Marianne Welmers

Alliance Discrepancies 97 CHAPTER 4 the theoretical framework of the SOFTA, coding guidelines, practice material from the developers and Dutch practice material taken from the Dutch drama series In Therapie ( In Therapy) . Trained coders independently coded at least 10 videotapes to increase their reliability as coders compared to the golden standard codings by first author and received feedback on each coding. As advised by Friedlander et al. (2005), training continued until coders differed no more than one point in their scale scores in 90% of the cases. After their training, each coder rated a random selection of the videotapes. Coding dilemmas were discussed and difficult parts were consensus coded duringmeetings with the coding team every two weeks. In total, 90 sessions were independently coded after training. Of these sessions, 15 random selected sessions (16.7%) were double coded by the first author, with coders blind to these double coded sessions. To assess interrater reliability, we calculated intra-class coefficients for the 15 double coded sessions using the singlemeasures of a two-way mixed effect model based on absolute agreement (Koo & Li, 2016). Because the present study focuses on discrepancies in individual alliances, we only used the Engagement and Emotional Connection subscales of SOFTA and averaged these two scales into one combined scale. Scores of these two subscales were significantly correlated (T1: r = .306, p = .003, T2: r = .421, p < .001). Intraclass Correlation Coefficient (ICC) for the combined scale was .44. According to Cicchetti (1994), ICC’s are fair when >.4, good > .6 and excellent >.8. Statistical Analyses Discrepancies were tested using a two level model to increase statistical power and account for dependency of data, as familymembers (level 1) were nestedwithin therapists (level 2). Level 1 concerns variance of alliance measures between family members within the family, while level 2 accounts for variance between families. To analyze discrepancies we extended this model with a categorical moderator, namely family member (mother, father, youth). To enable analysis of this categorical variable, we created dummy variables containing all the information included in the original categorical variable. To analyze the development of differences between family members, we calculated ICCs in the two- level model for T1 and T2. To analyze whether discrepancies differed across reporters, we analyzed interactions between twomoderating variables, namely, family member and alliance reporter (therapist, client, observer). To test for congruence between the therapist reports and family members’ self-reports of the alliance, we calculated pearson’s r for all therapist – familymember dyads. We compared these correlations by calculating the test statistic z and its corresponding p- value (Lenhard & Lenhard, 2014).