Most hiring decisions lean on a CV, a quick chat, and a feeling. Decades of research in industrial-organisational psychology suggest those signals are weak on their own. Some of the things recruiters trust most, like years of experience or a confident interview, predict how someone will do the job poorly.
This post walks through what the evidence says works and what does not. It is honest about an important point: the validity numbers many people quote came from a 1998 study, and those exact figures were revised downward in 2022. The rankings still hold in broad strokes, but the precise coefficients changed.
Key takeaways
- Structured, work-relevant methods predict job performance better than gut-feel signals.
- After the 2022 re-analysis, structured interviews rank as the highest-validity single method.
- Cognitive ability still matters but was moved down the rankings and revised lower than older research claimed.
- Years of experience, education amount, age, interests, and graphology are weak predictors.
- Combining methods beats relying on any single one, including the CV.
Why it matters (and a caveat)
Picking the wrong predictor wastes money and time. If you hire on years of experience or a relaxed conversation, you are using signals that research shows predict performance poorly. Structured, work-relevant methods give you a better read on whether someone can do the job. That is the difference between a hire that works out and one that does not.
There is an honest caveat. The Schmidt and Hunter (1998) meta-analysis summarised 85 years of research and reported high validity figures, for example general mental ability at roughly r = .51. In 2022, Sackett and colleagues found that earlier studies had systematically over-corrected for range restriction, which inflated many of those numbers. They concluded the true relationships are lower than once thought. Most methods that ranked high still rank high, but you should treat the old exact figures as outdated and focus on the relative order.
What the evidence says
Structured interviews now lead among single methods
Standardised questions and consistent scoring reduce interviewer bias and inconsistency. After the 2022 re-analysis, structured interviews emerged as the highest-validity single method, reported at about r = .42. They outperform unstructured interviews clearly.
Job knowledge tests are strong and practical
In the revised rankings, job knowledge tests came in just behind structured interviews, at about r = .40. They measure whether a candidate knows what the role requires, which maps closely to real work.
Work sample tests still help, but lower than older claims
Work samples ask candidates to do a slice of the actual job. They remain a good predictor, but the 2022 revision lowered their estimated validity because earlier range-restriction corrections had inflated the figure.
Cognitive ability still matters, moved down
Older research put cognitive ability near the top, around r = .51. The 2022 work revised its operational validity down to roughly .31, moving it to around fifth among single methods. It is still a real predictor, just not the clear leader it was once described as.
Combining methods beats any single one
Schmidt and Hunter found the strongest results came from pairing general mental ability with another structured method, for example a work sample, an integrity test, or a structured interview. The exact combination scores were later revised downward, but the principle stands: combine predictors rather than relying on one.
Unstructured interviews are weak and biased
A free-form chat is vulnerable to subjective error and inconsistency. Unstructured interviews show notably lower validity than structured ones, and the 2022 revision lowered their estimate further.
Experience, age, interests, and graphology predict poorly
Years of job experience, amount of education, age, and interests predict performance poorly. Graphology (handwriting analysis) has essentially zero validity. Do not weight these as if they tell you who will perform.
How to apply it in hiring
Step 1: Make your interviews structured
Write a fixed set of job-relevant questions and ask every candidate the same ones. Score answers against a defined rubric. This single change moves you from one of the weakest methods to the highest-validity single method in the research.
Step 2: Add a work sample or job knowledge check
Give candidates a small, realistic task or a test of the knowledge the role needs. This measures ability to do the work directly rather than a proxy for it. It also pairs well with a structured interview.
Step 3: Combine methods instead of trusting one signal
The evidence favours combining predictors, such as a structured interview plus a work sample or a cognitive measure. No single method, including the CV, should carry the whole decision.
Step 4: Stop over-weighting experience and gut feel
Do not rank candidates mainly on years of experience, education amount, age, or how well a casual conversation went. These are weak predictors. Skip graphology entirely, since it has essentially zero validity.
Step 5: Treat old validity numbers with care
If you cite research figures, know that the 1998 numbers were revised downward in 2022. Use the rankings to guide method choice, and avoid presenting the old exact coefficients as current fact.
Do this
- Use structured interviews with fixed questions and a scoring rubric.
- Add a work sample or job knowledge test tied to the real role.
- Combine two or more predictors rather than relying on one.
- Define what good looks like before you start scoring candidates.
- Present validity evidence as rankings, not as exact undisputed numbers.
- Ignore graphology and treat years of experience as a weak signal.
Common mistakes to avoid
Quoting the old numbers as current fact
Many articles still cite Schmidt and Hunter's 1998 figures, like cognitive ability at .51, as if they are settled. Sackett et al. (2022) showed those were inflated by over-correction for range restriction. Use the direction, not the dated coefficients.
