Introduction and Literature Review

Couple therapy is effective in improving couple relationships, as well as individual mental and physical health (Lambert, 1992; Roddy et al., 2020). However, many couples do not persist in treatment long enough to receive these benefits, with some studies indicating that over half of couples discontinue treatment early (Doss et al., 2011). Those who discontinue psychotherapy early fare poorer on several outcomes such as worsening of symptoms, lack of problem resolution, and lower satisfaction with the therapeutic process (Lampropoulos, 2010; Pekarik, 1992; Swift & Greenberg, 2015). In one study, only 13% of early discontinuers were classified as experiencing clinically meaningful change compared to over 70% of clients who completed treatment (Cahill et al., 2003). Clients also report experiencing a sense of failure (Ogrodniczuk et al., 2005), and are less likely to seek therapy elsewhere (D’Aniello et al., 2018). In addition to these poorer clinical outcomes, early discontinuation is associated with lower morale among therapists, wasted financial and time resources of insurers, government agencies, clients, and therapists (Bischoff & Sprenkle, 1993; Masi et al., 2003). Despite the importance of understanding early discontinuation, there is comparatively little research about what predicts discontinuation in couple therapy.

Discontinuation

Most of the research on discontinuation in systemic therapy has focused on demographic predictors often with ambiguous findings. For example, several studies have shown that older clients (Thalmayer, 2018), more educated (Bartle-Haring et al., 2007), or who have a higher income (Jurek et al., 2014) are more likely to complete treatment. Yet, for each of these factors, other studies have found no relationship with discontinuation. While understanding the client profile of couples who discontinue treatment may be beneficial, without knowing the causal mechanism between these factors and discontinuation, there is little this knowledge can do to help the therapist improve the likelihood of their clients continuing in treatment. The same can be said of many of the other predictors of early discontinuation that have been studied in couple therapy including severity of presenting problem (Heyman et al., 1999), initial satisfaction with the relationship (Graff et al., 2009), professional training (Hamilton et al., 2011), and therapist level of experience (D’Aniello & Tambling, 2016). None of these factors are within the ability of a therapist to change in their immediate interactions with the clients.

Therapeutic Alliance

The therapeutic alliance, on the other hand, is at least partially under the control of the therapist. The therapeutic alliance is one of the most consistent and robust predictors of client outcomes in psychotherapy research (Flückiger et al., 2018). In the most recent meta-analysis of the alliance-outcome relationship in systemic therapies, Friedlander et al. (2018) found a medium effect size (d = .62) with stronger alliances consistently associated with improved outcomes. This is notable as the therapeutic alliance is considerably more complex in systemic therapies.

In individual psychotherapy, the alliance can be thought of as the affective bond between the therapist and client related to the work of therapy, as well as the agreement on the goals and tasks of treatment (Bordin, 1979). In one conceptualization of the alliance in systemic therapies, the bond domain is referred to as “emotional connection” and the agreement on tasks and goals domains are referred to collectively as “engagement in the therapeutic process” (Friedlander et al., 2005). These two domains continue to be important in systemic therapy, but with added complexity. The expanded alliance in couple therapy requires that the therapist develop an alliance with each member of the couple or family system. These between-systems alliances can differ in strength, with one partner potentially closely aligned with the therapist while the other is more distant. Such split alliances are particularly detrimental to outcomes in systemic therapy (Friedlander et al., 2018). In addition to these multiple between-systems alliances, partners in the couple also develop a therapeutic alliance or “shared sense of purpose” with each other that allows them to work collaboratively and achieve their common goals (Friedlander et al., 2005). Finally, unlike individual psychotherapy where confidentiality of exchanges between the therapist and patient are guaranteed, what one partner says to the therapist is witnessed by the other partner and can impact the relationship long after the couple leaves the therapist’s office. As such, “safety within the therapeutic system” is an additional crucial element of the expanded therapeutic alliance, allowing the partners to take risks, be open, vulnerable, and flexible (Friedlander et al., 2005).

