Journal of Counseling Psychology 2015, Vol. 62, No. 1, 1-13

© 2014 American Psychological Association 0022-0167/15/$ 12.00 http://dx.doi.org/10.1037/cou0000044

Secure Attachment to Therapist, Alliance, and Outcome in Psychoanalytic Psychotherapy With Young Adults Peter Lilliengren

Fredrik Falkenstrom

Stockholm University

Uppsala University and Linkoping University

Rolf Sandell

Pia Risholm Mothander and Andrzej Werbart

Lund University

Stockholm University

Using a novel approach to assess attachment to therapist from patient narratives (Patient Attachment to Therapist Rating Scale; PAT-RS), we investigated the relationships between secure attachment to therapist, patient-rated alliance, and outcome in a sample of 70 young adults treated with psychoanalytic psychotherapy. A series of linear mixed-effects models, controlling for length of therapy and therapist effects, indicated that secure attachment to therapist at termination was associated with improvement in symptoms, global functioning, and interpersonal problems. After controlling for the alliance, these relationships were maintained in terms of symptoms and global functioning. Further, for the follow-up period, we found a suppression effect indicating that secure attachment to therapist predicted continued improvement in global functioning, whereas the alliance predicted deterioration when both variables were modeled together. Although limited by the correlational design, this study suggests that the development of a secure attachment to therapist is associated with treatment gains as well as predictive of posttreatment improvement in functioning. Future research should investigate the temporal develop­ ment of attachment to therapist and its interaction with alliance and outcome more closely. To ensure differentiation from patient-rated alliance, observer-based measurement of attachment to therapist should be considered. Keywords: attachment to therapist, therapeutic alliance, outcome, psychoanalytic psychotherapy, young adults

John Bowlby’s (1988) notion that the therapist optimally pro­ vides a “secure base” for the patient in successful psychotherapy has gained increased attention among psychotherapy researchers in the last two decades (Farber, Lippert, & Nevas, 1995; Farber & Metzger, 2009; Mallinckrodt, 2010). One important methodolog­ ical advance has been the development of specific measures for

assessing the quality of patients’ attachment to their therapist. Today, two self-report measures exist: the Client Attachment to Therapist Scale (CATS; Mallinckrodt, Gantt, & Coble, 1995) and the Components of Attachment Questionnaire-Therapist (CAQ-T; Parish & Eagle, 2003a, 2003b). Both measures have been used to explore how patients’ attachment to the therapist relates to impor­ tant in-session processes. For example, Mallinckrodt, Porter, and Kivlighan (2005) found that secure attachment to therapist, as measured with the Secure subscale of the CATS, was related to patients’ experience of depth, smoothness, and exploration in sessions. This finding has been replicated in two other studies using similar methodology (Janzen, Fitzpatrick, & Drapeau, 2008; Romano, Fitzpatrick, & Janzen, 2008). Also, in a study applying the CAQ-T, Saypol and Farber (2010) found that the quality of patient attachment to the therapist was associated with higher levels of patient self-disclosure in treatment. Taken together, these results support Bowlby’s hypothesis that secure attachment to therapist facilitates deeper emotional exploration in therapy. Still, a critical empirical question concerns whether secure at­ tachment to therapist is related to therapy outcome. From an attachment perspective, exploration of painful or frightening ex­ periences within the context of a secure therapeutic relationship should lead to changes in the patients’ internal working models. This, in turn, will promote change in the patients’ way of relating to self and others, leading to reductions in distress and growth in

This article was published Online First September 15, 2014. Peter Lilliengren, Department of Psychology, Stockholm University; Fredrik Falkenstrom, Center for Clinical Research Sormland, Uppsala University, and Department of Behavioural Sciences and Learning, Linkoping University; Rolf Sandell, Department of Psychology, Lund University; Pia Risholm Mothander and Andrzej Werbart, Department of Psychology, Stockholm University. This study is based on data from the Young Adult Psychotherapy Project conducted at the former Institute of Psychotherapy, Stockholm County Council, and the Psychotherapy Section, Department of Clinical Neurosci­ ence, Karolinska Institutet. The project was approved by the Regional Research Ethics Committee at the Karolinska Institutet, and all participants have given their informed consent. We give special thanks to Bjorn Philips. Department of Behavioural Sciences and Learning, Linkoping University, for valuable input during the study and preparation of the manuscript. Correspondence concerning this article should be addressed to Peter Lil­ liengren, Department of Psychology, Stockholm University, Frescati Hagvag 8, Room B116, Stockholm, Sweden 106 91. E-mail: peter.lilliengren@ psychology.su.se 1

2

LILLIENGREN ET AL.

functioning. To the extent of our knowledge, however, only two studies have specifically addressed the link between secure attach­ ment to therapist and outcome so far. Sauer, Anderson, Gormley, Richmond, and Preaco (2010) investigated the associations be­ tween patients’ attachment to the therapist, alliance, and outcome in a sample of 95 moderately distressed patients treated with brief therapy by trainee therapists at a university clinic. They found that higher levels of secure attachment to the therapists at Session 3 predicted reductions in psychological distress over time. In a recent study by Wiseman and Tishby (2014), however, secure attachment to therapist measured at Session 5 was unrelated to outcome for patients undergoing a 1-year long psychodynamic therapy. Nevertheless, lower relationship-specific attachment avoidance early in therapy predicted improvement, indicating that the absence of insecure attachment strategies in relation to the therapist may facilitate better outcomes. Clearly, more studies are needed to investigate the influence of both secure and insecure forms of attachment to therapist on therapy process and outcome. Another central research question concerns whether secure at­ tachment to the therapist can be meaningfully differentiated from the therapeutic alliance (Meyer & Pilkonis, 2002; Robbins, 1995). For example, in the study by Sauer et al. (2010), secure attachment to therapist correlated strongly with the alliance, and, hence, their respective impact on outcome could not be differentiated in the statistical analyses. Studies using the CATS typically report strong correlations (.60 -.80) between secure attachment to therapist and patient self-report measures of the therapeutic alliance (Bachelor, Meunier, Laverdiere, & Gamache, 2010; Lunsford, 2010; Mallinckrodt et al., 1995, 2005; Romano et al., 2008; Sauer et al., 2010). Addressing this overlap, Mallinckrodt, Porter, and Kivlighan (2005) concluded that “high quality working alliance and a secure attachment to one’s therapist appear to have many features in common and are probably perceived relatively similarly by clients” (p. 97). The strong covariation between measures of the alliance and secure attachment to therapist is problematic, however, because although the constructs overlap to some extent, there are also important differences (Mallinckrodt, Coble, & Gantt, 1995; Mallinckrodt, Porter, & Kivlighan, 2005; Obegi, 2008). For ex­ ample, contemporary notions of the therapeutic alliance are com­ monly derived from Bordin’s (1979) pan-theoretical definition, which emphasizes the patient’s conscious perception of the col­ laboration with the therapist (Horvath, 2006; Horvath, Del Re, Fliickinger, & Symonds, 2011). In contrast, attachment to the therapist is grounded in a particular sociodevelopmental theory and reflects the function of the relationship (i.e., how the patient uses the therapeutic bond on both conscious and unconscious levels). In particular, secure attachment to the therapist involves several specific relational components (Mallinckrodt, 2010; Parish & Eagle, 2003a, 2003b), such as using the perceived safety of the therapeutic relationship for exploration of novel or frightening experiences (i.e., “secure base”), turning to the therapist for com­ fort (i.e., “safe haven”), viewing the therapist as a resource for guidance (i.e., “stronger and wiser”), and reacting with distress when separated (i.e., “separation anxiety”), none of which are explicitly part of the alliance construct (Obegi, 2008). Thus, the­ oretically, a strong therapeutic alliance does not necessarily imply that the patient and therapist have also developed a secure attach­ ment relationship. For example, some patients may report a strong

