An assessment retake policy sets four things: how long a candidate waits before a second attempt, whether they see the same test form or a different one, which score counts, and how many attempts are allowed. It matters because second attempts are reliably higher — around a third of a standard deviation on cognitive tests — without underlying ability having changed.
Most hiring teams write that policy the week a candidate appeals. This walkthrough follows one such decision. The team and the numbers below are illustrative — a composite, not a client — but every piece of evidence used to settle each question is real and cited.
The situation
A services firm screens about 400 applicants a month for a customer-operations role. The funnel starts with a 35-minute online assessment: a numerical and verbal reasoning section, plus three AI-graded written scenarios. The cut score is 62, set by a modified Angoff panel the previous year.
A candidate scores 60. Her recruiter likes her, she has relevant experience, and she emails asking for another attempt. There is no retake policy. The talent acquisition lead has to invent one — and whatever she decides will apply to every candidate behind this one.
Six decisions follow, in the order they actually have to be made.
1. Allow retakes at all?
Refusing is the simplest option and the hardest to defend. Retesting is already the norm at scale: Hausknecht and colleagues' meta-analysis of 50 studies, 107 samples and roughly 134,000 participants (Journal of Applied Psychology, 2007) estimated that 25–50% of applicants in organisational and educational settings are retested. A single administration also carries measurement error, and a candidate two marks below a cut score sits inside the band where error plausibly explains the gap.
The honest case for no retakes is narrower than it looks, but it exists: if the assessment is the only hurdle, item exposure is high, and the role has thousands of applicants per opening, a blanket no-retake rule applied uniformly is cheaper and more consistent than a rule that gets bent. The team chose to allow retakes — but to write the rule down before granting the first one.
2. How long between attempts?
The instinct is a long cooling-off period, on the theory that practice effects fade. The evidence says the waiting period does less work than people assume. In Scharfen, Peters and Holling's meta-analysis of retest effects on cognitive ability tests (Intelligence, 2018), longer intervals were associated with slightly smaller gains, but the slope was roughly −0.0008 of a standard deviation per week — weak enough that a three-month wait buys you almost nothing in score terms.
What a waiting period actually buys is different and still worth having:
- Time for genuine preparation, which is the only kind of improvement you want to reward.
- Rotation of the item pool, so the second attempt is not a memory test.
- An administrative brake on candidates attempting repeatedly in a week.
For calibration, high-stakes admissions testing lands in the same range: GMAC requires 16 calendar days between GMAT appointments, with up to five attempts in a rolling 12-month period. The team set 21 days.
3. Same form, or a different one?
This is the decision that matters most, and it is usually made by default — by whichever form the platform happens to serve.
In the Scharfen meta-analysis, the first-to-second-test gain was 0.37 of a standard deviation when the identical form was reused, versus 0.23 with alternate forms. The difference is not merely cosmetic. Lievens, Reeve and Heggestad (Journal of Applied Psychology, 2007) examined 941 candidates in a high-stakes setting and found no measurement bias at the initial test — but retesting induced both measurement and predictive bias, with retest scores reflecting less general cognitive ability and more memory. In that sample, the shift eliminated the test's criterion-related validity.
Read plainly: reuse the same form and you may still get a number, but it is a number about a different construct. If your platform cannot draw a second attempt from a calibrated item bank, you do not have a retake policy — you have a memory test with a cut score. AssessAll serves second attempts from separate item pools for this reason, and keeps the AI-graded scenarios distinct across forms.
4. Which score counts?
Three defensible answers, and they are not equivalent.
Most recent score
- Simple to explain and administer.
- Supported by evidence that later scores are not worse measures: Van Iddekinge and Arnold's review (Annual Review of Organizational Psychology and Organizational Behavior, 2017) reports that retest scores tend to be higher, more variable, more reliable, and somewhat more strongly related to academic and job performance.
- In Lievens, Buyse and Sackett's study of Belgian medical school admissions (Personnel Psychology, 2005), knowledge-test scores from the second administration predicted academic performance better than first-administration scores.
Highest score
- Most candidate-friendly, and the most inflationary: it capitalises on every attempt's measurement error in one direction.
First score, with the retake as a tiebreaker
- Most conservative; hardest to explain to a candidate.
One finding should temper all three. In the same 2005 study, retesters gained about a third of a standard deviation on the knowledge test and the situational judgement test and about half a standard deviation on the cognitive test — and, crucially, the same score predicted lower subsequent performance for candidates who passed on a second attempt than for those who passed first time. A 64 on attempt two is not interchangeable with a 64 on attempt one. If you use the most recent score, you are accepting a slightly weaker signal in exchange for a fairer process. That is a reasonable trade, but make it knowingly.
The team chose the most recent score, from a fresh form.
5. How many attempts?
The returns run out faster than most policies assume. In the Scharfen meta-analysis, gains between consecutive administrations were 0.33 from the first to the second test, 0.17 from the second to the third, and 0.02 from the third to the fourth — not statistically significant. The authors conclude that "a plateau seems to be reached after the third test administration."
So a third attempt adds roughly half the improvement of the second, and a fourth adds nothing measurable. Two attempts per opening, with a longer window before a fresh cycle, captures almost all the legitimate upside.
6. What goes in the record
A retake policy is a selection procedure, and it needs the same paper trail as any other. The Uniform Guidelines on Employee Selection Procedures ask users to maintain records of the impact of their selection procedures (§1607.4(A)), treat a selection rate below four-fifths of the highest group's rate as evidence of adverse impact (§1607.4(D)), and make that determination at least annually (§1607.15(A)(2)(a)).
Retakes cut across all of that. If retakes are granted recruiter by recruiter, the rule is not the selection procedure — recruiter discretion is, and it will not pass an audit. Minimum record per attempt: attempt number, form ID, date, score, which score was used for the decision, and the proctoring integrity band for that sitting (AssessAll records these per attempt, which is what makes an attempt-level audit possible at all).
What the evidence does not settle
Three honest gaps. First, almost all the retest research is on cognitive and knowledge tests; practice effects on AI-graded open-response scenarios are far less studied. Second, the published gains come mostly from self-selected retesters, who differ from the full applicant pool. Third, nobody has established what retake policy optimises hiring outcomes — only what each option does to scores and validity.
The takeaway
A retake policy is four numbers and one sentence about which score counts, and it is cheaper to write before a candidate asks than after. Get the alternate form right first: reusing the same test is the one choice the evidence says can quietly convert your assessment into a memory test.