Marianne Welmers

Chapter 3 74 analysis of a dichotomous dropout variable at T2 on T1 client- and therapist reported alliance, T1 therapist observed alliance behaviors, therapist clinical experience and therapist neuroticism, extraversion, openness, agreeableness, and conscientiousness. We found no significant differences on any of these variables, except for therapists’ observed safety behaviors (β = 1.231, p = .001), indicating that therapists showed more safety behaviors at T1 in the dropout cases. Statistical Analyses First, we imputed missing values using expectation maximalization (Tabachnick, & Fidell, 2013). After removing one case with T1 alliance measures as well as therapist variables missing, a MCAR test as proposed by Little (1988) indicated that missing values were completely at random (χ² = 415.076 (370, N = 98), p = .053). Next, we tested for collinearity to enable a multivariate test of the relative importance of multiple independent variables by calculating a Variance Inflation Factor (VIF). All calculated VIF’s were below 3, indicating that there was no multicollinearity problem. For our main analyses we used a two level model to account for dependency of data (Tabachnick & Fidell, 2013), as family members (level 1) were nested within therapists (level 2). Level 1 concerns variance of alliance measures between family members within the family, while level 2 accounts for variance between families. Because of the short-term longitudinal nature of the study (i.e., client- and therapist reports as well as therapist’s alliance building behaviors were measured early and mid-treatment), we controlled for T1 alliance as well as T1 observed alliance building behaviors. We differentiated between parent-therapist and youth-therapist alliance by adding a parent/youth variable to the model. Because some studies indicate that the strength of the alliancemay differ between male and female therapists (e.g. Welmers- van de Poll et al., 2018), and that therapist age is associated both with years of clinical experience and observed interpersonal skills (Anderson et al., 2009), we also controlled for therapist age and gender. Within the two- level model, we performed two stepwise multiple regression analyses to sequentially identify whether family role (parent vs. youth), therapist age and gender, clinical experience, personality traits, mid-treatment alliance building behaviors, early alliance, and early treatment alliance building behaviors (independent variables) were significantly associated with therapist reported mid-treatment alliance or with client reported mid- treatment alliance (dependent variables).