--- title: "Speeded test" description: "A speeded test is one whose time limit is tight enough that a meaningful share of candidates never reach the last item, so the score reflects how much correct work was completed rather than how difficult a problem the person could solve. Speed and accuracy are measured together and cannot be separated afterwards." canonical: https://www.assessall.com/guides/glossary/s/speeded-test updated: 2026-09-09 source: AssessAll --- # Speeded test _Also known as: Speed test, Speeded form._ A speeded test is one whose time limit is tight enough that a meaningful share of candidates never reach the last item, so the score reflects how much correct work was completed rather than how difficult a problem the person could solve. Speed and accuracy are measured together and cannot be separated afterwards. ## How to tell whether the test you bought is speeded The time limit on its own tells you nothing. What decides it is the completion rate: the share of the cohort that reached the final item before the clock ran out. On a power form that figure is close to everybody. On a speeded form it is not, and the gap between the two is where the score stops meaning what the label says. That number is almost never published, and it is the single most useful thing to ask a vendor for. It is also a number they have, because it falls straight out of delivery data. A vendor who cannot say what share of the last cohort finished is telling you they have not looked at whether their instrument measures difficulty or pace. ## The reliability claim the field repeats, and the study that tested it The textbook rule is that internal-consistency reliability is inflated on a speeded test. The argument is mechanical: once a candidate runs out of time, every remaining item is scored wrong, so the unreached items agree perfectly with each other across everyone who ran out, and that spurious agreement pushes coefficient alpha up. On that reasoning, alpha on a speeded form should be discarded. Yigal Attali tested the reasoning against real number-right multiple-choice data at Educational Testing Service and found it does not hold, because the assumption behind it is wrong. Candidates who run out of time on a multiple-choice test do not leave the remaining items blank; they guess. Guessing is noise rather than agreement, so the responses to the last items are less consistent with the rest, not more. His report states it plainly: reliability estimates of number-right multiple-choice tests are not inflated by speededness. The consequence runs the other way, and it is more useful than the rule it replaces. Because the unreached items add noise, lengthening a speeded test can lower its reliability rather than raise it. Attali gives the condition: once more than half the candidates are forced to guess an item because there is no time left, adding that item may reduce the internal-consistency estimate. There is a harder half. An internal-consistency figure is computed inside a single sitting, so it cannot see how much a person's pace varies between sittings. Attali's point is that alternate-form reliability falls further than alpha does under speededness, because the noise from guessing is itself inconsistent across occasions. So on a speeded form alpha is not inflated, but it is still the friendliest reliability figure available, and it is the one a vendor will quote. ## What it changes about a decision Two scores from a speeded form and a power form of the same construct are not comparable, and the same person will not get the same result on both. Record which form was used, on every candidate, every time, and never rank a pool that sat a mixture. Extended time as an accessibility accommodation is a different thing on a speeded form than on a power form. On a power form extra time protects the meaning of the score. On a speeded form it changes what is being measured, because pace was part of the construct. This is not a reason to refuse the accommodation; it is a reason to record it and to stop treating the resulting score as interchangeable with the rest. The design question comes before all of that. A speeded form is the right instrument when the job is genuinely bounded by the clock, and the wrong one when it is not, in which case it will penalise the careful candidate you were trying to find. There is a regulator's sentence for this, and it is the one the market skips. Reviewing time limits across qualifications for Ofqual in November 2025, Stephen Holmes wrote that it is important from a fairness perspective to be clear and explicit when speed is intended to be measured as part of a test construct. Almost every commercial aptitude test carries a timer; almost none of them state whether pace is part of what is being measured or merely a scheduling constraint. That single missing sentence is the difference between speededness that was designed and speededness that happened. --- Source: https://www.assessall.com/guides/glossary/s/speeded-test