Patterns of Consistency and Deviation in Therapists' Countertransference Feelings

Abstract

The author addressed the question of consistency in psychotherapists' countertransference feelings. Research findings have indicated that the therapist's own personal feeling style may be more important than the patient's impact on the therapist's feelings. In this study, the feelings of 9 psychotherapists toward 28 patients were followed by using checklist self-report after each session during moderately long psychotherapies. ANOVAs and discriminant analyses showed that the therapists were very consistent in their feeling style over different patients and over time. The consistency in feelings toward the individual patients was smaller. Deviations from consistency are analyzed, and their importance for the understanding of different aspects of the countertransference is discussed. It is suggested that a meaningful use of the countertransference concept ought to be based on systematic identifications of recurrent and deviant patterns in the therapist's reactions.

The psychotherapist's emotional reaction is, for better or worse, an unavoidable aspect of every psychotherapy hour. Some therapists see it as a nuisance, some see it as a personal problem, and some choose to use it for reflection about the patient and the therapeutic situation. Authors in psychodynamic contexts encourage practitioners to use the potential information contained in the therapist's reactions, and some even encourage the communication of it to the patient. Reviews by Gabbard^1,2 and Sandler³ have noted that there is an emerging unity of opinion on the views about countertransference in the psychoanalytic literature. Reflection on the countertransference has become one of the basic tools of psychoanalysis and psychoanalytically informed psychotherapy.⁴ Schafer⁵ noted in 1992 that “we may install the analysis of countertransference alongside the analysis of transference and defensive operations as one of the three emphases that define a therapy as psychoanalytic” (p. 230).

Clinical authors have tried to distinguish the therapist's “normal reactions” from different varieties of deviating reactions. The less normal reactions have traditionally been denoted as countertransference. Some authors have described habitually or incidentally neurotic reaction patterns as typical of countertransference.^6–8 Others have described reactions to the patient's transference as being countertransference.^9,10 As a consequence of the elaboration of the concepts “projective identification”¹¹ and “enactment,”³ much theoretical and clinical work has been devoted to distinguishing between contributions to the countertransference from the therapist and from the patient.^12,13 Holmqvist and Armelius¹⁴ showed how contributions to the countertransference from the therapist, the patient, and the interaction between them could be estimated with two-way analyses of variance using data from inpatient units where each therapist reported feelings toward several patients and each patient evoked feelings from several therapists. It was found that the contributions to the countertransference by the therapist and the interaction were substantially larger than the patient contribution. The degree of helpfulness reflected in therapist feelings (e.g., therapist feels “helpful” versus “frustrated”) was more associated with the patient, whereas the degree of closeness in the feelings toward the patient (e.g., therapist feels “overwhelmed” versus “tired”) was more associated with the therapist.

Constructionist theories have emphasized the constructed nature of our experience. Beginning with Kelly,¹⁵ and continuing with theoreticians such as Gergen^16,17 and Shotter,^18,19 psychologists and philosophers of science have argued that our experience is largely constructed on the basis of our personal and social perspective. In psychoanalytic literature, this perspective has been advanced by Spence,^20–23 Schafer,^24,25 and Hoffman.^26,27 Using this perspective, the countertransference should be considered as a construction created by the therapist on the basis of his or her personal history in combination with impressions from the patient. Alternatively, the countertransference could be seen as socially constructed in the interplay between therapist and patient. In these perspectives, the question of normality in countertransference and the notion of definable contributions to the countertransference from therapist and patient lose much of their interest. Instead, the subjective evaluation of normality and deviance become emphasized. An empirical illustration of this view was given in Holmqvist,²⁸ where it was argued, based on empirical data, that it is inadequate to compare specific therapist reactions with some general norm. Instead, it is more meaningful to use the therapist's own habitual reaction pattern as the starting point for understanding processes in individual therapies.

Ogden^29,30 has described the unique and emerging nature of the psychoanalytic dialogue. The contemporary view of countertransference, as it has developed in the different variants of psychoanalytic theory, has moved toward an understanding of the countertransference as an interactional process.^31–42 The current conceptualization of countertransference thus emphasizes the interplay between therapist and patient in the creation of the therapist's (and the patient's) emotional reactions. More or less unconscious factors in both participants influence their feelings momentarily or permanently. It is the task of the therapeutic endeavor to discover as much as possible of the patient's influence on the therapist's reactions in the therapy, and for this, the therapist needs to know as much as possible about his or her own emotional world. Transference and countertransference, being based on the therapist's and patient's habitual emotional patterns, are largely unique to the therapeutic couple. Gabbard¹ notes that several psychoanalytic schools understand the analyst's countertransference as “a joint creation by patient and analyst. The analysand evokes certain responses in the analyst, while the analyst's own conflicts and internal self- and object-representations determine the final shape of the countertransference response” (p. 480; Gabbard's italics).

