Behavioural assessments for hiring: what they measure, and what they don't

One label, two kinds of evidence

Behavioural assessment covers any instrument intended to establish how someone acts at work rather than what they know or can recall. In practice that stretches from personality inventories through situational judgement tests to full work simulations. US federal guidance for hiring managers treats all of them as assessment methods to be chosen deliberately, and the anti-discrimination framework treats them alike as selection procedures — but the evidence they hand you afterwards could hardly be more different.

The useful split is not “personality versus skills”. It is the source of the record. A self-report inventory produces a set of agreement ratings in which the candidate describes their own tendencies. A behaviour-in-context measure puts a situation drawn from the job in front of the candidate and records what they chose, wrote or produced. One is a claim; the other is an observation of a performance under fixed conditions.

Neither is disqualified by that distinction, and a good process often uses both. The failure mode is collapsing them into one category on a scorecard, because then you end up defending a self-description as though someone had watched it happen.

• Self-report inventories — the candidate rates statements about themselves; the output is a trait profile.
• Situational judgement tests — the candidate is shown a work scenario and states how they would, or should, handle it; classified as low-fidelity simulations because nothing is performed.
• Work samples and simulations — the candidate performs a task that mirrors the job; the output is an artefact someone can open and re-read.

Self-report inventories: a careful record of a claim

A personality inventory asks candidates to rate their agreement with a series of statements designed to place them on relatively stable traits — most often some version of the five factors: extraversion, emotional stability, agreeableness, conscientiousness and openness. The result is a profile rather than a score on a task, and federal guidance is explicit that single-dimension use is ill-advised, because work outcomes are rarely explained by one trait in isolation.

The operational strengths are real. Administration is quick and cheap, the same items reach every candidate in the same order, and the scoring is entirely mechanical, so it does not drift with an interviewer's mood. Guidance for federal hiring managers also describes subgroup differences on personality measures as generally few, which is one reason these instruments remain in wide use.

The limit is structural rather than moral. The instrument records a self-description, collected from someone who knows what the job is advertised as wanting, at the moment they most want the job. That is a measurement condition, not an accusation, and it is the condition under which every score from an inventory is produced.

Impression management is a design property, not an accusation

Personality items are usually transparent: a candidate can read “I keep my commitments” and see exactly which answer looks better. Federal guidance says so directly, noting that item transparency can make it easy for applicants to fake or distort scores in their favour. This is not a fringe criticism of the method; it is written into the method's own description in a government reference for hiring managers.

The research question has never really been whether people can shift their answers — they can — but whether it changes who gets hired. That is what Rosse and colleagues examined in 1998, looking at response distortion in pre-employment personality testing and its effect on hiring decisions. Distortion matters most exactly where these instruments are used most aggressively: at the top of a ranked list, where a small shift reorders who reaches the next stage.

Vendors answer this with social-desirability scales and corrections applied to the raw score. Treat those with the same scepticism as the score itself — a correction is another measurement, with its own error, applied to a number you already could not observe. The practical position is simpler: use an inventory as one input among several, never as an automatic screen-out, and never as the evidence you cite when you explain a rejection.

Situational judgement: behaviour in context, at low fidelity

A situational judgement test presents a work problem drawn from the role and asks the candidate what they would do about it. Because the candidate is not placed in a simulated work setting and is not asked to perform the task, federal guidance classifies SJTs as low-fidelity simulations — closer to the job than a trait questionnaire, still short of doing the work. They are typically aimed at social-functioning competencies: conflict handling, prioritisation, escalation, working with a difficult customer.

The response format does more work than most buyers realise. “What would you do?” asks for a self-report of intention and carries some of the same transparency problem as an inventory. “Which response is best?” asks for judgement against a key and behaves more like a knowledge measure. Neither is wrong, but they are not the same instrument, and the difference should be visible in how you describe the result.

What makes an SJT defensible is where the scenarios and the key come from. Scenarios pulled from a job analysis and a key set by people who actually do the work can be explained to a candidate and to an auditor. A key set by whoever wrote the item, with no record of why one option scores higher, cannot. Published estimates of how well SJTs relate to later performance vary widely with the corrections applied, which is why this post quotes none of them; the argument you can actually make in a hiring meeting is about job relevance, not about a coefficient.

Work samples: behaviour you can re-read

A work sample asks the candidate to perform tasks or activities that mirror the ones employees perform on the job — triaging the ticket queue, replying to the awkward customer email, reconciling the numbers in the spreadsheet, walking through the call. Federal guidance is careful about when this fits: work samples suit competencies candidates are expected to hold on entry, and are a poor choice for anything you intend to train after hiring.

What they leave behind is the reason to prefer them. A work sample produces an artefact: a written reply, a decision trail, a file, a recording. Someone who was not in the room can open it three months later, read what the candidate produced, and form their own view. Scores from an inventory cannot be re-examined that way — you can only re-read the number.

The costs are equally real, and pretending otherwise is how work-sample programmes fail. Tasks take expertise to design and time to score, they demand more from candidates, and scoring drifts between reviewers unless a rubric fixes what each level means. A work sample without a rubric has not replaced a subjective judgement; it has removed the standardisation that made the questionnaire tolerable in the first place.

Where each one is defensible

Start from a job analysis rather than from an instrument. Under the US framework, a selection procedure that disproportionately excludes a protected group has to be justifiable as job related and consistent with business necessity, which is an argument about what the role requires and how the instrument samples it. Obligations differ by jurisdiction, this is not legal advice, and nothing here is a claim that any product makes an employer compliant — the use of the instrument is the employer's, and so is the justification.

Within that frame, the allocation is usually straightforward once the evidence question is asked out loud.

• Use a work sample for a competency the person must have on day one and that produces something a reviewer can look at.
• Use a situational judgement test for judgement in context that cannot practically be performed — escalation, prioritisation, handling conflict — with scenarios and a key drawn from people who do the job.
• Use a self-report inventory, if at all, for orientation and interview preparation rather than as a gate, and describe it in the record as what the candidate reported about themselves.
• Whatever the mix, write down which piece of evidence each decision leaned on — that sentence is what survives being questioned.

What this means for the record you keep

SkillCort is built on the behaviour-in-context half of this category: role blueprint, competency, task, rubric, and the candidate's actual output kept alongside the score that was given to it. What we can describe is content validity — that chain, expert-reviewed, with the reasoning visible — and reliability where enough scoring data exists to compute it. We do not claim that an assessment here predicts later job performance, and you should be wary of any assessment vendor who does so without showing you the study, the population and the date.

The same discipline applies to how a summary is produced. A model can draft the read of a transcript or flag where two evaluators disagreed, and that is genuinely useful when a panel is scoring forty candidates. It does not set a status, and it does not advance or reject anyone; a named human records the decision and the evidence it rested on. That separation is what lets the file answer a question months later without anyone having to reconstruct what happened.

Sources & further reading

• U.S. Office of Personnel Management, Assessment & Selection — Personality Tests
• U.S. Office of Personnel Management, Assessment & Selection — Work Samples and Simulations
• Rosse, Stecher, Miller & Levin (1998), The Impact of Response Distortion on Preemployment Personality Testing and Hiring Decisions — Journal of Applied Psychology
• McDaniel, Morgeson, Finnegan, Campion & Braverman (2001), Use of Situational Judgment Tests to Predict Job Performance: A Clarification of the Literature — Journal of Applied Psychology