Luppo Kuillman

Chapter 2 34 Multivariate analysis To assess the structural validity of the MSQ, we performed factor analyses and calculations of reliability estimates, as explained below. Model fit through Confirmatory Factor Analysis. We used the following goodness- of-fit indices to determine model fit using CFA: a) chi-square/degrees of freedom (χ2/ df ), b) root mean square error of approximation (RMSEA), c) standardized root mean square residual (SRMR), d) comparative fit index (CFI), and e) a goodness-of-fit index (GFI). The χ 2/df with a ratio between 0 and 2 is indicative of a good fit (Schermelleh- Engel, Moosbrugger, & Müller, 2003; Tabachnick & Fidell, 2007). For the RMSEA, a cut-off value less than or close to 0.06 was assumed to be appropriate (Hu & Bentler, 1999). The lower limit of the confidence interval (CI) should be close to 0, and the upper limit should not exceed 0.08. We also report the SRMR, as its standardized nature makes it easier to interpret. Values for the SRMR ranged from zero to 1.0, with good-fitting models having an acceptable threshold of less than 0.08. (Hu & Bentler, 1999) For the CFI, values equal to or greater than 0.95 are deemed indicative of a good model fit (Hu & Bentler, 1999). For the GFI, cut-off values greater than or equal to 0.95 are recommended for relatively low factor loadings and sample sizes (Miles & Shevlin, 2007). The Akaike information criterion (AIC) was used to compare different models. This criterion is a descriptive measurement, in which the preferred model is the one with the lowest value (Akaike, 1974). Internal consistency Cronbach’s alpha values were calculated to examine the reliability of all scales. In general, values equal to or greater than ≥0.70 are considered sufficient (Bernstein & Nunnally, 1994). Convergent and divergent validity Convergent validity refers to the extent to which a construct measures what it is purported to measure (Polit & Beck, 2004; Streiner, Norman, & Cairney, 2015). It is assessed according to data showing that different measurements of conceptually related dimensions of moral behavior are conceptually associated in the hypothesized direction. In this study, convergent validity was imputed according to statistically significant associations (linear associations between measurements of moral behavior), while divergent validity was assumed when there was no correlation (i.e., P > 0.05). The degree of overlap between constructs was estimated by calculating the nonparametric effect size of Rho (given the sample size). The statistically significant