In an attempt to evade the complicated issue of normality in countertransference, Kernberg⁷ suggested that all therapist reactions should be encompassed in the wide concept “total countertransference.” This was an attempt to bypass the difficult discussion of whether some feelings in some situations should be called countertransference and others should not. The effect of this encompassing concept has been, however, that the conceptual delineation of different aspects of the therapist's reactions has been put aside. Although it could be argued that all therapist reactions should be called countertransference because it is impossible to differentiate between normal and neurotic, or between reactions that are primarily evoked by the patient's transference and reactions that are due to the therapist's own history, such a usage of the concept may jeopardize the experience of the specific character of exceptional therapist reactions that the word originally was intended to catch. Such a word usage will probably lead to a search for a new word for those therapist reactions that are experienced as unusual or noteworthy in some way. The reason for noticing this type of countertransference clinically is that it may contain information about important phases and changes in the therapeutic process. It is thus important to find ways of distinguishing exceptional reactions from more normal reactions.

Four types of deviating reactions in therapists' feelings can be conceptualized:

1. Habitual feelings in a therapist that conspicuously differ from other therapists' habitual reactions (“therapist-characteristic countertransference”).

2. Feelings that differ continuously or repetitively in a whole therapy with one patient when this therapy is compared with the therapist's other therapies (“patient-characteristic countertransference”).

3. Feelings in individual sessions that are different from the therapist's general and usual feeling pattern over different patients (“session-characteristic countertransference with respect to therapist”).

4. Feelings in individual sessions that are different from the therapist's overall feeling pattern toward the specific patient (“session-characteristic countertransference with respect to patient”).

The purpose of this study was to analyze therapist reactions to patients in order understand the significance of the four types of deviating reactions described above. To do this, the consistency in therapists' emotional reactions to patients was assessed with respect to their own habitual patterns as well as to patterns evoked by the specific patient. Further, the significance of sessions where the reactions of the therapist deviated from the studied therapist's average patterns was analyzed.

METHODS

Therapists and Patients

Nine psychotherapists who were working at a psychotherapy unit reported their feelings toward patients after each therapy hour. There were 4 male and 5 female therapists. The therapists were authorized according to the Swedish law for psychotherapists and had passed the 5-year training for psychotherapists. All of them had worked for more than 10 years with psychotherapy. Their mean age was 44 years. All had a psychodynamic orientation.

The 28 patients were referred for psychotherapy from a general psychiatric clinic. All patients had an Axis I diagnosis; 2 had also an Axis II diagnosis (1 each had narcissistic personality disorder and borderline personality disorder), 13 were diagnosed as depressed (7 with major depression, 6 with dysthymia), 11 had generalized anxiety disorder, 1 had a sexual disturbance, and 3 had adjustment reactions with mixed mood. The average Global Assessment of Functioning (GAF) score for the patients when the therapies started was 62.

Therapists averaged about 3.1 patients each (range 1–6). No control was exerted over the therapists' selection of patients, but no therapist had a special interest in any group of patients that manifestly influenced the selection. The patients chosen for participation were those that the therapists successively took on for regular psychotherapy after the date when the project started.

Instrument

The therapists reported their feelings on a feeling checklist.^43,44 The checklist was a revised version of a previously used feeling checklist and contained 48 feeling words. The original checklist contained 30 feeling words. The additions were based on suggestions made from therapists who had used it and found that it lacked some important feelings. The respondent is asked to answer on a 4-point scale (0, 1, 2, 3) the question “Together with patient X, I felt… [helpful, angry, etc.]” The internal consistency (alpha) for the whole checklist was 0.82 in this sample. The checklist is intended to be sensitive to changes in therapists' feeling states, but it has been found to have rather strong stability over short periods of time. The average correlation for individual feelings among 3 days was 0.60, among 3 months 0.49, and among 6 months 0.33.⁴⁵ The concurrent validity of the checklist has been tested in several studies.^46–48 It was meaningfully correlated with ratings of respondents' self-image in a study where staff with positive images of self and mother reported more positive feelings toward psychiatric patients.⁴⁴ In a study of feelings toward patients with different psychiatric diagnoses, it was possible to discriminate between the diagnosis groups by the feelings evoked in the staff.⁴⁸ The feeling checklist could also be used to predict treatment outcome for different diagnostic groups on the basis of the feelings that the staff reported early during the treatment process.⁴⁹ Table 1 shows the words in the checklist.

