The therapeutic or working alliance has long been studied, and both clinicians and scholars assume that building strong alliances with clients is a significant contributor to positive outcomes of psychotherapy. Generally, alliance is defined as a professional relationship between a therapist and his or her patient, consisting of an emotional bond and agreement on what goals and tasks should be central in therapy (Bordin, 1979; Elvins and Green, 2008). Indeed, the alliance has proved to be a significant common factor in determining the effectiveness of psychosocial treatment of adults (Flückiger, Del Re, Wampold and Horvath, 2018), youths (Karver, De Nadai, Monahan and Shirk, 2018; Murphy and Hutton, 2018), and families (Friedlander, Escudero, Welmers-van de Poll and Heatherington, 2018; Welmers-van de Poll et al., 2018).

In conjoint family treatment, building alliances is complex, because the therapist simultaneously develops alliances with different family members (Friedlander, Escudero and Heatherington, 2006). Given the differences between family members in their expectations of the treatment and the therapist, the strength of these alliances might differ and, thus, discrepancies between alliances of different family members with the therapist might occur. Previous studies indicate that alliance discrepancies within the family contribute to treatment dropout (Muñiz de la Peña, Friedlander and Escudero, 2009; Robbins et al., 2008; Robbins et al., 2003). Moreover, in a meta-analytic study on alliance in couple and family therapy, such alliance discrepancies have been found to be an even stronger predictor of unfavourable treatment outcome as compared to problematic individual alliances with the therapist (Friedlander et al., 2018).

Although research on alliance in family treatment has burgeoned during the last two decades (for a meta-analysis, see Welmers-van de Poll et al., 2018), there is little research that captures aspects of the alliance typical of family treatment, such as alliance discrepancies. Important questions on how alliance processes in a systemic context evolve, interact, and can most favourably be assessed during treatment still remain unanswered. First, the development of discrepancies in alliance between the therapist and different family members over the course of treatment is still under-researched. Second, despite the notion that the therapist is likely to play a key role in the alliance/outcome association (Baldwin, Wampold and Imel, 2007), and should detect and repair problematic alliance discrepancies, it is as yet unclear if and how the therapist takes account of multiple alliances and their discrepancies in family treatment. To address these issues, in this study we aimed to examine the occurrence and development of alliance discrepancies in a Dutch home-based family treatment for youth problems, paying particular attention to the therapist’s evaluation of multiple alliances and their discrepancies.

Alliance discrepancies in family treatment

A first objective of this study was to investigate if discrepancies between alliances of different family members with the therapist occurred, and how they developed over the course of family treatment. Given differences between family members and their role during treatment, discrepancies in alliance might be expected to be the rule rather than the exception. For example, in family treatment parents are most often the initiating party for treatment, and children or adolescents do not always participate voluntarily (Friedlander, Escudero, Welmers-van de Poll and Heatherington, 2019; Shirk, Caporino and Karver, 2010). This might negatively affect the alliance between therapists and young people, as was illustrated in a study on family therapy by Robbins et al. (2006). In this study, mothers had significantly stronger alliances with the therapist than their sons. Furthermore, similar to many other cultures, mothers in the Netherlands are in general more involved in child rearing as compared to fathers (Sociaal Cultureel Planbureau, 2018), and prior research has indicated that youth and care professionals consult or involve mothers more often than fathers in matters concerning their children (Hoogeveen, 2018). Therefore, in family treatment therapists might form stronger alliances with mothers as compared to fathers.

Although alliance discrepancies are likely to occur during family treatment, the extent to which alliances are discrepant may vary during the treatment process. Several studies have shown that the strength of the alliance changes during treatment (Ardito and Rabellino, 2011; Horvath, 2006; Karver and Carporino, 2010; Weiss, Kivity and Huppert, 2014). Subsequently, the extent to which alliances of different family members are discrepant might also change during treatment. Ideally, as the treatment process evolves, the therapist attempts to balance different alliances and improve weak alliances, resulting in a decrease in alliance discrepancies.

