Assumption TestingThe data were tested for violations of the structural equation modelling assumptions of linearity and multivariate normality (Kline, 2005). There were no obviouscurvilinear trends in the data. There was, however, a violation of multivariatenormality. This was addressed by deriving the fit statistics for the saturatedand mediator models from the Satorra–Bentler chi-square, which adjusts for thenon-normality (Jöreskog, 2005; Jöreskog & Sörbom, 2004).It is assumed that the BOT-2 SF, the SSRS-T and the SDQ ESS-Teach load on a singlefactor. In order to confirm this assumption, a confirmatory factor analysis using robustmaximum likelihood was conducted on each of the three instruments. Robustmaximum likelihood estimation was used to account for violations of multivariatenormality. The results indicated that a one-factor solution provided a good fit for allthree instruments. These results are consistent with our proposed measurement modelin which each latent variable ‘drives’ a single indicator (Figure 1).