Modified Multigroup Ramsey RESET for Specification Detection in Semiparametric Path Models

Abstract

This study develops a modified multigroup Ramsey RESET (Regression Equation Specification Error Test) within a semiparametric multigroup path model, a statistical framework that combines parametric and nonparametric approaches across multiple groups. Developed using dummy-variable interactions, the method identifies both linear and truncated spline relationships and enables analysis of all groups within a single integrated model, eliminating the need for separate group tests. Simulation studies using empirical data on students’ AI literacy applied combinations of linear and truncated-spline relationship patterns. The method was evaluated using the p-value, Correct-to-Incorrect p-value Ratio, Dominance Ratio, and Accuracy. The results indicate that the modified multigroup Ramsey RESET generally produces larger p-values for models that match the underlying data-generating mechanism than for competing alternative models, demonstrating good discriminatory power in identifying appropriate model specifications. The empirical application further reveals that several relationships among variables exhibit truncated spline patterns rather than purely linear forms. Thus, the proposed method provides an alternative specification test for simultaneously identifying linear and nonlinear relationships within multigroup semiparametric path models.