Almost half a century ago, Kernberg published his seminal paper describing his view of the essential features of borderline personality organization—a broader construct than the DSM-defined borderline personality disorder (1). Among these features, he listed five defense mechanisms: devaluation, omnipotence, primitive idealization, projective identification, and splitting. Despite substantial interest by dynamically oriented clinicians, relatively little research has been conducted in this area in the ensuing decades. This gap has been due in large measure to the lack of reliable methods for assessing the presence of a range of defenses, or at least their conscious derivatives. In the past quarter century, only 10 studies have been published that attempted to delineate the mechanisms of defense used by borderline patients (2–10), and only eight of them sought to determine whether these defenses discriminate borderline patients from those with other diagnoses (2, 3, 5, 6, 8–10). Four of the studies (2, 6, 9, 10) relied on information obtained from videotaped clinical interviews rated according to reliable criteria developed by Perry (11). The other four (2, 3, 5, 8) used the Defense Style Questionnaire (12), a paper-and-pencil self-report measure developed by Bond and his colleagues that is designed to assess the conscious derivatives of unconscious mechanisms of defense.

Six of these eight cross-sectional studies found that borderline patients had significantly higher scores than axis II comparison subjects on scales measuring maladaptive action and image-distorting/borderline defenses (2, 3, 5, 6, 8, 9). One of the studies (3) found that female borderline patients had significantly higher scores than female axis II comparison subjects on the scale measuring adaptive defenses. Another (8) found that borderline patients had higher scores than axis II comparison subjects on self-sacrificing defenses as well as maladaptive action and image-distorting/borderline defenses.

The present study is an extension of the last cross-sectional study described above (8). It is distinguished by the large size of the patient groups being studied and the rigor with which their disorders were diagnosed. It is the first study to assess the presence of specific defenses in patients with criteria-defined borderline personality disorder and axis II comparison subjects longitudinally. It is also the first to use time-varying defense mechanism scores as a predictor of time to recovery from borderline personality disorder.

This study is part of the McLean Study of Adult Development, a multifaceted longitudinal study of the course of borderline personality disorder. The study methodology, which was reviewed and approved by the McLean Hospital Institutional Review Board, has been described in detail elsewhere (13). Briefly, all participants were initially inpatients at McLean Hospital in Belmont, Mass. Each patient was screened to verify that he or she was between the ages of 18 and 35; had a known or estimated IQ of 71 or higher; had no history or current symptoms of schizophrenia, schizoaffective disorder, bipolar I disorder, or an organic condition that could cause serious psychiatric symptoms; and was fluent in English.

All participants provided written informed consent after receiving a description of the study procedures. Each patient then met with a master’s-level interviewer blind to the patient’s clinical diagnoses for a thorough psychosocial and treatment history and diagnostic assessment. Four semistructured interviews were administered: the Background Information Schedule (14), the Structured Clinical Interview for DSM-III-R Axis I Disorders (15), the Revised Diagnostic Interview for Borderlines (16), and the Diagnostic Interview for DSM-III-R Personality Disorders (17). The interrater and test-retest reliability of all four of these measures has been found to be good to excellent (18–20).

At each of eight follow-up assessments conducted 24 months apart, psychosocial functioning and treatment utilization as well as axis I and II psychopathology were reassessed by interview methods similar to those used at baseline, by staff members blind to baseline diagnoses. After informed consent was obtained, our interview battery was readministered. The follow-up interrater reliability (within one generation of follow-up raters) and follow-up longitudinal reliability (from one generation of raters to the next) of the four instruments have also been found to be good to excellent (18–20).

