These initial effects of covariates focus on study, age and gender differences in the underlying factor (mean and variance) of depression. Now we return to the items themselves. To this point we have assumed, but have not yet demonstrated, that the items reflect the factor equivalently in all subgroups and across all studies. We identified potential sources of DIF based on our item plots due to participants’ age and study, previous literature on gender-related DIF for some items (e.g., Schaeffer, 1988; Steinberg & Thissen, 2006), and our sampling framework (our oversampled children of alcoholic parents might interpret or respond to some items differently). A clear advantage of the MNLFA framework is that it allows DIF testing across all of these covariates simultaneously, including continuous covariates like age. DIF testing is accomplished by allowing variables indexing these potential sources of DIF to moderate the intercept and factor loading for an item. Using a conservative alpha level of .01 to account for multiple testing, we detected DIF by study for six items, DIF by age for eight items, DIF by gender for
