Of the five CFA models tested, the models that did not adjust for method effects (models 1, 2 and 3) were worse fitting. Models with a TLI and CFI greater than approximately 0.95 and an RMSEA close to 0.06 are generally considered acceptable (Hu and Bentler, 1999). Models 1 and 2, which were consistent with the summed-scale approaches to using the PDD, did not fit the data well (CFI, TLI, RMSEA all under the criteria for good fit, < .95 for CFI, TLI and > .06 for RMSEA). Model 3, which tested for a latent positive and negative dimension of stigma, fit somewhat better than Models 1 and 2 but still fell short of optimal model fit criteria. Models 4 and 5, which adjusted for method effects, had good fit. The AIC and BIC statistics generated using MLR estimation were also lower in Models 4 and 5 compared to the other models, providing additional evidence of better fit when adjusting for parsimony. Nested model comparisons indicated that model 5, which included two substantive factors of perceived discrimination and perceived devaluation,
