Diversity tradeoff curves in personnel selection: Evaluating local study, meta-analysis, Bayes-analysis, and ensemble machine learning

Tang, Chen

Permalink

https://hdl.handle.net/2142/120523 Copy

Description

Title

Diversity tradeoff curves in personnel selection: Evaluating local study, meta-analysis, Bayes-analysis, and ensemble machine learning

Author(s)

Tang, Chen

Issue Date

2023-04-19

Director of Research (if dissertation) or Advisor (if thesis)

Newman, Daniel A.

Doctoral Committee Chair(s)

Newman, Daniel A.

Committee Member(s)

• Drasgow, Fritz
• Rounds, James
• Song, Q. Chelsea

Department of Study

School of Labor & Empl. Rel.

Discipline

Human Res & Industrial Rels

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Personnel selection
• Adverse impact
• Cross-validation
• Shrinkage
• Diversity
• Pareto-optimization

Abstract

One major advancement toward reducing adverse impact is the diversity-validity tradeoff curve methodology (De Corte, Lievens, & Sackett, 2007). The Pareto-optimal tradeoff curve provides sets of selection predictor weights that can often substantially enhance diversity (increase adverse impact ratio and number of minority job offers) with no loss of job performance, in comparison to unit weights (Wee, Newman, & Joseph, 2014). A chief limitation of this diversity-enhancing approach is the tendency for tradeoff curves to shrink, leading to lesser job performance and diversity outcomes upon cross-validation (Song, Wee, & Newman, 2017). Typical selection scenarios considered in Pareto-optimal shrinkage papers involve using a single local validity study as the calibration sample (see Rupp et al., 2020). The current project proposes to evaluate tradeoff curve shrinkage (both validity shrinkage and diversity shrinkage) using four types of validity evidence/calibration studies: (a) a local validity study, (b) a meta-analysis (Schmidt & Hunter, 1977), (c) a Bayes-analysis with empirical priors (Newman, Jacobs, & Bartram, 2007), and (d) an ensemble machine learning approach (Zhou, 2012). This dissertation consists of three studies. Study 1 examines conditions under which each approach performs best, offering recommendations on ideal methods for diversity improvement (reducing shrinkage and maximizing cross-validity) while using shrunken tradeoff curves in local selection settings. Study 2 evaluates effects of meta-analytic publication bias on the performance of meta-analysis, empirical Bayes-analysis, and ensemble learning. Finally, Study 3 considers potential biases due to the ignored violation of the assumption of independence between validities and artifacts.

Graduation Semester

2023-05

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/120523

Copyright and License Information

Copyright 2023 Chen Tang

Owning Collections