Adjusting Regression Models for Overfitting in Second Language Research

Abstract

Regression modeling is an increasingly important quantitative tool for second language (L2) research. While superior in many ways to more traditional methods, such as ANOVA, regression modeling, like all procedures, still has limitations, ranging from small sample sizes to a lack of screening for outliers and influential data points (Plonsky and Ghanbar, 2018). Since these limitations are common features in L2 research, this raises concerns that existing studies using regression may overfit the data, perhaps inflating effect size estimates. These issues can be partially alleviated via robust statistics, such as validation. This paper provides L2 researchers with an overview of these issues and an instructive look at one robust validation method: bootstrapping.