Sensitivity of Bayesian Truncated Spline Regression to Prior and Knot Configuration in Stunting Models

This study develops a Bayesian bi-response regression model using a truncated spline approach to examine nonlinear effects of economic, dietary, and environmental factors on nutritional and physical stunting. Sensitivity analysis was conducted to evaluate the influence of prior types and knot numbers on model performance using Deviance Information Criterion (DIC), Root Mean Square Error (RMSE), and bias. Results show that the informative Normal–Gamma prior combination yields the best performance, with the lowest DIC, smallest RMSE, and minimal bias. Models with three knots provide higher predictive accuracy, while noninformative Uniform priors cause instability and overfitting. Overall, the findings indicate that prior specification has a stronger effect on model robustness than the number of knots, emphasizing the importance of informative priors in Bayesian spline modeling for understanding complex, nonlinear determinants of child stunting.