Sign up free

Aptitude tests for hiring: what they measure and how to use them

What numerical, verbal, logical and mechanical reasoning tests measure, how well they predict job performance, and a fair way to use a score in a small hiring round.

The EdProcto editorial teamWritten and checked against the research it cites. Last checked . How the guides are made

Guides · Hiring, training and certification · · 13 min read

What do aptitude tests measure, and how should an employer use them in hiring?

Aptitude tests for hiring measure reasoning with numbers, words, patterns and mechanisms, which mostly reflects general mental ability. Recent re-analyses put its correlation with job performance at 0.22 to 0.31, not the 0.51 long quoted. Use the score as a low floor or one input beside a structured interview, with the same test, conditions and notice for every candidate.

Aptitude tests for hiring are short, timed tests of reasoning: a table of figures to read, an analogy to complete, a series to continue, a lever to balance. What a score tells you is narrower than it looks.

This guide is for a small or mid-sized company choosing or writing a pre-employment aptitude test, not for a test publisher: what the four common kinds measure and predict, adverse impact, how to use a score, and one small hiring round worked through.

EdProcto publishes this guide, and its page for corporates sets out its reasoning kits and tests written from a company's own material. The candidate's side is in aptitude test practice.

What do numerical, verbal, logical and mechanical reasoning tests measure?

Each measures reasoning on one kind of material, and for predicting performance the first three work largely as one measure of general mental ability. The US Office of Personnel Management (n.d.) describes cognitive ability tests as assessing reasoning, perception, memory, verbal and mathematical ability and problem solving.

Test What the candidate does In EdProcto's kits
Numerical Reads a table; works out a change, a share, a total "Numerical reasoning"
Verbal Completes an analogy; finds the odd word out "Verbal analogies", "Odd one out"
Logical Continues a series; tests a syllogism; fills a matrix "Number series", "Syllogisms", "Number matrices"
Mechanical Says which way a gear turns, or what balances a lever "Mechanical reasoning"

One ability or four?

For prediction, mostly one. Studying seven US Air Force jobs, Ree et al. (1994) found that general ability was the best predictor of every performance measure, and that specific abilities added a statistically significant but practically small amount. In a UK meta-analysis, verbal, numerical, perceptual and spatial tests and tests of general ability all predicted job performance, with validities of the same order (Bertua et al., 2005). So three short tests of different kinds work as one longer test of much the same thing.

Mechanical reasoning is the exception. Schmidt (2011) argues that technical aptitude, which these tests tap, grows out of experience in technical areas, and reports that at every level of general ability women scored lower on it than men, so such a test may underpredict women's performance. Use one only where the job involves mechanisms.

How well do aptitude tests for hiring predict job performance?

Less well than employers were told for two decades: current estimates put the correlation between general mental ability and job performance at 0.22 to 0.31, not 0.51.

The 0.51 is from Schmidt and Hunter (1998), whose summary of 85 years of research on 19 selection procedures paired general mental ability with each of the others. Then Sackett et al. (2022) found that the usual correction for range restriction had been overdone. About 80% of validity studies test people already in the job, yet they were corrected with a large factor taken from studies of applicants (Sackett et al., 2023). Revised, the methods that had ranked high lost 0.10 to 0.20, and ability fell to 0.31.

Sackett et al. (2024) pooled only studies from this century: 153 samples and 40,740 people. The observed correlation was 0.16, and 0.22 after correction. Squared, to give the share of variation in rated performance that scores account for, 0.51 is about 26% and 0.22 about 5%.

What still holds

Ability is still related to performance, and more strongly in complex work: the UK meta-analysis found higher validities in more complex occupations (Bertua et al., 2005), though its figures predate the re-analysis. But ability no longer stands out. In the revised ranking structured interviews come first, averaging 0.42 with an 80% credibility interval of 0.18 to 0.66, and work samples and job knowledge tests are among the leaders (Sackett et al., 2023). That interval is the caution for one employer: an average across studies is not what your test will achieve.

We read Sackett et al. (2023) in full and the other studies here in abstract only. The 0.51 and the 0.31 are as the 2024 abstract reports them.

What is adverse impact, and what does a fair process look like?

