data, there was no notable difference in fit between models that did and did not adjust for measurement non-invariance. This suggests that model fit was driven by variation in the data (e.g., current FTND had poorer model fit than lifetime FTND, studies differed in their inclusion criteria) and not by measurement non-invariance. Finally, variation associated with birth cohort could be due to the participants’ age at assessment. However, this concern is reduced by broad variation in the median birth year, even among studies with similar recruitment periods (Table S1)1. Post hoc analyses evaluating model fit while controlling for study sample, sex, and race further support that the variability across studies may be a primary driver of poor fit, as model fit improved modestly (RMSEA = 0.090 vs. 0.108 and CFI= 0.595 vs. 0.478 for the unadjusted vs. covariate-adjusted models). Moreover, model fit was good for regression models that included latent variable modeling of FTND and covariates.
