The direct effects of gender on AUD criteria in the sample as a whole were tested within the item-response model by regressing each criterion on the alcohol severity factor and gender simultaneously. A significant effect of gender on a criterion beyond that accounted for by underlying mean differences in the alcohol severity factor indicates that the criterion has different measurement characteristics in men and women, or measurement non-equivalence, and that the data should be grouped separately by gender. Similar tests were performed to evaluate direct effects of number of DUIs on AUD criteria using dummy variables representing individuals with 1, 2, and 3 or more DUIs. Based on the results of these tests, a 4-group model, with groups defined by DUI status (yes/no) and gender, was used to test the residual association of each of the 9 criteria with DUI status. In these models, factor means were set to zero for women without DUIs and were freely estimated in the other groups. As there was evidence for unequal variances (X2(4) = 128.54, p < .0001), variances were freely estimated across
