|Home | About | Journals | Submit | Contact Us | Français|
Using sequential analysis, the authors examined how therapists' actions related to the verbal disclosure and defensive patterns that followed therapists' interventions within a single therapy hour for 20 patients. At the same time, a new measure, the Psychodynamic Intervention Rating Scale (PIRS), was tested for reliability and construct validity. Results indicated that therapists fit their styles of intervention to patients' levels of distress and functioning. Within the session, patient's emotional elaboration was followed by therapist's defense interpretation, followed by more patient emotional elaboration. Patient elaboration of significance was followed by more transference interpretation, followed by more patient elaboration of significance. Noninterpretive interventions were followed by patient's disclosure of facts, not emotion. Both interpretive intervention process sequences and therapist's use of support predicted posttreatment symptom reduction. The PIRS was shown to have satisfactory reliability and construct validity. (The Journal of Psychotherapy Practice and Research 1999; 8:40–54)
Most studies of therapists' interventions have related therapists' actions to symptom outcome. Fewer have looked at the direct relationship between what a therapist does and how the patient reacts immediately following an intervention—the micro-outcome. In this study using sequential analysis, we examine how therapist actions relate to the verbal disclosure and defensive patterns that follow therapists' interventions in a single therapy hour for 20 patients. We also study how therapeutic techniques effective on the micro level are related to initial pretreatment evaluations of patient distress and functioning and to symptom change measured 5 months after termination of a brief therapy for severe grief reactions. Finally, we demonstrate reliability and construct validity for a new measure of psychodynamic interventions.
Orlinsky and Howard's1 generic model of psychotherapy strongly emphasizes the working relationship between therapist and patient in positive change of pathological patterns. But researchers have yet to agree on which parts of the therapeutic situation are responsible for creating an effective working alliance. The generic model divides the psychotherapeutic system into six aspects of process that can be studied: therapeutic contract, therapeutic operations, therapeutic bond, self-relatedness, insession impacts, and therapeutic course.
The study presented here examined two of these aspects of process in relation to treatment outcome: therapeutic operations and insession impacts. Our main aim was to study the effectiveness of different kinds of therapist techniques used in brief dynamic psychotherapies, from the perspective of insession impacts and treatment outcome. To accomplish this, we created a therapist action measure, the Psychodynamic Intervention Rating Scale (Cooper and Bond, 1992, unpublished manuscript, University of California, San Francisco), for looking at therapists' interventions. Reliability and validity data for this scale are presented for the first time in this report. We also considered several patient characteristics, particularly a patient's level of pretreatment distress and behavior prior to a specific type of intervention.
Two types of methods have been used for measuring therapist technique: macroanalytic methods that examine therapist techniques across a session of therapy, and microanalytic methods that examine therapists' interventions at the level of a phrase or speaking turn. When the interest is to study the global relationship of technique to session or treatment outcomes, then the macroanalytic methods work well. But if a researcher is interested in studying the insession impact of different therapist strategies in relation to therapeutic outcome, then the problem is best approached by focusing on immediate impacts revealed in the therapeutic discourse.2–4
A number of therapist intervention measures examine the grammatical structure of the therapist's verbal behavior independent of the topic or content of speech.5 The most widely used of these is the Hill Counselor Verbal Response Modes Category System (HCVRMCS).6,7 This system achieves its aims of independence both from content of the therapeutic dialogue and from therapeutic perspective, and it has been successfully applied as a microanalytic method. But the seven categories of verbal response modes are not always the specific interventions that researchers wish to measure when studying the impact of therapists' interventions on patients' defense mechanisms, moment-to-moment therapeutic alliance, or changes in states of mind. For example, the HCVRMCS does not distinguish types of interpretive interventions, such as transference and nontransference interpretations; it also includes a category, confrontation, that in our experience has too much conceptual overlap with defense and transference.
Other therapist ratings scales, such as the Therapist Intervention Rating Scale of Piper et al.,8 do give more focus to differentiating dynamic or interpretive from noninterpretive interventions. But the complexity of this scale—with its division of dynamic interventions into single, double, triple, and quadruple components and further division of dynamic interventions into impulses, anxiety, defenses, and dynamic expressions—makes its application cumbersome and overly detailed for many purposes. The Inventory of Therapeutic Strategies developed by Gaston and Ring9 makes important distinctions between types of interpretive interventions, but it is a macroanalytic or “molar” method, designed to assess the overall frequency of use in a given session rather than to pinpoint an intervention and examine its immediate sequelae. Gabbard et al.10 rated interventions according to seven meaningful categories but reported no interrater reliabilities and applied their scale to only 3 patients. Thus, although many scales for measuring therapist interventions already existed, our interest in examining therapy process at the microanalytic level of therapist–patient interactions made it necessary to create a new categorical rating scale, one that would allow researchers of psychodynamic psychotherapy to study therapy process at the level of therapist interventions and patient responses.
