Zero Truncated Models in Regression Analysis: An Examination of Their Advantages on Small Mean Values

Authors : Rıdvan Kara, Abdullah Yeşilova

Pages : 102-112

Doi:10.53433/yyufbed.1590611

View : 37 | Download : 34

Publication Date : 2025-04-29

Article Type : Research Paper

Abstract :In this study, two of the most commonly used zero-truncated regression models for modeling positive count data, namely Zero Truncated Poisson and Zero Truncated Negative Binomial, are compared with the classical Poisson and Negative Binomial regression models. The role of the mean of the dependent variable in model selection is examined. Simulations were first conducted using different mean values for the dependent variable, followed by a comparison of model performances using two different real data sets. The real data sets were constructed using crime data published by Turkish Statistical Institude (TSI). AIC, BIC, and residual plots were utilized to determine the most suitable model. The study found that zero-truncated models perform better when the mean of the dependent variable is below 5, compared to classical models.
Keywords : Negatif binom, Poisson, Sıfır değer kesilmiş negatif binom, Sıfır değer kesilmiş poisson