TABLE 1. Feeling checklist: mean score, standard deviations, and proportions of variance (η²_est) accounted for by therapists and by patients.

Statistical Methods

Two statistical methods were used: analysis of variance (ANOVA) and discriminant analysis. ANOVAs were used to study differences in feelings between therapists generally and between reactions toward individual patients. ANOVA was also used to analyze the proportions of variance in feelings attributable to therapists and to patients. The rationale for this method has been described in detail previously.¹⁴ It implies that the sums of squares of the between-groups factor (therapists, for instance) are divided by the total sum of squares. The quotient, eta-square (η²) indicates the proportion of variance accounted for by the differences between the therapists in relation to the total variance. The same procedure was used to assess the proportion of variance in therapist feelings attributable to differences between patients. But as there were only reactions to each patient from one therapist, the contributions to the total variance in feelings from the differences in the reactions to the individual patients were assessed in a hierarchical design with the patient factor nested within the therapist factor. As eta-square tends to give an overestimate of the factor's proportion of the total variance when N is high, a corrected eta-square (η²_est) was used.⁵⁰ This gives a more conservative estimate of the factor's contribution.

The discriminant analysis is a method used to test whether it is possible to discriminate groups of objects by a number of independent variables. In this study, the method was used to try to discriminate the therapists by the feelings that they reported and to discriminate the therapies (i.e., the treatment processes with the individual patients) by the feelings that were reported from them. The discriminant analysis also allows for an estimation of the certainty of the discrimination, and it makes it possible to identify checklists that do not fit into their “true” class. For this purpose, a test called “leave-one-out” is used. With this method, each case is classified into a group according to the discriminant functions computed from all the cases except the case that is being studied. Some cases will be classified into their true class, and others will be classified into wrong classes. The proportion of misclassified cases is the leave-one-out estimate of misclassification. Based on the classification of cases into groups, the discriminant analysis can also give a classification map called a “territorial map.” This map is similar to a factor-analytic object plot in that the independent variables (feelings) define the dimensions by their loadings. In contrast to the factor-analytic object plot, this map indicates groups of cases, with boundaries separating the groups from each other. A measure of the importance of each variable (feeling) in creating the dimensions is given by a coefficient called the discriminant function coefficient.

The ANOVA method assumes that the data have normal distribution, that the groups have homogeneous variances, and that the cases are independent. Discriminant analysis in addition requires that the within-group variance-covariance matrices are equal over groups. Inspection of the normal Q-Q plots (a method for assessing normality) for each feeling revealed only small deviations from normality. But Levene's test of homogeneity of variances indicated significant deviations from the assumption of equal variances for most of the feelings. And Box's M, a test measuring the equality of the variance-covariance matrices, indicated that the feelings significantly deviated from equality. In addition, the checklists cannot be considered to be independent, as each therapist may in some sense be influenced by the previous reports. Despite these problems, the methods were nevertheless used, as the intention was not to assess absolute values for differences or variance proportions, but to assess the relative influence of different factors on the therapists' feelings. The implications of the problems mentioned here are discussed at the end of the paper.

RESULTS

Therapist and Patient Contributions to Feelings

The 9 psychotherapists reported their feelings toward the 28 patients by completing, in all, 1,340 checklists. There were thus on the average 48 checklists completed for each patient (range 9–88) and an average of 149 checklists per therapist (range 27–320).

In order to analyze the proportion of the total variance in therapist feelings that was accounted for by average differences between the therapists, ANOVAs were performed for each feeling. Corrected eta-square values for each feeling were estimated as indicators of amount of variance accounted for by the main factor Therapist. In order to study the amount of variance added by average differences between the patients for each therapist separately, hierarchical ANOVAs were performed with the patient factor nested within the therapist factor. Table 1 reports the mean scores for each feeling, the standard deviation, and the amount of variance accounted for by the therapist factor and by the nested patient factor (see under Methods).

