REVISITED AND COMPARED WITH A LATENT TRAIT APPROACH Wim

Macready and Dayton (1977) introduced two probabilistic models for mastery assessment based on an idealistic allor-none conception of mastery. Although these models are in statistical respects complete, the question is whether they are a plausible rendering of what happens when an examinee responds to an item. First, a correction is proposed that takes account of the fact that a master who is not able to produce the right answer to an item may guess. The meaning of this correction and its consequences for estimating the model parameters are discussed. Second, Macready and Dayton's latent class models are confronted with the three-parameter logistic model extended with the conception of mastery as a region on a latent variable. It appears that from a latent trait theoretic point of view, the Macready and Dayton models assume item characteristic curves that have the unrealistic form of a step function This content downloaded from 130.89.45.231 on Tue, 22 Dec 2015 15:07:23 UTC All use subject to JSTOR Terms and Conditions