Relying on the unstructured chat
A relaxed, free-form interview feels informative but predicts poorly and invites bias. Without standard questions and scoring, you are measuring rapport, not job ability.
Treating years of experience as proof of skill
Experience length predicts performance poorly. Two candidates with the same years can perform very differently. Test what they can do instead of counting their tenure.
Assuming cognitive ability is the single best predictor
It was described that way for years. The 2022 revision moved it down to around fifth among single methods. It still matters, but it is no longer the clear top signal.
Using graphology or other zero-validity gimmicks
Handwriting analysis has essentially no validity. Any method without evidence behind it adds noise and risk, not insight.
Frequently asked questions
What is the single best predictor of job performance?
After the 2022 re-analysis by Sackett and colleagues, structured interviews emerged as the highest-validity single method, reported at about r = .42. Job knowledge tests followed closely at about .40. This reshuffled older rankings, which had placed cognitive ability at the top. These are still estimates, and combining methods predicts better than any single one.
Did cognitive ability stop being a good predictor?
No. Cognitive ability is still a real predictor of job performance. The 2022 research revised its operational validity down to roughly .31 and moved it to around fifth among single methods. Older studies reported it near .51, but that figure was inflated by over-correction for range restriction. It matters, just less than once claimed.
Why were the validity numbers revised in 2022?
Sackett, Zhang, Berry, and Lievens (2022) found that earlier meta-analyses had systematically over-corrected for range restriction. That over-correction inflated the criterion-related validity of many selection methods. After addressing it, the true predictor-criterion relationships came out considerably lower. Most methods that ranked high still rank high, but the exact numbers dropped.
Is years of experience a good way to screen candidates?
No. Research shows years of job experience predict job performance poorly. The same is true for amount of education, age, and interests. These feel meaningful but tell you little about future performance. Test relevant skills and knowledge directly with structured methods instead of ranking people by tenure.
Does graphology or handwriting analysis work for hiring?
No. Graphology has essentially zero validity as a predictor of job performance. It is one of the clearest examples of a popular method with no evidence behind it. Using it adds noise to your decisions without adding insight. Stick to structured, work-relevant methods that research supports.
Should I rely on the CV to predict performance?
The CV is a weak predictor on its own. It is useful for screening and for understanding background, but it should not carry the hiring decision. The stronger predictors come later in the process: structured interviews, work samples, and job knowledge tests. Combine methods rather than trusting any single signal, including the resume.
The bottom line
The pattern in the evidence is clear in direction, even where the exact numbers shifted. Structured, work-relevant methods predict job performance well. Structured interviews, job knowledge tests, work samples, and cognitive ability all carry real signal, and they work best in combination. The weak predictors are the familiar ones: the casual interview, years of experience, age, interests, and graphology.
Be honest about the 2022 revision when you talk about this. The 1998 figures were inflated and were corrected downward, so lead with rankings, not precise coefficients. A clean, well-formatted CV helps you read a candidate quickly and decide who to advance, which is where a tool like RefineCV helps. The stronger predictors come later in the process, once you put structured methods to work.
Read every candidate clearly
The strong predictors come later in the process, but a clean CV helps you decide who to advance. RefineCV formats every candidate CV into your branded template. Try it free with 10 CVs, no credit card.
Related reading: recruitment vs talent acquisition and CV parsing explained.
Sources
- Schmidt & Hunter (1998), Psychological Bulletin 124(2), 262-274 (1998): The Schmidt & Hunter (1998) meta-analysis, summarising 85 years of research, reported general mental ability as among the most valid single predictors (about r = .51), with the strongest results from combining GMA with a structured method. These point estimates were later revised downward and should not be presented as current undisputed values.
- Sackett et al. (2022), Journal of Applied Psychology 107(11), 2040-2068 (PubMed) (2022): Sackett, Zhang, Berry & Lievens (2022) found prior studies systematically over-corrected for range restriction, overestimating criterion-related validity, and concluded true relationships are considerably lower, though most high-ranking procedures still rank high after correction.
- SIOP, TIP article summarising Sackett et al. (2022) (2022): After Sackett et al. (2022), structured interviews emerged as the highest-validity single method (about r = .42), followed by job knowledge tests (about .40), with cognitive ability revised down to roughly .31 and moved to around fifth among single methods.
- I/O at Work, The Best Employee Selection Methods According to Research (2020): Structured, work-relevant methods predict job performance reliably, whereas years of experience, education amount, age, interests, and graphology predict it poorly, with graphology having essentially zero validity.
- eSkill, The Best and Worst Predictors of Job Performance (2023): Structured interviews consistently outperform unstructured interviews because standardised questions and scoring reduce bias and inconsistency; unstructured interviews show notably lower validity.