A Missing Data Approach to Correct for Direct and Indirect Range Restrictions with a Dichotomous Criterion: A...

Area of Science:

• Psychometrics
• Statistical modeling
• Organizational psychology

Background:

• Selection methods often face range restriction, biasing predictive validity estimates.
• Existing correction methods are inadequate for dichotomous criteria due to unknown base rates.
• Lack of research on accurate correction methods for dichotomous criterion variables.

Purpose of the Study:

• To develop and evaluate a new correction method for range restriction with dichotomous criteria.
• To address the deficiency in scientific research regarding predictive validity estimation for dichotomous outcomes.
• To compare the accuracy of a novel missing data approach against traditional correction formulas.

Main Methods:

• Viewing range restriction as a missing data problem, utilizing multiple imputation by chained equations.
• Conducting Monte Carlo simulations to assess accuracy based on selection ratio, predictive validity, and base rate.
• Comparing the proposed missing data approach with Thorndike's correction formulas.

Main Results:

• The missing data approach significantly improves the accuracy of predictive validity estimation compared to Thorndike's formulas.
• Correction accuracy increases with higher selection ratios and predictor-criterion correlations.
• The proposed method provides a valid estimation of the unknown base rate of success.

Conclusions:

• Multiple imputation by chained equations is a superior method for evaluating predictive validity with dichotomous criteria.
• This approach overcomes the limitations of traditional methods in scenarios with restricted criterion data.
• Recommendations are made for adopting this missing data technique in selection method evaluations.