Models 1 and 2 sought to answer the question, do the data support perceived devaluation-discrimination as a single factor (model 1), or as two factors (model 2)? Model 3 attempted to determine the plausibility of positive and negative dimensions of perceived stigma as proposed by others (Interian et al. 2010). Finally, motivated by evidence from prior factor analytic studies that extracted separate “method factors” to adjust for positively- and negatively-worded items (Marsh et al. 2010), we specified Models 4 (a single PDD factor) and 5 (two PDD factors) as analogs of Models 1 and 2, while statistically adjusting for method effects. We accomplished this by specifying the positively worded items to load on a latent method factor (LMF), which ‘parses’ the item wording effects from the ‘true’ factor structure underlying the items (Marsh et al. 2010). As depicted in Figure 1, the LMF was specified to be uncorrelated with the latent factors.