The Effect of Alliance on Discontinuation

Given the robust alliance-outcome relationship across models of therapy, it is not surprising that a strong therapeutic alliance is a consistent predictor of clients’ decision to continue in therapy in both individual psychotherapy (Swift & Greenberg, 2015) as well as couple therapy (Anderson et al., 2019; D’Aniello et al., 2018; Harris et al., 1988; Sheehan & Friedlander, 2015; Yoo et al., 2016). Even after only one session of relational therapy, the alliance has been shown to predict early termination (McCrady et al., 1996; Yoo et al., 2016). As expected, the early research shows that split alliances are uniquely predictive of discontinuation in couple therapy (Bartle-Haring et al., 2012; Glebova et al., 2011; Jurek et al., 2014; Muniz de la Peña et al., 2009; Shnaider et al., 2015). Other research on the alliance has provided evidence that it is one of the mechanisms by which other predictors of discontinuation act. Anderson et al. (2019) found that alliance fully mediated this association. While more research is certainly needed to understand the alliance-discontinuation relationship, it seems clear that therapists would be wise to build strong between-systems alliances with all members of the system, avoid split alliances, and monitor the quality of within-system alliance or shared sense of purpose.

Monitoring the Alliance

There are three ways a therapist might monitor the therapeutic alliance. First, a therapist might rely on their intuition. While there is certainly space for a therapist’s intuition in therapy, research has indicated that the client’s perception of the alliance is the most clinically useful (Horvath, 2001) and that therapists are not all that good at identifying ruptures in the alliance (Talbot et al., 2019). Second, a therapist might rely on client-reported measures of the alliance. Several such measures exist- for example the four-item Intersession Alliance Measure is a reliable and valid measure of the expanded therapeutic alliance that can easily be used to track the alliance over time (Anderson et al., 2023). Other valid measures of the expanded alliance include the Integrative Psychotherapeutic Alliance Scales (Pinsof et al., 2008) and the SOFTA-S self-report measure of the alliance (Friedlander et al., 2005). Routinely monitoring alliance is certainly a best practice. However, we believe that even such monitoring of alliance following a session is not sufficient. A measure of the alliance that is completed after the therapy session is over does not provide the immediate feedback a therapist could use to address ruptures in the alliance, split alliances, or other problems in alliance development as they occur. The third way therapists can track the alliance is by using observation skills to recognize client behavioral cues that indicate the strength of the alliance. The best observational measure of the expanded therapeutic alliance (System for the Observation of Family Therapy Alliance or SOFTA-O; Friedlander et al., 2005) provides behavioral indicators for each of the four dimensions of the alliance. Therapists could become adept at spotting behavioral indicators of problematic alliances to address potential problems during the session. This, however, would require that we first know (1) which dimensions of the alliance are most strongly associated with discontinuation and (2) which behavioral indicators within these dimensions may be most important to monitor.

Previous research on the alliance in couple therapy has not provided answers to either of these two questions. The goal of this study was to explore the answers to these two unknowns. One additional problem with much of the alliance-dropout research is that it uses relatively weak post-hoc correlational designs. A second goal of this study was to use a design that would provide more confidence in any findings from the study. The two research questions that guided this study were: (1) Which dimensions of the alliance differ between those who drop out of therapy and those who continue? (2) Are there specific behaviors within these dimensions that are particularly relevant to dropout? We hypothesized that in examining these questions that couples who discontinued would show lower alliance quality, fewer positive alliance behaviors, and more negative alliance behaviors than couples who continued successfully in therapy.

Methods

To answer these questions, we employed a matched-cases design to compare the quality of the four dimensions of the alliance and the prevalence of behaviors within these dimensions among 15 couples who discontinued from therapy (discontinuers) and among a matched sample of 15 couples who completed therapy successfully (completers). The associate clinic director at the on-campus clinic of a COAMFTE accredited MFT training program provided a list of all couple cases that (1) consented to allow video recordings of their sessions for research and (2) were treated by therapists who were student trainees but who had since graduated from their clinical program. A member of the research team reviewed the termination summary and final case note to determine whether the coupled discontinued therapy unsuccessfully. Two questions on the termination summary helped arrive at this conclusion. Therapists identified whether additional services were recommended and their rating of the success of treatment.

Discontinuation was operationalized as a therapist (1) recommending additional treatment and (2) rating treatment as unsuccessful. After the initial categorization, the research team read through the final case notes to confirm that there was no evidence of a plan to discontinue or delay treatment. Fifteen cases matched the discontinuation criteria and had video recordings of the final sessions of therapy. Next, we reviewed cases that completed treatment (defined as the therapist rated the treatment as successful or very successful and did not recommend additional treatment) to find matching cases. The primary matching criteria was having the same therapist. When multiple successful cases by the same therapist were available, the match was determined by the number of matching criteria in the following priority: marital status, length of the relationship, age, income, ethnicity, employment status, and the number of children. If there was not a matching successful case with the same therapist, the match was determined by the remaining matching variables.

