Page 167 - Demo

Implementing MDET: Evaluating adaptations and effectiveness1656Table 4 continuedFixed effects B SE %u0394DSTime*Start MDET implementation-0.004 0.02Time*Start MDET implementation *Care home condition0.01 0.02Note. %u0394DS = difference in deviance statistic (Log-likelihood) compared to the previous model.* p < .05 ** p < .01 *** p < .001First, an unconditional means model (model 1) was fitted, including only a random intercept for care home. The ICC of this model was 0.68, meaning that 68% of the variance in involuntary care recordings was accounted for by differences between care homes. To test if the level of organisations should also be included as a random effect, we also fitted an unconditional means model with a random intercept for organisation. The ICC of this model was 0.065, meaning that 7% of the of the variance in involuntary care recordings was accounted for by differences between organisations. Because of the low ICC, we pursued the analyses with only a random intercept for care home.In the next model (model 2), an autoregressive autocorrelation component was added (AR1). This model showed a significantly better fit to the data than the unconditional means model (%ud835%udf122 (1) = 16.804, p < .001). Next, time was added as a fixed effect (model 3). Time was not significantly associated with weekly recorded involuntary care recordings (B = 0.03, SE = 0.02, p = 0.074), and adding time did not significantly improve the model fit (%ud835%udf122 (1) = 2.86, p = .091). Then, the start of implementation was added to the model (model 4), which was not significantly associated with involuntary care recordings (B = -0.002, SE = 0.44, p = 0.997). The model fit did also not significantly improve (%ud835%udf122 (1) = 0.18, p = .674). The interaction between time and the start of implementation was then added to the model (model 5). The interaction term was not significantly associated with involuntary care recordings (B = -0.002, SE = 0.01, p = .854). Including this term made the model fit significantly worse (%ud835%udf122 (1) = 7.02, p = .008). Lastly, we included the interaction between time, start of implementation, and care home