Years of Experience Is the Weakest Requirement You Will Write

Does a years-of-experience minimum predict performance?

No, or close enough to no that the number cannot carry a filter. A meta-analysis of 81 independent samples put the correlation between prehire work experience and later job performance at .06, with training performance at .11 and turnover at .00, and experience with relevant tasks, jobs or occupations did no better, at about .07 ¹.

Three things about that estimate matter more than its size. It is a corrected correlation, so it is not near zero because somebody forgot to adjust for range restriction. It measures experience at the moment of hire, which is exactly what a minimum-years field asks about, and not tenure in a job somebody already holds, which is a different and more favorable literature. And the authors name their own exceptions: prehire experience predicts a little better when somebody first starts the job, and task-level experience does predict training performance, both from small subsets of studies ¹.

The most-quoted table in hiring gives experience a better number, and its provenance is the point. Schmidt and Hunter's 1998 summary put years of job experience at .18, near the bottom of a table led by .54 for work sample tests and .51 for cognitive ability ². That .18 rests on tenure among people already in the job, and the 2022 re-analysis re-conceptualised the predictor, moving the question to experience measured at the point of hire ³.

The same 1998 table also calculated what each screening field adds on top of a cognitive ability test: years of job experience raised the multiple correlation from .51 to .54, a gain of .03 or 6%, and years of education added .01, a gain of 2% ². The authors warned in a footnote that the education figure assumed no correlation between years of education and ability inside an applicant pool, and that any positive correlation pushes that increment toward .00 ².

Two caveats keep this honest. That increment assumes an employer already runs a cognitive ability test, which most do not. And the whole 1998 table was revised downward in 2022: all five of its top-ranked methods survive in the new ranking, but the mean across Schmidt and Hunter's top five was .49 and the mean across the 2022 top five is .37 ³. A requirement built on remembered numbers from the older version is working from figures the literature has since revised.

None of this touches licensure. Where a regulator or a statute sets a minimum, the evidence about prediction is beside the point, and that case gets its own section below.

What is the number actually doing in your posting?

Sorting on career shape. A minimum-years line rewards continuous employment, an early start in the field and a first job that happened to carry the right title, and it penalizes a career break, a late switch, and anyone whose relevant work sat under a title your industry does not use. None of that is capability, and all of it is legible to whoever wrote the field.

It used to do a second job, and that one has quietly stopped. When a person read every application, a stated minimum was a real deterrent and a real sorting device, because writing an application was expensive and most people did not bother with a role that had told them they were ineligible. An applicant working with a model now pays almost nothing to apply and almost nothing to phrase five years of adjacent work as five years of the work you named. The field still filters, on whether the applicant chose to satisfy it.

That is the same failure that took the keyword screen apart, and it has the same shape: a criterion everyone can meet on paper stops separating anyone. If the resume stage is where the years line does its work, the next questions are whether the resume screen is still throwing away the right people and what replaces a keyword filter once every resume matches.

One measured signal exists about what requirement text does to reader behavior, and it favors naming skills. LinkedIn's analysis of paid postings from employers running 100 or more jobs, March 2021 to March 2023, found that posts listing skills in the requirements section were correlated with an 11% higher view-to-apply rate than posts that did not ⁴. That is correlational, on one platform, from a company that sells skills-based hiring products, and a view-to-apply rate says nothing about whether the applicants were any good. Treat it as the weakest kind of support for a change worth making for stronger reasons.

Replace the number with the behavior it stands in for

Take the highest number in the posting and ask the hiring manager one question: which observed behavior does it stand in for? The answer is usually concrete once somebody has to say it out loud. Five years often means has carried a full annual cycle of this work without a supervisor checking it. Eight years often means has been the person the escalation lands on.

Then write that sentence into the posting and delete the number. Three rewrites, from lines you have probably shipped:

• Was: 5+ years of financial analysis experience. Now: Has owned a close or a forecast cycle end to end, including the month it went wrong.
• Was: 7+ years in enterprise sales. Now: Has carried a quota through a full renewal cycle and can walk through a deal that was lost late.
• Was: 3+ years working on data pipelines. Now: Has been on call for a pipeline in production and can describe a failure they were paged for.

Each rewrite does something the number could not: it names a thing an interviewer can ask about and a former manager can confirm. Asking about it well is its own skill, and reading a candidate's level off the decisions they owned is that conversation written out. A requirement that survives that test is a requirement, and the rest of the list deserves the same question, which is the whole of what makes a requirement testable.

If nobody can name the behavior, delete the line. A number whose meaning the hiring manager cannot state in one sentence was never a bar that removing it lowers, because it was never holding anything up.

Keep a number for two reasons, and name which one

Two reasons survive. The first is a statutory, licensure or contractual minimum, which is not yours to edit: where a regulator, an accrediting body or a customer contract sets the years, the line stays. The second is a real exposure threshold, meaning there is a cycle in this work that takes calendar time to complete once, and the job needs somebody who has completed it.

Both belong in the posting with the reason attached. A line reading five years, because the client contract requires five years, is a fact a reader can act on. A line reading 5+ years required with nothing after it tells the reader only that a number exists, and gives the hiring team nothing to apply consistently when a strong candidate arrives at four. What a licence, an accrediting body or a contract actually requires is a question for counsel on your own roles.

The exposure threshold is the one people get wrong, so write it as the cycle rather than as the clock. An auditor who has closed two year-ends has the exposure. An auditor who spent three years on one quarter's testing does not, and the years field cannot tell them apart. Write the cycle and let the count fall out of it.

Everything else moves out of the filter and into the evaluation, and that trade has a real cost worth saying out loud: you are exchanging a free field on an application form for reviewer minutes at a later stage. It is only worth making if the later stage exists. Deciding which requirements earn a stage is the same problem as telling a must-have from a nice-to-have, and it belongs in the conversation that opens the requisition.

See a sample report >