Heterogeneous Returns to Earning Sub-Baccalaureate Credential: A Machine Learning Approach

• In Event: Noncredit Workforce Training in Community Colleges

Abstract

In this paper, we estimate the heterogeneous labor market returns to earning sub-baccalaureate credentials (e.g., credit (associate degrees, certificates), and non-degree programs) using Kentucky’s comprehensive administrative data on postsecondary education and earnings. Sub-baccalaureate credentials play a critical role in the postsecondary education system and workforce development. They serve as important pathways for individuals seeking to enhance their employment opportunities without pursuing a traditional four-year degree. By offering skill-based training aligned with industry demands, these credentials appeal to a diverse range of students, including in-school youth, displaced workers, and those looking to upskill or reskill.

A substantial body of literature has examined the economic returns to postsecondary credentials, with a growing focus on sub-baccalaureate degrees and certificates (e.g., Bahr et al., 2015; Bettinger and Soliz, 2016; Carruthers and Sanford, 2018; Dadgar and Trimble, 2015; Darolia, Guo, and Kim, 2024; Jepsen, Troske, and Coomes, 2014; Liu, Belfield, and Trimble, 2015; Minaya and Scott-Clayton, 2022; Stevens, Kurlaender, and Grosz, 2019; Xu and Trimble, 2016). These studies typically employ a difference-in-differences design with individual fixed effects to compare within-student changes in post- vs. pre-program labor market outcomes between those who earned a credential and those who pursued but did not complete. While highlighting that sub-baccalaureate credentials can lead to wage gains and improved employment prospects, these studies also show that the employment outcomes vary significantly across dimensions such as gender, race/ethnicity, age, field of study, and credential length.

Our study contributes to two distinct strands of the literature. First, we advance the methodological approach by employing machine learning methods to produce more precise "apple-to-apple" comparisons and estimate the treatment effects of completing sub-baccalaureate credentials. Specifically, we classify students who are otherwise very similar before enrolling in a sub-baccalaureate program, with the only difference being whether they completed the credential. We conduct the analysis for both credit and non-degree programs by dividing the credential type into associate’s degrees, long-term certificates, short-term certificates, and non-degree programs.Second, we leverage machine learning techniques to uncover complex, non-linear patterns of heterogeneity in labor market returns. Specifically, we explore how different factors, such as individual demographics and pre-program labor market experience, interact to produce diverse employment outcomes. We estimate the Conditional Average Treatment Effect using the Causal Forest DML model to capture how the estimated labor market returns differ by various groups of individuals.

Authors

• Kangwook Noh, University of Kentucky
  Presenting Author

• Chuanyi Guo, University of Kentucky
  Non-Presenting Co-Author