Figure 1 depicts a unidimensional MIMIC-interaction model.1 A continuous response process is
assumed to underlie each categorical observed item response, and thresholds (ts) on the response continuum determine whether a person responds in one category or the next (Figure 1 shows the
case of binary items; generally, the number of thresholds is one less than the number of
responses). The MIMIC model is parameterized as an IRT model, and EM MML is used here
to fit the model to the item data directly. The latent scale is identified by fixing the mean and
variance of y to 0 and 1, respectively, and the grouping variable is dummy (0-1) coded