Paper Summary
Share...

Direct link:

Penalized Regression Versus Forward Stepwise Regression in Variable Selection and Significance Testing: Evidence From PISA 2015

Thu, May 4, 8:00 to 9:30am CDT (8:00 to 9:30am CDT), Division D Virtual Sessions, Division D - Section 2: Quantitative Methods and Statistical Theory Virtual Paper Room

Abstract

Despite the popularity of LASSO for variable selection in machine learning, statistical inference was a challenging task due to its biased estimates. The recent considerable progress in statistical inference with penalized regression was the motivation of this study. Specifically, we employed the post-selection inference (PSI) in predicting students’ belonging to school, and hundreds of Programme for International Student Assessment (PISA) 2015 variables were explored, identified, and statistically tested. PSI selected fewer but theoretically relevant predictors than forward stepwise regression (FS). This study empirically supports that PSI is an improvement from traditional methods such as FS. Further investigation of divergent findings between PSI and FS will contribute to new hypothesis generation and help close the gap between explanatory and predictive modeling.

Authors