Modeling Semiparametric Geographically Weighted Regression Using Linear Model Coregionalization to Specify Variable’s Type

Abstract

Abstract Semiparametric geographically weighted regression (SGWR) is a regression model that contains two types of variables, namely global and local variability. The grouping of variables in this study utilizes the partial sill (psill) value that obtained from the output of the linear coregionalization model (LMC) resulting from the combination of three variogram functions. This study aims to identify variables that affect poverty in Papua Province in 2020, including literacy rate (LR), life expectancy (LE), school participation rate (SPR), RREB realization, population, per capita income, dependency ratio and labor force participation rate (LFPR). Based on the psill value, LR variable, RREB realization, population, per capita income, and LFPR as global variables while LE, SPR and dependency ratio variables are grouped as local variables. The five global variables have a significant influence on the number of poor people, while of the three local variables, only the dependency ratio has a significant influence in each location and others are not significant in certain locations. The results of the analysis showed that the SGWR1 model with its local variables had a proportion of non-nugget psill above 80% more in accordance with the data owned than the SGWR2 model which included variables with a proportion above 70%. This is indicated by the AICc value of the SGWR1 model of 76.504 while the SGWR2 model is 81.588.