The therapist’s evaluation of alliances in family treatment

A second objective of this study was to investigate the therapist’s perspective on multiple alliances and their discrepancies in family treatment. As the therapist is expected to play a key role in improving alliance processes, it is important that he or she should be able to assess and monitor alliances during family treatment. However, the therapist’s perception of alliances may not always be accurate due to personal involvement with the family and may be biased by strong personal reactions to family members (Muñiz de la Peña, Friedlander and Escudero, 2009). In this study, we highlighted two particular aspects of the therapist’s perspective on alliances in family treatment.

A first aspect is the correct evaluation of alliance discrepancies by the therapist in order to repair problematic alliance discrepancies. A study in which therapist reports of the alliance were compared with individual client self-reports showed that therapists systematically overestimated weak alliances and underestimated strong alliances (Hartman, Joos, Orlinsky and Zeeck, 2015). This could also be the case in family treatment, leading to an underestimation of alliance discrepancies by the therapist. Furthermore, the most overt indicator of the strength of the alliance for a therapist is the behaviour of family members during sessions. In family treatment, however, family members observe each other in their interactions with the therapist, which might cause feelings of shame and a reluctance to display inner thoughts and feelings (Friedlander et al., 2006). This was illustrated in a small-scale study on in-family differences in the emotional bond with the therapist. In this study, discrepancies in the emotional bond with the therapist were reported by clients as well as by observers, but the discrepancies were larger for self-reports, indicating that clients’ observed interactions with the therapist only partially mirrored their self-reports (Muñiz de la Peña, Friedlander and Escudero, 2009).

A second aspect of therapists’ evaluations of alliances is the extent to which their perspective is congruent with family members’ perspectives on the alliance. Studies of adult and youth psychotherapy indicate that congruence between therapist and client alliance evaluations leads to more favourable treatment outcomes (Bachelor, 2013; Fjermestad et al., 2016; Kivlighan, 2007; Rozmarin et al., 2008; Zilcha-Mano, Snyder and Silberschatz, 2017). In family treatment, monitoring the alliance with less involved family members might be more difficult than monitoring the alliance with highly involved family members. This may result in differences between family members in the extent to which the therapist and a family member agree upon their alliance. As we argued above, parents most often initiate treatment and mothers are generally more involved in child rearing matters than fathers. Therapists might, therefore, be primarily focused on the alliance with parents (especially mothers) instead of youth, and subsequently be more congruent in their evaluation of the alliance with parents (especially mothers) than with youth.

The present study

The aim of the present study was to gain insight regarding alliance discrepancies and therapist’s evaluation of multiple alliances and discrepancies in family treatment. Our first research question was: do discrepancies between alliances of different family members in family treatment occur, and how do they develop over the course of treatment? Based on the reviewed literature we formulated the following hypotheses: (1) there are significant discrepancies in the strength of the alliance between family members, more specifically: parents have stronger alliances with the therapist as compared to young people, and mothers have stronger alliances with the therapist as compared to fathers; and (2) discrepancies in alliance decrease over the course of treatment.

Our second research question was: how does the therapist evaluate multiple alliances and alliance discrepancies? Here, we formulated the following hypotheses: (1) when therapists report on multiple alliances with different family members, discrepancies between these alliances are less than alliance discrepancies in observer reports and family members’ self-reports; and (2) therapist-reported alliance is associated more strongly with parent-reported alliance as compared to youth-reported alliance, and therapist-reported alliance is correlated more strongly with mother-reported alliance as compared to father-reported alliance.

Results of this study may serve as a framework to further investigate the alliance as a common therapeutic factor in family treatment, and help therapists to monitor and optimise alliance processes in order to enhance treatment outcome.

