Date of Award

5-2024

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

College/School

College of Humanities and Social Sciences

Department/Program

Psychology

Thesis Sponsor/Dissertation Chair/Project Chair

Michael Bixter

Committee Member

Kevin Askew

Committee Member

John T. Kulas

Abstract

This dissertation explores the ongoing debate between traditional statistical regression models and machine learning (ML) algorithms in predictive modeling, focusing on the impact of sample size and the number of variables. Study 1 investigates the relationship between sample size and predictive accuracy, proposing hypotheses regarding the advantages of ML over regression as sample size increases. Additionally, the study examines the influence of the number of variables on predictive accuracy, emphasizing the trade-off between ML and regression models. Using data from the Federal Employee Viewpoint Survey, the research aims to contribute insights into the conditions favoring each modeling approach. Study 2 shifts the focus to incremental validity, exploring whether work-related psychological constructs enhance ML models' predictive accuracy in turnover intention when compared to biodata alone. The proposed hypotheses suggest that incorporating psychological constructs will improve predictive accuracy, addressing the "garbage in garbage out" concern prevalent in ML applications. The methods involve diverse datasets, including responses from federal employees in an online survey through Amazon's MTurk, with machine learning algorithms such as Gradient Boosting Trees, Random Forest, Neural Networks, and Support Vector Machines being compared to linear and logistic regressions. The dissertation seeks to advance understanding in the field, offering practical insights for researchers and practitioners navigating the dynamic landscape of predictive modeling.

File Format

PDF

Share

COinS