Remaining differences in item trends were less stark but still may suggest possible multidimensionality or DIF, which we explored further using a nonlinear exploratory factor analysis model to account for the binary nature of the items. Because the factor analysis procedures we used assume that observations are independent of one another, we performed our analyses on a cross-sectional calibration sample drawn by randomly selecting one repeated observation per participant from the longitudinal data.4 Initial exploratory factor analyses estimated within and across studies suggested that the item set was indeed multidimensional. Factors generally consistent with anxiety and depression emerged; item plots grouped by these two factors also displayed greater homogeneity (i.e., anxiety items showed more similar patterns to one another than to depression items), corroborating a multidimensional structure. Although we could choose to develop separate commensurate measures for both anxiety and depression, here we focus our efforts on developing a commensurate measure of depression from the 17 items that most consistently loaded on the depression factor.
