The data were analyzed in a two stage structural equation modeling (SEM) process (Anderson & Gerbing, 1988). First, a measurement model was assessed using confirmatory factor analysis (CFA) which specified all of the survey items loading on their respective latent constructs. As there were some items with non-normal distributions, all SEM analysis utilized maximum likelihood estimation with robust standard errors (MLR), which addresses the potential violation of multivariate normality in such circumstances (Muthén & Muthén, 1998-2012). To assess the convergent validity of the seven constructs, we assessed both the construct reliability (CR) and average variance extracted (AVE) levels as outlined by Fornell and Larcker (1981). All the constructs had CR and AVE levels above .7 and .5, respectively, with only one exception. Contrary to previous findings, harmonious passion indicated inadequate convergent validity (CR = .64; AVE = .38) and was therefore removed from the analysis (Fornell & Larcker, 1981). Another measurement model was then assessed which specified the remaining six constructs and their associated items. All factor loadings were significant and above the recommend .5 level (Hair, Black, Babin, Anderson, & Tatham, 2006; Hulland, 1999), and the CFA indicated acceptable fit with the data (x2 = 842, df = 215, CFI = .93, TLI = .92, RMSEA = .06, SRMR = .05). With only one exception (AVE = .49 for affective commitment), all the latent constructs met the recommended levels of CR and AVE (Fornell & Larcker, 1981). As AVE is a conservative measure which is often slightly below .5 (Jiang, Klein, & Carr, 2002), this lone exception was deemed acceptable for the current analysis. The CR, AVE, and Cronbach alpha values for the constructs are provided with the correlation matrix in Table 1.