Methods

Participants and treatment setting

The participants were 92 parents (n = 57 (step)mothers, n = 35 (step)fathers) and 61 children from 61 families. The mean age of the child for whom the treatment was indicated was 10.2 (SD = 4.5; range 0.3–17.8). The mean age of the children participating in the study was 11.0 (SD = 3.1; range 4–17), and the mean age of parents was 39.6 (SD 8,3; range 25–57). In three families, one or both parents were born in another Western country, and in four families one or both parents were born in a non-Western country. Participating families received home-based family treatment for youth problems, called Intensieve Pedagogische Thuishulp (IPT, Van der Steege, 2007). They were seen by 36 IPT workers (14 per cent male, M age 42.7; SD = 9.6), with an average of 8.4 (SD = 4.6) years of experience as an IPT worker. All had a social work-related (post-)bachelor’s degree.

Intensieve Pedagogische Thuishulp (IPT, Van der Steege, 2007) is a home-based family treatment in the Netherlands for families dealing with complex child behaviour and parenting problems. Most families experience problems in other domains as well, such as financial problems, parental psychopathology or lack of a supporting social network. The treatment adopts an empowering, solution-focused and systemic approach, focusing on improving parenting skills and enhancing social support (Van der Steege, 2007). During treatment, an IPT worker visits the family once or twice a week or once every two weeks, depending on the family’s needs and stage of treatment. Families in this study received IPT for an average period of forty-nine weeks (SD = 29,7; range 12–168).

Procedures

Participating families were drawn from four teams specialized in IPT from two Dutch youth care organisations. When a family started treatment with an IPT worker of a participating team, they were informed about the research project by the IPT worker or institution and received a letter with information. In one team, all IPT workers directly asked their clients to participate. In the remaining three teams a member of the research team called the family to ask them to participate. Children were asked to complete the Working Alliance Inventory (WAI) and were included in coding observations using the System for Observing Family Therapy Alliances (SOFTA) when they were aged 8 or older. Therapists were asked to reflect on alliances with all family members involved in treatment, regardless of age. Participating family members of 12 years and older signed an informed consent letter and the project was approved by the ethical board of the Faculty of Social and Behavioral Science of the University of Amsterdam. All participating families received a €10 gift card and two families were randomly selected to receive a voucher for visiting a zoo or amusement park of their own choice.

When a family participated, two sessions with an IPT worker at the family’s home were videotaped. For T1 – in the early treatment phase – the third session (and exceptionally the fourth or fifth) was filmed. We chose the third session because families were informed about the research and asked to consider participation in the first session. By choosing the third session they had some time to consider participation, but treatment was still in its starting phase, which in this model of treatment lasts about six weeks (Van der Steege, 2007). The second video observation (T2) was two months later, when treatment was in the phase of active change. For fourteen families there was no T2 measure available because the treatment had already ended (n = 5), or because the therapist or clients no longer wished to participate because the situation had changed (n = 9). There was no T1 measure for two families, and four families only participated by completing the WAI and had no video observations. None of these four families responded to the request to complete the WAI at T2. Study dropouts were not excluded because this might have reduced the clinical representativeness of our study. We compared alliance measures at T1 for dropouts and completers by performing a multilevel regression analysis of a dichotomous dropout variable at T2 on client-reported, therapist-reported and observer-reported alliance at T1, and found no significant differences (p = .586), which indicates that selective attrition was unlikely.

Measurements

Working Alliance Inventory – Short Form (WAI-s). To assess the alliance as perceived by therapists and family members, we used the Working Alliance Inventory, Short Form (WAI-s; Horvath and Greenberg, 1989; Killian, Forrester, Westlake and Antonopoulou, 2017). The questionnaire consists of twelve items (e.g. ‘My family counselor and I agree on what I should do in order to improve the way things are going in my family’ for the client version, or ‘This client agrees on what family members should do in order to improve the way things are going in the family’ for the therapist version) and measures task, goal and bond elements of the alliance on a 5-point Likert scale ranging from 1 (never) to 5 (always). In the current sample, Cronbach’s alpha for the therapists’ version was .85 at T1 and .88 at T2, and for the clients’ version .91 at T1 and .92 at T2.