The most common therapist feelings (those with the highest mean scores) were present, interested, respected, receptive, sympathetic, relaxed, and helpful. The least common therapist feelings were cynical, hurt, threatened, embarrassed, afraid, disliked, and absent. The feelings with the largest variance were motherly, fatherly, strong, insufficient, overwhelmed, forceful, and heavy, and the feelings with least variance were cynical, threatened, hurt, embarrassed, afraid, disliked, and absent. The feelings where the average differences between the therapists accounted for the largest shares of the variance were fatherly, helpful, surprised, motherly, objective, strong, and suspicious. The feelings where the therapist factor accounted for the least proportion of variance were relaxed, cynical, locked-out, uncomfortable, hurt, tense, and embarrassed. For all feelings, the differences between the therapists were significant with P<0.0001. The feelings where differences between patients added most to the variance were bored, anxious, manipulated, suspicious, questioned, and locked-out, and the feelings where the patients added least to the variance were surprised, hurt, embarrassed, strong, helpful, vital, angry, rigid, and fatherly. The mean amount of variance accounted for by the therapist factor was 20%, with an addition of 8% for the patient factor. When all feelings were summed up to one scale, thus measuring the total amount of feelings, differences between therapist means accounted for 55% of the variance in feelings, and differences between patient means added 7%.

Consistency in Therapist and Patient Patterns

Three variants of discriminant analyses were made. In the first analysis, the question was asked whether the therapists could be discriminated by the feelings that they reported, irrespective of patient and session number. The discriminant analysis gave four discriminant functions with eigenvalues over 1.0. They accounted for 31.3%, 28.8%, 14.8% and 10.8% of the variance, respectively (cumulated value 85.7%) and had the canonical correlations 0.91, 0.88, 0.81, and 0.77. In order to test to what extent the therapists were consistent in their ways of completing the checklists, the cross-validation classification test “leave-one-out” was used. As described above, in this test a class model is created for each checklist, where the tested checklist is left out. The left-out checklist is then classified into the model (i.e., the particular therapist) where it fits in best. The result showed that 91.2% of the checklists were correctly classified (range 86.2% to 94.8% between individual therapists).

In the next analysis, an attempt was made to discriminate the reactions to the individual patients, irrespective of therapist, by the feeling they had evoked. Four discriminant functions with eigenvalues over 1.0 were created. They accounted for 26.3%, 21.7%, 12.5%, and 9.0% of the variance, respectively (cumulated value 69.4%), and had the correlations 0.90, 0.89, 0.82, and 0.77. In the classification test, 57.9% of the checklists were correctly classified as to patient (range 25.7% to 92.3% between individual therapies).

In a third set of discriminant analyses, the question was asked whether it was possible to discriminate between the therapies of different patients for each therapist separately. For all the therapists with more than one patient, discriminant analyses created significant functions that discriminated the therapies from each other. With the leave-one-out classification system, on the average 53.0% of the therapies were correctly classified as to patient (range 48.3% to 80.9% for individual therapies).

These analyses thus showed that it was possible to discriminate checklists from the individual therapists with high certainty, indicating that the different therapists had distinct feeling patterns overall. But each therapist's reactions toward his or her different patients were more varying. A way of expressing this is to say that the therapists were quite homogeneous in their reactions across patients and over time, and more heterogeneous in their reactions toward specific patients.

Common and Individual Perspectives

In order to explore this difference one step further, the results of discriminant analyses based on the checklists from all therapists were compared with discriminant analyses based on checklists from two of the therapists separately. Figure 1 shows the territorial map for the discrimination of all therapies. The territorial map shows the location of each therapy in comparison with the other therapies, based on the two first discriminant functions. On the map, heavy lines are drawn around each therapist's therapies. It should be observed that the discrimination was demanded for the individual therapies, irrespective of who the therapist was. The chances for the therapy areas to become located together in a specific way was thus not influenced in advance. The fact that for each therapist, all of his or her therapies were located together indicates the consistent nature in the therapists' feeling patterns toward different patients.

FIGURE 1.

Location of the reactions in the 28 therapies on a territorial mapTherapists are designated A through I, and each therapist's patients are numbered. Boundaries between therapists are marked with heavy lines.

The map shows that two therapists contrasted with each other. Therapist A's therapies are all to the left on the map, and therapist C's therapies are all to the right. The feelings loading strongly on the first function were fatherly (discriminant function coefficient=0.55), helpful (0.39), sympathetic (0.23), and strong (0.20), and on the second function, objective (0.37), motherly (0.31), receptive (0.30), strong (0.30), surprised (0.30), manipulated (0.26), and affectionate (0.24).

This means that both directions on the map were mainly unipolar. (The first dimension could be interpreted as measuring feelings of helpfulness and the second as measuring warmth, but the small number of therapists and patients make such general interpretations less interesting.) The upper right-hand corner was defined by many feelings, the lower left-hand corner by few feelings. Thus, therapist C reacted with more feelings than therapist A. This was validated by the fact that the mean frequency of all feelings among the therapists was lowest for therapist A (0.67) and highest for therapist C (1.06). This difference was significant at P<0.0001.

