Personnel selection is one of the better-studied areas in applied psychology, and the findings have had limited effect on how most organisations actually hire.
What the research measures
Predictive validity — the correlation between a selection method's assessment and subsequent job performance.
Which is measured by assessing candidates, hiring regardless of the assessment where possible, and comparing later.
Meta-analyses pooling many such studies have been conducted across decades, with periodic reanalysis correcting earlier methodological issues.
What predicts reasonably well
Structured interviews, where every candidate is asked the same predetermined questions and answers are scored against defined criteria.
Work sample tests, where candidates perform tasks resembling the actual job.
Cognitive ability tests, which have consistently shown predictive validity while raising legitimate concerns about adverse impact on some groups.
Job knowledge tests where the role requires specific knowledge.
What predicts poorly
Unstructured interviews, where the interviewer converses freely and forms an impression.
Which is the most widely used method by a large margin and among the weakest predictors.
Years of experience, which correlates weakly with performance beyond a modest initial period.
Education level beyond meeting a threshold requirement.
References as typically conducted, which are constrained by legal caution to the point of carrying almost no information.
Why unstructured interviews persist
They feel informative, which is the core problem.
Interviewers form confident impressions and those impressions predict poorly, which is a documented pattern.
Research has found that unstructured interview information can actually reduce accuracy, by adding noise to more valid information already available.
They also serve purposes beyond selection — the candidate assessing the organisation, relationship building, and a sense of fairness from having met someone.
Making interviews structured
The practical improvement available to any organisation.
Define the competencies the role requires before writing questions.
Ask every candidate the same questions in the same order.
Use behavioural questions asking about specific past situations rather than hypothetical ones.
Score against a defined rubric immediately rather than forming an overall impression.
And have interviewers score independently before discussing, since discussion produces convergence toward the most confident voice rather than toward accuracy.
Work samples
Among the strongest predictors and the most resented by candidates when handled badly.
Which means keeping them short, paying for substantial time, and never using candidate work commercially.
Exercises resembling the actual work predict better than abstract puzzles, which is intuitive and frequently ignored in technical hiring.
Bias
Structure reduces the influence of similarity and impression, which is where much bias operates.
Anonymising applications at the screening stage has shown effects in several studies, with results varying by context.
Diverse interview panels are commonly recommended, with the evidence more mixed than the recommendation implies.
The most robust finding is that reducing discretion reduces the opportunity for bias to operate, which is an argument for structure regardless of anything else.
The realistic constraint
Small organisations cannot run validated selection systems, and they can adopt the cheap parts.
Same questions, defined criteria, independent scoring, and a work sample. That is most of the available gain and costs an afternoon to set up.
Candidate experience
Which affects who accepts and what they say afterwards.
Long processes with many stages lose candidates, particularly the strongest ones with alternatives.
Which means process length is a selection mechanism operating in the wrong direction, and reducing it generally improves the pool without reducing accuracy if the remaining stages are the valid ones.
Prompt communication, including to unsuccessful candidates, costs little and affects reputation measurably.
Job descriptions
Requirements lists that exceed what the role needs deter qualified applicants, with evidence that the effect falls unevenly across groups.
Which is a straightforward fix — separating genuine requirements from preferences, and keeping the list short.
Salary disclosure in advertisements is now required in several jurisdictions and improves applicant quality by removing mismatched candidates early.
Probation
The final selection stage and the most valid, since it observes actual performance.
Which means using it — with defined expectations and a genuine review — rather than treating it as a formality that passes automatically.
Employment protections during and after probation vary considerably by jurisdiction and should be understood before relying on it.
Internal mobility
Hiring from within has better information available — actual observed performance — than any external assessment provides.
Which makes internal candidates lower risk, and the common failure is not considering them systematically.
Organisations that post roles internally first, and that track internal applications, fill more roles internally and retain people longer.
Onboarding
Early departures are frequently caused by onboarding rather than by selection.
Which means the hiring process succeeded and the following weeks did not, and the cost is attributed to the wrong stage.
Clear expectations, an assigned point of contact and early meaningful work are the elements consistently associated with better retention.
Which means selection and retention should be assessed together rather than as separate functions with separate owners.