Negative Binomial Regression in Overcoming Overdispersion in Extreme Poverty Data in Indonesia

Abstract

Indonesia's extreme poverty status in 2021 was recorded to be high at 4% or 10.86 million people. One of the efforts in poverty alleviation is to analyze the factors influencing extreme poverty. Although the number of studies on poverty in Indonesia continues to grow, the findings are inconclusive because they are often discussed qualitatively. This study aimed to analyze the factors that influence extreme poverty in Indonesia using negative binomial regression. The data used was the amount of extreme poverty in 34 provinces of Indonesia as the response variable. Then, the explanatory variables used consist of 8 from the Central Bureau of Statistics. The analysis stage sought data exploration, the correlation between variables, Poisson regression model specification and assumption test, handling overdispersion with negative binomial regression, and model feasibility test. Based on the AIC value and dispersion ratio, the negative binomial model obtained an AIC value of 920.03 with a dispersion ratio 1.372. It shows that the negative binomial regression model is good enough to model extreme poverty in Indonesia. Furthermore, the factors significantly influencing extreme poverty in Indonesia are households with proper drinking water, housing status, and families with access to appropriate sanitation.