Therapist A:

The six therapies of therapist A are marked A1, A2, A3, etc., on Figure 1. They are all located to the left on the map. In this analysis, in which the data from all the therapies were used to create the discriminant functions, therapy A5 had an extreme location in relation to all other therapies and also in relation to therapist A's therapies. The therapy was extreme in both the horizontal and the vertical dimensions. It was characterized by a low level of feelings generally, and specifically by a low level of those feelings that were primarily used to create the two discriminant functions (fatherly, helpful, objective, motherly).

Figure 1 thus gives a picture of the locations of the therapies when they were all compared with each other, irrespective of who the therapist was. In contrast to this picture is a different set of territorial maps showing the location of the therapies when data from only one therapist were used. Figure 2 shows the results of a discriminant analysis using feeling checklists from only therapist A's therapies. The map shows the location of the therapies according to the two first discriminant functions. On this map, therapy A5 is located in the middle of the map. The first function primarily discriminates therapies A2 and A6. The second function discriminates therapy A4 from therapy A3. In this analysis, based on the reactions of only therapist A, the discriminating feelings in the first two functions were sad (0.63), anxious (0.41), confused (0.41), afraid (0.34), insufficient (0.27), and motherly (0.26), and for the second function happy (0.43), indifferent (–0.42), bored (–0.41), interested (0.41), enthusiastic (0.40), relaxed (0.39), and forceful (0.38). Thus, only one feeling, motherly, was important both in the general discrimination (Figure 1) and in the discriminant analysis made from only therapist A's therapies (Figure 2).

FIGURE 2.

Location of therapist A's reactions to her six patients, based on checklists from therapist A only

Asked about her memories after the therapies were finished, therapist A commented:

The common thing about therapies A2, A5, and A6 was that I reacted otherwise than I usually do. Normally, I tend to feel rather calm as a psychotherapist even if the patient tells about intensely emotional experiences or shows strong feelings. In therapies A2 and A6, I fell prey to strong countertransference feelings that made me more or less stuck. In therapy A2 I had strong feelings of contempt, whereas in therapy A6 I had strong sexual fantasies and fear of them. Usually I don't get this upset by such subjects. In therapy A5, on the other hand, the trouble was that the patient talked about intensely emotionally loaded situations involving hurt and desertion, without showing any emotional reaction. She was used to handling difficulties by not letting herself be touched. My attempts to reach her and to give her possibility to react were met with astonishment and confusion. The lack of congruence made me feel empty.

Therapist C:

Therapist C's five therapies on Figure 1 (C1, C2, etc.) were characterized by high scores on the feelings in function 1 (fatherly, helpful, sympathetic, and strong).

A separate discriminant analysis was made using reactions only from therapist C. In this analysis, the feelings suspicious (–0.46), frightened (–0.37), anxious (–0.34), overwhelmed (–0.33), manipulated (–0.31), disappointed (–0.26), and threatened (–0.25) were most important in the first function, and respected (0.33), cautious (–0.30), amused (–0.29), enthusiastic (0.28), disliked (–0.28), and sad (–0.24) in the second. The word fatherly had the value –0.13 in the first function and 0.18 in the second. Thus, although the word fatherly had a strong influence in discriminating the reactions of this therapist from the others' (0.55 in the first function in Figure 1), this word had no large influence in the discriminant analysis based on only her own reactions. None of the feelings that served to discriminate the reactions of therapist C from the other therapists' reactions was important in the discrimination between the therapies based only on her own feelings.

These comparisons thus show that the common territorial map using data from all therapies was quite different from the discriminant analysis maps based only on the therapist's own therapies. It seems as if each therapist creates his or her own feeling map for the therapies.

Identifying Deviating Reactions: Illustrations

The discriminant analyses aim at separating groups of objects (here therapists or therapies) from each other. If the analysis succeeds, it is also possible to identify those objects (sessions) that statistically do not belong to the proper group (therapist or therapy). The discriminant functions may predict for each object whether it belongs to the group model (i.e., therapist or therapy) or not. We can thus identify those sessions in the group that are unlike the others to that extent that the discriminant analysis predicts them to belong to some other group. This possibility was used here to further the understanding of the meaning of deviant session reports. As indicated previously, on the average about 91% of the checklists were correctly classified in the “leave-one-out” procedure according to therapist. The predictions according to therapy within therapist were correct, on the average, in 53% of the cases.

