How to Test Skills Without Rebuilding Bias

Skills-first assessment is the subject here, and the practical question is how to replace a broad credential proxy without creating a new opaque gate. The reader can judge each method by four tests: does it represent the work, is it applied consistently, can people access it on fair terms, and does it give managers evidence they will actually use? No single tool wins on every test, which is why inclusion and assessment have to be designed together.

Direct evidence is better only when it is relevant evidence

The appeal of a work sample is easy to understand. Instead of inferring ability from a degree or years in a similar job, the employer asks a candidate to perform an activity related to the role. A coding exercise can reveal how someone approaches debugging. A project simulation can show whether they can find relevant information, organise tasks and communicate a plan. A structured interview can ask every candidate comparable technical, behavioural or situational questions. These methods move the decision closer to observable performance.

That does not make them automatically superior. A measure can be direct and still measure the wrong thing. A timed test may reward speed when the job rewards careful judgment. A simulation may recreate an artificial environment rather than the conditions employees face. An interview may be called competency-based while allowing each manager to ask different questions and rely on different impressions. The crucial standard is not whether a tool carries the label “skills assessment,” but whether the evidence has a defensible relationship to competent performance in the role.

The OECD cites research in which general cognitive ability tests and work-sample tests had the highest average validity for predicting job performance among 19 recruitment methods, while years of education and years of similar-role experience had substantially lower predictive power. It also cites a later meta-analysis finding that pre-hire work experience is generally a poor predictor of future performance. These findings support looking more closely at what a candidate can do. They do not establish a universal ranking of hiring tools. Predictive validity in one context does not prove fairness, accessibility or job relevance in another.

In many roles, a defensible assessment is a bundle of evidence rather than a single score. A practical exercise can show task performance. A structured interview can probe reasoning, communication and the ability to handle situations the exercise did not cover. Relevant work or learning history can explain how the candidate acquired the capability. A qualification can add evidence of knowledge or satisfy a legal requirement. The employer’s job is to decide in advance which evidence answers which question, so that a candidate is not judged by an unannounced standard that changes from one interviewer to the next.

The choice involves a real trade-off between depth and scale. A carefully designed work sample may give rich information but take time to administer and evaluate. An online test can handle many applicants quickly, but the format may favour people with better equipment, more practice or greater familiarity with the testing environment.