Obviously, one cannot establish a linear relationship between the amount of a certain seemingly helpful intervention and the effectiveness of a psychotherapeutic treatment.11 Nevertheless, there is little doubt that “what is done by therapists and how they act, decisively affect the process patients pass through.” (Sachse,12 p. 274) On theoretical grounds, interpretive interventions have been assumed to be central in producing patient insight and change. But while most investigators report positive associations between successful outcome and interpretive interventions for short-term psychodynamic and patient-centered therapy,13–15 some researchers have failed to find a positive relationship with treatment outcome.16,17 Nevertheless, Orlinsky et al.18 conclude that the record for the relationship of interpretive interventions to outcome is one of the more consistently positive.
Evidence for the effectiveness of other kinds of therapeutic strategies has been less compelling. Techniques such as exploration, support, reflection, and clarification have not shown a predominantly positive relationship with treatment outcome across studies, but they also do not appear to be harmful.18 A more recent study19 contrasting psychodynamic-interpersonal and cognitive-behavioral therapists suggested, however, that although interpretive interventions were closely tied to client change for both orientations, reflection was linked with client change only in the psychodynamic therapies. This was despite the frequent use of reflection by therapists of both orientations.
More troubling have been the findings for transference interpretation. Henry et al.20 note that Freud cautioned against its misuse in cases where resistance had not yet had time to develop. Nevertheless, early pioneers of short-term psychodynamic therapy took the position that “more is better.”21,22 Although a few early studies had found a positive relationship between transference interpretations and outcome,23 more recent attempts to demonstrate this beneficial relationship have generally failed. McCullough et al.24 found that when a transference interpretation was followed by emotion, the outcome was positive, but when it was followed by patient resistance, it was not. McCullough et al. based their findings on large time sequences, so the possibility that patient defenses were elicited by their own insight-oriented disclosures was not examined.
Not only has frequency of transference interpretations failed to predict positive outcome, in many cases it has been related to poor therapeutic alliance and treatment outcome.25–28 For example, in a 5-year study of outcomes of time-limited psychotherapy, the Vanderbilt group found that therapeutic alliance deteriorated as transference interpretations increased; this was particularly true when such interpretations were measured later in the therapy, when presumably resistance would have had some time to develop.27 Gabbard et al.10 found that transference interpretations were the most potent in changing therapeutic collaboration, both increasing and decreasing it, with borderline patients.
These studies not only suggest that patient characteristics ought to be taken into account, but also raise the question of whether and how therapists “shape” their interventions to fit their patients' interactive or cognitive styles. In Beutler's model of eclectic psychotherapy,29,30 the therapeutic stance is seen as an integral plan of intervention tailored to a specific patient, instead of as a static quality. Beutler emphasizes tailoring treatment to a patient's specific style of interacting and to altering a treatment plan if the pattern of interaction changes during the therapy. Jones et al.31,32 studied 40 cases of stress response syndrome and bereavement. For patients with low pretreatment levels of disturbance, good outcome was associated with therapists using transference interpretations and focusing more generally on affect and the meaning associated with a patient's experiences. But for more disturbed patients, those with high levels of pretreatment disturbance, good outcome was associated with therapists offering supportive comments and explanations for a patient's specific problem behaviors.
The belief that transference interpretations should lead to positive outcome was based on the assumption that such interpretations should elicit more affect from patients regarding meaningful conflicts. But is this in fact true? Does any type of therapist intervention reliably elicit greater affective patient disclosure?33 McCullough et al.24 studied several types of therapist interpretations, including transference interpretations, and the subsequent responses of patients. When patients responded defensively, the association between therapists' interpretations and outcome was negative. It was only when therapists' interpretations were followed by affective responding on the part of the patient that positive associations with outcome were found. However, transference interpretations were twice as likely to provoke a defensive response on the part of the patient. In addition, the positive association with treatment outcome was not unique to transference interpretations, but held for nontransference interpretations as well. Henry et al.20 have concluded that “Transference interpretations do not seem uniquely effective, may pose greater process risks, and may be counter therapeutic under certain conditions” (p. 479).
The question of whether or not different strategies produce different outcomes is probably best resolved by focusing on immediate insession impact as revealed by the therapeutic discourse.2–4 Results from the Penn Psychotherapy Project showed that outcome was more successfully predicted from information about the communication between therapist and patient than from therapist or patient variables not including the therapeutic interaction.34
The primary purpose of this research was to study the link between therapists' interventions and patients' verbal responses within a session and treatment outcomes measured 5 months after termination of therapy. Although verbal responses alone do not reveal all aspects of the patient's emotional state, they were sufficient for rating on the patient process measure used in this study.
First, at the level of insession impacts, we predicted that although supportive strategies would be followed by disclosure, dynamic interpretive interventions would be followed by more emotional patient disclosures. We wondered if transference interpretations, in particular, would be followed by both more affective expression, but also more defensiveness.