Therapy A2:

Therapy A2 lasted for 32 sessions. After 11 sessions, the therapy was broken off on the patient's initiative, to be started anew more than a year later and continue for another 21 sessions. For therapist A, 92.8% of the reactions were generally correctly classified as to therapist. In therapy A2, 78% of the sessions were correctly classified as to therapist. The 1st, 3rd, 5th, 6th, 7th, 10th and 11th sessions were misclassified as belonging to some other therapist. Thus, in the first phase of the therapy only 36% of the checklists were correctly classified. In the later part of the therapy only one session, number 28, was wrongly classified.

In order to get an idea of how these deviant sessions differed from the therapist's usual pattern, the six sessions in the beginning of the therapy that were incorrectly classified were analyzed for differences in frequency of feelings. In general, the therapist reported fewer feelings from these sessions. The general frequency of feelings was 0.52 in these sessions, versus 0.67 for all other sessions from this therapist (t=3.1, P<0.05). The following feelings had lower frequencies in the six misclassified early sessions in comparison with all other session reports from this therapist (all differences with P<0.001): affectionate, motherly, cautious, happy, insufficient, objective, disappointed, surprised, aloof, anxious, mute, confused, frightened, and rigid. For two feelings, there were higher frequencies in the deviating sessions: angry and uncomfortable (both with P<0.05).

When a discriminant analysis was made based solely on checklists from this therapist, thus discriminating between her therapies, on the average 44.3% of the sessions were correctly classified as to therapy. In therapy A2, 40% of the sessions were correctly classified. Among the first 11 sessions, 7 (64%) were correctly classified. In this perspective, the first phase of the therapy did not stand out as exceptional.

After the therapy's end, therapist A commented:

The start of therapy A2 was characterized by feelings of contempt for the patient. His narcissistic problems were striking, and his contempt toward others marked the sessions. He was not ready to reflect on his problems, and I felt insufficient when I could not help him or use my countertransference. I was totally locked in my feelings, and I stopped reflecting about the content of the therapy and about our relation. Then the patient stopped coming. He wrote a letter informing me that he was afraid of coming too close. He came back a year later, and then he could talk about his fear of contact and we could establish a working alliance. Although he still devalued other persons, he now was open for scrutinizing himself and his own share in relations that did not work. Now, I started feeling empathic with the patient and I could handle the situations when his contempt broke through. It was also possible to work with his feelings of superiority without feeling contempt for him.

The feelings reported from the deviant sessions in the beginning of the therapy, in comparison with the therapist's general feeling pattern (but not in comparison with the pattern in this therapy), was thus corroborated by the therapist's memory of the exceptional nature of this phase.

Therapy B1:

For therapist B, 87.5% of the sessions were correctly classified as to therapist. In therapy B1, going for 68 sessions, 12 sessions (18%) were wrongly classified as to therapist, namely sessions 1, 4, 5, 14, 21, 31, 47, 51, 55, 57, 62, and 65. Out of these, 7 sessions (5, 14, 47, 55, 57, 62, 65) were classified as belonging to the same (i.e., to another) therapist. Five of these sessions, from the later part of the therapy, sessions 47, 55, 57 ,62 and 65, were compared with all other session reports from this therapist. In comparison with these, the selected sessions were characterized by less feelings overall (0.48 vs. 0.71; P<0.05). Particularly, the therapist felt less forceful, strong, locked-out, confused, happy, angry, cautious, frustrated, rigid, amused, overwhelmed, disappointed, manipulated, uncomfortable, motherly and embarrassed (all differences with P<0.001).

In the analysis using only this therapist's checklists, on the average 68.8% of the sessions were correctly classified as to therapy. In therapy B1, 69.1% of the sessions were correctly classified. In the latter part of the therapy, a fairly large number of sessions were misclassified as to therapy, namely sessions 53, 56, 61, 62, 64, 66, 67, and 68. Thus, only session 62 was misclassified in both analyses. Out of the misclassified sessions, all but session 64 were classified as belonging to therapy B2. In order to understand the pattern, the average feelings in sessions 53, 56, 61, 62, 66, 67, and 68 were compared with all other sessions from the therapy. It was found that in these sessions, the therapist felt less angry, strong, forceful, objective, bored, and fatherly, and more relaxed and sympathetic (all differences with P<0.001).

Therapist B commented on the therapy afterwards:

The countertransference in this therapy was characterized by rather strong negative feelings, more than what is usual for me. During the first phase, I often felt irritated, sometimes angry, from time to time uninterested and sleepy. The patient for a long time resisted my attempts to make her interested in reflecting on the therapeutic relationship. Instead, she insisted on talking about emotional injuries that she had experienced. At the time of the 31st session, I noted that the patient for the first time was interested in a comment from me about our relationship. It felt as if it was the first time she noted that we had a working relationship. After this, there was a dramatic change. If the first part of the therapy was marked by feelings of irritation, feeling locked-out, and boredom, I now started to feel that I was rewarded for my work. There was much more warmth and mutuality in the relationship. During the finishing part of the therapy, I remember that I felt warm and close to the patient. When we parted, we both had the feeling that we had done a good job and that the patent had reached her treatment goals.

Thus, according to the checklists, therapist B differed from his general pattern mainly in several of the last sessions in this therapy, by feeling less forceful, strong, confused, and locked-out. In his memory, it was in the first part of the therapy that he differed from his usual pattern by feeling more irritated and less interested. He differed from the pattern set in this specific therapy in several other sessions in the later part of the therapy by feeling less angry, strong, and forceful, and more relaxed and sympathetic. This seems to be in accordance with his memory, where he also underlines the positive change during the later part of the therapy.

DISCUSSION

One aim of this study was to analyze the consistency in psychotherapists' feelings toward patients, both with regard to the general pattern for the therapist and with regard to reactions in individual therapies. Another was to understand the significance of different varieties of deviance from consistent patterns. It was found that the therapists' individual feeling styles were quite stable and consistent. Checklists could, irrespective of patient, to a large extent be correctly predicted according to which therapist they came from. The possibility of classifying feeling checklists according to which patient they referred to was more moderate. Based on these findings, several ways of understanding the significance of deviating reactions were explored.

It is important to be aware of the limitations in the results. The analyses were based on data from one psychotherapy unit. Different kinds of processes at such a unit may influence reaction patterns. Therapists' ambitions to develop personal profiles might inflate habitual differences in feeling patterns or feeling reports. It is also possible that the answers reflect personal styles of responding to the checklist rather than the “true” feeling. Another limitation pertains to the words in the feeling checklist. Although it has been revised several times in accordance with suggestions from therapists who insisted on specific feelings as being important in therapy relations, it should nevertheless be seen as a sample of possible feelings. Another sample might have given somewhat different results.

The reports from the therapists were taken at face value. No attempt was made to compare these reports with any other information about the therapist's feelings. Future studies on this topic might need to validate these subjective reports. Another obvious shortcoming is that there was no control of the selection of patients. All patients taken on for psychotherapy were included in the study, but the distribution of the patients to the different therapists was not controlled for. There may have been selection factors both of a chance nature and in the therapists' choice of patients that influenced the therapists' reactions. Replication of the study with a controlled design would strongly enhance the generalizability of the results.

As noted previously, another serious problem was the violation of several of the assumptions for the ANOVAs and the discriminant analyses. Some of these problems may be due to the small sample size, but they may also be inherent in these kinds of populations. Although the methods are rather robust concerning assumption violations, these problems imply that the results should be interpreted with caution. The most serious problem in this regard is the lack of independence between the session reports in the individual therapies. Although the subjective impression may be that each therapy hour evokes its own feelings, it is nevertheless a matter of statistical dependence. It should, however, be noted that the results are interpreted as indicating patterns of relationships between different factors that influence the countertransference and as identifying deviations from the patterns. No attempt is made to assess the absolute importance of these factors.

The central result about the pervasive consistency of therapists' feeling patterns confirms previous results, where it was found that psychiatric staff tended to be rather homogeneous in their personal styles in their reactions toward patients.¹⁴ In comparison, and also in congruence with previous results, the consistency in reactions to individual patients was weaker. Thus, reactions toward patients should primarily be seen as belonging within the emotional universe of the individual therapist. Within this universe there was moderate consistency in patterns of reactions toward individual patients, but there were also differences between sessions, the meanings of which need to be further analyzed.

The notion of unique emotional universes for the therapists was also corroborated by the comparison of discriminant analyses that were based on all checklists and analyses that were based on checklists from only one therapist. The patterns of relationships between the therapies for an individual therapist were quite different when the discriminant analysis was made for all therapies than when similar analyses were done using only data from each therapist separately. Therapist A, for instance, had one kind of deviating reactions in a therapy that differed from the others in the common discriminant analysis, and another kind of deviance in the therapies that were strongly discriminated in her own perspective. Therapy A5 differed in a way that underlined the difference between this therapist and the other therapists. The therapist commented that she tends to keep calm in most situations. The common analysis also showed that she tended to react with less feeling than the others. Therapy A5, where she felt “emptiness,” was an extreme example of this tendency. However, the discriminant analysis based on only her own reactions emphasized differences that within her own world were more important. These incongruencies point to the need to differentiate “How do I react in comparison with other therapists?” and “How do I react in comparison with myself?” as two quite different questions.