System for Observing Family Therapy Alliances (SOFTA-o). The SOFTA-o (Friedlander, Escudero and Heatherington, 2006) has been developed to assess four dimensions of alliance-related behaviour from videotaped family or couple therapy sessions. Two dimensions reflect the individual alliance between a family member and the professional based on Bordin’s (1979) classical definition of the alliance, i.e. Engagement in the Therapeutic Process, representing task and goal elements of the alliance, and Emotional Connection. The two other dimensions, Safety within the Therapeutic System, and Shared Sense of Purpose within the Family, reflect aspects of the alliance that are unique to conjoint family treatment.

When using the SOFTA-o, a trained coder observes a session and notes the frequency of specific positive and negative alliance-related behaviours along the four dimensions. Some examples for positive and negative behavioural indicators include: ‘Client describes or discusses a plan for improving the situation’ or ‘Client shows indifference about the tasks or process of therapy (e.g. paying lip service, “I don’t know”, tuning out)’ for Engagement; ‘Client shares a lighthearted moment or joke with the therapist’ or ‘Client avoids eye contact with the therapist’ for Emotional Connection; ‘Client implies or states that therapy is a safe place’ or ‘Client expresses anxiety non-verbally’ for Safety; and ‘Family members ask each other for their perspective’ or ‘Family members blame each other’ for Shared Sense of Purpose. After observing the session, based on the frequency, intensity and clinical meaningfulness of the marked behaviours, coders assign global ratings on each dimension on a 7-point Likert scale ranging from –3 (extremely problematic) to +3 (extremely strong). For the purpose of this study, the SOFTA-o was translated from English to Dutch, following guidelines as prescribed by Van Widenfelt, Treffers, De Beurs, Siebelink and Koudijs (2005).

For coding the videotaped IPT sessions with the SOFTA-o, we used the training manual by Friedlander et al. (2005) as a guideline. The first author received training from Professor Escudero, one of the developers of the instrument. After training and translation of the manual, she coded twelve videotapes with at least two family members in the session to use as a gold standard. Coding dilemmas were discussed with Professor Escudero. Next, three master’s students of Educational Sciences at the University of Amsterdam were trained, receiving fifteen hours of coder training over five weeks. They were introduced to 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 ten videotapes to increase their reliability as coders compared to the gold standard codings by first author and received feedback on each coding. As advised by Friedlander et al. (2005), training continued until coders differed by no more than one point in their scale scores in 90 per cent of cases.

After their training, each coder rated a random selection of the videotapes. Coding dilemmas were discussed and difficult parts were consensus coded during meetings with the coding team every two weeks. In total, ninety sessions were independently coded after training. Of these sessions, fifteen random selected sessions (16.7 per cent) were double coded by the first author, with coders blind to these double coded sessions. To assess inter-rater reliability, we calculated intra-class coefficients for the fifteen double coded sessions using the single measures of a two-way mixed effect model based on absolute agreement (Koo and 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 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. To analyse 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 analyse the development of differences between family members, we calculated ICCs in the two-level model for T1 and T2. To analyse whether discrepancies differed across reporters, we analysed interactions between two moderating 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/family member dyads. We compared these correlations by calculating the test statistic z and its corresponding p-value (Lenhard and Lenhard, 2014).

Results

Preliminary analyses

To start with, we calculated alliance means and standard deviations for all family members and all reporters. Results are shown in Table 1. Overall, across different reporters (therapists, clients, and observers) alliances with mothers were strongest both at T1 and T2, followed by fathers and then youth. Clients provided the highest alliance ratings, followed by observers and then therapists, except for alliances with fathers and mothers at T2 (therapists had higher scores than observers, but still lower than clients). To test if differences between reporters were significant, we extended our two-level model with alliance reporter as a moderating variable. As shown in Table 2, we found significant differences between different reporters of the alliance, indicating that therapists generally reported weaker alliances as compared to clients and observers both at T1 and T2.

Alliance discrepancies

Table 3 shows the results of multilevel regression analyses of alliances with family role as moderating variable. We found significant differences in alliances with the therapist between mothers, fathers, and young people. In line with our first hypothesis, alliances with parents were stronger as compared to youth both at T1 and T2, and alliances with mothers were stronger as compared to fathers both at T1 and T2.