Second, we hypothesized that post-therapy outcome would be better in patients who responded to therapists' interpretive interventions with deeper levels of disclosure and would be related to therapists' overall use of both supportive and interpretive therapeutic strategies.
Third, we also predicted that therapists would tailor their therapeutic strategies to accommodate individual patient characteristics, offering more support to poorer functioning patients and reserving interpretive strategies for better functioning patients.
Subjects were treated by faculty therapists in a university clinical research center. Subject status was assessed first at a pretherapy evaluation, then at the fourth therapy hour, and finally at a 5-month follow-up session, using a series of self-report and clinician ratings of distress and functioning. Psychotherapy process measures were derived from the transcribed text of the fourth therapy hour. Predictions about therapeutic outcome were assessed by relating process variables from the fourth therapy hour to measures of posttreatment distress and functioning. Predictions about specificity of therapeutic strategies were assessed by relating process variables from the fourth therapy hour to measures of pretreatment distress and functioning.
The subjects for this study comprised 20 cases of bereavement ranging from normal to pathological grief reactions. These subjects were randomly selected from a larger population of bereaved subjects, with the stipulation that they were matched for gender (all female) and age (mean = 40.2, range 26–52 years) across two patient groups: a treated parental bereavement group and a treated spousal bereavement group. The average educational attainment for the group was a high school education (Hollingshead and Redlich educational classification, mean = 4, SD = 1.78, range 2–7). Eighty percent of the subjects were white, 5% were Hispanic, and the rest did not indicate their ethnicity.
Subjects had been randomly selected from two archived samples of subjects (N = 31, all women, and N = 52, 50 women, 2 men; 10 subjects randomly chosen from each sample)35,36 involved in psychotherapy process outcome research for pathological grief reactions. The psychotherapy research subjects had been voluntarily recruited through public notices or from referrals by other mental health professionals. Exclusion criteria for the archived samples can be found in the published articles; generally these included past or present psychotic illness, previous psychiatric hospitalization, and history of alcohol or drug abuse with substantial impairment. As part of the research protocol, subjects received treatment in a university clinic research center for their bereavement reactions. After learning more about the studies, subjects who did not wish to participate or who fit the exclusion criteria were referred to other community therapists for treatment.
Nine faculty therapists with a mean of 8.8 years of experience after completing clinical training served as therapists for patients in this study. Four were clinical psychologists, 3 were psychiatrists, and 2 were psychiatric social workers; 4 were male and 5 were female. Eight of the patients were treated by the psychiatrists, 8 by the clinical psychologists, and 4 by the psychiatric social workers.
Prior to therapy, all subjects were interviewed by an independent clinical evaluator who did not provide the treatment for that subject but who was part of the faculty therapist team described above. The therapies were brief (12 sessions), time-limited, once-a-week dynamic psychotherapies for stress response syndrome. They were conducted according to a manual37 for brief treatment of stress response that drew on different therapeutic approaches.22,38 Approximately 5 months post-therapy, all subjects received a follow-up evaluation interview, which was generally conducted by the same evaluator who had done the initial evaluation interview.
Subjects and interviewers or therapists filled out batteries of rating scales at the first evaluation session; after sessions 4, 8, and 12; and after termination of the therapy. The evaluation interview occurred approximately 6 months after the loss (mean = 191.75 days; SD = 142.73). The fourth therapy hour was conducted approximately 1 to 2 months later (mean = 226.2 days; SD = 134.05), and the follow-up session 5 months after brief therapy termination. For the purposes of the present study, two symptom self-report measures and two clinician rating scales were chosen as measures of pretreatment distress and treatment outcome.
Therapy process measures were derived from the fourth therapy hour. The fourth therapy hour of the treated bereavement subjects was transcribed from archived audiotapes and scored for therapist's intervention, patient disclosure, and verbal defensiveness. Pilot work on different therapy hours from some subjects in the sample revealed that the fourth therapy hour had the full range of both disclosure and defensive categories. Research27 also indicates that by hour 5, therapist and patient have established a working therapeutic relationship. In fact, Sexton et al.39 found that therapeutic alliance was formed in the first session of a brief treatment and changed little thereafter. We recognize, however, that although we chose hour 4 as representative of the early sessions, it might not have been. It is possible that different patient–therapist pairs proceed to deepening at different rates. This limitation must be recognized.
Twenty sessions, one for each subject, were transcribed and segmented by thematic unit (TU). A TU consists of a single idea, usually one to a few sentences long. In our studies of TU segmentation, reliability agreement among 3 segmenters ranged from a conservative estimate of 0.77 to a less conservative estimate of 0.87.40 Patient speech was rated with our elaboration measure for disclosure and our dyselaboration measure for verbal defensiveness.41 Therapist speech was rated with a new measure, the Psychodynamic Intervention Rating Scale (PIRS), developed by Cooper and Bond (unpublished manuscript, 1992).