alliance even though they do not use the therapeutic relationship as a secure base or experience it as a safe haven (Lunsford, 2010). It is also possible that some patients who have developed a secure attachment to the therapist may report a temporary weak alliance, for example, in connection with an alliance rupture (Safran & Muran, 2000) or when exploring negative transference reactions (Woodhouse, Schlosser, Crook, Ligiero, & Gelso, 2003). Thus, one might hypothesize that the therapeutic alliance may vary somewhat during therapy, whereas secure attachment to the ther­ apist probably develops more progressively and is more likely to stay relatively stable once established. However, in order to in­ vestigate such hypotheses, the constructs need to be measured distinctly. One possible reason for the considerable overlap found when measuring the alliance and secure attachment to therapist is com­ mon method bias (Podsakoff, MacKenzie, & Podsakoff, 2012). Because patient self-report measures have been used to assess both constructs, at least some of the observed covariance may be due to shared measurement method. Self-reports may be the most appro­ priate strategy for assessing the alliance because it predominantly refers to the conscious and collaborative aspects of the therapeutic relationship. Further, research suggests that it is the patients’ self-reported perspective of the alliance that is the strongest pre­ dictor of outcome (Horvath et al., 2011). Attachment to therapist, however, involves unconscious, implicit-procedural processes that reflect the emotional function of the relationship. Therefore, an assessment method based on observer ratings might be more appropriate (lacobvits, Curran, & Moller, 2002; Maier, Bernier, Pekrun, Zimmermann, & Grossmann, 2004). Measures that aim to assess attachment to therapist from an observer perspective have just recently been developed (Lilliengren et al., 2014; Talia et al., 2014), but their utility for differentiating alliance and attachment in relation to outcome has not yet been investigated. In the present study, we aimed to explore three main issues regarding the relationships between secure attachment to therapist, alliance, and outcome, (a) Is secure attachment to therapist at termination associated with improvement? Although secure attach­ ment to therapist has been shown to predict theoretically relevant in-session processes, few studies have it linked with outcome, and the results so far have been somewhat inconsistent. Thus, this is still a fundamental empirical issue to address. In line with an attachment-informed perspective on psychotherapy process and change, we hypothesize that the level of secure attachment to the therapist at termination will be related to the amount of change during therapy, (b) Does secure attachment to therapist relate to outcome once alliance is controlled for? This question concerns the overlap between secure attachment to the therapist and the alliance and their respective impact. Because secure attachment to therapist indicates that the patient uses the therapeutic relationship for emotional exploration and regulation, we hypothesize that it will be more strongly associated with improvement compared with self-reported alliance. In order to address the issue of method bias, we also introduce a novel approach for assessing patient attach­ ment to therapist from patient narratives, the Patient Attachment to Therapist Rating Scale (PAT-RS; Lilliengren, 2011; Lilliengren et al., 2014), while using a standard patient self-report measure to assess the alliance, (c) Does secure attachment to the therapist or the alliance predict any changes after treatment termination? The­ oretically, an attachment-informed perspective on psychotherapy

SECURE ATTACHMENT TO THERAPIST, ALLIANCE AND OUTCOME

process suggests that changes produced within the context of a secure attachment relationship would also be associated with sus­ tained or even increased improvement posttreatment. Because a strong alliance does not necessarily imply an emotionally secure therapeutic relationship, we hypothesize that the alliance will not be as closely linked with maintenance of gains. Predictors of posttreatment changes is a central but largely neglected issue in psychotherapy research (Blatt, Zuroff, Hawley, & Auerbach, 2010; Falkenstrom, Grant, Broberg, & Sandell, 2007), and, to our knowl­ edge, no previous study has looked at attachment to therapist in this regard.

Method Setting Data from the Young Adult Psychotherapy Project (YAPP) was used for the purposes of this study. The YAPP was a naturalistic, prospective, and longitudinal study of young adults (age 18-25) in psychoanalytic psychotherapy, conducted at the former Institute of Psychotherapy, Stockholm, Sweden. A total of 134 self-referred patients were included; 92 allocated to individual therapy and 42 to group therapy. The patients’ psychological well-being was mea­ sured with standardized self-report and interview instruments ad­ ministered before therapy, at termination, and at two follow-up points: 1.5 and 3 years after termination, respectively. The overall design and outcome of the YAPP has been described in detail elsewhere (Lindgren, Werbart, & Philips, 2010; Philips, Wennberg, Werbart, & Schubert, 2006).

Participants Patients who received individual therapy and participated in the interview at termination (thereby enabling assessment of attach­ ment to therapist; see below) were selected for this study. Of the total of 92 patients allocated to individual therapy, nine never started therapy, one dropped out from treatment, and 12 never showed up for the interview at termination. Thus, the available sample encompassed 70 patients. Fifty-six of these (80%) were female, and their mean age was 22 (SD = 2.1; range = 18-26) at the start of therapy. At intake, 23 (32.8%) reported living alone, 17 (24.2%) with a partner, eight (11.4%) with a friend or in a dormitory, 17 (24.2%) reported living with one or both of their parents, and five (7.1%) with another close relative (e.g., sibling, cousin, or grandparent). None of the patients was married or had children. Most of the patients were full-time students (n = 23; 32.8%), combined part-time studies with part-time work (n = 14; 20%), or studied part time without any additional work (n = 9; 12,8%). Sixteen (22.8%) worked full time and four (5.7%) worked part time. Only three (4.2%) were unemployed or on longer sick leave; however, 18 (25.7%) reported being unable to work or study due to poor mental and/or physical health at start of therapy. In terms of cultural and socioeconomic background, the majority of the patients were bom in Sweden (n = 65; 92.8%), but 18 (25.7%) had at least one parent of foreign origin. Most came from highly educated families where at least one parent had an academic degree (n = 51, 72.8%). Half the sample (n = 35, 50%) had experienced parental divorce during childhood.

