6.5 Partial Identifiability of Attribute Profiles

When a model is said to be identifiable, all model parameters need to be identifiable in CDMs, however, researchers are sometimes more interested in classifications.

Here we consider the partial identifiability of population parameters when item parameters are known.

A possible application is that when item in a test bank have been calibrated but used to assemble new test forms.

It is important to note that even when the CDMs are identifiable, parameter estimation still depends on the data. For instance, we need to be careful about sample size, balance of the attribute distribution, misspecifications of the dimensionality of the attribute profile.