Operational definitions for the various categories of the elaboration and dyselaboration measures were developed through a series of pilot applications and were consolidated into content analysis manuals for training judges. Categories of elaboration and dyselaboration are briefly defined in Table 1. The reliability and validity of these scales has been studied by Horowitz and co-workers.41–43
Two teams of four clinical psychology graduate students rated either elaboration or dyselaboration. Elaboration and dyselaboration judgments were made on each TU. Each judge worked independently, and after scoring each session, the four judges in each team came together to reconcile differences. Reliability data were taken from the individual scores of the four raters before a consensus rating was assigned, but only consensus ratings were used to test the study hypotheses. A transcript averaged 476 TUs (range 380–613; SD = 69.9). To determine reliability of the two measures, 150 TUs were randomly selected so that 75 consecutive units were from the first half and 75 were from the second half of each therapy hour. Because some categories were rated more frequently than others, the marginal homogeneity of the interrater matrix could not be ensured. In such cases, Light's44 formula for generalized kappa is preferable to Cohen's kappa. All categories were entered into a single matrix to obtain an overall kappa for each measure by using Light's formula. The averaged kappa coefficients were 0.73 (86% agreement) for elaboration and 0.72 (84% agreement) for dyselaboration.
A number of interesting therapeutic intervention scales9,45 have been created in recent years, but very few of these are specific about the types of interpretive interventions. This limits their utility for examining microprocesses in expressive and psychodynamic therapies. The Psychodynamic Intervention Rating Scale (PIRS) was developed by Cooper and Bond (1992) to exclusively and exhaustively code each TU for all therapist utterances. (Gaston and Ring's9 Inventory of Therapeutic Strategies served as an important source of categories, but their interpretive interventions were simplified by limiting them to two categories.) More important, categories were defined in such a way that each therapist utterance (segmented into TUs) could be exhaustively classified. Interpretive interventions include defense and transference interpretations. Noninterpretive interventions include direct questions, clarifications, supportive and work-enhancing statements, associations, acknowledgments, reflections, and contractual arrangements. Definitions of each intervention and examples to help in discrimination are provided in a manual. In addition, there is a discussion of how to discriminate any given category of intervention from possibly overlapping or confounding categories. Categories of therapists' interventions are defined in Table 2.
Two raters (S.C. and M.B.) scored the transcripts independently for therapists' interventions and then arrived at consensus ratings by discussing differences until an agreement was reached. Interrater reliabilities were based on the independent assessments, whereas predictive data analyses were based on the consensus data. Interrater reliabilities were calculated with Light's kappa for each individual category and for the measure as a whole. The overall kappa was 0.85 (87% agreement) with a range from 0.83 to 0.99 across the categories. The high kappas most likely result from the fact that the inventors of the scale were also the raters. A great deal of prerating discussion and practice had left these two raters with a high degree of clarity as to the rules for categorizing interventions. For a truly adequate test of interrater reliability, the scale has to be taught to others. This phase is being carried out, but results are not yet available. Because this is a new measure, we present the means, standard deviations, and reliabilities for each category, based on the single therapy hour, in Table 3.
There were some areas of difficulty in making judgments about a therapist's actions. Difficult points of discrimination included the transition from a reflection, clarification, or association into a defense or transference interpretation. For example, the difference between an association such as “I remember that it was hot the day your brother was killed” and a defense interpretation such as “The hot weather today seems to remind you of the day your brother was killed, and I think that's why you've been so distraught all day,” can be confusing. The link to unconscious processes (i.e., the reason for the feelings being out of awareness) is necessary for a statement to be rated as an interpretation. A rater should wait to score the defense interpretation until the therapist actually comments on the process of keeping an affect out of awareness or on an affect itself that is out of consciousness. Transference interpretations are defined broadly, with the idea that even if the therapist is addressing nontransference aspects of the patient–therapist relationship, the intervention is qualitatively different from interpretations that do not include this relationship. Because it is often very difficult to claim that a comment on the “real relationship” has no transferential connotation, we made only one category for interventions addressing any aspect of the therapeutic relationship.
It can also be difficult to tell if a question is simply aimed at eliciting facts or if it is part of an interpretation. Unfinished interventions often create scoring difficulties. Another source of error arises when multiple interventions occur in one turn of therapist speech. The different interventions can be made clear by dividing the turn of speech into TUs.
In order to evaluate pretreatment, current, and posttreatment distress and functioning, we chose two self-report and two interviewer measures that have been widely used as outcome measures.
Subject self-report measures included the Depression subscales and the Brief Symptom Inventory (BSI), from the 53-item short form of the Symptom Checklist-90 (SCL-90).46 The Depression sub-scale was included because depression is one of the most commonly reported bereavement symptoms.47,48
Interviewer ratings included the Brief Psychiatric Rating Scale (BPRS)49 and the Global Assessment Scale (GAS).50 The BPRS was used with eight of the 16 items omitted (the eight items applicable to psychotic disorders were omitted, because psychotics had been excluded from our sample). The items used covered somatic concerns, anxiety, guilt feelings, tension, depressive mood, hostility, suspiciousness, and uncooperativeness. A total symptom score was derived by summing the ratings on these categories. The GAS50 is a rating of functioning in the last seven days on a continuum of mental health to illness from 1 to 100. The higher the rating, the better an individual's overall functioning has been judged.
