Abstract
The complex development of socioeconomic data requires an analytical approach that is capable of adequately capturing nonlinear structural relationships and inferential uncertainty. This study aims to compare Bayesian and non-Bayesian approaches in truncated spline-based semiparametric path analysis, both in terms of statistical inference and methodological implications, using financial literacy as a case study. The non-Bayesian approach is estimated using weighted least squares (WLS) with significance testing based on double bootstrap resampling, while the Bayesian approach is constructed through likelihood, prior, and posterior specifications with inference based on credible intervals. In addition to the empirical analysis, a simulation study evaluates the performance of both approaches across a range of sample sizes, noise levels, degrees of nonlinearity, knot specifications, and error structures. Within the simulation scenarios considered in this study, the Bayesian approach was more stable at small to medium sample sizes, whereas the non-Bayesian approach showed greater efficiency at large sample sizes. Methodologically, these findings confirm that the superiority of inferential approaches is contextual and highly influenced by data characteristics and analysis objectives. Rather than proposing a new estimator, this study contributes a systematic comparative framework in which Bayesian and non-Bayesian inference are applied to an identical truncated spline semiparametric path model and evaluated side by side. It also extends the application of truncated splines in structural models. The proposed framework is expected to serve as a methodological reference for socioeconomic research involving nonlinear relationships and indirect effects under conditions of data uncertainty.

This publication has 32 references indexed in Scilit: