ABSTRACT
Much of psychometric testing theory has followed in the tradition of Lord and Novick (1974). The initial motivation for the research reported in this chapter was to analyze the performance of a stochastic knowledge structure model (Falmagne, 1989) on data previously analyzed with the normal ogive model. The normal ogive model is unidimensional along one latent ability, and is intended to model performance on items that measure complex, continuously distributed abilities. The knowledge structure model, like other latent class models, assumes discrete dichotomously measured abilities and is better suited to model items testing curricular mastery, rather than general ability (see Molenaar, 1981). The analysis presented in this chapter demonstrates that even though the above is generally true, the performance of the stochastic knowledge structure model is comparable to that of the normal ogive model even on items measuring continuously distributed abilities.
