We utilized multigroup IRT to model the effect of age, sex, and birth order on endorsement rates for each substance/class, which are shown in Table 1. For the sake of completeness, models representing all possible combinations of constraining either locations and/or discriminations equal across ages, sexes, and/or twins are reported. The best fitting model (see Table 2) was the original unconstrained model (Model 1b) in which substance locations and discriminations were estimated independently for each age and sex group but were constrained equal across twins (i.e., there were no birth-order effects). The good fit indices resulting from this model (CFI=0.99, TLI=0.99, RMSEA=0.068) support the use of a single-factor model. This resulted in six (3 ages × 2 sexes) sets of substance IRT parameters (locations and discriminations) being estimated.
