Elien Neimeijer

52 A measurement model was examined using the pooled within-level covariance matrix. In this model, group climate was specified as a latent variable, using four indicators: support, growth, repression, and atmosphere. Second, a structural model was examined in which a direct effect from group climate on aggressive incidents and coercive measures (both represented by observed [composite] variables) were specified, as well as an indi- rect effect such that aggressive incidents mediated the relation between group climate and coercive measures. Third, the pooled between-level covariance matrix was used to examine the hypothesised measurement and structural models at the between-group level. The variable repression was recoded such that a higher score was indicative of less repression because research on MSEM has found that reversely scored variables may cause convergence problems (Gustafson, & Stahl, 2005). Also, negative residual variance at level-2 is a common problem in MSEM, which can result in non-convergence of the model (Kim, Dedrick, Cao, & Ferron, 2016). The variable growth at the between-part of the model (level-2) displayed negative residual variance. Because the residual variance was close to zero and non-significant, it was fixed to zero, which is a recommended prac- tice when using multilevel SEM (Hox, 2010). Exact model fit was calculated with a Chi-squared test. Because the Chi-squared test is sensitive to sample size, fit measures that are less sensitive to sample size were Support Group size Aggressive incidents Coercive measures Growth Security level Repression Care intensity Atmosphere Group climate e3 e4