Open AccessTHE STABILITY OF EQUATED TEST SCORES 1
Author(s) -
Karon Bertram P.
Publication year - 1954
Publication title -
ets research bulletin series
Language(s) - English
Resource type - Journals
eISSN - 2333-8504
pISSN - 0424-6144
DOI - 10.1002/j.2333-8504.1954.tb00252.x
Subject(s) - equating , percentile , statistics , mathematics , stratified sampling , sampling (signal processing) , test (biology) , population mean , standard error , computer science , biology , paleontology , filter (signal processing) , estimator , rasch model , computer vision
The effects of sampling error were studied empirically with respect to four methods of equating scales of tests administered to non‐overlapping groups of subjects: 1. mean and sigma method, 2. equi‐percentile method, 3. maximum likelihood method using an “anchor” test, 4. standard reference group method using an “anchor” test. The methods were compared under both random and stratified sampling. Results showed that sampling error was 1) smaller for those methods which make use of an anchor test than for those which do not, 2) smaller for equated scores closer to the mean of the total population than for those further from the mean, 3) was not decreased by stratification by institution when the “anchor” test methods of equating were used.