PrepClubsPrepClubs
PrepClubs ResearchReport 2026-02Edition 1Published

What an extra thirty seconds on a question actually buys

Every prep guide tells candidates not to dwell on a hard question. The usual justification, that deliberating makes you wrong, is not what the data says. The real argument is arithmetic, and it is stronger.

Sample 533 analysed attempts on speeded forms carrying per-item timing, drawn from a corpus of 575 distinct candidates. 17,825 individual item sittings in the controlled comparison.

1Finding

3.1 points

is all that separates fast and slow answers to the same question

72.5% against 69.4%, across 190 items sat at least 20 times each. Compared without holding the question constant, the same comparison appears to be worth 20.2 points.

Spending longer on a question does not meaningfully damage your chance of getting that question right. The apparent penalty is almost entirely an artefact: long answers land on hard items, and hard items are answered wrong more often whoever is answering them.

The cost is somewhere else, and it is large. The median attempt on a speeded form spends 102 seconds beyond a thirty-second-per-item cap. At the median item time of 14 seconds, that surplus would have covered 7.3 further questions. Report 2026-01 measures what happens to those questions: on a speeded form they are simply never answered.

2The gradient that looks like a penalty

Group every answered item by how many seconds it took and the pattern is dramatic. Accuracy peaks in the 5–10s band at 82.4% and falls to 62.2% at the slow end. Read naively, that is a 20.2-point penalty for thinking.

Accuracy on an aptitude test question against the seconds spent answering it, before controlling for question difficulty. Accuracy rises from 63.5% in the 0–5s band to a peak of 82.4% at 5–10s, then falls steadily to 62.2% in the slower bands.
Figure 1. The apparent penalty for deliberating. Section 3 shows how much of it survives a control. Vector copy
Table 1. Accuracy by seconds spent on the item
Seconds on the itemItem sittingsAnswered correctly
0–5s2,02963.5%
5–10s4,85182.4%
10–15s4,06974.6%
15–20s3,00571.5%
20–30s3,48367%
30–45s2,23363%
45–60s77762.2%
60s+78164.7%

Answered items only, speeded forms only. An unanswered item has no dwell time to record.

The dip in the fastest band is worth noting on its own. Items answered in under five seconds run at 63.5%, below the peak, which is what carelessness looks like. But the size of that dip is the point: being too fast costs about 18.9 points on the items it affects, while never reaching an item costs everything on it.

3The same comparison inside the same question

The comparison above is between different questions. To remove that, take every individual question sat at least 20 times, split its own sittings at its own median time, and compare the two halves. Item difficulty is then held exactly constant, because it is the same item.

The same comparison inside a single aptitude test question, which holds difficulty constant. Across 190 questions, the faster half of sittings answers 72.5% correctly and the slower half 69.4%, a gap of 3.1 points.
Figure 2. Holding the question constant leaves a gap of 3.1 points, against the 20.2 points in Figure 1. Vector copy
Table 2. Within-item comparison
MeasureValue
Questions with enough sittings to split190
Minimum sittings per question20
Item sittings in the comparison17,825
Faster half, answered correctly72.5%
Slower half, answered correctly69.4%
Gap3.1 points
Mean gap within a question0.2 points
Questions where the faster half scored higher102 of 190 (54%)

The faster half wins on 54% of questions. That is not a rule, it is a coin landing slightly heavy on one side. Whatever is left after the control is small enough to be explained by stronger candidates answering faster, which this design does not rule out and does not need to.

4What the surplus time costs

If deliberation does not damage the item, the case for moving on has to be made in time rather than in accuracy. It is easily made.

What time spent over a thirty-second-per-question cap costs on a speeded aptitude test. The median attempt spends 102 seconds past the cap, against a median question time of 14 seconds, which is 7.3 further questions never reached.
Figure 3. The cost of a long dwell is not the item it is spent on. It is the items further down the paper. Vector copy
Table 3. Time spent past a thirty-second cap, speeded forms
MeasureValue
Attempts measured533
Median seconds spent beyond the cap102s
Mean seconds spent beyond the cap249s
Median seconds per item14s
Questions the median surplus would cover7.3
Share of the clock spent on over-cap items40.7%
Attempts with at least one item over 30 seconds98%
Attempts with at least one item over a minute63%

The mean is far above the median because a minority of attempts sink several minutes into a handful of items.

