How much of a timed aptitude test is never answered

The shortfall is not spread across the paper. In the opening fifth, 0.2% of items go unanswered. By the fourth fifth it is 31.9%, and in the closing fifth it is 57.4%. The curve is not a decline, it is a cliff, and it arrives in the same place on every speeded form measured.

None of these tests penalises a wrong answer. That is what makes the number a finding rather than a description. An unanswered question is the only response on the paper that is guaranteed to score nothing. Every other response, including one chosen at random with two seconds left, has positive expected value.

Each attempt is divided into fifths by item position and the share of items left blank is counted in each. The two lines in Figure 1 differ in one respect only: how many seconds the form allows per question.

Subject matter barely moves this number. Verbal, numerical, abstract and mechanical forms all appear on both sides of the divide. Seconds allowed per question moves it.

McQuaig is the case worth sitting with. It allows 18 seconds a question, the same as the CCAT, and runs at 0.2% blank against the CCAT’s 20.6%. The difference is not the clock in the abstract but the clock relative to the work each item demands: McQuaig items are quick word and number association, CCAT items are multi-step reasoning. What binds is seconds per unit of work, and most candidates never establish which kind of paper they are sitting until it is too late to change how they pace it.

A guess on a four-option item is worth a quarter of a mark and a guess on a five-option item a fifth. Our forms mix both, so the value of one blank is measured rather than assumed: across the speeded forms it averages 0.2 of a mark.

The practical instruction that follows is narrow and easy to execute: stop working with about thirty seconds left and put an answer on everything you have not reached. On a fifty-question form that is worth roughly two marks, which is 8% of a typical Cubiks score and 7% of a typical CCAT score, for no additional preparation and no risk.

• It is not causal. The speeded and generous groups are different tests sat by different people. The clock is the obvious difference and the only one we can point to, but it is not a randomised comparison.
• It does not price a guess against an employer’s cut score. Marks forfeited are raw marks on our own forms. Whether two marks changes an outcome depends on a scaled score and a threshold neither we nor the candidate can see.
• It does not apply to negatively marked tests. Every form here scores a wrong answer as zero rather than as a deduction. A test with a penalty inverts the advice entirely.
• These are practice attempts. They were sat on PrepClubs forms written to each test’s published format, not on the publishers’ own instruments, and sat with the stakes of practice rather than of a job application.

Every discard is published. The large one is the re-grade: banks were edited after launch, so an attempt whose stored score no longer matches the form on disk cannot be scored honestly and is dropped.

Candidates who failed to finish were not excluded. Not finishing is the measurement. Filtering on it would be selecting on the outcome being reported, which is the error this report exists to avoid.

The clock filter is half, not ninety percent. A 90% threshold selects for time pressure, so it would thin the generous-clock comparison group of exactly the people who found the paper easy and handed in early. At half the clock the comparison is fair and the result still holds: 57.4% against 12.1%. As a robustness check, among the 488 attempts that ran the clock down to 90% or more the final fifth runs at 61.2%.