Page 90 - Reduction of coercive measures

Chapter 4
Multilevel analyses
The first step of the analyses showed an ICC correlation of .08 at the care unit level. The variance in total number of coercive measures across units was significant (.27; z = 2.80, p = .005). This shows that the hierarchical data structure should not be ignored and multilevel analysis is indicated.
In the second step each predictor of either the level of resident or the level of unit was added separately to the model. A negative association was found between level of social adaptive behavior skills and total coercive measures (b = -0.01, t = -3.0, p = .003). In addition, a positive association was found between attribution of stability and the total of coercive measures (b = 0.063, t = 0.03, p = .032). Also, a positive association was found between challenging behavior and coercive measures applied to prevent from direct unforeseen danger (b = 0.06, t = 3.04, p = .003) (Table 3).
The third step to include all factors simultaneously was carried out in two phases. First the resident related variables were entered in the model and second the staff related variables were entered. Findings showed a negative association between social adaptive behavior and coercive measures (b = -0.01, t = -2.25, p = .027) and a positive association between challenging behavior and coercive measures applied to prevent from direct unforeseen danger (b = 0.04, t = 2.14, p = .035) (Table 4).
88