Bagging Based Heteroscedasticityadjusted Ridge Regression for District Level Poverty Modeling in West Nusa Tenggara Province

Abstrak

This study develops an applied bagging-based heteroscedasticity-adjusted ridge (HAR) regression framework for district-level poverty modeling in West Nusa Tenggara Province, Indonesia. The response is the percentage of people living below the poverty line, while the predictors represent life expectancy, expected years of schooling, mean years of schooling, adjusted expenditure per capita, and the open unemployment rate. Because a verified district-year microdataset was not supplied, the empirical section is explicitly presented as a reproducible illustration using 50 synthetic observations for the ten regencies and cities during 2020-2024, calibrated to plausible ranges reported in recent official publications. The predictor correlations range from 0.957 to 0.979 for the four human-development variables, with variance inflation factors between 17.47 and 44.26. The Breusch-Pagan statistic is 14.619 (p = 0.012), indicating heteroscedastic residuals. Leave-one-out cross-validation shows that the bagged HAR-HK estimator achieves the smallest predicted residual sum of squares (PRESS = 213.307), compared with 218.997 for HAR-HK and 231.562 for ordinary least squares. The bagged estimator therefore reduces PRESS by 2.60% relative to its HAR counterpart and by 7.88% relative to OLS. The results illustrate how bootstrap aggregation can stabilize ridge-parameter selection when regional development indicators are strongly collinear and residual variance is unequal. The numerical findings are methodological illustrations and must be re-estimated using quality-controlled BPS data before policy use.