Defensive style was measured with the Defense Style Questionnaire, an 88-item self-report measure that assesses for the presence of both defensive styles and specific defense mechanisms. It has been found to be internally consistent and to have criterion validity (12). Each item is rated on a 9-point Likert scale. Individual defenses are assessed with one to nine questions. We added three items to more fully measure the defense of emotional hypochondriasis, which we have described elsewhere (21). These three items (“No matter how often I tell people how miserable I feel, no one really seems to believe me”; “No matter what I say or do, I can’t seem to get other people to really understand how much emotional agony I’m in”; and “I often act in ways that are self-destructive to get other people to pay attention to the tremendous emotional pain that I’m in”) were combined with the three existing items to measure the related defense of help-rejecting complaining (“Doctors never really understand what is wrong with me”; “My doctors are not able to help me really get over my problems”; and “No matter how much I complain, I never get a satisfactory response”). The combined defense of emotional hypochondriasis was found to have a Cronbach’s alpha (measuring internal consistency) of 0.77, compared with an alpha of 0.64 for the defense of help-rejecting complaining.

Statistical Analysis

Data obtained from the Defense Style Questionnaire were assembled in panel format (i.e., multiple records per patient, with one record for each follow-up period for which data were available). Random-effects regression modeling methods assessing the role of group (borderline versus other personality disorder), time, and their interaction, and controlling for gender (as a significantly higher proportion of the borderline patients than axis II comparison subjects were female) were used in analyses of mean defense score data over time. If tests of group-by-time interactions were not significant, indicating that the patterns of change were the same for both groups, these analyses were rerun with main effects of group and time only. Because these defense scores were positively skewed, they were logarithmically transformed prior to modeling analyses to achieve more symmetrical distributions. Because analyses are based on logarithmically transformed scores, the results are interpreted in terms of relative, rather than absolute, differences. Given the large number of comparisons, we applied the Hochberg correction (22) for multiple comparisons. Finally, for administrative reasons related to funding, the Defense Style Questionnaire was administered to only a subset of patients at the 2- and 4-year follow-up assessments. As a result, these data were collected for 135 of 342 participants (106 with borderline personality disorder and 29 with nonborderline axis II diagnoses) at the 2-year follow-up and 120 of 333 participants at the 4-year follow-up (97 and 23, respectively). A multiple imputation procedure (with 10 imputations of missing defense data) was used to conduct analyses that included the observed 2- and 4-year follow-up data. The imputation procedure incorporated both group and baseline and follow-up Defense Style Questionnaire data as predictors of the missing defenses data.

Discrete time survival analyses were used to assess the relationship between the 19 defense mechanisms studied and the outcome of recovery from borderline personality disorder. This outcome has previously been defined as concurrent symptomatic remission from borderline personality disorder and good social and vocational functioning (23, 24). Good social and vocational functioning has been defined as having at least one emotionally sustaining relationship with a friend or partner and vocational performance that is consistent, competent, and full-time (including work as a houseperson). Time-varying values for the defenses were used in these analyses. These values were not transformed for ease of interpretation. Each defense mechanism was assessed individually, and those that were significant were then entered into a multivariate survival model. Using a backward deletion method, the most parsimonious model for predicting recovery was obtained.

A total of 290 patients met criteria for borderline personality disorder according to both the Revised Diagnostic Interview for Borderlines and DSM-III-R, and 72 met DSM-III-R criteria for at least one nonborderline axis II disorder (and neither set of criteria for borderline personality disorder). The following primary axis II diagnoses were found for these comparison subjects: antisocial personality disorder (N=10, 13.9%), narcissistic personality disorder (N=3, 4.2%), paranoid personality disorder (N=3, 4.2%), avoidant personality disorder (N=8, 11.1%), dependent personality disorder (N=7, 9.7%), self-defeating personality disorder (N=2, 2.8%), and passive-aggressive personality disorder (N=1, 1.4%). Another 38 comparison subjects (52.8%) met criteria for personality disorder not otherwise specified (which was operationally defined in the Diagnostic Interview for DSM-III-R Personality Disorders as meeting all but one of the required number of criteria for at least two of the 13 axis II disorders described in DSM-III-R).