These findings were used in an attempt to shed some light on the complicated question of the informative value of the therapist's reactions. It was found that within the overarching influence of the therapist's style, deviating tendencies or deviating reactions in individual sessions could be regarded as indicators of important processes in the therapy. The statistical method does not give any information about the reason for the deviating reactions. What the method does achieve is to point out those reactions that deviate from the therapist's normal reaction pattern. The discriminant analysis does this without relying on the therapist to report that the reaction is unusual. In this way, it was possible to bypass one of the problems in studying countertransference. The therapist's consciousness of countertransference reactions as clinically described may vary, and methods of mapping countertransference that presuppose that the therapist reports the reaction as unusual may consequently be less apt. With the method presented here, this question is irrelevant. The deviating session reports may be perceived as unusual or not by the therapist. Their unusualness is detected by the statistical procedure. This became clear in the description of therapy B1, where it was found that the therapist tended to consider the latter part of the therapy as typical for his way of reacting emotionally, whereas the checklist data indicated that in this phase, his reactions often were less negative than what was usual for him.

In the introduction, different ways of conceiving countertransference were described. In therapist-characteristic countertransference, the therapist's general pattern deviates from other therapists' reaction patterns. Such differences between therapists were obvious in the results. Substantial proportions of the variance in feelings were associated with differences between therapists (Table 1). All therapies fell within therapist-specific areas in the common discriminant analysis (Figure 1). The discriminant analyses for individual therapists created other patterns between the therapies than the general discriminant analysis (Figure 2). Almost all checklists could be accurately predicted as to therapist. These findings point to the pervasive nature of the therapists' feeling patterns. They do not, however, say whether a specific reaction style is neurotic or not. It may well be, but a more constructive perspective would be to emphasize the advantage for each therapist in being aware of his or her own emotional style, regardless of ideas of a normal therapeutic emotional stance.

The second variant of habitual countertransference, the patient-characteristic countertransference, was less conspicuous. The average patient contribution to the variance in feelings was smaller. The percentage of correctly classified checklists with regard to patients in the discriminant analyses was also smaller. Thus, the therapists' reaction patterns in individual therapies were less constant than were their general patterns across patients. The third and fourth variants of countertransference described in the introduction, session-characteristic countertransference with respect to therapist and to patient, both implying deviations from recurrent patterns, were studied by picking out sessions that were incorrectly classified with respect to therapist and to patient. As the consistency in general therapist patterns was stronger than the consistency in individual therapies, it probably would be more meaningful to look for deviations from the therapist's general pattern than to look for deviations from the therapy-specific pattern. This was so in therapy A2, where the troublesome feelings in the start of the therapy were clearly unusual for the therapist, both in her own memory and in the discriminant analysis based on all therapies. The two types of deviation did, however, give different signals about important phases and developments in the therapies. In therapy B1, deviating sessions according to the two models were found to alternate with each other in the latter part of the therapy. The therapist's view that the latter part of this specific therapy was characterized by warmer feelings was congruent with checklist data. His view that the first part of the therapy was unusual in that it evoked angry and uninterested feelings was not, however, corroborated by the checklists. According to his memory, this phase was “session-characteristic with respect to therapist,” but this was not so according to the discriminant analysis.

The findings in this study thus suggest that it would be useful to delineate different variants of consistency and deviation within the therapist's total emotional reaction. The consistent pattern is the therapist's habitual feeling style, and the general pattern in the therapy is the patient-specific feeling style. In contrast to these, deviating reactions in individual sessions could be called “countertransference” in a more restricted sense. All four variants of therapist reactions may be useful for therapist reflections about processes in the therapy.

Consistency, as it is described here, implies stability over time. Deviation from the established pattern in an individual therapy implies that a stable pattern over time has been broken. Sometimes an individual session may stand out as different from the habitual pattern; sometimes a series of sessions may deviate in a trendwise fashion. The model outlined here makes it possible to develop analyses of the balance between stability and change in the therapist's reactions, in relation to, for instance, therapy outcome. It would also make it possible to develop techniques for giving individual therapists feedback on their reactions in individual sessions, in order to increase their perceptiveness regarding important countertransference developments.

The results in this study give arguments for considering a therapist's reactions as belonging primarily to his or her own emotional universe. Within this universe, some reactions deviate from the normal pattern and can meaningfully be focused for special consideration. Sandler's³ description of the countertransference as a combination of influences from both therapist and patient was not refuted by our results, but a better way of expressing the interplay leading to the therapist's emotional reaction would be to say that within the therapist's rather stable and unique emotional universe, different patients, some consistently and some occasionally, evoke different patterns of reactions.