After identifying 15 matched pairs of couples, we selected and coded the final session from the discontinuer cases and the matching session for the completer cases. For example, if the first couple discontinued after session 10, we coded session 10 from their matched completer couple. This was a naturalistic study. No specific treatment protocol was used, and the treatment approach was not monitored. However, all therapists received supervision from faculty supervisors.

Sample Characteristics

Twelve therapists contributed cases for this study: Ten females and two males. Ten were Master MFT students and two were clinical psychology doctoral students. All cases were seen at the on-campus clinic that serves the training needs of both the MFT and Clinical Psychology programs. All were practicing under the supervision of credentialed supervisors. Five therapists contributed one dropout and one continuing case, two contributed two dropout and two continuing cases each, one contributed one dropout and two continuing cases, one contributed three dropouts and two continuing cases, one contributed two dropouts, and two contributed one continuing case each.

Thirty heterosexual couples (60 individuals) participated as clients in the study. Approximately 87% were White, 13% were Hispanic, 3% were Asian, and 3% marked “other” for their race/ethnicity. Age ranged from 24 to 62 with the median age of 28. Income ranged from $1200 to $90,000 with a mean of $38,000. Twenty-three couples were married and reported an average length of marriage of 6.35 (SD = 8.03) years. Data on the length of relationship of non-married couples was not provided on the clinic’s intake form. Of the fifteen couples that discontinued treatment, ten did so during the assessment phase of treatment (3 after session 1, 2 after session 2, and 4 after session 3). The remainder discontinued treatment in the treatment phase (1 following session 5, 2 after session 6, 1 after session 7 and two after session 10).

Measures

Alliance dimensions and behaviors were coded by six MFT master’s students using the System for Observing Family Therapy Alliances (SOFTA-o, Friedlander et al., 2005), which was designed to measure the working alliance in conjoint family therapy. The SOFTA-o examines the four dimensions of the therapeutic alliance (Friedlander et al., 2005): Engagement in the therapeutic process, emotional connection to the therapist, safety within the therapeutic system, and shared sense of purpose within the family. Each dimension includes specific behaviors that contribute to or weaken the working alliance. For example, the behavior, “client expresses interest in the therapist’s personal life” is a positive indicator in the emotional connection dimension; and “client expresses anxiety nonverbally (e.g., taps or shakes)” is a negative indicator of the safety within the therapeutic system dimension (Friedlander et al., 2005). The SOFTA-o has both a client and therapist version to capture each person’s contribution to the overall alliance. In the client version, male and female partners are assessed separately. After reading through the coding manual, coders met weekly to receive further training and practice coding. Questions that the coding team could not resolve were graciously answered by Dr. Friedlander, one of the developers of the SOFTA-o. Coders watched each session at least twice: Once to review the couple’s behaviors and another time to review the therapist’s. For every video, coders rated each dimension on a 7-point Likert scale, with anchors − 3 (extremely problematic), 0 (unremarkable or neutral), and (+ 3) extremely strong) following rating guidelines in the SOFTA-o. The videos were split into three even sections and given a score for each section. SOFTA has been found to be reliable and valid in studies with diverse couples and families in at least three countries (Friedlander et al., 2006). SOFTA ratings correlate with therapists’ perception of in-session behavior, alliance strength, session value and ease, client improvement, and goal attainment posttreatment (Friedlander et al., 2006).

An advanced MS student in the MFT program served as the criterion coder under the direction of the second author. Five coders were female, and one was male. One coder was a Latina from Chile, and the rest were White from the US. Training occurred over the course of 24 weeks (1.5 hour meetings with two-four hours of practice at home). Once coders consistently achieved an interclass correlation coefficient (ICC) of .70 on training tapes, coding of research tapes commenced. Coders and the lab manager were blind to whether the couple was a dropout or continuing couple. Every five sessions, coders double coded the same video to guard against coder drift and assess reliability. ICCs of coded tapes ranged from .75 to .89.