A comparison of symptoms at the pretreatment evaluation and fourth therapy hour showed that clinicians' evaluations remained unchanged but self-report symptoms on the SCL-90 Depression scale and the BSI had declined (t = 2.40, df = 17, P < 0.05, and t = 2.89, df = 17, P < 0.01, respectively). Comparing pretreatment symptoms with those at 5-month follow-up indicated a significant reduction in symptoms on the SCL-90 Depression scale, the BSI, and the clinician-rated BPRS (Table 4). When Reliable Change Index51 scores were calculated for each measure, it was found that between 53% (SCL-90 Depression) and 67% (BPRS) of the sample showed clinical improvement between the pretreatment evaluation and posttreatment follow-up assessments.
The means, t-values, and percentages showing clinical improvement for each measure are presented in Table 4.
The proportions for each category of therapists' interventions were correlated with the assessments of patient functioning and level of distress measured at the pretreatment and first follow-up evaluations using two-tailed Pearson product-moment correlation coefficients or partial correlation coefficients. A limitation of the use of proportions is that it may fail to detect if there is an optimal proportion of any intervention for any individual subject.
A series of lag sequential analyses were computed to examine the therapy interaction process between therapist and patient. The analyses used therapists' interventions during the first to fifth TU (lag 1, 2, 3, 4, and 5) prior to a patient elaboration or dyselaboration. Each lag was a separate analysis. The SPSS-X log linear program was used.52 This type of analysis looks at the probabilities associated with first-order sequences for each observed therapist and patient sequence (contingent) given the observed frequency of each therapist and patient verbal category in the discourse (the base rate). The analysis yields an overall likelihood ratio chi-square (L2), which, if significant, indicates that some of the observed sequence cell frequencies are significantly different from the expected sequence cell frequency values. When the overall L2 for an analysis is significant, each observed sequence cell frequency can be evaluated further for its significance by reference to the adjusted residuals, which are an acceptable approximation to Z-scores53 for individual cell goodness of fit. Only dependencies with Z-scores equal to or greater than ±1.96 (P = 0.05) would be considered significantly different from expected sequence cell values. Because each analysis contains multiple Z-score assessments the values for the Bonferroni-adjusted Z-scores are also presented for each analysis. An effect size (ES) calculation based on a formula derived from Bakeman and Gottman54 (pp. 151–154) was then applied to each Z-score value of ± 1.96. Effect sizes were evaluated using Cohen's55 conventions for small (0.10), moderate (0.30), and large (0.50) effect sizes.
Therapists appeared to fit their style of intervention to a patient's level of distress and functioning. There was a trend for therapists to use support (r = 0.53, df = 19, P < 0.05) and work-enhancing strategies (r = 0.39, df = 19, P < 0.10). (We chose to include correlation coefficients with P = 0.10 because such values indicate a medium effect size that may have practical significance. The ratio of a type II to a type I error for N = 20 and r = 0.30 and P < 0.10 is 6, roughly comparable to the ratio of 7 obtained for r = 0.50 and P < 0.05 with the same N-value.56) with subjects self-reporting higher levels of symptoms (BSI) and to use questions with subjects self-reporting greater depression (r = 0.39, df = 19, P < 0.10). Similarly, the interviewer's evaluation of signs of distress (BPRS) appeared moderately associated with work-enhancing strategies (r = 0.37, df = 20, P < 0.10). In contrast, therapists were more likely to use defense interpretations with subjects receiving higher interviewer ratings of functioning on the GAS (r = 0.47, df = 20, P < 0.05).
Lag analyses of therapists' interventions were carried out for up to 5 patient thematic units following a therapist's interventions. Results from the lagged loglinear analyses indicated, that for all five lags, therapists' interventions were followed by significant changes in patient elaboration (maximum at lag 1 [n = 1,584], L2 = 170.30, df = 18, P < 0.01; minimum at lag 5 [n = 1,383], L2 = 36.66, df = 18, P < 0.01). (Since TUs were used as the sequential unit, we were not surprised that when we looked at the effects of patients' prior elaborations on their current elaborations, strong effects were obtained. Nevertheless, loglinear analyses that included both patients' prior elaborations and therapists' prior interventions showed that even after the effects of patient elaborations were accounted for, therapists' interventions still had a significant effect on elaboration for all but lag 4 analyses (maximum at lag 3, L2 = 66.73, df = 20, P < 0.01; minimum at lag 1, L2 = 32.60, df = 20, P < 0.05). Table 5 contains the significant Z-values and effect sizes for therapist interventions and patient elaboration sequences. Many effect sizes for these results were in the moderate to strong range, but because the number of TUs used in these analyses was quite high, a number of effect sizes for significant values were small. (The n of TUs represents the total number of contingent TUs summed across the 20 subjects.) Questions, associations, clarifications, and contractual therapist statements were interventions that were followed by patients conveying facts. The pattern for associations was the most stable of these, with significant results obtained for 3 out of the 5 lagged TUs.
