An examination of psychometric bias due to retesting on cognitive ability tests in selection settings.

941 real candidates for the Belgian medical and dental school entrance examination

Real high-stakes admissions candidates who sat the entrance examination more than once; young adults, with a real criterion (subsequent academic performance) available

The strongest design in the score-versus-skill cluster and the reason it is graded B. It is not a laboratory manipulation and not a self-report survey: these are actual candidates with a real consequence riding on the test, and the outcome is not another test score but subsequent grade point average - an external criterion the test was built to predict. That combination lets it ask the question every other coaching study has to duck: does the gain predict anything? Limitation: observational (candidates choose whether to retake), single national selection system, and unreplicated in this exact form.

Retesting introduced BOTH measurement bias and predictive bias. Retest scores loaded more heavily on memory and less on g than first-attempt scores - the second sitting is measuring something different from the first. And the decisive result: the retest gain ELIMINATED the test's criterion-related validity, because the gain did not predict later academic performance. A candidate who improved by retaking did not thereby become a better bet. This is the cleanest available answer to "does a raised score mean raised ability", obtained on real candidates with real stakes and a real criterion, and the answer is no.

Low for the central claim. Whatever selects candidates into retaking, it cannot explain why the gain fails to predict the criterion - a real ability gain would predict it regardless of who chose to retake.

No direct replication of the criterion-validity result on another selection system was identified. It converges with te Nijenhuis et al. (2007) from a completely different methodological direction, which is the strongest form of corroboration available here.