Elien Neimeijer

30 Statistical analyses Construct validity of the GCI was examined by means of confirmatory factor analysis (CFA). We used the lavaan package (Rosseel, 2012) in the R environment (version 3.4.1; R Core Team, 2017). A multifactor model was specified in which each item loaded on only one factor. The fit of the model was examined using the Comparative Fit Index (CFI), the Tucker–Lewis Index (TLI), Root Mean-Square Error of Approximation (RMSEA), and the Standardised Root Mean Square Residual (SRMR). For a good-fitting model, cut-off values of CFI > 0.90, TLI > 0.90, RMSEA < 0.05, and SRMR < .08 are required (Hu & Bentler, 1999; Kline, 2005). We used the robust MLR maximum likelihood estimation procedure to account for non-normality. A non-significant Chi-Square indicates exact model fit, a ratio between the χ2 statistic and the degrees of freedom (df ) lower than 2.5 indicates a close fit to the data (Hu & Bentler, 1999). A modification index, giving the expected drop in chi-square if the parameter in question is freely estimated, was used to improve model fit. Thus, parameters that could improve model fit by freeing those parameters were identified. Further improvement of model fit was achieved by removing one item that did not load significantly on the factor (one item of the repression scale). Next, convergent validity was examined by calculating Pearson r correlations between the subscales of the GCI and the report marks (between 1 and 10). A positive moder- ate to strong correlation between the subscales support, growth and atmosphere and the corresponding report marks is seen as indicative of convergent validity of the three subscales. A negative moderate to strong correlation between the repression subscale and the corresponding report mark for repression indicates convergent validity of the subscale repression. Pearson’s correlations of r = .10 - .30 are seen as small, r = .30 - .50 are seen as a moderate, and r > .50 are seen as a large (Cohen, 1988). Reliability analyses were conducted in SPSS 24 (both Cronbach’s alpha and Guttman’s Lambda-2). Alpha’s above .70 and 0.79 were fair; between 0.80 and 0.89 were good (Cicchetti, 1994). For in- terpreting reliability estimates, including Guttman’s lambda-2 (λ-2), there are some gen- eral rules of thumb; λ-2 above 0.70 are sufficient for group-level studies (Guttman, 1945; Osburn, 2000). In order to determine what proportion of the variance in each of the four group climate subscales could be attributed to the group level and the individual level we computed the intraclass correlation coefficient (ICC) which is calculated by dividing the level-2 variance by the total variance (Raudenbush & Bryk 2002). Items with an ICC close to zero indicate that variation is mainly within clients, instead of between groups. On