Martijn van Teffelen

Imagery-enhanced cognitive restructuring: Efficacy 83 4 trait scores were entered as dependent variables. Condition, time and condition by time indicators were entered as independent variables. We differentiated between immediate intervention (T1) and sustained intervention efficacy (T2). Treatment integrity analyses (see Supplemental Material 2) showed that one therapist relatively underperformed compared to the others. For this reason, the variable `therapist` was treated as a random effect within our analyses. This deviates from pre-registration. Assuming that a ‘therapist-effect’ was constant across all repeated measures, the therapist variable was entered as a random intercept. To gain two extra degrees of freedom, baseline indicators were removed from themodel under the assumption that there are no baseline group differences for the dependent variable. 2 RESULTS Means and standard deviations are shown in Table 2. Four cases were excluded from further analyses, as they were not able to formulate a belief that was related to their autobiographical situation. An examination of group differences of study variables at baseline revealed no significant differences ( p ’s > .198), except for the QMI, H (2) = 6.97, p = .031 (lower at I-CR). Visual inspection of outliers revealed four outliers in the CR and AC conditions with values outside the ± 1.5 * interquartile range. When these outliers were removed, the QMI group difference was no longer significant, H (2) = 4.72, p = .094. Nonetheless, analyses were run with and without QMI scores as covariate 3 . As the pattern of results did not differ in any of the analyses when QMI score was entered as a covariate, results are presented without using QMI scores as covariate. Immediate intervention efficacy Hostile beliefs To test the main hypothesis that I-CR reduces hostile belief ratings to a greater extent than CR and AC (HI) we ran a mixed regression model. Fixed effects are shown in Table 3. Time and condition variables were reference coded, using AC as reference category. The random effect of therapist was not significant ( p = .594). Results showed that I-CR ( b = -0.55, t = -3.96, p <.001) and CR ( b = -0.28, t = -2.05, p = .044) resulted in greater reductions in hostile belief ratings from pre- (M1) to post-intervention (M3) compared to AC (see Table 3 and Figure 2). Also, when the CR condition was selected as reference category, I-CR led to a larger reduction in hostile belief ratings than CR ( b = -0.28, B = -11.03, SE = 5.03, t = -2.19, p = .031) from pre- (M1) to post-intervention (M3). 2 Models were compared to each other using Χ 2 -tests . 3 In the pre-registration we planned to report results that included QMI as covariate. As the QMI did not influence the pattern of results we decided to exclude it to gain statistical power.