Poverty Modeling in Indonesia using Geographically and Temporally Weighted Regression (GTWR)

Abstract

Poverty is a big problem that must be resolved by the government and the people of Indonesia. Various programs are designed and implemented to alleviate poverty in Indonesia. Research is needed to find out what factors influence the problem of poverty. One statistical method that can be used to analyze this effect is the geographically and temporally weighted regression (GTWR) method. This method combines the effects of spatial and time simultaneously. The formation of the model begins with determining the weighting matrix. In determining the weighting matrix, a fixed kernel function is used where the bandwidth value for each location and time of observation is the same. Weighting matrix with kernel functions used are gaussian, bi-square, exponential and tricube kernel functions. The selection of the best model is done by comparing the GTWR model from each of the weighting matrices of the four kernel functions. The best model is determined by looking at the largest R2 value and the smallest AIC. Based on the results of the data processing, the GTWR model with the weighting matrix of the exponential kernel function has the largest R^2=71,05% value and the smallest AIC=718,5934. Variables that have a significant effect on the model differ in each location and time of observation. Significant predictor variables were determined by comparing the values of t and values t in statistic . The predictor variable is significant when t values  are bigger than values t in statistic. The results of data analysis show that the variable life expectancy (UHH) has an influence in most provinces in Indonesia in each year of observation.