The single most important thing a small employer can know about hiring is which selection methods actually forecast who will do the job well — and that answer changed in 2022. For roughly two decades the field’s reference point was Schmidt & Hunter (1998), which ranked general mental ability (GMA, i.e. cognitive ability) at the top with an operational validity around .51 and work samples at .54. Sackett, Zhang, Berry & Lievens (2022) showed that a statistical correction for “range restriction” had been applied too aggressively in the meta-analyses those numbers came from, inflating many validities. When they reanalysed the evidence using more defensible corrections, the magnitudes fell and the ranking reshuffled.

The revised operational validities (Sackett et al., 2022). Higher numbers mean better prediction of job performance; the maximum possible is 1.0.

Structured employment interviews — .42 (down from .51)
Job-knowledge tests — .40 (down from .48)
Empirically-keyed biodata — .38 (up slightly from .35)
Work-sample tests — .33 (down sharply from .54)
General mental ability / cognitive ability — .31 (down from .51)
Integrity tests — .31 (down from .41)
Assessment centres — .29 (down from .37)
Interests, defined as person-job fit — .24 (up from .10)
Conscientiousness, overall — about .19-.21 (down from .31)
Unstructured interviews — .19 (down from .38)
Years of job experience — .07 (down from .18)

The headline. Cognitive ability is no longer the stand-out single predictor. Structured interviews emerged with the highest mean validity, and job-knowledge tests, empirically-keyed biodata, and work samples sit at or above GMA. Sackett et al. (2023) put it bluntly: one could now treat structured interviews, not cognitive ability, as the focal predictor against which others are compared.

The trap to avoid. A great deal of HR content online — vendor blogs, “best predictor” listicles, even some training decks — still repeats the 1998 figures (GMA on top at r ≈ .51, work samples at .54). Treat those numbers as outdated. The reason is specific and technical: the older meta-analyses corrected for range restriction using artifact distributions that assumed far more restriction than the data actually supported, which systematically overstated validity. If you see “cognitive ability is the #1 predictor at .51,” you are reading pre-2022 numbers.

What did not change. The revision lowered magnitudes; it did not overturn the core practical lesson. The strongest predictors are still job-specific methods built on a real understanding of the role (structured interviews, job-knowledge and work-sample tests, biodata), and combining methods still beats any single one. “Lower than we thought” is not “useless.”

A note on certainty. The .42 for structured interviews carries a wide spread (standard deviation .19; an 80% credibility interval roughly .18-.66), meaning a structured interview can be designed well or badly. The revised numbers are also contested in follow-up literature, so treat any single decimal as an estimate rather than a fixed fact.

For an Ontario SMB, the takeaway is that the cheapest high-validity upgrade available is not a fancier test — it is converting your existing interview into a structured one and adding a second job-relevant method. The dollar case for getting this right lives in the numbers cluster (the cost of a bad hire); the legal limits on what you may test live in the Compliance cluster.

Related notes

Do structured interviews predict performance better than unstructured ones — and what makes an interview "structured"? — Structured interviews substantially out-predict unstructured ones — McDaniel et al. (1994) put corrected validity at about .44 vs .33, and Sackett et al. (2022) at .42 vs .19 — and "structure" means job-analysis-based questions asked identically of every candidate, scored on anchored rating scales by trained raters.

Work-sample, ability, and job-knowledge tests: how well they predict and when to use them — Job-knowledge tests (.40) and work samples (.33) are strong, job-specific predictors and now sit at or above cognitive ability (.31) — but work-sample validity was revised down sharply in 2022, work samples and knowledge tests only work for candidates who already have the skills, and cognitive tests carry the largest adverse-impact risk.

Why combine selection methods? Incremental validity and the cost of the interview-only hire — No single method predicts performance well enough on its own, but methods that tap different things add incremental validity — a structured interview plus a cognitive or work-sample measure pushed composite validity above .60 in the classic data — which is why hiring on one unstructured interview is the weakest defensible basis for a decision.

Which hiring / selection methods actually predict job performance? — On the best current peer-reviewed evidence (Sackett, Zhang, Berry & Lievens, 2022), structured interviews are the single strongest predictor of job performance (operational validity ≈ .42), ahead of job-knowledge tests (.40), empirically-keyed biodata (.38), work samples (.33), and general mental ability/cognitive tests (.31) — a major reordering from Schmidt & Hunter's widely-cited 1998 ranking, which over-stated cognitive ability at .51 due to range-restriction overcorrection.