preferable. In this study, there was overdispersion in the number of mites. Therefore, parameter estimations were interpreted according to negative binomial regression. Effects of all independent variables were statistically significant on number of mites (p<0.01) in the negative binomial regression. The number of mites in Bardakci area was higher than that in Edremit and Samranalti (p < 0.01). Mite density in July was higher than that in other months (p < 0.01). The number of mites in the Golden variety was also found to be less than that in the Starking variety (p < 0.01).