3

Psychiatric diagnoses were not formally assessed in the YAPP, but main problem areas were categorized on the basis of data from the intake interviews (Wiman & Werbart, 2002). The most com­ mon complaints were low self-esteem (97%), depressed mood (66%), anxiety (55%), and conflicts in close relationships (66%). About one third had self-reported personality disturbance accord­ ing to the Diagnostic and Statistical Manual o f Mental Disorders, fourth edition, text revision (American Psychiatric Association, 2000) and International Classification of Diseases-10 Personality Questionnaire (Ottosson et al., 1995). The patients in this sample were treated by 32 therapists (21 female and 11 male) with an average of 10 years of clinical experience after licensure. Twelve of the therapists treated one patient, 13 had two patients, and seven had three or more patients. The therapists shared a psychoanalytical frame of reference, de­ spite varying preferences regarding specific theories and tech­ nique. All therapists met weekly in clinical teams, where treatment problems and clinical experiences were discussed. No treatment manual was used; rather, the therapies were carried out in accor­ dance with the naturalistic setting of the project. Goals, duration, and frequency of therapy were adjusted to each patient’s needs with the possibility of renegotiation during treatment. The average length of treatment in the present sample was 23 months (range = 2-55; SD = 13.0).

Independent Variables Secure attachment to therapist. The level of secure attach­ ment to the therapist was assessed with the PAT-RS (Lilliengren, 2011; Lilliengren et al., 2014). The PAT-RS is a newly developed observer-rated instrument that measures the quality of the patient’s attachment to his or her therapist based on his or her verbal description of the therapist as a person, experience of the thera­ peutic process, and reactions to attachment-related issues in the therapeutic relationship (closeness, separation, etc.). The instru­ ment is grounded in the two-dimensional model of adult attach­ ment (Bartholomew, 1990; Bartholomew & Horowitz, 1991; Hazan & Shaver, 1987; Mikulincer & Shaver, 2007), which relates adult attachment-system function and interpersonal dynamics to the underlying dimensions of anxiety and avoidance. Following this model, the PAT-RS is designed to assess the quadrants of attachment relatedness (see Figure 1) in four subscales: Security (low anxiety, low avoidance), Hyperactivation (high anxiety, low avoidance), Deactivation (low anxiety, high avoidance), and Dis­ organization (high anxiety, high avoidance). To assess the subscales, the rater compares patient narratives with prototypical descriptions of how secure, hyperactivating, de­ activating, or disorganized attachment strategies may typically be presented in the context of the therapeutic relationship. This pro­ cedure is repeated in relation to nine components that have been suggested to define an attachment relationship (Parish & Eagle, 2003a, 2003b): 1. Secure base, 2. Stronger/wiser, 3. Safe haven, 4. Proximity seeking, 5. Particularity, 6. Responsiveness, 7. Strong feelings, 8. Separation anxiety, and 9. Mental representation (for an example of the rating sheet for one of the components, see the Appendix). The prototypical descriptions of each component are rated on a 5-point Likert scale ranging from 1 (does not fit at all) to 5 (fits very well), and the final subscale score is obtained by

LILLIENGREN ET AL.