To test whether discrepancies between family members decrease over the course of treatment, we calculated the ICC between all different alliances across reporters in the multilevel model. At T1 the ICC was .12 (p < .01), indicating a small in-family correlation between alliances with different family members. At T2, the ICC was .35 (p < .05), indicating that in-family alliance differences became smaller, which was in line with our hypothesis.

Therapists’ evaluation of alliances

With regard to the therapists’ evaluations of alliance discrepancies, results of multilevel interaction analyses between family member and alliance reporter showed no significant differences in the reports of alliance discrepancies between therapists, clients, and observers. In contrast to our hypothesis, this indicates that therapists report alliance differences between family members comparable with clients’ self-reports and observer reports.

Our hypothesis that the therapists’ perspective is more congruent with self-reports for parents’ alliances as compared to youth, and for mothers’ alliances as compared to fathers’ was partially confirmed. As shown in Table 4, therapist reports of the alliance with mothers had a moderate positive correlation with mothers’ self-reports at T1 and a large correlation at T2, with both correlations significant. Therapists’ ratings of alliances with fathers were slightly negatively correlated with fathers’ self-reports at T1, just as therapists’ ratings of alliances with youth were slightly negatively correlated with the young people’s self-reports at T1. However, both correlations were not significant. The difference between mothers and fathers in the strength of the correlation between therapist- and self-reports was trend-significant (z = 1.533, p = .063). The differences between mothers and youth (z = 1.064, p = .144) and between fathers and youth (z = −0.246, p = .403) in the strength of the correlation between therapist- and self-report were both not significant. These findings were partially in line with our hypothesis, as they indicate to some extent that therapists were more congruent in their alliance perspective with mothers as compared to fathers, but do not indicate that therapists were significantly more congruent in their alliance perspective with parents as compared to youths.

At T2, correlations between therapist- and self-reported alliance increased for both mothers and fathers, while the difference in the strength of these correlations was not significant (z = −0.632, p = .264). In contrast to our hypothesis, this indicates that the therapist’s perspective on the alliance was just as congruent with the perspective of mothers as with fathers when treatment was in the active change phase. For alliances with youths, the correlation between therapist-report and self-report at T2 was small and not significant. Differences in therapist-report and self-report correlations were significant both between mothers and youths (z = 1.793, p = 0.036) and fathers and youths (z = 1.771, p = .038). This indicates, as hypothesised, that in the active change phase of treatment therapists were more congruent in their perspective on the alliance with parents as compared to youths.

Discussion

We investigated (a) if discrepancies between alliances with different family members in family treatment occur, and how they develop over the course of family treatment, and (b) how the therapist evaluates multiple alliances and their discrepancies in a Dutch home-based family treatment for youth problems.

Alliance discrepancies in family treatment

Our study findings confirm our hypotheses that discrepancies between family members occur, but become somewhat smaller during treatment. We found that, regardless of who reported on the alliance, mothers had stronger alliances with their therapist compared to fathers. This finding might partially be explained by the impression that contacts during treatment tend to be most intensive with mothers, that is: in our study mothers were present in 93 per cent of the observed sessions, whereas fathers were present in only 35 per cent of the observed sessions. When contacts with mothers during treatment are more frequent as compared to those with fathers, alliances with mothers may also have more time and opportunity to evolve, and therefore might be stronger.

Differences in alliance strength between fathers and mothers might also be explained by a gender match between mothers and therapists, given the fact that most therapists (86 per cent) in the current sample were female. This hypothesis is further supported by findings of previous studies on the effect of gender match on alliance or on satisfaction with the therapeutic relationship, indicating that female client-therapist dyads have the strongest alliance in adult and adolescent substance abuse treatment (Kuusisto and Artkoski, 2013; Wintersteen et al., 2005), marriage and family therapy (Johnson and Caldwell, 2011), and a collection of varying psychosocial treatments for adults (Bhati, 2014). Bhati (2014), however, speaks of a general ‘female effect’ rather than a gender match effect, as in her study dyads with female therapists – even dyads with male clients – had stronger alliances compared to dyads with male therapists. It should be noted that in all these study samples, as in our study, male therapists were under-represented or even absent. Therefore, further research including more male therapists is needed to draw valid conclusions on the effect of gender match on alliance.

