Shortening a hiring assessment to reduce candidate drop-off usually fails on both counts. Across roughly 250,000 applicants, assessment length did not drive incremental attrition past the first 20 minutes, and the applicants who quit scored lower than those who finished. Shortening cuts reliability immediately — and reliability is what your pass/fail decision rests on.
The case below is illustrative. It is a composite of a decision that comes up constantly in volume hiring, written to show the reasoning rather than to report any one organisation's data. No client figures appear in it.
The decision on the table
A shared-services team screening for voice and chat roles runs a 38-minute assessment: English comprehension, a short cognitive section, and two AI-graded scenario items. Completion sits at 71%. The hiring head wants it cut to 15 minutes, on a simple premise — the test is long, candidates are abandoning it, so a shorter test recovers the lost 29%.
The premise contains two claims. The first is causal: length is what drives the drop-off. The second is implicit: the candidates being lost are ones worth keeping. Both are testable, and the published evidence runs against both.
Claim one: length drives the quitting
Hardy, Gibson, Sloan and Carr (2017), in the Journal of Applied Psychology, examined applicant attrition across roughly 250,000 job-seekers drawn from dozens of live selection systems. Their finding was not that length is irrelevant, but that it is badly misunderstood: quitting clusters heavily in the opening minutes, and extending an assessment beyond about the 20-minute mark did not produce the proportional increases in attrition that practitioners expect. The authors' own recommendation was that reliability and validity, not fear of drop-off, should determine how long an assessment is. The paper record is in Journal of Applied Psychology, 102(7), 1148–1158.
The practical reading: a candidate who has invested 22 minutes is not the one at risk. The one at risk left before minute five. A 2025 vendor analysis of aggregated client data — a provider's own figures rather than peer-reviewed research, so treat it as directional — reports the same shape, with over half of dropout occurring in the first five to ten minutes and 7.7% of starters leaving on the contact-information page, before seeing a single test item.
If that pattern holds in your funnel, cutting 23 minutes off the back end removes time that almost nobody was quitting during.
Claim two: you are losing good candidates
This is the assumption that does the most damage, and it has been tested directly at scale. Hardy, Gibson, Carr and Dudley (2021), in the International Journal of Selection and Assessment, set out to examine the belief that stronger candidates are the ones who abandon long batteries. Across two studies totalling about 2.19 million applicants, the relationship ran the other way:
- Study 1 — 327,517 job-seekers across eight hiring contexts. Assessment performance was negatively associated with subsequent attrition: the better someone was doing, the less likely they were to quit.
- Study 2 — 1,844,604 applicants for hourly associate roles and 18,937 applicants for nursing roles. Same pattern, replicated.
The authors' conclusion is worth quoting in substance: some degree of attrition may benefit rather than harm organisations using pre-hire assessments. Candidates who sense they are performing poorly are the likelier quitters.
That reframes the 71% completion rate entirely. It is not 29% lost talent. It is a mix of genuine friction losses — the contact-information wall, a device problem, an interrupted session — and self-selection that was going to be a rejection anyway.
Pricing the shortening
The cost side is arithmetic, not opinion. The Spearman–Brown prediction formula tells you what happens to internal-consistency reliability when you change test length. Take a 40-item section with an alpha of .88, scored on a scale with a standard deviation of 10 points:
- 40 items — reliability .88, standard error of measurement 3.5 points, 95% band around a score roughly ±6.8 points
- 24 items — reliability .82, SEM 4.3, band ±8.4 points
- 20 items — reliability .79, SEM 4.6, band ±9.1 points
- 15 items — reliability .73, SEM 5.2, band ±10.1 points
Halving the test widens the uncertainty around every candidate's score by about a third. If the cut score is 70, the 40-item version cannot reliably separate a 63 from a 77; the 15-item version cannot reliably separate a 60 from an 80. At high volume, that band is the difference between a defensible decision and a coin toss dressed as a number. The AERA/APA/NCME *Standards for Educational and Psychological Testing* require reliability and standard error to be reported for the scores actually used — which means a shortened form needs its own reliability evidence, not the old form's.
Recovery is expensive, too. Running the same formula in reverse: to get a 20-item test with reliability .79 back up to .88, you need twice the items. There is no cheap route back.
The legal exposure moves in the same direction. EEOC guidance on employment tests and the Uniform Guidelines at 29 CFR Part 1607 tie the defensibility of a selection procedure to the validity evidence for the procedure as used. A shortened form is a different procedure.
Where the drop-off actually was
The team instrumented the funnel by step instead of arguing about total minutes. Four things surfaced:
- A contact-information page before any content. Candidates were asked for details before they had any reason to care.
- No stated duration. Candidates budgeted for ten minutes and abandoned at twelve.
- A session that did not survive interruption. An unrecoverable session punishes anyone taking the test between shifts.
- A hand-off from the tracking system that silently failed on some mobile browsers.
None of those is a length problem. All four are fixable in a fortnight.
What they changed
- Moved contact capture after the first scenario item.
- Stated the duration up front, rounded generously — "about 40 minutes" for a test with a median completion in the low thirties.
- Made sessions resumable, with proctoring integrity bands recording the interruption rather than voiding the attempt. Flagging a break as a band rather than a disqualification is how platforms including AssessAll keep a resumable session honest.
- Fixed the mobile hand-off.
- Left the measurement alone.
The total test length did not change. The decision quality did not degrade, because nothing about the scores changed.
When shortening is the right call
Shortening is sometimes correct, and pretending otherwise is just a different bias:
- Reliability is already well above what the decision needs. A sift that only has to separate the top 40% from the rest does not need .90. Cut the surplus.
- Sections are redundant. Two subscales correlating at .85 are measuring one thing twice. Drop one and lose nothing.
- Adaptive delivery is available. Item-level adaptive testing holds reliability at fewer items — but only if the bank is deep enough to support it.
- The content is not doing work. Items with near-zero discrimination are length without measurement. Remove them and reliability can rise.
- Per-candidate cost, not drop-off, is the real constraint. That is a budget problem with a budget answer. Pay-as-you-go pricing — AssessAll charges ₹30 / US$0.50 per assessment credit, with free credits on signup — changes the economics without touching the instrument.
What these have in common: each one removes length that was not buying measurement. None of them removes measurement to buy completion.
Takeaway
Before shortening an assessment to fix drop-off, find out where candidates are actually leaving and whether the leavers were scoring well — the published evidence says they were not. Length that buys reliability is not friction, and a completion rate is not a quality metric.