Nearly two in three attempts contain at least one question that ate a full minute, and the median attempt gives 40.7% of its entire clock to items running over thirty seconds. That is where the unanswered final fifth in Report 2026-01 comes from. It is not a failure of knowledge at the end of the paper. It is a budget spent in the middle.

5The rule this supports

  • Cap each item at about thirty seconds on a speeded form. When the cap is reached, answer with whatever you have and move on. You give up around 3.1 points of accuracy on that item and buy back the rest of the paper.
  • Do not agonise about the answer you leave behind. The evidence that a longer look would have rescued it is thin: half of the questions here are answered no better by the people who took longer over them.
  • Read the item before answering it. The one genuine speed penalty in the data is at the very fast end, under five seconds, which is the signature of answering without reading.
  • Practise the cap, not the knowledge. A candidate who already knows the material and cannot pace it will keep producing the same score.

6What this does not show

The within-item split is observational. It holds the question constant but not the candidate, so a residual gap of a few points is consistent with stronger candidates answering faster as well as with any effect of speed itself. The report does not claim a causal penalty for thinking; it reports that the gain from thinking longer is too small to pay for what it costs elsewhere.

  • The candidate is not held constant. The split is within a question, not within a person. A residual gap of 3.1 points is consistent with abler candidates simply answering faster.
  • Timing is per item as recorded by the test player. An item revisited later accumulates time; an item abandoned and never answered contributes no reading at all.
  • Thirty seconds is a choice, not a discovery. It is roughly twice the median item time on these forms, which makes it a usable instruction. Nothing in the data identifies an optimal cap.

7Sample, filters and discards

The corpus and the filters are shared with every other report in this series, so the two can be cited side by side. Speeded forms only: CCAT, Wonderlic, PI Cognitive, Cubiks Logiks.

Attempts on record1,473
Discarded: re-grade disagreed with the stored score556
Discarded: no answers recorded26
Discarded: used under half the allowed clock164
Of which on speeded forms, carrying per-item timing533
Analysed727
Distinct candidates behind those attempts575

8The forms this report is computed from

Only forms allowing under thirty seconds a question carry per-item timing worth splitting, so the comparison is drawn on these four.

9Common questions

How long should you spend on one aptitude test question?
On a speeded form, no more than about thirty seconds. Not because thinking longer makes you wrong, which this report finds it largely does not, but because the time has to come from somewhere. The median attempt here spends 102 seconds past that cap across the paper, which at a median of 14 seconds an item is 7.3 questions never seen.
Is it true that your first instinct is usually right on aptitude tests?
Not as told. The raw data looks like it supports the claim: accuracy peaks in the 5–10s band and falls by 20.2 points at the slow end. But that comparison is between different questions, and slow answers land on hard questions. Inside the same question the faster half is ahead by only 3.1 points, and the faster half wins on just 102 of 190 items, which is close to a coin flip. The case for moving on is the clock, not intuition.
Does rushing make you careless?
At the very fast end, yes, a little. The 0–5s band runs at 63.5%, below the 82.4% of the 5–10s band, which is consistent with answers given without reading the item properly. The cost of being too fast is real but it is small, and it is bounded: an item answered in four seconds still has positive expected value, while an item never reached has none.
What counts as a speeded test here?
A form that allows under thirty seconds a question: the CCAT, Wonderlic, PI Cognitive and Cubiks Logiks forms on this platform. 533 analysed attempts on those forms carry per-item timing, which is what makes this comparison possible at all.

Suggested citation

“Holding the question constant, candidates in the faster half of sittings answer correctly 72.5% of the time against 69.4% for the slower half, a gap of 3.1 points across 190 items and 17,825 sittings. The median speeded attempt spends 102 seconds beyond a thirty-second-per-item cap, enough for 7.3 further questions. PrepClubs Research, Report 2026-02.”

Download the figures as CSVMethod and standardsAsk us to check a figure

Reproduce any figure or table with attribution to PrepClubs Research and a link to this page. The underlying per-question records are not published: joined to an attempt they are re-identifiable at this sample size.

Other reports in this series