With regard to discrepancies in alliance, we also found that parents had stronger alliances as compared to young people. This was in line with our hypothesis and with meta-analytic findings of Roest et al. (2020). They argue that building alliances with children and adolescents (hereafter referred to as 'youth') can be complicated by developmental issues, such as cognitive limitations in understanding the necessity of treatment for younger children. For teenagers, conflicts with parents on the need for and aim of treatment and problems with accepting authority might be a complicating factor. In family treatment, forming an alliance with children or adolescents might be even more complex, as they are less powerful interaction partners compared to adults, and the presence of parents in sessions (even when only in part) might cause feelings of shame and a hesitation to be open (Escudero and Friedlander, 2017). Although being heard during treatment appears to be very important for young people in family therapy (Strickland-Clark, Campbell and Dallos, 2000), playing an active role in a conjoint treatment process with adults is challenging for youth, and should therefore be carefully guided by the therapist both verbally and non-verbally (Escudero and Friedlander, 2017; O’Reilly, 2008).

Therapists’ evaluation of alliances in family treatment

We found no significant differences in alliance discrepancies between therapist-, client-, and observer-reports, indicating that therapists assess differences in alliances comparable with both observers’ and family members’ self-reports. This finding underlines both the robustness of the occurrence of alliance discrepancies and the therapists’ ability to detect differences between family members’ alliances. Notably, our study is probably the first to include observed, self-reported, and therapist-reported measures when investigating alliance discrepancies. More research is needed to build a stronger evidence base on therapists’ ability to accurately recognise alliance discrepancies.

Therapist reports of the alliance were moderately associated with mothers’ self-reported alliance in the early stage of treatment, but there was no therapist- and self-report correlation for fathers and youths. This indicates that at the beginning of treatment, therapists lack a shared perspective on the alliance with both fathers and youths. It seems that in the early stage of treatment therapists’ perspectives on the alliance are closest to mothers’ perspectives, who in our sample were the most involved party in treatment. These results should, however, be interpreted with caution, as differences in the therapist-self-report correlation at T1 were only trend significant for the difference between fathers and mothers, and not significant for the differences between parents and youths.

It was interesting to find that the association between therapist reports of the alliance with fathers and their self-reports became notably stronger during treatment. At T2 there was a significant and large correlation, which was comparable with the association between therapist- and self-reports of mothers’ alliances at T2. Apparently, when therapists spend more time in treatment with fathers, a more shared perspective on their alliance evolves. In contrast, correlations between therapist- and self-reports of the alliance with youths barely increased during treatment, indicating that even when the treatment process evolves, therapists still lack a shared perspective on the alliance with youths. This finding is in contrast to meta-analytic findings of Roest et al. (2020), who found a moderate correlation between therapist-reported and self-reported alliances with youths in youth psychotherapy. Perhaps our finding on the lack of a shared perspective between therapists and youths is specific to the alliance with youths in conjoint family treatment, where therapists might have a stronger focus on the alliance with parents. It should be noted, however, that in our sample the number of youths with self-reports of the alliance was small, and further research is required to draw valid conclusions about the association between therapist- and self-reports of the alliance with young people in family treatment.

Limitations, strengths and future directions

Several limitations of this study should be considered when interpreting our study findings. First, we analysed data from a highly heterogeneous sample in terms of targeted problems, age of referred children, and family members present at observed sessions. Also, only a small number of fathers and youths with self-reports and observer reports of the alliance was included in the sample. Furthermore, in comparing alliance reports of therapists, clients, and observers, it could be that results are to some extent confounded by informant bias. That is, differences between alliances from family members’ self-reports might be a result of different informants reporting on different alliances (e.g. mother reporting on mother’s alliance, father reporting on father’s alliance, etc.), whereas for therapist- and observer-reports different alliances are evaluated by the same person (e.g. therapist reporting on both mother’s and father’s alliance).

