MODELING STUNTING PREVALENCE IN INDONESIA USING SPLINE TRUNCATED SEMIPARAMETRIC REGRESSION

Abstract

Semiparametric regression combines parametric and nonparametric regression approaches. It is employed when the relationship pattern of the response variable is known with some predictors, while for other predictors, the relationship pattern is uncertain. The parametric regression component in this study is linear regression, while the nonparametric component utilizes a spline truncated estimator, resulting in a semiparametric spline truncated regression model. The case study focuses on the prevalence of stunting across 34 provinces in Indonesia in 2022, revealing a relatively high prevalence of 21.60%. The research aims to determine the optimal number of knots, the best model, and factors influencing stunting prevalence in Indonesia. The findings indicate that the optimal three-knot model with a GCV of 9.30 yields an RMSE of 1.70 and R2 of 92.71%. Significance tests for simultaneous and partial parameters reveal that all predictor variables significantly influence stunting prevalence.