Ridge,LASSO,and elastic net regression: Fathers’ psychological well-being in familial contexts

Abstract

This study examined the applicability of regularized regression in family science by focusing on fathers’ psychological well-being as an applied example. Using cross-sectional survey data from 277 U.S. fathers, we compared Ordinary Least Squares (OLS) regression with Ridge, LASSO (Least Absolute Shrinkage and Selection Operator), and Elastic Net in regularized regression to evaluate how diverse familial factors—including father involvement, father role identity, parenting competency, parent–child relationships, coparenting relationships, and work–family conflict—concurrently predict fathers’ well-being. Results indicated that while the three regularized approaches produced similar overall performance, they provided advantages over OLS by reducing overfitting, improving generalizability, and enhancing interpretability. We also compared the analysis results of the three approaches in terms of the prediction accuracy and interpretability. The findings reveal how regularized regression can highlight key factors, such as father involvement, coparenting, and parenting competency. This study also illustrates how regularized regression complements traditional regression in family science by addressing multicollinearity, reducing overfitting, and supporting feature selection. These results highlight the potential of computational approaches such as machine learning to strengthen both theory and practice in family research.