4 HIGH AVOIDANCE

S u b scale 3:

S u b scale 4:

D e a c tiv a tio n

D is o rg a n iz a tio n

LOW

HIGH

ANXIETY

ANXIETY

S u b scale 1:

S u b sca le 2:

S e c u rity

H y p e ra c tiv a tio n

LOW AVOIDANCE

Figure 1. The subscales of the Patient Attachment to Therapist Rating Scale represented in the two-dimensional model of adult attachment.

calculating the average score across all components for each subscale. For the specific purposes of this study, we only used the Secu­ rity subscale. An initial study of the psychometric properties of the PAT-RS (based on the same materials as in the present study) indicated excellent internal consistency (Cronbach’s a > .90 across three raters) as well as close to excellent interrater reliability (ICC [2, 1] = .74; Cicchetti, 1994; Shrout & Fleiss, 1979) for this subscale (Lilliengren et al., 2014). Construct validity was also indicated by moderate correlations in the expected directions with measures of the alliance (the Helping Alliance Questionnaire [HAq-II]; Luborsky et al., 1996), developmental levels of internal representations (Differentiation Relatedness Scale [DRS]; Blatt & Auerbach, 2001), self-concept (Structural Analysis of Social Be­ haviour [SASB]; Benjamin, 2000), symptom severity (the Global Severity Index [GSI]; Derogatis, 1994), and global functioning (the Global Assessment of Functioning [GAF]; American Psychi­ atric Association, 2000). Further, Security correlated strongly (r = -.92) with an index based on all three insecure subscales in the PAT-RS, indicating that the subscale captures a general secureinsecure dimension of patients’ attachment to their therapist. Because the intake interviews in the YAPP were conducted before treatment had started, we were only able to assess the level of secure attachment to therapist from the interviews at termina­ tion. Two semistructured interview methods, the Private Theories Interview (PTI; Werbart & Levander, 2005, 2006) and the Object Relation Inventory (ORI; Auerbach & Blatt, 1996), provided in­ formation on the patients’ view of the therapeutic process as well as their experience of the therapist as a person. In the PTI, the patients are asked to elaborate their own ideas of what constitutes their problems, the background to those problems, and what could be of help. At termination, the patients were specifically asked about their retrospective view on what in their therapy contributed to change, what had been the obstacles, and what could have been different. In the ORI, the patient is requested to give a description of his/her mother, his/her father, him/herself, and the therapist. The

material for this study consists of answers to the ORI question, “Please give a description of your therapist.” The spontaneous response was followed by an “inquiry” in which the interviewer repeats adjectives or descriptive words mentioned and asks the patient to elaborate, for example, “You said understanding?” The PTI and ORI interviews were administered consecutively on the same occasion (shortly after termination) by 19 experienced psychotherapists and psychologists trained in the PTI and ORI techniques. All interviews were audio-recorded and transcribed verbatim. Three raters rated the interview material with the PAT-RS: the first author (main developer of the PAT-RS) and two master’slevel students who received training by the first author. Twelve of the 70 interviews were rated with consensus discussion among the raters (training and calibration), 37 interviews were rated indepen­ dently (single measure ICC = .74; average scores were used in the final variable), and the remaining 21 interviews were rated by a single rater (first author). Raters were kept blind with regard to outcome measures in the YAPP. Alliance. The HAq-II (Luborsky et al., 1996) is a self-report instrument with 19 items rated on 6-point Likert-scales ranging from 1 (/ strongly feel it is not true) to 6 (I strongly feel it is true). The HAq-II has demonstrated excellent reliability and good con­ vergent validity with other measures of the alliance (Luborsky et al., 1996). The Swedish version of the patient questionnaire showed excellent internal consistency (a = .91) in the whole YAPP sample (Lindgren, Werbart, & Philips, 2010). The HAq-II was distributed to patients after an agreement for therapy had been reached with the therapist (around the fifth session) and then repeatedly every third month. Because we were only able to measure secure attachment to therapist at termination and one of our questions concerned the relative impact of the alliance and secure attachment, we decided to use the last available alliance rating (i.e., closest to the measurement of secure attachment) to indicate the alliance. However, when no alliance rating was avail­ able within 3 months of the termination date, the data were regarded as missing (n = 9).

Outcome Measures Symptoms. Self-reported psychiatric symptoms were mea­ sured with the Symptom Checklist-90 (SCL-90; Derogatis, 1994; Derogatis, Lipman, & Covi, 1973). This questionnaire consists of 90 items referring to symptoms experienced over the last 7 days, rated on a 5-point Likert scale ranging from 0 (not at all) to 4 (very much). The GSI was used as an indicator of overall symptom load. The Swedish version of the SCL-90 has demonstrated adequate reliability and validity (Fridell, Zvonomir, Johansson, & Mailing Thorsen, 2002). In the present sample, excellent internal consis­ tency (a > .90) was observed across the measurement points used. Global functioning. The GAF (American Psychiatric Associ­ ation, 2000) was used to measure overall symptomatic and social functioning. The scale ranges from 1 to 100 with 10-point intervals describing discrete levels of functioning. The ratings were based on the total interview material, and a group of trained raters (different from the raters of the PAT-RS) did all assessments. To increase reliability, ratings were obtained through discussion among the raters until consensus was reached. The therapists were not involved in the rating of their own patients.