Adverse impact is a substantially different rate of selection that works to the disadvantage of a race, sex or ethnic group, as the US Uniform Guidelines on Employee Selection Procedures (1978) define it, and reasoning tests are the method most prone to it. The Office of Personnel Management (n.d.) says cognitive tests typically produce larger racial and ethnic differences than other valid predictors. Roth et al. (2001) found a Black-White difference of about one standard deviation among job applicants overall, and less in many instances, varying with the job's complexity and the study's design.

The height of the bar matters. On two normal curves one standard deviation apart (arithmetic, not a finding about your applicants), a cut one standard deviation below the higher mean passes 84% of one group and 50% of the other, a ratio of 0.59. A cut at that mean passes 50% and 16%: 0.32.

The Guidelines' rule of thumb is four-fifths: a selection rate below 80% of the highest group's rate is generally regarded as evidence of adverse impact, though a difference that rests on small numbers may not be. Sackett et al. (2023) add that combinations of methods with and without an ability test can reach comparable validity, so a process can be as valid with smaller differences.

From those sources and the Chartered Institute of Personnel and Development (2026), a fair process is:

  • Job-related. Name what the job needs before choosing a test.
  • The same for all. One paper, one time limit, one set of rules.
  • Announced. Candidates hear the requirements and the duration beforehand, and can ask for adjustments.
  • Adjusted where reasonable. GOV.UK (n.d.) says a UK employer must make reasonable adjustments, such as letting a candidate complete a written test on a computer.
  • Never alone. The Institute says test results should never be the sole basis for a decision.
  • Decided by a person. The UK Information Commissioner's Office (n.d.) gives a recruitment aptitude test run on pre-programmed criteria as an example of an automated decision, so let a person, not a pass mark, turn a candidate down.
  • Recorded. Who sat, who passed, and the rule that decided.

None of this is legal advice. The rules differ by country, so take your own advice on employment law.

How should an employer use an aptitude test score?

Use it as a low floor or as one input among several, never as the decision.

A cut score, a ranking or one input?

A cut score asks least of a test. The Uniform Guidelines say a cut-off should normally be consistent with normal expectations of acceptable proficiency in the workforce, and that evidence good enough for pass or fail may not support ranking. A practical floor sits just under the lowest score of people who do the job well, set before any candidate sits. See setting a pass mark.

Ranking asks most, and a short test cannot bear it. Treat 12 questions as 12 independent chances. A candidate who gets three in four right in the long run scores exactly 9 about a quarter of the time, between 8 and 10 in two sittings in three, and 6 or below about once in twenty. One mark is noise.

As one input, the score joins a structured interview and a job knowledge test, two kinds of evidence the revised research ranks higher.

Why tell candidates and send practice material?

Because practice moves scores, and not everyone has had it. Hausknecht et al. (2007) pooled 50 studies with 134,436 people: scores rose by 0.26 of a standard deviation with practice, and by more with coaching or an identical form. So send every candidate the same practice material, such as the guide on preparing for a pre-employment test, and give a retake a new paper.

What can an untested in-house quiz claim?

It can claim consistency, and nothing about prediction. Under the Uniform Guidelines any measure used as a basis for an employment decision is a selection procedure, a home-made quiz included. They add that a test said to measure a trait such as intelligence or aptitude cannot be supported by its content alone: it needs other evidence, such as a study showing that scores go with performance. The US Equal Employment Opportunity Commission (2007) advises that tests be properly validated for the positions and purposes they are used for.

You can say You cannot say
Everyone sat one test under one limit: questions made by one rule, or the very same questions The score predicts who will perform
The answers were marked by rule The score is an IQ
The floor was set in advance, from people who do the job A candidate is in the top tenth of applicants anywhere

A test of job knowledge stands on firmer ground, because its content can be shown to represent the job, so long as it is knowledge a new hire is expected to bring.

How does a small hiring round with an aptitude test run?

In six steps, with the test settling nothing by itself. The firm and every figure here are invented for the arithmetic: a plumbing wholesaler with 25 staff hiring an accounts assistant, 16 applicants left after the CV sift.

