Conditional Standard Error of Measurement

The ordinary (unconditional) SEM is a single number summarizing average measurement error across the whole test, assuming constant precision. The conditional SEM recognizes that precision varies by score level and reports the error that applies at each point on the scale. Because tests usually measure unevenly — more precisely in some ranges than others — the conditional SEM gives a truer picture for an individual examinee, especially at the extremes or near a decision threshold where the average can be quite misleading.

Because the amount of information a test provides depends on where an examinee falls relative to the items. On the number-correct metric, scores near the floor and ceiling have restricted variability. Under IRT, the test information function peaks where items are concentrated and tapers where items are sparse, so the conditional standard error is small near well-targeted regions and large elsewhere. A test built with most items of middling difficulty measures middling examinees precisely but high and low performers poorly, and the CSEM makes this explicit.

Because the consequence of a test often hinges on a single threshold — pass or fail, proficient or not — and the relevant question is how precisely the test measures right there, not on average. If the conditional standard error is large at the cut score, many borderline examinees could be misclassified, and that uncertainty should inform the decision and its reporting. The Standards therefore emphasize reporting measurement error conditional on score level, particularly at cut scores, so classification accuracy can be evaluated honestly.