Book a meeting
All articles
Assessment Science20 November 2026·6 min read

Extra Time Added 26 Points for High Scorers and Nothing Below 400: Six Myths About Assessment Time Limits

The time limit on your assessment is a scoring decision, not an administrative setting. Six beliefs about speededness, tested against the measurement evidence, plus a completion-rate check you can run on data you already hold.

By AssessAll Editorial

A test is speeded when its time limit is tight enough that candidates score lower than they would with more time. Speededness is not difficulty. It adds work rate to the score as a second, usually unintended trait. The long-standing convention is to treat a test as essentially unspeeded only when at least 80% of candidates reach the final item.

Almost every hiring assessment has a clock on it, and almost nobody has checked what the clock does to the scores. The limit is usually inherited from a template, defended on instinct, never tested. Six beliefs drive those decisions. Here is what the evidence shows.

Myth 1: A tighter time limit makes the test more selective

The intuition is that less time separates the strong from the weak. The evidence says a tight clock mostly compresses the people at the top.

When ETS researchers gave SAT candidates roughly time-and-a-half, verbal scores rose by fewer than 10 points on the 200 to 800 scale, and scores for the lowest group, below 400, actually fell slightly. On mathematics, extra time was worth about 20 points overall and 17 to 26 points for candidates already scoring above 600, while candidates at 400 or below gained nothing at all (ETS, *Testing and Time Limits*, 2004).

A later College Board study of 1,929 students found the same shape: in mathematics only the middle-ability group gained significantly, roughly 43 scaled points, while low-ability students showed virtually no difference and high-ability students did worse with double time than with time-and-a-half. The authors' summary: "Some extra time improves performance, but too much may be detrimental" (Mandinach, Bridgeman, Cahalan-Laitusis & Trapani, College Board Research Report 2005-8).

Shortening a test to raise selectivity therefore does little to the candidates you are screening out. It acts on the ones near the top, where your decisions are tightest.

Myth 2: High reliability proves the time limit is fine

This is the most expensive error on the list, because the number used to defend the test is partly produced by the problem.

Internal-consistency estimates such as coefficient alpha are inflated on time-limit tests. Candidates who run out of time leave a block of unanswered items at the end, and that block behaves like a highly consistent extra factor, so the coefficient rises for reasons unrelated to the construct. Cronbach and Warrington set this out in *Psychometrika* in 1951 and proposed separate formulas for speeded tests. Seventy-five years on, it is still routinely ignored in vendor reliability claims.

The modelling problem follows. As ETS's review of speededness methods puts it, speeded responses "do not depend solely on a test taker's ability" and are therefore not appropriate for traditional item response theory (Cintron, ETS Research Report RR-21-03, 2021). An alpha of .92 on a test where a third of candidates never reach the last item is not evidence of a good test. It is evidence that you measured something consistently, and you do not yet know what.

Myth 3: Time pressure hurts everyone equally, so it is fair

Equal treatment is not the same as equal measurement. If the clock interacts with anything that differs between groups, the total score carries that difference.

A 2023 study in Educational and Psychological Measurement modelled this directly. Among 794 university students taking 25 GRE-derived mathematics items under a 20-minute limit, the raw gap in items correct was d = −0.33. Fitting a two-dimensional model that separated latent mathematics ability from a latent completion factor, the ability gap was no longer statistically significant while the completion gap, the work-rate component, was (Stoevenbelt et al., 2023, 83(4), 684–709). Ability and completion correlated only about .35: related, clearly not the same trait.

The authors are careful: their model had convergence problems and they call for replication. The structural point stands regardless. A timed total score blends two traits, and an observed group difference may sit largely in the one you never intended to select on.

Myth 4: Non-native English speakers are penalised by the clock, so remove it

Half right, and the half that is wrong matters.

GMAC's study of speededness by first language found large raw differences. Korean-first-language test takers completed all verbal items only 47.4% of the time, against 95.4% for native English speakers; on the quantitative section, French speakers at 78.7% were the only other group below the 80% convention (GMAC Research Report RR-12-01).

Once ability was held constant, however, differential speededness never exceeded 0.09 on any comparison. Non-completion was heavily concentrated in some language groups, yet candidates of equal measured ability were not treated very differently by the clock itself. The SAT research points the same way: racial, ethnic and gender differences neither widened nor narrowed when extra time was given.

So measure rather than assume. Non-completion by subgroup is a design alarm worth investigating, not proof of unfairness on its own. In India, where many candidates sit an English-medium aptitude test in a second or third language, that distinction separates a design review from a guess.

Myth 5: A tight clock stops cheating

Preknowledge shows up as unusual speed, not slowness, which means a tight limit hides the signal instead of suppressing it.

In a study of computer-based licensure exams, researchers flagged responses far shorter than a response-time model predicted. One candidate answered five items correctly in extremely short time, including one whose timing residual had a chance probability of about .000021, while getting only 60% of the other 106 items right (Qian, Staniewska, Reckase & Woo, *Educational Measurement: Issues and Practice*, 2016). In simulation the method caught 67 of 100 true preknowledge cases when candidates knew 10% or fewer items. Earlier work found detection rates of .8 when a known item was answered within 10 seconds, falling to .2 at 30 seconds.

Those methods need headroom. If your limit forces everyone to rush, fast-and-correct stops being anomalous. Integrity is better served by a workable limit plus response-time and behavioural monitoring — what AssessAll's AI proctoring reports as integrity bands — than by a clock that makes every candidate look suspicious.

Myth 6: If the job is fast-paced, a timed test measures that

Sometimes true, but only when the choice was deliberate. Time limits are essential where speed is part of the construct, typing being the obvious case. For tests of knowledge or reasoning, the ETS view is that speed "should play a minor role, at most."

If work rate matters in the role, it belongs in the job analysis and should be scored as its own result, not smuggled into a knowledge score where you can neither see it nor defend it. That is the logic of the *Standards for Educational and Psychological Testing* and of selection guidance requiring that what you select on be job-related (EEOC; Uniform Guidelines, 29 CFR 1607). Timing is also where accommodation duties bite, extended time being among the most commonly requested adjustments (ADA testing accommodations).

A speededness check you can run this week

  1. Compute the completion rate. What percentage of candidates reach the final item? Below 80%, treat the test as speeded until shown otherwise.
  2. Compute partial completion. What percentage reach at least 75% of items? A low figure here means the clock is shaping the score, not trimming it.
  3. Separate omitted from not-reached. A skipped item mid-test is a decision; a blank run at the end is a time limit. Scoring them alike destroys the evidence.
  4. Break completion down by subgroup, location and device. Mobile and second-language candidates are the usual surprises.
  5. Pilot at 1.5 times the limit. Expect gains concentrated in the upper-middle range. If low scorers gain a lot, your limit is doing something the research would not predict.
  6. Decide whether work rate is in your construct. If it is, score it separately and say so in the candidate instructions. If not, set the limit so the clock is a container, not a variable.

Steps 1 to 4 need only the response data you already hold. Step 5 needs one extra form run on a sample, which on pay-as-you-go credits at ₹30 per assessment is cheap evidence to own.

The takeaway

A time limit is not an administrative setting; it is a scoring decision that adds work rate to every result and inflates the reliability figure you would use to defend the test. Check your completion rates before you defend your cut score, because a test nobody finishes is measuring something you never chose to select on.

#time limits#speededness#psychometrics#test fairness#response time

Measure it, don't guess it.

Start free with 100 credits — or write to solutions@bodhih.com.

Start free