Results

Dimension Totals

To assess whether there was a difference between the dropout and continue group on each of the four broad dimension of the alliance, a series of one-tailed independent t-tests were conducted. One-tailed tests were appropriate given the directional nature of the hypotheses and the small sample size. We also opted not to correct for family-wise error because of the exploratory nature of the current study and the limited power from our small sample size. While this may increase the likelihood of Type I error, at this stage in the research process we were more concerned about committing Type II errors. Results for each of the SOFTA-o dimensions are presented in Table 1. We hypothesized that those who continued in therapy would have a higher observed alliance scores in each dimension of the SOFTA-o. Results only partially confirmed this hypothesis. Five of the dimensions had a significant difference in the expected direction: female emotional connection (t(52) = − 2.32, p = .01, d = − 0.63), male emotional connection (t(52) = − 2.66, p < .01, d = − 0.73), therapist emotional connection (t(52) = − 2.93, p = .003, d = − 0.78), male safety (t(52) = − 2.80, p < .01, d = − 0.76), and shared sense of purpose (t(52) = − 1.87, p = .03, d = − 0.51). All other differences in dimensions were not statistically significant.

Table 1 Differences in alliance dimensions between discontinuers and completers
Full size table

Although the male engagement and female safety dimensions were not found to be significant, their effect sizes (respectively, t(52) = − 1.67, p = .05 d = − 0.45; t(52) = − 1.36, p = .09, d = − 0.37)) suggest that if the sample size were larger, there might have been a significant difference between the means for discontinuers and continuers. Though these results were not statistically significant, they may still be worthwhile to pursue in future research.

Individual Behaviors

Next, we examined whether there were differences in the individual alliance behaviors that comprised the different dimensions of the expanded therapeutic alliance. Independent samples t-tests were conducted to examine mean differences in the two groups on each separate behavior (43 behaviors for each partner and 43 behaviors for the therapist). Table 2 presents the results of these analyses for observable client behaviors and Table 3 presents the results for observable therapist behaviors. Five items differed significantly for the female partners: two in the engagement dimension (expresses optimism and complies with enactment), two were in the emotional connection dimension (shares light-hearted moment and shows interest in the therapist’s personal life). Three additional items were significantly different in the shared sense of purpose dimension which was coded jointly for male and female partners (offer to compromise, share a joke, and validates other’s point of view). Each of these behaviors was in the predicted direction except for the “complies with enactment” which found that the female partner complied with enactments significantly more often in the discontinuer group than in the completer group.

Table 2 Clients’ behavioral indicators of the alliance for discontinuer and completer cases
Full size table
Table 3 Therapists’ behavioral indicators of the alliance by discontinuer and completer cases
Full size table

Six male partner behaviors were significantly less likely to be observed among couples who discontinued treatment. Among the engagement in the therapeutic process items, male partners were less likely to complete homework or express optimism and were more likely to exhibit indifference about the tasks or processes in therapy. Like their partners, males were also less likely to share a light-hearted moment during the session (emotional connection domain) or vary their emotional tone (safety in the system domain). Male partners who discontinued were also significantly less likely to show vulnerability during the session (safety in the system domain).

Only one therapist behavior differed significantly between couples who successfully completed treatment and those who discontinued. Therapists were less likely to share a light-hearted moment with their clients in couples who discontinued treatment prematurely. No other behavioral indicators differed significantly between the groups.

Discussion

Past research has shown that the therapeutic alliance is an important predictor of premature discontinuation in couple therapy. This exploratory study examined two research questions to provide a more nuanced understanding of this relationship: (1) Which dimensions of the alliance differ between couples who discontinue and successfully complete treatment and, (2) are there specific behaviors within these dimensions that are particularly relevant for dropout.

The data suggest that three dimensions of the alliance may be particularly important: emotional connection, safety in the system, and shared sense of purpose. First, the emotional connection dimension was rated as significantly lower for couples who discontinued prematurely. This was true regardless of whose behaviors were being rated-male partner, female partner, or therapist. Furthermore, the size of the effect ranged from moderate when rating the couples’ alliance to large when rating the therapist’s contribution to the alliance. It comes as no surprise that the affective bond between each client and the therapist is an essential component of treatment. Past research in individual therapy has found that a strong and stable bond is associated with successfully completing treatment (Janeiro et al., 2018) and that difficulty forming a bond indirectly led to dropout through alliance ruptures (Tschuschke et al., 2022). This research provides initial evidence that the same is true in couple therapy as well.

