RASCHING AN ACHIEVEMENT TEST

In this paper I relate experiences gained in the process of moving data from Lertap, a classical test theory (CTT) system, to ConQuest and Winsteps, two item response theory (IRT) systems which have a particular focus on the Rasch model. The work reported below involves matters of dimensionality, indices of model fit, measurement precision, and scaling. To a considerable extent this research was prompted by comments brought back by the Dean of a large engineering faculty after she had attended a recent conference. “We might have a look at Rasch scaling”, she wrote on a memo distributed to staff. The memo quoted from a conference paper: “Test scores are mere enumeration, a tally of number of items correct. They are not measures of proficiency. Rasch scales are objective fundamental measures expressed on an interval scale”. Of course test scores have a bit more going for them than simple enumeration. If on a sixty-item test, scored on a number-correct basis, Jorge has a score of 60, Milagros a score of 45, and Jaime a score of 30, we can say that Jorge has answered twice as many items correctly as Jaime. If we imagine a horizontal scale, with number-correct scor