Elien Neimeijer

100 Given the nested nature of the data (clients are nested in living groups), a multilevel approach was also used to examine the relation between Group climate and Work cli- mate. Instead of using group mean scores for both group climate and work climate, client scores of group climate were used (level-1 variable) in relation to group mean scores of work climate as a characteristic of the group in which clients resided (level-2 variable). Hierarchical linear modeling (HLM) was used to account for violation of the independ- ence assumption of regression. HLM allows for examination of how variation in the de- pendent variable is attributed to differences within-group (i.e., individual level, level-1) or between-group (i.e., living group-level, level-2; Raudenbush & Bryk, 2002). The anal- yses were conducted using the “lme4” package (Bates, Maechler, Bolker, & Walker, 2015) in the R environment. The “lmerTest” package (Kuznetsova, Brockhoff, & Christensen, 2015) was used for the calculation of p-values, which uses the Satterthwaite approxima- tion procedure for calculating degrees of freedom. Several models were fit for each dependent variable, i.e. Support, Growth, Repression, and Atmosphere. First, a random intercept only model (null model) was fitted without predictors to estimate the Level-2 variance and ICC (intraclass correlation coefficient) for the dependent variables. When significant Level-2 variance is demonstrated, multi- level analysis is warranted, and Level-2 predictors were examined in a multilevel model. If no significant Level-2 variance in the dependent variable (i.e., aspect of group climate) was found, it was concluded that work climate was not related to that aspect of group climate. Subsequently, three multilevel models were fitted. The first model included only main effects of Level-1 predictors (gender, age, and placement status). A second model in- cluded Level-2 predictors (security level and gender composition of the living group). The third model added the group mean work climate scores of staff (Team functioning, Workload, Work Environment, and Job Satisfaction) as Level-2 predictors. The variable security level consisted of three categories (low security, medium security, and high se- curity) and was dummy-coded, using low security as the reference category. The fit of the models was compared using likelihood-ratio tests. Parameter estimates and statistical tests are reported for the initial and the final model.