SECURE ATTACHMENT TO THERAPIST, ALLIANCE AND OUTCOME

Interpersonal problems. The level of interpersonal difficul­ ties was assessed with the Inventory of Interpersonal Problems (HP; Horowitz, Rosenberg, Baer, Ureno, & Villasenor, 1988). This inventory consists of 64 items self-rated on 5-point Likert scales ranging from 0 (not at all) to 4 {very much), and the sum of all items comprises a measure of overall interpersonal distress. The Swedish translation of the IIP has shown adequate psychometric properties (Weinryb et al., 1996), and the internal consistency for the total score ranged between .93 and .95 across the measurement points used in the present study.

Data Analysis Because about two thirds of the therapists treated more than one patient in this sample, a considerable amount of the data is nested within therapists. This violates basic assumptions regarding inde­ pendence of observations, and ignoring the possible influence of therapist effects may lead to faulty conclusions (Wampold & Serlin, 2000). In order to examine the possible presence of thera­ pist effects, we first used linear mixed-effects models procedures (Heck, Thomas, & Tabata, 2014) to separate out the variance due to differences between therapists from the total variance in each variable. This enables the calculation of an intraclass correlation (ICC; see Wampold & Serlin, 2000), which, in this context, may be interpreted as the percentage of the total variability that may be attributable to therapist differences. In terms of the independent variables, between-therapist variability in Security (as well as length of therapy) proved to be zero in this sample (i.e., ICC = .00). However, we found possible presence of therapist variability in patient-rated alliance (ICC = .13), indicating that about 13% of the variance in alliance ratings was due to differences between therapists, but the variance component at the therapist level was not significant (Wald Z test; z = 0.977, p = .329). We also examined the presence of therapist effects in each outcome at termination while controlling for score at intake, enabling the calculation of a conditional ICC for therapist differences in effec­ tiveness. These calculations indicated the possible presence of therapist effects for the GSI (ICC = .07), GAF (ICC = .05), and IIP (ICC = .10), respectively, suggesting that about 5%-10% of the variance in patient change may be attributable to betweentherapist variability. The estimates of the variance components at the therapists level were nonsignificant, however (GSI, z = 0.617, p = .537; GAF, z = 0.426, p = .670; IIP, z = 0.863, p = .388). Thus, our obtained ICCs for between-therapist variability in alli­ ance and outcome were in line with the literature on therapist effects in naturalistic studies (Baldwin & Intel, 2013), but the power was probably too low for the variance components to reach significance in this sample. Because even small amounts of between-therapist variability may lead to biased estimates (CritsChristoph & Mintz, 1991; Wampold & Serlin, 2000), we decided to proceed with our analyses within a linear mixed-effects models framework and include a random intercept at the therapist level, providing a more conservative test of our regression models. Additionally, because we were only able to measure secure attachment to therapist at termination, we decided to perform analyses separately for two time periods: intake to termination and termination to 1.5 years follow-up. Accordingly, the results of our within-therapy analyses (i.e., intake to termination) are referred to as associations, whereas the results for the follow-up period are

5

regarded as predictions. In order to address our three main research questions, we first modeled change for each time period using outcome at termination and follow-up as dependent variables while controlling for score at intake and termination, respectively. Next, Security was entered together with length of therapy as a covariate because, arguably, longer therapy may both lead to better outcomes and a more secure attachment to the therapist. This constituted “Model 1.” Lastly, the alliance was entered together with the other variables, constituting “Model 2.” Before the main analyses, all data were inspected following the recommendations of Tabachnick and Fidell (2013). One univariate outlier in IIP total score at termination was corrected to the next score within three standard deviations from the mean. Significantly skewed variables were either square-root or log-10 transformed depending on which adjustment improved normality most. After transformation, only the IIP at follow-up was still significantly, but not extremely, skewed (skewness = .624; SE = .306; z = 2.07; p = .039). In order to obtain more interpretable outputs given that some variables were transformed, we calculated the standardized |3 coefficients for each independent variable in our final models using the standard formula: (3 = B (SDJSDy). Primary signifi­ cance level was set to p < .05 and due to the primarily exploratory aim of the study (Bender & Lange, 2001), all tests were performed without any correction for familywise error rate (e.g., Bonferroni). All statistical calculations were performed with the SPSS (v. 19) software package.

