adult outpatients with early-onset PDD, CBASP leads to greater improvements in depression at the end of treatment [3] and 1 year later (but not beyond that) [4]. We conducted a sequential mediation analysis testing the hypothesis that the treatment effect (CBASP vs. SP) on outcome is mediated by a sequential indirect effect of treatment-specific improvements in the therapeutic relationship and a subsequent decrease in social inhibition.The inclusion and exclusion criteria of this RCT are described elsewhere [3]. Briefly, the acute-phase treatment consisted of 24 sessions of CBASP or SP over 20 weeks. In the continuation phase, 8 further sessions were administered between week 20 and 48. In CBASP, the therapeutic relationship is explicitly targeted using techniques that are collectively labeled disciplined personal involvement. For example, positive in-session experiences with the therapist are contrasted with hurtful reactions from significant others to promote safety in the therapeutic dyad (interpersonal discrimination exercise). SP calls for a far less structured approach to the therapeutic relationship and relies on the broader common factors of psychotherapy.Our analysis was performed in a modified intention-to-treat population that included all randomized patients for whom data were available for the mediating variables for at least 1 of the assessments (256 out of 268 patients, 95.5%). These mediating variables were the Helping Alliance Questionnaire (HAQ) and the Inventory of Interpersonal Problems (IIP). Outcome was measured using the clinician-rated Hamilton Rating Scale for Depression (HRSD-24). Out of the subsample analyzed in the current report, 130 patients (50.8%) were randomized to CBASP and 126 patients (49.2%) to SP. The mean age of the patients was 45.05 years (SD 11.8), 64.8% were female (n = 166). We imputed missing values for the HAQ, the IIP, and the HRSD-24 by using the mean of the posterior distribution from the fully conditional specification method obtained by the iterative Markov Chain Monte Carlo estimation.We examined sequential mediation via an ordinary leastsquares path analysis using a bootstrapping approach facilitated by the PROCESS macro version 2.16. The independent variable was group allocation (CBASP vs. SP), the dependent variable was the total score of the HRSD at week 48 (end of treatment), and the putative mediating variables were change of the HAQ between week 1 and the end of the acute treatment phase at week 20 (M 1 ) and subsequent change in the social inhibition subscale of the IIP during the continuation phase from week 20 to 48 (M 2 ). These time points were chosen to establish a temporal sequence of treatmentinduced changes in the therapeutic relationship followed by improvement in interpersonal functioning. For further details of the analysis, see Figure 1.This analysis revealed that a significant part of the association between treatment and outcome is mediated through the sequential indirect effect of improvement in HAQ (M 1 ) and reduction in the IIP social i...