Elien Neimeijer

57 Structural equation modelling First, a measurement model was examined using the pooled within-level covariance ma- trix. A model with group climate represented as a latent variable showed an acceptable fit to the data: χ 2 (2) = 5.56, p = .062, CFI = .986, TLI = .959, RMSEA = .097, SRMR = .032. Second, the structural model was specified, in which a direct effect was specified from group climate on aggressive incidents, which was represented by an observed (compos- ite) variable, as well as direct effects from group climate and aggressive incidents on co- ercive measures, which was also represented as an observed variable. An indirect effect was specified such that the relation between group climate and coercive measures was mediated by aggressive incidents. Results showed a good fit to the data: χ 2 (8) = 12.63, p = .125, CFI = .986, TLI = .974, RMSEA = .055, SRMR = .043. Next, a measurement mod- el was fitted using the pooled between-level covariance matrix, in which group climate was specified as a latent variable, which resulted in poor model fit: χ 2 (2) = 25.71, p < .001, CFI = .894, TLI = .682, RMSEA = .452, SRMR = .065. Modification indices suggested a correlation between residual variances of the indicators atmosphere and support , which resulted in good model fit: χ² (1) = 0.94, p = .331, CFI = 1.00, TLI = 1.001, RMSEA = .000, SRMR = .013. The structural model indicated a good fit to the data, based on the majority of fit indices: χ² (6) = 8.56, p = .200, CFI = .991, TLI = .979, RMSEA = .086, SRMR = .040. Subsequently, a two-level model was fitted, in which the within- and between-lev- el models were examined simultaneously. Also, gender and legal status were included as within-level covariates. Between-level covariates group size and care intensity were considered, however, including these variables in the model showed a significant deteri- oration in model fit of the between-level part of the model. The final model (see Figure 2) showed a good fit to the data: χ² (24) = 38.71, p = .029, CFI = .970, TLI = .947, RMSEA = .050, SRMRW = .076, SRMRB = .068. However, results indicated a standardised beta co- efficient greater than 1 between aggressive incidents and coercive measures. Therefore, a covariance between aggressive incidents and coercive measures was specified at the between-level part of the model. Results indicated that, at the within-group level, group climate was negatively related to aggressive incidents ( β = -.21, p = .005), but not significantly related to coercive meas- ures. Aggression ( β = .51, p < .001) and legal status ( β = .17, p = .011) were significantly related to coercive measures. The relation between group climate and coercive meas- ures was significantly mediated by aggression (indirect effect, β = -.11, p = .010). At the