Glossary · Assessment Science
Cut Score
A cut score is the minimum score a test-taker must achieve to pass an assessment, advance to the next selection stage, or meet a defined competency standard. It transforms a continuous score distribution into a binary outcome — pass or fail, advance or screen out — and is one of the most consequential, and most frequently arbitrary, decisions in a selection process.
Why it matters in hiring and assessment.
The cut score is where abstract psychometrics meets concrete human outcomes. Set it too high, and qualified candidates are rejected; too low, and the test provides no useful screen. Set it arbitrarily — by rounding to 60 %, or because "that felt right," or because an HR manager wanted a specific shortlist size — and it is both indefensible under challenge and likely to produce adverse impact patterns that were never intended.
There are two broad categories of cut-score setting:
- Norm-referenced cut scores: "Pass the top 30 % of applicants." The threshold floats with the quality of the applicant pool. Practical for competitive selection when you need a shortlist of a fixed size. Problematic when pool quality varies across hiring cohorts — a low-quality pool in a slow market means weaker candidates advance at the same "percentile" threshold.
- Criterion-referenced cut scores: "Pass all candidates whose score corresponds to a predicted performance rating above a defined standard." Requires criterion validity data — the correlation between scores and performance outcomes — to ground the threshold in something real. Methods include the Angoff method (subject matter experts estimate the probability that a "minimally competent" worker answers each item correctly), the Bookmark method, and contrasting-groups analysis. These are more work to establish but far more defensible under legal or regulatory challenge.
The measurement error around a cut score — quantified by the standard error of measurement derived from the test's reliability coefficient — means that candidates near the threshold cannot be confidently classified. A defensible process acknowledges this uncertainty with a "borderline band" and may apply additional review for candidates in that zone rather than treating the cut as a hard binary.
Cut-score decisions should be documented at the time they are made, with the rationale recorded, before any candidates are assessed. Post-hoc cut-score adjustment to achieve a desired pass rate or shortlist size — even if it looks innocuous — is a significant legal risk.
Example.
A company uses an Angoff standard-setting panel of five experienced Java developers to set the cut score for a Java assessment. Each panellist estimates the probability that a minimally competent developer — someone they would hire but not consider a star performer — would answer each of the 40 items correctly. The mean of all panellist estimates across all items is 0.62, giving a cut score of 0.62 × 40 = 24.8, rounded to 25. Candidates scoring 25 or above advance. This score is documented, tied to a subject-matter-expert judgement of job-relevant minimum competency, and can be defended if challenged.
Related terms.
- Criterion Validity
Criterion validity evidence allows cut scores to be anchored to predicted performance levels rather than arbitrary round numbers.
- Adverse Impact
The cut score level directly determines pass rates and therefore affects whether disparate impact thresholds are triggered for protected groups.
- Reliability Coefficient
The standard error of measurement derived from the reliability coefficient determines the uncertainty band around the cut score — low reliability means more candidates are misclassified near the threshold.
- Norm-Referenced Scoring
One approach to cut-score setting uses the norm distribution to define a percentile threshold — contrasted with criterion-referenced methods that anchor to job performance.