Speed or power? What time limits on psychometric tests really measure

Almost every online ability test has a time limit, but the limit does very different jobs depending on how the test is designed. Understanding the difference between speed and power tests helps employers choose the right test, read the results correctly and handle requests for extra time fairly.

Two kinds of test

Psychometricians have long distinguished between two ideal types of test.^[1] A pure speed test is made of items that almost everyone could answer correctly given enough time, with a time limit so tight that few or no candidates finish. Scores depend mainly on how many items a person gets through. Classic examples are clerical checking and simple coding tasks. A pure power test is the opposite: items vary in difficulty, the hardest are beyond most candidates, and there is enough time for nearly everyone to attempt every question. Scores depend on how difficult a problem a person can solve.

Most real ability tests sit between these poles. A numerical or verbal reasoning test with a time limit is usually intended to be primarily a power test, with the limit there for practical reasons, but it will be speeded to some extent if a meaningful proportion of candidates run out of time. Testing specialists use rules of thumb to check this, such as the proportion of candidates who complete the test and the proportion of total score variance explained by items not reached.

Why time limits are set

There are good reasons for time limits. They make administration predictable and standardised, so every candidate has the same conditions, which is essential when scores are compared with a norm group. They limit the opportunity to look up answers or seek help in unsupervised online testing. And for some roles, working quickly and accurately under time pressure is a genuine part of the job, so measuring speed is the point rather than a side effect.

The difficulty arises when speededness is unintended. Lu and Sireci reviewed the validity issues this creates: if a test meant to measure reasoning is more speeded than its designers intended, scores start to reflect working speed, test-taking strategy and anxiety as well as reasoning ability, and items left unreached may leave parts of the intended content unmeasured.^[2] Research on reasoning tests supports this concern. Wilhelm and Schulze found that time-limited reasoning scores were more strongly related to mental speed than unlimited ones, suggesting that a tight limit turns a reasoning score into a blend of reasoning and speed.^[3]

Speed and accuracy are different things

Speed and accuracy are related but distinct. A candidate who answers twenty questions with 95% accuracy and one who answers thirty with 65% accuracy might end up with similar raw scores, yet they are working in very different ways. Modern psychometric models treat response time and response accuracy as separate, correlated aspects of performance, and can estimate each person's speed alongside their ability.^[4]

This is why reporting both speed and accuracy can be useful. The number of questions attempted, the percentage of those answered correctly and the proportion of the available time used each add information to the main score. A high score achieved by working fast but carelessly may call for a different conversation at interview from the same score achieved slowly and precisely, and which pattern suits the role depends on the job. None of these indicators should become a hurdle in itself; they are context for interpreting the score in bands alongside other evidence. Tests such as error checking make the trade-off between speed and accuracy especially visible, because both matter in the work they simulate.

Time limits, fairness and extra time

Time limits interact with fairness in two ways. First, if a test is more speeded than it needs to be, any group that tends to work more slowly for reasons unrelated to the job, such as candidates working in a second language or those less familiar with timed testing, may be disadvantaged. Generous limits on tests intended to measure power reduce this risk without changing what the test measures.

Second, under the Equality Act 2010, employers have a duty to make reasonable adjustments where a provision, criterion or practice puts a disabled person at a substantial disadvantage.^[5] Extra time is one of the most commonly requested adjustments, for example for candidates with dyslexia, visual impairments or some physical conditions. Research on extra time helps explain how it works. A major review by Sireci, Scarpati and Li found that extended time tended to improve scores for all test takers, with larger gains for students with disabilities.^[6] A 2025 research review by Ofqual, the exams regulator in England, reached a similar conclusion: extra time helped most students somewhat, largely because many of the tests studied were speeded, but where students could work close to their potential within the standard time, the benefits were largely confined to those with an access need.^[7] In other words, extra time can raise scores for anyone if the standard limit is tight, which is why it fits most naturally where the limit is not part of what is being measured.

The implication is that extra time is easiest to justify, and least likely to distort comparisons, on tests designed primarily to measure power. Where speed is part of what is being measured, adjustments need more careful thought, and alternatives to extra time may be more appropriate. Decisions should be individual, based on the candidate's needs and the test's design; our guide to reasonable adjustments in online testing covers the practicalities.

Choosing and reading timed tests

For employers, three questions help. Is speed a genuine requirement of the role, or is the time limit there mainly for practical reasons? Does the publisher provide evidence on how speeded the test is, such as completion rates? And does the report show how many items were attempted and how accurately, as well as the overall score? Good answers to these mean the time limit is doing the job it was designed for, rather than quietly changing what the test measures.