What is a work sample test?
A work sample test is a hiring assessment that asks a candidate to perform a task that mirrors the actual work of the job, then evaluates the output against a defined rubric. An engineer writes code, a salesperson builds a mock account plan, an editor revises a draft. It measures what a person can do, not what a resume claims they once did.
That definition is the whole point of the method. Most hiring signals are proxies. A degree is a proxy for skill. Years of experience is a proxy for skill. A polished interview answer is a proxy for skill. A work sample is not a proxy. It is a direct sample of the behavior you are trying to predict, which is exactly why it predicts so well.
Why work samples predict performance
The single most cited paper in personnel selection is Schmidt and Hunter's 1998 meta-analysis, which summarized 85 years of research across thousands of studies. It ranked selection methods by their correlation with later job performance. A perfect predictor would score 1.0. Pure chance scores 0.
Work sample tests came in at .54, near the top of the entire field and slightly ahead of general mental ability at .51. Unstructured interviews, the default method at most companies, scored .38. Years of job experience scored .18. Years of education scored .10. Read those last two numbers again. The two things most resumes lead with are among the weakest predictors ever measured.
The numbers were updated in 2022. Sackett, Zhang, Berry, and Lievens revisited the same literature and argued that decades of meta-analyses had overcorrected for range restriction, inflating every coefficient. Their corrected estimates are lower across the board: work samples drop to .33, GMA to .31, structured interviews to .42. The headline finding is that the ranking shuffled. Structured interviews now sit at the top, and cognitive ability is no longer the uncontested king.
Whether you trust the 1998 numbers or the 2022 revision, the conclusion holds. Methods that observe the work predict the work. Methods that talk around it do not.
Why the direct sample wins
There is a principle underneath all of this called behavioral consistency. The best predictor of future behavior in a situation is past behavior in a similar situation. A work sample manufactures that similar situation on demand. You do not have to wait for a reference to recall it or trust a candidate to describe it accurately. You watch it happen.
Work sample vs other methods
Here is the full hierarchy from both studies, side by side. The 1998 column is the classic Schmidt & Hunter estimate. The 2022 column is the Sackett et al. revision. Where the revision did not publish a comparable corrected figure, the cell is left blank.
| Selection method | 1998 validity | 2022 revision |
|---|---|---|
Work sample test Candidate performs a realistic slice of the job | .54 | .33 |
General mental ability (GMA) Cognitive ability test | .51 | .31 |
Structured interview Same questions, anchored rating scale | .51 | .42 |
Job knowledge test Tests domain knowledge directly | .48 | .40 |
Unstructured interview Free-form conversation, no rubric | .38 | lower |
Reference checks Past-employer verification | .26 | — |
Years of job experience Tenure as a proxy for skill | .18 | — |
Years of education Degree level as a proxy for skill | .10 | — |
Notice what stays true across both columns. Methods that put the candidate in contact with the actual work, work samples, job knowledge tests, and structured interviews built around real scenarios, cluster at the top. Methods that infer skill from a distance, unstructured chats and resume credentials, sink to the bottom. The revision changed the magnitudes. It did not change the lesson.
How to build a work sample test
A work sample is only as good as its design. A vague task scored on gut feel is just a slower interview. Follow these steps to build one that actually predicts.
- Start from the real job, not the job description. List the two or three tasks the person will spend most of their time on in the first 90 days. Your sample should reproduce one of those, not a clever puzzle that no one will ever do again.
- Shrink it to a self-contained slice. Carve out a piece that can be completed in 30 to 90 minutes without insider context. The candidate should be able to start cold and produce something real. If it needs three internal systems and a week of ramp, it is too big.
- Write the rubric before anyone takes the test. Define what strong, solid, and weak output look like, in concrete observable terms. Decide the criteria first so you are scoring the work, not your impression of the person.
- Give every candidate the identical task. The same prompt, the same materials, the same time limit. Standardization is what turns a sample into a measurement instead of an anecdote.
- Score blind where you can. Strip names and remove identifying details before review. Have at least two evaluators rate independently, then compare. Disagreement is information, not a problem to paper over.
- Respect the candidate's time. Keep it short, be explicit about the time expectation, and pay for anything substantial. A bloated unpaid take-home filters for free time, not skill, and it quietly drives away your strongest candidates who have other offers.
- Watch the reasoning, not just the result. The final artifact tells you what they produced. How they got there, the questions they asked and the tradeoffs they made, tells you how they will perform on the next problem you have never seen.
Common mistakes
Testing trivia instead of the job
The most common failure is reaching for a clever brain teaser or an algorithm puzzle that has nothing to do with the daily work. It feels rigorous. It predicts almost nothing. A work sample earns its validity from resemblance to the real task, and a puzzle resembles a puzzle.
Scoring on vibes
A work sample with no rubric is not a work sample. It is an interview with extra steps. Without criteria defined in advance, evaluators drift toward candidates who remind them of themselves, and the whole predictive advantage evaporates.
Making it too long
Multi-day take-homes look thorough and quietly sabotage your funnel. Your best candidates have other options and will not spend ten unpaid hours. You end up selecting for desperation and free time. Keep it tight, or pay for the larger effort.
Forgetting to validate it
A work sample is a hypothesis: people who do well here will do well on the job. Check it. Once you have enough hires, compare their sample scores to their later performance. If there is no relationship, your test is measuring the wrong thing and needs redesign.
FAQ
What is a work sample test?
A work sample test is a hiring assessment that asks a candidate to perform a task that mirrors the actual work of the job, then evaluates the output against a defined rubric. Examples include a coding exercise for an engineer, a mock account plan for a salesperson, or a writing assignment for an editor. It measures what a person can do, not what they claim on a resume.
How accurate are work sample tests at predicting job performance?
Schmidt & Hunter (1998) reported a corrected validity of .54 for work sample tests, among the highest of any single method. Sackett et al. (2022) re-estimated it lower at .33 after correcting for overcorrection of range restriction, but work samples still rank among the strongest predictors. Both studies agree they beat unstructured interviews (.38 and lower) and years of experience (.18).
What is the difference between a work sample test and a skills assessment?
A skills assessment measures an isolated capability in the abstract, such as a multiple-choice grammar quiz or a logic puzzle. A work sample test reproduces a realistic slice of the job itself, with the same ambiguity and judgment calls the role demands. The closer the task is to the real work, the more predictive it tends to be.
Are work sample tests fair to candidates?
Well-designed work samples tend to produce smaller subgroup differences than cognitive ability tests and are easier for candidates to see as job-related, which improves perceived fairness. Fairness depends on design: a clear rubric, the same task for every candidate, reasonable time limits, and a task that does not advantage people with insider context.
How long should a work sample test take?
Long enough to surface real judgment, short enough to respect the candidate. Most effective work samples run 30 to 90 minutes. Anything that demands many unpaid hours hurts your completion rate and your reputation, and it filters for free time rather than skill. If the task must be large, pay candidates for it.
Sources
- Schmidt, F. L., & Hunter, J. E. (1998). The validity and utility of selection methods in personnel psychology. Psychological Bulletin, 124(2), 262-274.
- Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. (2022). Revisiting meta-analytic estimates of validity in personnel selection: Addressing systematic overcorrection for restriction of range. Journal of Applied Psychology.