Two behavioral indicators in this group occurred less frequently for those who discontinued treatment. Females in the continuing group showed more of an interest in the therapist’s personal life. This indicator suggests that the client has an interest in the therapist as a person and not just a professional. Interestingly, even though the total dimension was significant for men and therapists, the only significant individual item for both men and therapists was “shares a lighthearted moment.” That item was significant for females as well and was the only item that was significantly lower among all three members of the system. Shares a lighthearted moment was defined as “a behavioral connection through humor or goodwill, typically signaled by laughter” (Friedlander et al., 2005). Under the dimension Shared Sense of Purpose, there was a significant difference between groups on the item “family members share a joke or a lighthearted moment with each other.” This item is similar with the only difference being that the couple shared a connection while laughing, like making eye contact (Friedlander et al., 2005). Coders were trained to only code these items if the parties involved were genuinely amused rather than laughing out of discomfort. Positive affect appears to be particularly important in continuing treatment and is consistent with previous findings that couples who continue in treatment have more positive exchanges than couples who drop out of therapy but do not differ in the frequency of negative exchanges (D’Aniello et al., 2021). Lighthearted moments may also be meaningful because of what Fife et al. (2014) call the therapist’s “way of being.” The way of being is how the therapist shows their “own genuineness and aliveness” with their clients (Corey, 2005, p. 17). Therapy is a human rather than technical endeavor, so those who value the personhood of the clients and put the client’s needs first demonstrate a way of being that fosters a good therapeutic relationship (Fife et al., 2014). By creating an environment where the clients and therapist can connect through humor, the therapist is showing their way of allowing the client to be more comfortable and connected with the therapist.

A couple’s shared sense of purpose, also known as the within system alliance, was also significantly lower in couples that discontinued treatment prematurely. This dimension of the expanded therapeutic alliance is uniquely related to discontinuation. When partners do not feel like they are a team working together toward a common goal, they are less likely to stay engaged in therapy. The within system alliance has been shown to be a particularly strong predictor of other clinical outcomes in relational therapy (Anderson & Johnson, 2010; Friedlander et al., 2018) with some research suggesting that it is a prerequisite for change (Friedlander et al., 2008). Given these findings, it would benefit therapists to pay particular attention to this unique dimension of the expanded alliance.

Couples who discontinued treatment scored significantly lower on three indicators of a shared sense of purpose. They offered to compromise with each other less frequently than couples who successfully completed therapy. As mentioned earlier, they were also less likely to share a joke with their partner during the session. Finally, couples who discontinued treatment were less likely to validate the other person’s point of view.

Finally, a sense of safety is also likely an important dimension to monitor. When observing the male partner’s alliance behaviors, couples who drop out exhibit significantly less safety than do couples who complete therapy successfully. The size of this effect was moderately large (Cohen’s d = − .76) and was the largest difference between groups when considering only the clients’ contribution to the alliance.

Within this dimension, two indicators differed between the two groups. The specific significant items were “varies emotional tone”, which was lower among the discontinuer group for both men and women, and “shows vulnerability”, which only differed significantly for men. Varies emotional tone is defined as shifting from one emotion to another, such as anger, sadness, fear, or happiness (Friedlander et al., 2005). This item was most often shown when the client began to cry. Showing vulnerability was marked when a client talked about something difficult and painful to talk about, often noted by crying (Friedlander et al., 2005). These two items were often connected; men would start talking about a vulnerable subject and then begin to cry. These findings suggest that men who felt safe enough in the therapy room to express their emotions and discuss things that were otherwise difficult to verbalize were more likely to continue treatment. This agrees with the literature that feeling safe has been shown as a precursor for positive change (Christensen et al., 1998; Friedlander et al., 2014). It might also be that men feel pressure from society to avoid expressing emotions. When they go to therapy and feel a safe connection with the therapist, they might feel as though they can share what they have already been feeling but have been unable to express. This outlet might encourage them to continue treatment.

It is interesting that the safety dimension as a whole and the shows vulnerability item were only statistically significant when considering the male partner’s alliance indicators. A general pattern that emerges in the data is that the effect sizes at the dimension level and for many of the individual items are larger when considering the male partner’s alliance indicators than when rating the female partner’s alliance behaviors. The importance of the male partner’s alliance in couple therapy has been noted in several studies, including in relation to discontinuation (Bartle-Haring et al., 2012; Bourgeois et al., 1990; Friedlander et al., 2010). There may be several explanations for this. First, men generally have more power in the U.S. and may therefore have more power over decisions to continue in treatment. Another potential explanation is that therapy asks both partners to be emotionally vulnerable and expressive in treatment, tasks that challenge traditional gender stereotypes for masculinity. Research in the field of sex roles shows that traditionally, the emotional well-being of the marriage and the family has been largely undertaken by women (Bourgeois et al., 1990). As such, married men are less likely than their wives to talk about or consult with others about intimate issues (Bourgeois et al., 1990). It is possible that these two factors make male partner “buy-in” to therapy particularly important. These explanations, however, are conjecture and should be explored through future research. Until we fully understand why the male partner’s alliance is uniquely predictive of outcomes, therapists would be wise to pay particular attention to their alliance with the male partner.

