Reading focus
Syllabus: What Can Individual Differences Predict?
Connect correction assumptions and differences across criteria with the proposed uses of cognitive, personality, and other selection measures.
Why you’re reading it. Finish by separating evidence of predictive value from the further case for a particular decision rule.
A question to consider. Would evidence that a score predicts completion justify using it for admissions, allocating extra support, or both? Specify what additional evidence and fairness judgments each use would require.
Why this reading
Choices about statistical correction can change the apparent usefulness of selection measures. This conceptual reappraisal connects those choices with practical questions about combining predictors, defining outcomes, and weighing consequences.
Before you read
Range restriction occurs when a studied sample has less variation than the population of interest, often after selection. A validity coefficient here is an association between a predictor and an outcome. Corrected estimates target specified hypothetical conditions; they are not automatically the accuracy an organization will observe.
Read for
- Trace the assumptions behind corrections for restricted range and criterion unreliability, including differences between applicant and incumbent studies.
- Compare average validity with variation across studies, jobs, outcomes, and combinations of predictors.
- Identify where the argument moves from estimating associations to choices about diversity, costs, and selection priorities.
Consider afterwards
- A measure has the highest average corrected validity in a table. What information about the intended outcome, local applicants, other measures, and consequences would you need before preferring it?
Editorial context
This paper interprets the authors’ 2022 reanalysis and discusses objections to its correction strategy; it is an argued methodological revision, not a universal ranking of tests. Training performance and routine job performance need not favor the same predictors. Estimates of group differences do not establish their causes or determine what fairness requires. The article introduces selection tradeoffs, rather than providing a complete account of fair decision making.