Page 160 - Demo

Chapter 6158model with a random intercept for care home. The intraclass correlation coefficient (ICC) was calculated as an indicator of the variability in weekly recorded involuntary care between different care homes. We explored the unconditional means model with a random intercept for organisation, and calculated the ICC to examine the variability of registrations between the four organisations. Successive growth models were built as follows: a model with a random intercept for care home and an autoregressive autocorrelation structure (AR1), to account for the dependency of observations over time within care homes (model 2); time (continuous) added as a fixed effect (model 3); the start of the MDET method (coded 0 or 1) as a fixed effect (model 4); the interaction between the start of the MDET method and time as a fixed effect (model 5); and the interaction between start of the MDET method, time, and care home condition (CAU or intervention, coded 0 or 1) as a fixed effect (model 6). Confidence intervals (90%) were calculated for the effect size (beta coefficient) of the interaction between time, start of the MDET method, and care home condition on weekly recorded involuntary care (model 6). For research question 3, the same models as for research question 2 were developed, but with weekly recorded incident reports as the outcome variable. Then, we conducted equivalence tests as follows: 90% confidence intervals were calculated for the effect size of the interaction between time, start of the MDET, and care home condition on incident reports. The smallest effect size of interest was a priori set at [-0.2; 0.2]. If the values in the 90% confidence interval would fall within this range, we would conclude that there is no meaningful difference between care home conditions (CAU versus intervention) regarding the effect of MDET on weekly recorded incident reports over time. For all linear mixed effects models, differences between goodness of model fit were calculated using log-likelihood ratio tests (package lmtest; Zeileis & Hothorn, 2002).