While the engagement in the therapeutic process dimension did not differ significantly between the two groups, there were four individual indicators within this dimension that did occur with differing frequencies. Men in the group who discontinued treatment were less likely to complete homework tasks and showed greater indifference about the tasks or process of treatment. This indicator was marked when clients either showed indifference nonverbally (e.g., doing something else, not following the flow of conversation, looking around the room), or verbally (e.g., displaying a lack of energy or enthusiasm or making comments that indicate they do not believe something would help; Friedlander et al., 2005). As discussed earlier, it is particularly important for therapists to engage the male partner in therapy.

The item “client expresses optimism or indicates that a positive change has taken place” was significant for both men and women, with couples who completed treatment showing more optimism than those who discontinue therapy. This item was noted when the client described feeling hopeful or seeing that change is possible (Friedlander et al., 2005). Hope has been linked as a precursor to positive change (Christensen et al., 1998), and lack of hope is related to dropout (Jurek et al., 2014; Shnaider et al., 2015). It makes sense that clients with hope would continue therapy because they have some inner belief that change is possible, and therapy is a means to bring about the progress they are looking for. Clients may have felt more motivated to work on their presenting problem because they have hope that therapy will work for them. This increased motivation can result in actions that improve the clients’ presenting problem.

Finally, women in the discontinuer group were more likely to comply with the therapist’s requests for enactments. This item was defined by the client responding to the therapist’s explicit request to do something in the session such as facing each other or talking to one another (Friedlander et al., 2005). Interestingly, this was significant for females in the opposite direction than was expected. Those in the dropout group complied more frequently than those in the continuing group. This could be in line with what D’Aniello et al. (2018) found when they suggested that clients do not want tasks to go beyond where they are in treatment. If the client felt that the tasks they were given were too much, they considered discontinuing treatment (D’Aniello et al., 2018). It could be that those who complied with enactments in the dropout group did not have a good experience or felt that the therapist was asking them to do something they were not comfortable with. It is also possible that therapists were shifting focus to just the female partner in the group who discontinued treatment, reflecting a potential split alliance.

Together, these results suggest that therapists should monitor the therapeutic alliance in couple therapy. As they monitor the alliance, these exploratory results suggest that paying specific attention to emotional connection, safety in the therapeutic system, and the couple’s within system alliance or shared sense of purpose are particularly important. In addition, they should be concerned if there is a lack of positive affect such as optimism, sharing light-hearted moments, or partners validating each other’s points of view. Finally, therapists would do well to give increased weight to the quality of the alliance with the male partner. This is particularly true in the safety dimension. When male partners are not showing vulnerability or varying their emotional tone we recommend that therapists address the safety of therapy with the client.

Limitations and Future Directions

This study had several limitations that should be considered when reviewing the findings. The most important of these limitations revolve around the sample for the study. First, the sample for this study came from a university training clinic where the therapists and perhaps the clients are not representative of therapy practiced in the community. Second, the sample size is small. To compensate for the small sample, we made two decisions that demand caution when interpreting the results. We opted to use one-tailed t-tests and to not adjust for family-wise error. It is possible, and perhaps even probable, that some of the significant findings we outline in this study are Type I errors. We believe that these decisions were appropriate given the directional nature of the hypotheses and the exploratory nature of the study. We have also included effect-size estimates with all the analyses to help readers better interpret the findings. However, these results should be seen only as preliminary, with further replication in larger and more diverse samples needed to be confident in the findings. In addition, the clinic provided only limited demographic data and no information on relationship satisfaction or other variables that have been shown to be linked to dropout in couple therapy. It is possible that the differences in the alliance that we observed in this study may be related to one of these confounding variables. Despite these limitations, this study has offered important insights in the specific dimensions of the alliance that are associated with discontinuation as well as specific behaviors that therapists can use to identify clients who may be at risk of discontinuation.