What Is a Work Sample Test?

A work sample test is a hiring assessment that asks a candidate to actually perform a piece of the job, a short, realistic task done under timed and observed conditions, and scores the result against a rubric instead of a self-report or an interviewer's gut read. A customer service candidate answers a mock support ticket. A data analyst cleans a messy spreadsheet. The test does not ask what someone knows about the job. It watches them do a slice of it.

For most of the last four decades, one number followed the format everywhere it was discussed: a validity of 0.54 against job performance, high enough to rank work samples among the best predictors personnel psychology had. That number traces to Hunter and Hunter's influential 1984 review. It also turned out to rest on a narrow, decades-old set of studies that nobody could fully reconstruct.

The work sample test number that got corrected

Roth, Bobko and McFarland (2005), published in Personnel Psychology, went back through the accumulated research rather than taking the 1984 figure on faith, pooling 54 studies and 10,469 people to find an observed correlation between work sample scores and job performance of 0.26, rising to 0.33 with an 80% credibility interval of 0.24 to 0.42 once the standard correction for unreliability in the performance ratings is applied, and to 0.39 once error on both sides of the equation is corrected for, with neither corrected estimate's credibility interval reaching the widely cited 0.54.

The gap is not a rounding difference. The authors call it roughly a third lower than the figure the field had been citing for twenty years, and trace it to newer studies, many from service-sector jobs that were barely represented in the original 1984 data, pulling the pooled estimate down.

Why it still holds up

A corrected validity of 0.33 sounds like a step down, but it lands in respectable company. Sackett, Zhang, Berry and Lievens' 2022 correction to decades of personnel-selection meta-analyses puts structured interviews at 0.42 and cognitive ability tests at 0.31, with work samples again landing at 0.33, the same figure Roth and colleagues reached seventeen years earlier through an entirely separate route. Two independent corrections, run on different literatures at different times, converge on the same number.

The same 2005 paper also summarizes something a raw validity coefficient does not capture: work sample tests are reported in the literature it reviews to show smaller score gaps between ethnic groups than cognitive ability tests, and candidates tend to rate the format as fairer than a generic interview or a personality inventory, likely because the task in front of them plainly resembles the job itself.

What this means for a candidate

A work sample test is hard to talk your way through, because there is no separate step where a claim gets converted into a number. The task and the score are the same event. That is the same logic behind a situational judgement test, which presents a realistic scenario instead of a real task, and behind scoring either one against a behaviorally anchored rubric rather than a rater's impression. It is also why a self-rated personality trait loses much of its predictive value once an applicant has a motive to look good: a rating can be adjusted before it is given, but a recorded response to a real task cannot be un-performed after the fact. The six-domain check behind this site follows the same rule, scoring what someone actually does with a scenario instead of what they say about themselves.