MODELING POVERTY IN WEST JAVA PROVINCE USING NEGATIVE BINOMIAL REGRESSION WITH PENALIZED SMOOTHLY CLIPPED ABSOLUTE DEVIATION

Abstract

The number of poor people is an example of discrete or count data. One commonly used regression model for count responses is the Negative Binomial regression. Regression modeling with many predictor variables results in the problem of multicollinearity. This condition causes the parameter estimator to become unstable. One method to overcome this problem is to use the penalty function to optimize the selection of predictor variables. This study aims to analyze the factors influencing the number of poor people in West Java Province using Negative Binomial regression with the Smoothly Clipped Absolute Deviation (SCAD) penalty function. The research data was sourced from the Central Bureau of Statistics in 2022, covering 27 districts/cities in West Java Province with 21 predictor variables. The method applied selects variables and estimates parameters simultaneously in the Negative Binomial regression model. Based on the AIC value, it was found that the Negative Binomial penalized SCAD model (AIC = 628.12) had better performance than the Negative Binomial regression model (AIC = 634.34). The Negative Binomial penalized SCAD regression model yielded five significant predictor variables with value of 92.8%. This model is simpler than the Negative Binomial regression model with six predictor variables. The regional minimum wage, number of cooperatives, percentage of the population who have health insurance, the pure college enrollment rate, and non-food expenditure are important variables as factors affecting the number of poor people in West Java Province.