Robust Mixed Geographically and Temporally Weighted Regression to Modeling the Percentage of Poverty Population in Java in 2012-2018

Abstract

Poverty is a fundamental problem in Indonesia, as Java the island with the largest number of poor people. To monitor poverty in regency/municipality in Java Island a model is needed that can explain the diversity of locations and times, such as the GTWR model. Adding global effects to the GTWR model provides a more flexible model and makes interpretation easier, this model is known as MGTWR. Outlier problems in MGTWR as well as in linear regression models can produce biased coefficients. This problem can be solved by adding weight to the error, resulting in the model called RMGTWR. This study aims to model the percentage of poor population data in regency/municipality on Java Island in 2012-2018 and find out the factors that influence it. RMGTWR with bisquare kernel has pseudoR² 72.73%. Factors that influence significantly on global variables are literacy rates. Whereas, significant local variables are education completed by primary schools, per capita expenditure, recipients of Raskin households, population groups in the age group of 15-64 years, and the average school year are local variables. The coefficient of RMTGWR produces a parameter estimate that is relevant to the trend between poverty and predictors.