Results Table 1 presents descriptive statistics and correlations for secure attachment to the therapist, alliance, length of therapy, as well as the outcome measures at intake, termination, and follow-up, re­ spectively. As expected, Security was positively correlated with the alliance (r = .47, p < .001), but the correlation is notably weaker compared with studies that have used patient self-reports to assess both constructs (.60-80; Bachelor et al., 2010; Lunsford, 2010; Mallinckrodt et al., 1995, 2005; Romano et al., 2008; Sauer et al., 2010). Of further note, secure attachment to the therapist at termination was unrelated to scores in outcome measures at intake. In contrast, the alliance was significantly associated with lower symptoms {r = -.26, p = .041) and fewer interpersonal problems (r = -.30, p = .023) at intake, suggesting that patients who were less distressed when entering therapy also reported more positive alliances toward termination. Also, length of therapy was signifi­ cantly related to higher scores on the IIP (r = .28, p = .023) at intake, indicating that patients with more interpersonal problems at start of therapy tended to receive longer treatments. Thus, although the patients’ initial level of distress seems related to both the alliance and length of therapy, it was not directly associated with the development of a secure attachment to the therapist in this sample. In terms of outcome at group level, effect sizes (Cohen’s d) indicated moderate to large within-group effects between intake and termination (GSI, d = 0.78; GAF, d = 1.32; IIP, d = 0.55) but no or only small effects during follow-up (GSI, d = 0.16; GAF, d = -0 .0 8 ; IIP, d = 0.17), mirroring the results of the YAPP project as a whole (Lindgren, Werbart, & Philips, 2010). Further, as expected, the outcome measures were both correlated with each other as well as autocorrelated across the assessment points. Still,

LILLIENGREN ET AL.

6

the associations vary considerably in strength, and the effect sizes suggest differential rates of change for each outcome. Therefore, we decided to analyze each outcome separately rather than col­ lapsing them into an underlying dimension of overall distress.

o 00

Is Secure Attachment to Therapist Associated With Improvement?

i *IY1*VO i

\0

in

The results of our linear mixed-effects model analyses for associations with outcome at termination are presented in Table 2. Our first question was addressed in Model 1, which indicated that Security was significantly associated with change in all three outcomes (GSI, (3 = -.2 9 ,p = .007; GAF, p = .45, p < .001; IIP, P = —.22, p = .024), supporting our first hypothesis. Of note, length of therapy was unrelated to change in all outcomes (GSI, P = —.02, p = .831; GAF, p = -.1 1 , p = .306; IIP, p = -.08, p = .451). The conditional ICC for therapist effects indicated that approximately 11% of the total variability under Model 1 might be attributable to differences between therapists, but the variance component estimates for therapist variability were not significant (GSI, z = 0.632, p = .528; GAF, z = 0.842, p = .400; IIP, z = 0.830, p = .407).

I I I

I I I

I I

I5 U

I I I

Does Secure Attachment to Therapist Relate to Outcome Once Alliance Is Controlled For? I I

Model 2 indicates that Security maintained its relationship with change in the GSI (P = -.28, p = .033) and GAF (P = .45, p = .002) when alliance was accounted for, but the association with reduction in the IIP was no longer significant (P = -.17,p = .160). Of note, the alliance was not significantly related to change in any outcome under Model 2 (GSI, P = -.13, p = .294; GAF, p = .18, p = .177; IIP, P = -.10, p = .454). Thus, our hypothesis that secure attachment to therapist would be more strongly associated with change compared with patient-rated alliance was supported. Therapist variability decreased somewhat when the alliance was entered in Model 2, but changes in the estimates must be inter­ preted with caution because the variance components estimates were nonsignificant under Model 1.

I I I

I I

* Ci)/3 13 X jD a

Does Secure Attachment to Therapist or the Alliance Predict Any Improvements After Treatment Termination?

a

NO o o

ON OO r-~

Secure attachment to therapist, alliance, and outcome in psychoanalytic psychotherapy with young adults.

Using a novel approach to assess attachment to therapist from patient narratives (Patient Attachment to Therapist Rating Scale; PAT-RS), we investigat...
8MB Sizes 2 Downloads 4 Views