Comparison of dichotomous and polytomous response models

In some experimental situations, it is feasible to classify the outcome of the experiment into more than two ordered categories. To compensate for the extra effort required in the more refined classification, it is important to know whether it will lead to more efficient estimates than the binary classification. It is shown that the Fisher information for the common parameters in the more elaborate model is greater, which is further supported in a numerical study. Based on a simulation study, a similar conclusion is reached regarding the extent to which such gains are achieved in small sample situations. Some empirical results are given theoretical justification. © 1986.