In contrast, defense interpretations were followed by disclosure of emotion. This finding was stable, with significantly more emotional elaborations for up to 3 TUs after a therapist's defense interpretation. Defense interpretations also were followed by conveying significance and making connections between different topics and the self for lags 1 and 3. There is a decrease in conveying facts following defense interpretation, which is, of course, consistent with an increase in conveying significance and making connections. Effects of transference interpretations and work-enhancing strategies were also very stable. Results for transference interpretations were most stable. Transference interpretations were followed by conveying significance and making connections for up to five sequences after the therapist's intervention. Work-enhancing strategies were associated with conveying significance and making connections for lags 3 through 5.
Far fewer significant associations were obtained for dyselaboration, and these only appeared for the first TU following an intervention. Overall values for the effects of therapists' interventions on dyselaborations were only significant for lag 1 (n = 628; L2 = 40.87, df = 18, P < 0.01). Dyselaborations beyond filler and hedging are rare occurrences, so despite the general absence of significance, any effects of therapists' interventions on individual dyselaboration sequences are still of considerable interest. Only transference interpretations, associations, and work-enhancing strategies showed direct positive effects on dyselaboration. Patients moved away from emotional concepts immediately following therapists' associations (Z = 4.96, ES = 0.42). Transference interpretations (Z = 3.13, ES = 0.55) and work-enhancing strategies (Z = 2.00, ES = 0.38) were followed by dyselaborations that distorted the significance of the interpretation or intervention. Interestingly, transference interpretation was also negatively associated with filler and hedging dyselaborations (Z = –2.84, ES = 0.50). These Z-scores can be contrasted with a Bonferroni-adjusted Z = 2.88 for significance (P = 0.002). The effect sizes are all in the moderate to strong range.
The absence of clear effects on dyselaboration was puzzling. It seemed likely that because these types of events were relatively rare, and more especially because dyselaborations were likely used to dampen or mask emotion and avowed significance, the effects of therapists' interventions on dyselaboration might be mediated through elaboration. In order to examine this possibility, the effects of patient elaborations in the first through fifth TUs prior to a dyselaboration were studied. The significant results are presented in Table 6. Overall loglinear values were significant for all lag analyses (maximum at lag 1 [n = 2,308], L2 = 36.78, df = 4, P < 0.01; minimum at lag 5 [n = 2,241], L2 = 11.86, df = 4, P < 0.05). Significant effects were found for a number of sequences. Conveying only facts was followed by moving away from emotional aspects of a topic within 1, 2, 3, or 4 TUs. Emotional elaborations increased the probability of filler and hedges within 2 TUs. Conveying significance led to distorting dyselaborations within 2 and 3 TUs. The possibility, therefore, that therapist's interventions might have an indirect effect on patient dyselaboration by affecting patient elaboration could not be ruled out.
The context of a specific therapeutic intervention was analyzed by examining patient verbalizations prior to a therapist's intervention, using the same sequential approach. Results from the lagged loglinear analyses indicated that although therapists did not respond differentially to patient dyselaboration, they did respond reliably to patient elaboration with several types of interventions (maximum at lag 1 [n = 1,523], L2 = 97.41, df = 18, P < 0.01; minimum at lag 5 [n = 1,395], L2 = 43.85, df = 18, P < 0.01). Table 7 contains the significant Z-values and effect sizes for these analyses. The most consistent relationships were found between elaboration and interpretive interventions. Therapists responded to patients' emotional elaborations with defense interpretations and to patients' elaborations of significance with transference interpretations for all lagged sequences.
Outcome was assessed 5 months after the termination of therapy by patient self-report of symptoms on the BSI and the SCL-90 Depression subscale, and an interviewer's clinical ratings on the BPRS and GAS. On the basis of the literature reviewed above, we expected that greater use of dynamic interpretations and work-enhancing or support interventions would lead to better outcome on the 5-month follow-up measures. We did not have specific predictions for the other therapist intervention strategies. Patients self-reported lower follow-up symptoms on the BSI when therapists had used defense interpretations (r = –0.48, df = 15, P < 0.05), even when pretreatment symptom level was statistically controlled. Therapists had used reflection with subjects rated in the follow-up interview as lower in signs of distress (BPRS, r = –0.61, df = 15, P < 0.01) and higher in level of functioning (GAS r = 0.48, df = 16, P < 0.05) when pretreatment levels of distress or functioning were controlled.
