The analysis uses a newly developed method for exploratory structural equation modeling (ESEM), which integrates exploratory factor analysis (EFA) into structural equation modeling such that factor indicators load on all factors, avoiding the requirement of zero cross loadings in CFA (Asparouhov and Muthén, 2009; Eaton et al., 2012; Marsh et al., 2009, 2010). We opted for EFA over CFA because in a strict CFA, the presence of zero cross-loading does not always yield a well-fitting model, and misspecification of zero loadings in factor identification may yield distorted factors with overestimated factor correlations and subsequent distorted structural relations. The ESEM model with covariates estimates both the measurement and structural parts simultaneously (Asparouhov and Muthén, 2009). Specifically, the lifetime Axis I mental disorders were included as manifest variables for the 3 latent factors presumed to represent one externalizing and two internalizing (i.e., fear and distress) dimensions in the ESEM, and PDs were included as covariates in the structural model together with gender, age, race/ethnicity, and marital status. However, to conserve space, specific demographic effects are not shown, as these associations have been well documented in other NESARC studies (Grant et al., 2004a).
