A1294
Title: Multinomial logistic regression utilizing external machine learning predictions
Authors: Chi-Shian Dai - National Cheng Kung University (Taiwan) [presenting]
Abstract: In many modern applications, a carefully designed primary study provides high-quality individual-level data for interpretable modeling, while additional external information is available only through black-box, nonparametric machine-learning predictions. Although summary-level external information has been studied in the data integration literature, there is limited methodology for leveraging external nonparametric predictions to improve statistical inference in the primary study. A general empirical-likelihood framework is proposed that incorporates external predictions through moment constraints. A key advantage of nonparametric prediction is that it induces a rich class of valid moment restrictions that remain robust to covariate shift under a mild overlap condition without requiring explicit density-ratio modeling. The focus is on multinomial logistic regression as the primary model and addresses common data-quality issues in external sources, including coarsened outcomes, partially observed covariates, covariate shift, and heterogeneity in generating mechanisms. Large-sample properties of the resulting fused estimator are established, including consistency and asymptotic normality under regularity conditions. Moreover, sufficient conditions are provided under which incorporating external predictions delivers a strict efficiency gain relative to the primary-only estimator. Simulation studies and an application are provided.