The contingencies between each therapist intervention and patient elaboration were summed for TUs 1 through 5 and correlated as a proportion with patient outcome measures. Significant correlations were found between outcome measures and several therapeutic contingencies, some of which had been statistically reliable in the sequential analysis. All correlations reported below were controlled for pretreatment symptom levels. The sequential contingency between a therapist's defense interpretation and a patient's emotional elaboration had proved reliable in the sequential analysis. This contingency was significantly correlated with patient follow-up BSI scores (r = –0.61, df = 15, P < 0.01) and depression scores from the SCL-90 (r = –0.55, df = 15, P < 0.05). When the correlation between follow-up BSI and this contingency was controlled for the significant association between follow-up BSI and therapists' overall use of defense interpretation (as well as pretreatment BSI scores), the correlation dropped but still approached significance (r = –0.45, df = 14, P = 0.08).
Although the links between therapists' use of support strategies and patient disclosure were not reliable, the contingency between a therapist's use of support strategies and a patient's disclosure showed a number of significant associations with outcome. The contingency with patient emotional elaborations was significantly associated with follow-up BPRS scores (r = –0.78, df = 15, P < 0.01). The contingency between a therapist's use of support strategies and conveying significant elaborations was also associated with follow-up BSI and depression scores (r = –0.58, df = 15, P < 0.05, and r = –0.59, df = 15, P < 0.01, respectively). Significant associations were also found for the contingency between a therapist's use of support strategies and a patient's factual elaborations for the follow-up BPRS scale and GAS scales (r = –0.62, df = 15, P < 0.01, and r = 0.77, df = 16, P < 0.01, respectively). Finally, even though the contingencies of therapists' associations and patients' emotional and significant elaborations were very unlikely (see Table 5), they were negatively associated with follow-up GAS scores (r = –0.52, df = 15, P < 0.05, and r = –0.53, df = 15, P < 0.05, respectively).
Our study offers preliminary evidence that the Psychodynamic Intervention Rating Scale is a reliable measure of therapists' interventions and can be used to study the sequence of therapist–patient interaction at a micro level. The kappas measuring interrater reliability were sufficiently high to demonstrate that raters can agree on their categorizing of interventions.
Some construct validity for the scale has been demonstrated by the fact that when it was used in this process research, all of the findings made clinical sense.
First, the findings that therapists 1) tended to use more supportive strategies with patients showing more symptoms and distress and lower functioning and 2) tended to use more interpretive interventions with higher functioning and less distressed and symptomatic patients, are consistent with usual clinical practice.
Second, the findings 1) that patients' self-report of lower symptoms 5 months post-termination was associated with therapists' use of defense interpretations and support strategies and 2) that interviewer-rated outcome was significantly associated with reflection, also make clinical sense.
Third, transference interpretations were not significantly correlated with outcome in any way. This finding could be explained by the base rate being too low and the variance too high to achieve statistical significance. However, we will discuss other possible explanations later in this section.
The PIRS seems to be a reliable measure of therapists' interventions that is useful to study interactional sequences at a micro level. It is clear that more work has to be done on this scale. We must show that we can successfully train raters to achieve reliability. We must modify the manual as areas of confusion emerge with other raters. We must apply the scale to larger samples and to different types of samples. We must demonstrate that therapists using different orientations use a different profile of interventions. Further studies of predictive validity showing that specific interventions are followed by specific, anticipated consequences are also needed. This study should be considered a preliminary demonstration of the Psychodynamic Intervention Rating Scale, a measure that can offer certain advantages over other scales for psychodynamic process research but that requires further testing and refinement.
Our results support prior studies in suggesting that therapists do tailor their intervention strategies to a patient's level of distress and functioning. Therapists in our study used more supportive approaches rather than challenging the defensive organization of patients who were experiencing high initial levels of distress; in contrast, when a patient appeared to be higher functioning, therapists employed defense interventions. Implicit in this result is the possibility that therapists view defense and transference interpretations as potentially disrupting an intrapsychic or interpersonal equilibrium, a consequence that can be useful only when a patient is functioning well and experiencing manageable levels of anxiety or stress.
When the therapeutic process was examined at the micro level, the various intervention techniques were followed by differential responses on the part of patients. Interventions oriented toward asking specific questions, associating to the patient's comments, clarifying patient remarks, or addressing contractual aspects of the therapeutic situation were followed by fact-giving and information-giving on the part of the patient. On the other hand, actively interpreting patient disclosures was followed by more emotional discourse in which patients either shared emotions and significant realizations with the therapist or made insightful connections. It is interesting in this regard that the context in which therapists most consistently made such interpretive interventions was when patients conveyed emotions or made significant associations and insightful connections. These interactive sequences may represent particularly important moments in the therapy hour, when both patient and therapist are actively engaged in significant therapeutic work. Specific work-enhancing interventions also were followed by significant realizations and insightful connections on the part of patients. The stability of these results over TUs, even when a patient's prior discourse was taken into account, suggests a strong conclusion in this regard.
