A Semiparametric Geographically Weighted Regression Approach to Modeling Human Development Index Disparity in South Sulawesi

Source abstract

Human Development Index (HDI) values in the 24 regencies and cities of South Sulawesi ranged from 69.45 to 85.23 in 2024. This study asks which indicators account for that gap and whether their associations with HDI vary by location, using Semiparametric Geographically Weighted Regression (SGWR), which keeps some coefficients fixed province-wide and lets others vary locally. Nine predictors from BPS-Statistics of South Sulawesi Province were used, including three HDI dimension indicators (life expectancy, expected years of schooling and per-capita expenditure), whose coefficients are read as the weight of each dimension in the observed disparity. Distances were great-circle distances between regency capitals. Each coefficient was classified as local or global with a Monte Carlo test of spatial variability, bandwidths were chosen by AICc, and local t-tests used the da Silva–Fotheringham adjusted level. Only the intercept and the construction cost index varied significantly (p = 0.013 and 0.003), with the same result for polygon centroids. The resulting SGWR fitted slightly better than OLS (AICc 18.34 against 20.03, a difference below 2), a fully local GWR overfitted (AICc 40.13), and no local coefficient was significant. Expenditure per capita, expected years of schooling and life expectancy were the significant province-wide associations. HDI disparity across South Sulawesi's regencies is therefore driven mainly by the expenditure and education dimensions, in much the same way across the province.