When the dependent variable is continuous, the linear regression analysis is performed by using Least Squares Method (OLS). However, if the dependent variable is discrete or count data, analysis using linear regression models will yield ineffective, inconsistent and contradictory results. Therefore, different regression models have been developed for count data. Among these, the best known regression models are Poisson and negative binomial regression models. Poisson regression model is used in case of equal dispersed in the application. In case of over-dispersed, generalized Poisson regression model or negative binomial regression model is preferred. This study investigates the comparison of Poisson and negative binomial regression models in case of over-dispersed. Empirical results show that the negative binomial regression model gives better results if the dependent variable shows over-dispersed. To confirm this, both models were compared with the AIC, BIC and G2 information criteria. In addition, marginal effects and incidence ratio (IRR: Incidence Ratio Rate) values were calculated to interpret the coefficients of the models. As a result, the presence of over-dispersed should be checked in cases to be analysed by Poisson regression and it should be taken into consideration that the analysis can be continued with negative binomial regression when it exists.
Primary Language | Turkish |
---|---|
Journal Section | Research Articles |
Authors | |
Publication Date | July 15, 2020 |
Submission Date | March 14, 2020 |
Published in Issue | Year 2020 Volume: 2 Issue: 1 |