PEMODELAN ANGKA HARAPAN HIDUP DI INDONESIA DENGAN PENDEKATAN BAYESIAN: BAYESIAN ADAPTIVE SAMPLING DAN BAYESIAN MODEL AVERAGING DALAM SELEKSI VARIABEL

Abstract

This study applies Bayesian Adaptive Sampling (BAS) and Bayesian Model Averaging (BMA) to model life expectancy in Indonesia, addressing model uncertainty and improving predictive accuracy. The analysis incorporates Zellner’s g-prior, which enhances variable selection by balancing prior information and data-driven learning, ensuring more stable and reliable parameter estimation. Bayesian methods provide greater flexibility compared to classical regression, particularly in managing heterogeneous demographic data. The results identify poverty rate, healthcare professional ratio per 1,000 residents, percentage of infants receiving exclusive breastfeeding, and regional health expenditure as key determinants of life expectancy. The poverty rate negatively impacts life expectancy, whereas the other factors contribute positively, highlighting the importance of healthcare access, infant nutrition, and government investment in public health. The final model achieves an R² of 78.1%, indicating that these variables collectively explain a substantial proportion of life expectancy variability. By integrating Zellner’s g-prior, Bayesian inference facilitates a robust selection of influential predictors, leading to more precise policy recommendations. The study suggests that enhancing healthcare distribution, promoting breastfeeding awareness, and optimizing health budget allocation can significantly improve life expectancy outcomes. Bayesian methods provide a powerful framework for demographic modeling by incorporating uncertainty and refining estimation accuracy.