Education Commons^™

Accounting For And Quantifying Dependencies In Dichotomous Test Data, Barry Sheridan

Research outputs pre 2011

This paper reports on a strategy employing the Extended Logistic Model of Rasch to quantify dependencies in dichotomous test data by providing standard errors of measurement which are commensurate with the loss in information arising from dependencies among subsets of items. Data collected for the Australian Scholastic Aptitude Test were analysed and shown to exhibit strong dependencies among the items involved. A technique is presented for estimating the number of equivalent independent items (found to be 74 in this study) to the original test number (100).