Do Reference Checks Actually Work?

Do reference checks actually work, or is that final phone call mostly theater at the end of a hiring process? A 2016 update to decades of personnel-selection research puts a number on it, and the legal rules that shaped that number have only tightened since the studies behind it were run.

What a reference check actually measures, according to the research

Frank Schmidt, In-Sue Oh, and Jonathan Shaffer's working paper, an update to Schmidt and Hunter's landmark 1998 meta-analysis, reports a validity of 0.26 for reference checks predicting job performance, and an older estimate of 0.23 for predicting performance in training, drawn from Hunter and Hunter's original 1984 meta-analysis, cited within the same paper. Both figures sit well below the corrected validity of a structured interview (0.42) or a work sample test (0.33), the two methods that top the field's 2022 recalculation of selection-method validity.

The gap makes sense once you look at what a reference actually is: one person's memory of how someone performed, delivered secondhand, with no shared rubric and no way for a hiring manager to check it against anyone else's account of the same work.

Why that number is more optimistic than today's reference call

The 0.26 figure is probably too generous for a reference check placed today. The same working paper explains that the underlying studies were run before employers grew wary of being sued over a bad reference: starting in the 1970s and 1980s, former employees began suing past employers over negative references, and by the turn of the millennium, 36 states had passed laws granting employers immunity for a good faith reference.

Immunity laws were supposed to bring detailed references back. The paper's own reading of the evidence is that they mostly have not: most former employers are still willing to confirm only dates of employment and a job title, not a disciplinary record, the reason someone left, or how they actually performed, the exact content a validity study needs in order to measure anything. A reference check run under that policy is not testing the same thing the 0.26 figure was built from.

A reference check is not the same as an employment check

It helps to separate two things hiring processes often blur together. An employment check confirms a claim already on the resume: that a role existed, the dates line up, the title matches. A 2026 survey of 1,500 US employers found only 39.0% verify previous employment for every hire, so even that narrower check does not run consistently. A reference check goes further and asks a former colleague to vouch for how someone actually performed, exactly the part today's legal climate leaves most former employers unwilling to discuss.

That leaves a gap in the middle of most hiring processes: a claim about the past that nobody is fully checking, sitting between a resume the candidate wrote and a job the candidate has not started yet.

Do reference checks actually work in practice?

They work as a weak, final sanity check, not as proof. A structured interview or a scored work sample test asks a candidate to do something in front of an evaluator using the same rubric applied to everyone else, the same logic behind a documented, checkable proof of skill instead of a claim someone else has to vouch for secondhand. A work sample test does not depend on whether a past employer's legal team allows a former manager to say anything beyond a job title. Neither does a check an employer can actually verify directly, and neither does a scored assessment run under the same conditions for every candidate.

None of this means a reference call is worthless. It can still catch an outright fabrication, confirm a role existed, or surface something a candidate would rather not mention. What it cannot do, at the validity the research actually finds, is carry the weight a lot of hiring processes still put on it.