An Application Comparison of Two Negative Binomial Models on Rainfall Count Data

Counts data models cope with the response variable counts, where the number of times that a certain event occurs in a fixed point is called count data, its observations consists of non-negative integers values {0,1,2,}. Due to the nature of the count data, it is generally considered that response variables do not follow normal distribution. Therefore, because of the skewed distribution, linear regression is not an effective method for analyzing counting results. And hence, the use of the linear regression model to analyse count data is likely to bias the outcomes, “Negative binomial regression” is likely to be the optimal model for analyzing count data under these limitations. Researchers may sometimes count more zeros than expected. Going to count data with several Zeros gives rise to the “Zero-inflation” concept. In health, marketing, finance, econometrics, ecology, statistical quality control, geographical and environmental fields, data with abundant zeros is common when counting the incidence of certain behavioural and natural events, such as the frequency of alcohol consumption, drug consumption, the amount of cigarettes smoked, the incidence of earthquakes, rainfall, etc. The Negative Binomial, “Zero-Inflated Negative Binomial” (ZINB), and “Zero-Altered Negative Binomial” (ZANB) models were used in this paper to analyse rainfall data.