An Old Problem with a New Solution, Raising Classical Questions: A Commentary on Humphry

摘要

Researchers applying the Rasch model are frequently confronted with a problem, namely, that its core requirement-non-intersecting item response functions (i.e., uniform item discrimination parameters)-is violated by empirical data. Consequently, Birnbaum (1968) formulated a more general, two-parameter item response model (2PL model), allowing for variations in item discrimination parameters and creating a more flexible model that could provide more adequate data description. However, as the 2PL model lacks the scientifically required property of specific objectivity, it is not a desirable measurement model (cf. Scheiblechner, 2009, for a detailed account). Moreover, by introducing an item discrimination parameter for each item, the 2PL model gains flexibility while losing testability in comparison to the Rasch model. This means that Rasch's claim to "general laws and the meaning of measurement in psychology" (Rasch, 1961) has been converted into a method of data description that is not tied to the scientific requirement of invariance.