Statistical Evidence in Salary Discrimination Studies: Nonparametric Inferential Conditions

Area of Science:

• Social Sciences
• Statistics
• Economics

Background:

• Salary allocation fairness is often assessed using statistical methods.
• A key principle is equal pay for equal merit across demographic groups.
• Complete merit data is frequently unavailable, potentially biasing analyses.

Purpose of the Study:

• To establish theoretical, nonparametric conditions for valid salary fairness inferences.
• To differentiate between fairness concerning true merit and observed merit.
• To review and illustrate latent variable models for salary equity studies.

Main Methods:

• Development of theoretical, nonparametric conditions for fairness inference.
• Definition of two distinct types of fairness: true merit vs. observed merit.
• Review and application of latent variable models as parametric special cases.

Main Results:

• Identified conditions under which observed merit data can support fairness inferences.
• Demonstrated the specification of latent variable models as structural models with latent means.
• Illustrated model application using real-world salary data.

Conclusions:

• Inferences about salary fairness are possible even with imperfect merit measures.
• Distinguishing between true and observed merit is crucial for accurate assessments.
• The proposed conditions and models offer a framework for empirical salary equity studies.