Margit Kooijman

Therapist effects | 105 Data analysis Descriptive analyses were calculated for all variables. Because patients were nested within physiotherapists, multilevel linear analyses were performed to explore factors associated with change in severity of shoulder complaints. Data was checked for normal distribution. All analyses were conducted using Stata version 13. Two models were developed: model 0: intercept-only, model 1: variables at patient level and physiotherapist level (the personality trait dimensions of the BFI, therapists’ age and gender). These were entered based on their univariate significance level (p < 0.25) 13 . All variables were centered around their grand means for interpretable meaning. For model 1, associations were expressed as regression coefficients along with standard errors, 95% confidence intervals and p values. Analyses were performed with complete data and alpha was set on 0.05. Results Study sample The BFI questionnaire was returned by 77% of the therapists (n = 56). Non-response analysis revealed that responding therapists did not differ from non-responders with regard to gender (p = 0.59) but were older (p = 0.04). After matching patients with shoulder complaints from the NPCD database with participating physiotherapists, 2,814 patients were included in the study of whom 2,116 had complete data on independent variables (figure 1).