Handling Imbalanced Data With Weighted Logistic Regression and Propensity Score Matching methods

Abstract

The adoption of empirical methods for secondary data analysis has witnessed a significant surge in IS research. However, the secondary data is often incomplete, skewed, and imbalanced at best. Consequently, there is a growing recognition of the importance of empirical techniques and methodological decisions made to navigate through such issues. However, there is not enough methodological guidance, especially in the form of a worked case study that demonstrates the challenges of imbalanced datasets and offers prescriptive on how to deal with them. Using data on P2P money transfer services, this article presents a running example by analyzing the same dataset using several different methods. It then compares the outcomes of these choices and explicates the rationale behind some decisions such as inclusion and categorization of variables, parameter setting, and model selection. Finally, the article discusses certain regressions models such as weighted logistic regression and propensity matching, and when they should be used.