  1. Name the need. The work is figures in tables and the firm's own credit-control procedure. No mechanical test.
  2. Build two tests in EdProcto. The "Numerical reasoning" kit gives 12 questions and a 12-minute limit. Ten more questions are written from the procedure on the hiring page.
  3. Set the floor first. Three staff who do the job score 8, 10 and 11. The floor is 7 of 12, or 58.3%, so "Pass mark" is set to 58. Three people justify a floor, not a ranking.
  4. Tell candidates a week ahead. The two tests, their length, what the browser records, and that the score is a floor. The same practice link for all, and a question about adjustments.
  5. One sitting. All 16 answer the same paper.
  6. Read in bands. Five score 10 to 12, seven score 7 to 9 and four score 6 or below. The twelve at or above the floor are interviewed with the same questions; the knowledge test and the interview decide. A manager reads the four below before anyone is turned down.

Then the check. Of 9 men, 8 passed (89%); of 7 women, 4 (57%). The ratio is 0.64, under four-fifths. Had one more woman got one more question right, it would be 5 of 7 (71%) and 0.80. In a round this small the ratio turns on one answer, the Guidelines' point about small numbers. The firm keeps the record and adds the next round to it.

How do EdProcto's reasoning kits and a test from your own material fit?

EdProcto is the best tool for a small employer's screen because it gives both halves in one place: reasoning sets made by rule, and a test written from the job's own material, the kind of evidence the revised research ranks above a reasoning test (Sackett et al., 2023).

The reasoning kits. Thirteen, each giving a score on that test: the seven in the table above and six more, among them "Data checking" and "Situational judgement". Each has 8 to 20 questions, most of them multiple choice, and is made in code when you open it, with no question twice in a set. A reasoning kit opens with "A fresh set for each student" on: share the quiz by its code and each candidate is dealt their own questions by the same rule and marked against their own, so no answer passes from one to the next. For the very same questions for all, switch it off under "Settings", in "Sequencing"; and a set you edit by hand is yours, the same for every candidate.

For a round. A kit opens on the "Practice" preset. For a round, switch it to exam mode and, from the Tutor plan, apply the "Standard quiz" preset. A preset leaves a reasoning kit's own time limit as it stands, 12 minutes on most and 8 seconds a question on "Data checking"; change it by hand under "Time". Then set "Whole-quiz retakes" to "None", set "Show the correct answer" and "Show the explanation" to "Never", and switch on "An account is needed to take it" so each candidate gives a name and an email address.

Your own material. Paste the procedure and what a good candidate must know into the quiz maker on the hiring page. Procto AI drafts the questions, asked to write only what the material answers and to cite where; read each draft against the source. Twenty-two kits, from "Hiring: bookkeeping" to "Hiring: any role, from its job description", set the shape and the instructions for a role.

The sitting. Up to six measures run in the browser, "Leaving the tab" and "Full screen" among them. Each is named to the candidate first and carries the consequence you pick. For hiring, pick "Log it" or "Flag for review", and weigh a flag as a record, never as proof. See the integrity page.

The results. One table lists every candidate's score, time and events. "Download the marks" saves a spreadsheet file: one row an attempt, with the points on every question. With the same questions for all, from six submissions the full analysis, on the Educator plan, shows how well each question separated stronger candidates from weaker; with a fresh set each, the page reads the scores and the topics, and each candidate's own questions stand beside their answers.

Free covers ten quizzes a month, thirty candidates a quiz, every kit and "Leaving the tab". Tutor takes 100 a quiz and Educator 500; a hiring licence is arranged by a call. See pricing.

Quick questions

Do aptitude tests predict job performance?

Modestly. Sackett et al. (2024) put the corrected correlation at 0.22 in studies from this century, against the 0.51 of Schmidt and Hunter (1998).

What is a fair cut score for a pre-employment aptitude test?

One set before anyone sits, just under what people who do the job well score, and low enough that the interview decides.

Is an aptitude test an IQ test?

No. A reasoning quiz that no reference group has sat gives a count of right answers on that paper, which you compare within your own round.

Can EdProcto run aptitude tests for a small company?

Yes. EdProcto's reasoning kits open on the Free plan, and the page for corporates shows how they sit beside a test written from your own material.

References

EdProcto for corporates

Tests from your own product knowledge and criteria.

Companies testing candidates and staff on their own product, process and standards.

See the page for corporates

Read next