What does Sackett et al. (2023) say about designing a selection process?
A 2023 follow-up drawing out what the revised validity estimates mean in practice. Its central figure: in a six-predictor composite, giving cognitive ability zero weight reduces overall validity from .61 to .56. Under the older estimates the same removal cost .20, falling from .66 to .46.
Sackett, P. R., Zhang, C., Berry, C. M. & Lievens, F. (2023). Revisiting the design of selection systems in light of new findings regarding the validity of widely used predictors. Industrial and Organizational Psychology, 16(3), 283–300.
Primary source opened and quotes confirmed on .
In its own words
“Sackett et al. (2022) identified previously unnoticed flaws in the way range restriction corrections have been applied in prior meta-analyses of personnel selection tools. They offered revised estimates of operational validity, which are often quite different from the prior estimates. The present paper attempts to draw out the applied implications of that work.”
What it does not say
Each of these is a claim made in this market and attributed to the source above. None of them is supported by it.
Commonly claimed: That cognitive ability testing should be dropped.
The paper prices the choice rather than making it. .05 of validity is a real cost; it is simply far smaller than the .20 the older framework implied, which changes how a validity-versus-adverse-impact trade-off should be argued.
Commonly claimed: That .61 → .56 applies to dropping a test you use on its own.
It is a composite of six predictors. An employer whose entire screen is one cognitive test is not in that arithmetic — removing it there removes the whole predictor, not one of six contributions.
Commonly claimed: That this is a separate finding from the 2022 paper.
It is the same data set read for its applied implications. Cite the 2022 paper for the estimates and the 2023 paper for what to do about them; citing the 2023 figures as a new meta-analysis overstates what was found.
The number worth carrying into a procurement conversation
Convert .05 of composite validity into outcomes and it stops being abstract. At a base rate of 50% and one hire in ten, a composite at .61 produces a success rate of 90.2% among those hired and a composite at .56 produces 87.4% — a difference of under three good hires per hundred.
That is the price of removing the predictor with the largest subgroup differences from a well-built composite. It is a number a buyer can weigh against a named fairness risk, which is a materially different conversation from the one the 1998 framework supported, where the same removal looked like giving up a fifth of the process's accuracy.
What .05 of validity is worth at four different selection ratios
The .61 → .56 figure is a correlation, and a correlation is not a quantity anybody hires on. Converting it into the proportion of successful hires makes it decidable, and the conversion depends on one thing the paper does not fix: how selective the process is.
Computed on the bivariate-normal model at a base rate of 50%, the success rate among those hired runs as follows. Hiring one applicant in two: 70.9% at a composite validity of .61, against 68.9% at .56. One in five: 84.5% against 81.7%. One in ten: 90.2% against 87.4%. One in twenty: 93.7% against 91.2%. **So the cost of giving cognitive ability zero weight is between 2.0 and 2.8 successful hires per hundred across the whole plausible range of selection ratios.**
The comparison that matters is with what the 1998 framework implied for the same decision, because that is the framework most procurement conversations are still arguing from. Under the older .66 → .46, the same removal cost 7.7 points of success rate at one applicant in two, 11.5 at one in five, 11.5 at one in ten and 10.6 at one in twenty. **The revision shrinks the priced cost of dropping the predictor with the largest subgroup differences by roughly four-fold, and it does so at every selection ratio rather than at a convenient one.**
Two honesty notes on this arithmetic, because it is this site's calculation and not the paper's. The two composite validities are the paper's; the conversion is the ordinary Taylor-Russell model and it reproduces the 90.2% and 87.4% figures printed above it exactly, which is the check that the method is the same one. And the model assumes a normally distributed criterion, a fixed base rate and top-down selection, none of which is precisely true of a real pipeline — so read the differences between the columns rather than the levels inside them.
Why the framing changed as well as the figures
The older literature tabulated each predictor's incremental validity over cognitive ability, which presumes ability is the anchor everything else adds to. On revised estimates it is not the anchor: structured interviews sit above it, and the composite arithmetic shows it contributing modestly once several other predictors are present.
For anyone designing a screen, that reorders the build. Start from the best-structured interview the organisation can actually run consistently, then add the instrument least correlated with it, and treat the ability test as one contributor to be justified on its own terms rather than as the foundation.
Where this source is used here
These pages argue from the source above. If it is ever superseded, these are the pages that have to change.
Read next
- Sackett et al. (2022) — the validity re-estimates — The paper that cut the published validity of most hiring predictors by .10–.20 and put structured interviews on top.
- Schmidt & Hunter (1998) — the table everyone still quotes — The 1998 meta-analysis behind almost every validity figure in assessment marketing — and the reason quoting it in 2026 is a dated claim.
Founder of AssessAll and of Bodhih Training Solutions, a corporate training company in Bangalore. Works on assessment design, scoring and reporting across hiring, L&D and certification programmes.
Last reviewed