A final limitation of the study is that the inter-rater reliability of observer-reports of the alliance was small (ICC = .44), albeit sufficient (Cicchetti, 1994). However, according to Koo and Li (2016), a relatively low ICC score could be explained by low variability among sampled subjects, a small number of subjects (k < 30) or a small number of raters (n < 3). They suggest that a reliability study should include at least thirty heterogeneous samples being tested by at least three raters before low ICC scores should be interpreted in terms of poor reliability. In our study, we single-coded most sessions, which automatically generates lower agreement as compared to consensus coding by two or three coders.

Despite its limitations, our study adds to the body of knowledge on alliance in family treatment in several ways. To our knowledge, it is the first study on alliance in the unruly context of home-based family treatment focusing on multiple alliances with different family members, including observations as well as both therapist-reports and client self-reports. Furthermore, we used a multi-level model to account for statistical dependency, allowing us to compare alliances of different family members regardless of who reported on the alliance, and to compare different perspectives (i.e. therapist, client, observer) on alliance discrepancies. Finally, the use of two different measure moments enabled analyses of the development of alliance discrepancies between family members and of congruence between therapist-reports and client self-reports of the alliance.

Future research on alliance in family treatment should focus on investigating the effect of alliance discrepancies between family members as well as the effect of congruence between therapist- and self-reports of the alliance on treatment outcome. Ideally, in this kind of research a developmental approach is applied, as our study findings demonstrate that alliance discrepancies and congruence between therapist-report and client self-report of the alliance evolve during treatment. With regard to alliance discrepancies, previous studies provide some evidence that discrepancies between family members within the family contribute to treatment dropout (Friedlander et al., 2018; Muñiz de la Peña et al., 2009; Robbins et al., 2008; Robbins et al., 2003). However, to date there is only a handful of studies on alliance discrepancies in relation to treatment outcome, with various definitions and often problematic methodologies (Welmers-van de Poll et al., 2018), and none of these studies investigated how the development of alliance discrepancies (i.e. increase or decrease of discrepancies) can affect treatment outcome.

Studies on the effect of congruence in therapist- and self-reported alliance have shown somewhat ambiguous results, ranging from no effect in adult psychotherapy (Fitzpatrick, Iwakabe and Stalikas, 2005) to better outcomes when ratings are more congruent in adult psychotherapy (Bachelor, 2013; Kivlighan, 2007; Rozmarin et al., 2008; Zilcha-Mano, Snyder and Silberschatz, 2017) or when congruence increases during treatment in youth CBT (Fjermestad et al., 2016). However, studies on the effect of congruence between therapist and client ratings of the alliance in conjoint family treatment are sparse. In family treatment, perhaps congruence of one family member with the therapist might be more important for positive treatment outcome as compared to other family members. Our finding that therapist- and self-reports of the alliance with parents were more congruent as compared to those of youths underlines the importance of this issue.

Our findings underline the complexity of building and monitoring multiple alliance processes in conjoint family treatment. Our findings demonstrate that the therapist’s perspective on the alliance with a specific family member is not necessarily comparable with that family members’ own perspective, especially with youth. To gain a more shared perspective on the alliance, therapists could consider asking for family members’ feedback on the alliance, which has proven to be effective in several settings (e.g. Mihalo and Valenti, 2018; Van Hennik, 2020; Zilcha-Mano and Errazuriz, 2015). In doing so, it is recommended to pay particular attention to children’s and adolescents’ feedback on the alliance (for an example, see Rober, Van Tricht and Sundet, 2020). Our findings also demonstrate that differences in alliances with the therapist between family members are the rule rather than the exception. The awareness that building alliances with some family members (most often youths) demands an extra effort might enhance the process of building and balancing multiple alliances in family treatment.