There are several methods available for deriving conditional standard errors of measurement, that is, standard errors of measurement at different score levels. Thorndike (1951) suggested that one could take results from all test takers having a given observed score, compute an estimate of the measurement error variance for each respondent by partitioning the respondent’s item responses into parallel half tests, computing the variance between the halves, and then averaging the result over all respondents at the given score level.