A comparison method of m distributions in item response theory
Hideki Toyoda · Japanese Psychological Research · 1991
The purpose of this study is to propose a comparison method for comparing m normal latent distributions in IRT (Item Response Theory). The input data is the sets of the item response patterns to the set of items whose item parameters are known. The process of this method is as follows. First, some likelihood models whose parameters are equal or different between the populations, are constructed. Next, the best model among them is selected by the minimum AIC (Akaike's information criterion). Lastly, equality of latent trait populations are judged. Through the result of computer simulations in case of two samples, this method is shown to be effective for comparing the latent populations.