Recruitment is one of the most expensive decisions organisations make repeatedly, and one of the least measured. Applying basic statistics to your own hiring process usually reveals that several stages predict nothing - and that the ones everybody trusts most are the weakest.
What the research consistently shows
Decades of validity research in occupational psychology point the same way. Structured methods - work sample tests, structured interviews with consistent questions and scoring, cognitive and job-knowledge assessments - predict job performance substantially better than unstructured methods. The traditional unstructured interview, conducted differently with each candidate and scored on impression, performs notably poorly, despite being the method most hiring managers trust most.
Years of experience and educational attainment, taken alone, are weaker predictors than most organisations assume.
The practical implication: structure is the intervention. Same questions, same order, agreed scoring criteria, scores recorded before discussion.
How to measure your own process
You do not need published research - you have your own data:
- Record scores at each stage for every candidate, numerically
- Record performance ratings at 6 and 12 months for those hired
- Correlate them. Which stages predict later performance and which are noise?
Most organisations that do this discover one or two stages carry nearly all the predictive power, and that a favourite stage - often a culture-fit chat - correlates with nothing except the interviewer's preferences.
The catch to be aware of: you only observe performance for people you hired, which is a textbook case of survivorship bias. The correlation you measure is therefore understated. It is still far more informative than not measuring.
Statistics as a guard against bias
Unstructured judgement is where bias enters most easily, because it is unrecorded and unaccountable. Statistical monitoring is how you find it:
- Track progression rates by stage and by demographic group. A stage where one group's pass rate is markedly lower deserves investigation.
- Watch out for base rate effects. Small applicant numbers in a group produce wildly variable rates - do not over-interpret a difference in a handful of candidates.
- Check for proxies. A criterion that appears neutral (postcode, continuous employment, a specific institution) can act as a proxy for protected characteristics.
Note that UK employers have legal obligations here under the Equality Act; measurement supports compliance but does not replace proper HR and legal advice.
The metrics worth tracking
- Applications per hire, and pass rate at each stage
- Time to hire, and offer acceptance rate
- Quality of hire - performance at 6 and 12 months
- First-year attrition, which is where bad selection shows up most expensively
- Cost per hire by source, and quality by source (cheap sources often produce poor retention)
That combination tells you whether to fix your attraction, your selection or your onboarding - three very different problems that all present as "our hiring isn't working".
Where judgement remains essential
Statistics narrows the field and removes noise; it does not choose colleagues. The aim is to stop wasting predictive information you already have, not to reduce hiring to a formula.
Building the capability
The Statistics for Business course at London School of Business UK covers correlation, bias and measurement design that applies directly to people analytics, alongside our HR-focused courses such as Develop & Implement Recruitment Strategies. Enquire today.