Cognitive ability tests in recruitment

What cognitive ability tests measure

Cognitive ability tests, also called aptitude or reasoning tests, measure how well someone takes in new information, sees relationships and solves problems under standardised conditions. They do not measure job knowledge, effort or personality. In recruitment, three types of subtest are common:

• Numerical reasoning: drawing conclusions from figures, tables and charts. The arithmetic is usually simple; the difficulty lies in choosing the right data and the right steps.
• Verbal reasoning: judging whether a statement follows from a short text, is contradicted by it or cannot be determined from it.
• Abstract or logical reasoning: finding the rule in series of shapes, figures or numbers, with little dependence on language or prior knowledge.

General and specific abilities

Scores on different reasoning tests tend to correlate. Most models of intelligence therefore describe a general factor at the top, with broader and narrower abilities beneath it. For hiring, this means that a combination of subtests gives a more reliable estimate of general ability than any single one, while a specific subtest earns its place when the role clearly calls for it, such as numerical reasoning for a financial analyst.

Choose subtests from a job analysis rather than from habit. The cognitive ability tests page shows the subtests and formats available on Match.hr.

What they predict, and how strongly

Cognitive ability is one of the most researched predictors of job performance. For a long time the reference point was a meta-analysis by Schmidt and Hunter (1998), which estimated the validity of general mental ability for job performance at about .51, among the highest of the selection methods they compared.

Sackett, Zhang, Berry and Lievens (2022) re-examined how the earlier studies corrected for range restriction: validity can only be calculated for people who were hired, and that group varies less than the full applicant pool. They concluded that these corrections were often too large. Their revised estimate for general cognitive ability is about .31, with structured interviews at about .42.

Validity coefficients are averages across many studies and jobs. Whether a particular test predicts well in your setting also depends on its reliability, how relevant the subtests are to the role and how the scores are used. The guide on reliability and validity explains how to read these figures.

Fairness and adverse impact

On some cognitive tests, groups differ in their average scores, for example by educational background, or between people tested in their first and in their second language. Averages say nothing about individuals, and scores vary widely within every group. Yet when a high cut-off or strict top-down ranking is applied, even moderate average differences can produce different selection rates between groups. This is called adverse impact.

You cannot remove this risk entirely, but you can reduce it and justify your choices:

• Set the required level from the job, rather than selecting the highest scorers by default.
• Use cut-offs sparingly. A modest minimum, or no cut-off at all with the score weighed alongside other evidence, usually has less adverse impact than a high fixed threshold.
• Combine methods. Structured interviews, personality assessments or work samples broaden the evidence and lower the weight of any single test.
• Test in the right language. A verbal test in a second language partly measures language proficiency. Offer a validated version in the candidate's own language where possible, or give more weight to abstract reasoning.
• Plan adjustments. Agree on accommodations such as extra time for dyslexia or a screen-reader-friendly version in advance, and note them when you interpret the scores.
• Check for bias. Ask the test provider for fairness analyses per language version and, where lawful, monitor selection rates in your own process.

The page on fairness and bias explains how item bias, measurement invariance and adverse impact are analysed, and what these analyses cannot show.

Testing conditions online

Most cognitive tests in recruitment are now taken online, at home and without supervision. That is quick and convenient for candidates, but it raises two questions: is the person taking the test really the candidate, and did they work alone?

Item banks and randomised versions

Well-built online tests draw their items from a large bank, so candidates receive different but equivalent versions. Shared answers are then of little use, and memorised items are unlikely to reappear.

Time limits

Most reasoning tests are timed. A time limit keeps results comparable and makes looking things up harder, but it also adds speed to what is being measured. Check that the time limit is the one used for the norm group, and tell candidates in advance how long they will have.

Verification at decisive steps

When an unsupervised score carries real weight in a decision, consider a short verification test under supervision later in the process, on site or in a proctored session, and compare it with the first result. Tell candidates about this at the start.

Candidate experience

Tell candidates what the test is for, how long it takes, which device works best and how the results will be used. Clear practice items at the start and a calm, accessible interface help to reduce test anxiety, which can lower scores without saying anything about ability.

Practice and preparation

Candidates prepare for reasoning tests, often with practice tests on websites such as Assessment-Training.com. Familiarity with the question types reduces anxiety and confusion about the instructions, so the score is more likely to reflect actual reasoning.

Practice does have an effect. Scores tend to rise somewhat when someone has already taken the same or a similar test, mainly through familiarity with the format and better use of time. This is one reason to keep a waiting period between retakes and not to compare a retake directly with a first attempt.

For employers the conclusion is reassuring. With a large item bank and randomised versions, practising means getting used to the format, not learning the answers. Being open about the test and its question types also puts candidates with and without access to preparation on a more equal footing.

Using the results well

A score only becomes meaningful when you know what it is compared with. Reports usually express a result as a percentile, stanine or similar score against a norm group. Choose a norm group that fits the job level: candidates for a graduate role should be compared with a similar population, not with the working population as a whole.

• Treat the score as one source of evidence, alongside interviews, work samples and references.
• Look at confidence ranges: small differences between candidates often fall within measurement error.
• Decide in advance how the score will be weighed, so every candidate is treated the same way.
• Let a person decide. Do not use a cognitive score to reject candidates automatically.

Where software or AI ranks or filters candidates, the EU AI Act classifies it as high risk and requires human oversight; see the EU AI Act and assessments.