Support for direct links between therapists' interventions and dyselaboration was considerably weaker than for elaboration. This led us to wonder whether such effects might be mediated through elaboration. According to control process theory,57 this might be the case. As an individual struggles with difficult and emotional topics that challenge prevailing self-views, there are both a desire to disclose and examine these conflicting ideas and feelings and a desire to instantiate the existing self-schemas. Therefore, disclosures that may be overwhelming might subsequently be blunted by verbal retractions, misattributions, or misinterpretations. When this possibility was examined by looking at the effects of a patient's prior elaborations on subsequent defensive verbalizations, the results were highly suggestive. A patient's experience and disclosure of emotion were followed by verbal reactions that minimized their impact. Elaborations that conveyed the significance of disclosures and made insightful links among ideas and topics were followed by dyselaborations that distorted their significance. Therefore, it appears likely that as disclosures struck closer to the emotional core of a topic and became more transparent to a patient, the patient instituted controls that were reflected in the subsequent dyselaborative verbalizations.
Gabbard et al.10 found that in long-term psychotherapy with borderline patients, transference interpretations were high-risk but high-gain interventions and were most powerful in stimulating enhanced collaboration with the therapist. This finding may also reflect that transference interpretations work best when the patient is ready, primed by a supportive preparation, and is able to invoke a short defensive state to counterbalance the immediate anxiety stimulated by the interpretation.
Such a conception is consistent with Bales's58 equilibration model, which suggests that the therapeutic environment must both promote disclosure and allow for the reduction of anxiety, with therapist and patient participating in the equilibration process. But can strategies that promote this process be related to more positive therapeutic outcomes? On the macro level, the results were equivocal. Direct associations between therapists' interventions and patient self-report follow-up scales seemed to affirm a conclusion that therapists' strategies that interpreted defenses were related to symptom reduction 5 months after completion of therapy. Independent posttreatment clinician evaluations of symptom reduction and overall functioning, however, were related only to therapists' reflections. Reflection can be seen as an induction of working through the event, linking affects to memories or associations connected to the loss. Further, one might interpret this as a deepening intervention strategy: brief, noninterpretive statements with affective tone have been found to be helpful to patients in experiencing and expressing feelings and seem to have a much greater impact on patients' ability to understand therapists' intentions than do more complex interventions.59
When the therapist intervention and patient disclosure sequences studied at the micro level were related to posttreatment outcomes, the effects of interpretive interventions at the macro level were strengthened. When therapist use of dynamic conflict interpretations resulted in patients' disclosure of emotional material, patient self-report follow-up measures showed a reduction in symptoms. But therapist use of support strategies followed by any type of patient disclosure also predicted follow-up symptom reduction. This finding suggests that the support strategy itself, rather than a specific sequence, might be the active variable. An alternative explanation is that patients who elicit supportive behavior have a better response. Since therapists were most likely to use support strategies with patients who self-reported higher symptoms, it is likely that improvements occurred differentially within this patient group because direct relationships with support interventions and posttreatment outcome (with BSI r = 0.53, df = 19, P < 0.01) were found only when pretreatment symptom levels were not controlled.
The lack of significance regarding transference interpretations is interesting because many theorists22,37,60 consider transference interpretations to be the most powerful pool in the therapeutic armamentarium. However, our research might contain examples of what Frances and Perry33 describe concerning the range of relevance of transference interpretations. It is possible that in a sample of patients experiencing grief, the subjects benefit most by examining their relationships with the lost person and/or with people in their current situations with whom they have unresolved issues. These subjects may not develop very significant transferences in these brief therapies or may not find comments about their relationship with the therapist as relevant as comments about their difficulty in mourning the lost person and emotionally connecting to their living attachments and their current situations.
Our findings about the importance of supportive interventions are consistent with the findings of the Psychotherapy Research Project of the Menninger Foundation. Wallerstein61,62 studied processes and outcomes of psychoanalysis and psychoanalytic psychotherapy. Supportive mechanisms, in contrast to the analytical interventions, accounted for most of the successful outcomes. Even under the more nearly ideal condition of long-term treatment, the importance of the role described in psychoanalytic theory for interpretations designed to trigger patient insight was not found. Nevertheless, other studies that have looked at the effectiveness of interpretive interventions in short-term psychodynamic or client-centered therapies15,63 confirm our finding for short-term psychodynamic therapy, that these types of interventions can be related to positive therapeutic outcome. In our study, we have additionally demonstrated how interpretive interventions may function by promoting both more significant disclosures and control mechanisms that keep disclosures from overwhelming an individual.
Dean Sonneborn, M.A., and Clyde Sugahara, Ph.D. of UCSF provided statistical consultation. This research was supported by the Program on Conscious and Unconscious Mental Processes of the John D. and Catherine T. MacArthur Foundation. The third author is supported by the Swiss National Foundation (SNF8210-037067).