Baseline demographic data for the sample were reported previously (13). Briefly, 77.1% (N=279) of the participants were female and 87% (N=315) were white. The participants’ mean age was 27 years (SD=6.3), their mean socioeconomic status rating was 3.3 (SD=1.5) (where 1=highest and 5=lowest) (25), and their mean Global Assessment of Functioning score was 39.8 (SD=7.8) (indicating major impairment in several areas, such as work or school, family relations, judgment, thinking, or mood). Data on co-occurring axis I and II disorders at baseline and over 6 years of prospective follow-up for both study groups were reported previously (26, 27).

Attrition was relatively low. A total of 275 borderline patients and 67 axis II comparison subjects were reinterviewed at the 2-year assessment, 269 and 64 at the 4-year assessment, 264 and 63 at the 6-year assessment, 255 and 61 at the 8-year assessment, 249 and 60 at the 10-year assessment, 244 and 60 at the 12-year assessment, 238 and 59 at the 14-year assessment, and 231 and 58 at the 16-year assessment. All told, 87.5% (N=231/264) of the surviving borderline patients (13 died by suicide and 13 of other causes) were reinterviewed at all eight follow-up waves. A similar participation rate was observed for the axis II comparison subjects, with 82.9% (N=58/70) of surviving patients in this group (one died by suicide and one of other causes) reassessed at all eight follow-up waves.

Table 1 contains information related to the four defense styles derived through factor analyses of the items of the Defense Style Questionnaire (12). Mean scores for adaptive defenses, self-sacrificing, image-distorting, and maladaptive action defenses are reported for both study groups. Borderline patients had significantly higher scores than axis II comparison subjects on the two lower-level defensive styles: image-distorting ([1.15–1]×100%=15% higher) and maladaptive action ([1.21–1]×100%=21% higher) defenses, which are similar but not identical to the image-distorting and immature defense levels described below. Both groups exhibited a significant increase in the mean score for the adaptive style ([1.05–1]×100%=5% higher) (which is similar to the mature defenses described below) and a significant decrease in the mean score for image-distorting ([1–0.83]×100%=17% lower) and maladaptive action defenses ([1–0.86]×100%=14% lower).

Tables 2, 3, and 4 contain information related to mature, neurotic, and immature defense mechanisms as defined by Vaillant’s classification system (28), and Table 5 contains information related to the image-distorting or borderline defenses as defined by Kernberg (1). Table 2 details Defense Style Questionnaire scores for mature defenses over time for both study groups. Suppression scores were significantly lower for borderline patients than for axis II comparison subjects ([1–0.89]×100%=11% lower). In terms of change over time, anticipation scores increased significantly for those in both groups by 11% ([1.11–1]×100%).

Table 3 details Defense Style Questionnaire scores for neurotic defenses over time for both groups. No between-group differences were observed for isolation or reaction formation. However, both groups had significantly lower scores on isolation over time ([1–0.77]×100%=23% lower). For the defense of undoing, the relative difference of 1.41 for diagnosis indicates that the mean Defense Style Questionnaire score reported by the borderline patients at baseline was approximately 40% larger than the corresponding mean for axis II comparison subjects. The significant interaction between diagnosis and time indicates that the relative decline from baseline to 16-year follow-up is approximately 22% ([1–0.97×0.80]×100%) for borderline patients, in contrast to the nonsignificant 3% decline for axis II comparison subjects.

Table 4 details Defense Style Questionnaire scores for the six immature defenses assessed. There were no between-group differences for either denial or fantasy. However, both groups reported a significant decline in the mean scores for denial (5% lower) and fantasy (24% lower) over time. There were significant between-group differences for the defenses of emotional hypochondriasis, passive aggression, and projection, with borderline patients reporting significantly higher scores of 40%, 16%, and 23%, respectively. Both study groups also underwent a significant decline in mean scores for these three defenses over time (declines of 26%, 15%, and 16%, respectively). In addition, a significant baseline difference was observed for acting out, with scores for borderline patients at study entry 43% higher than those for axis II comparison subjects. The significant interaction between group and time indicates that the relative decline in acting out from baseline to 16-year follow-up was approximately 28% ([1–0.87×0.83]×100%) for borderline patients, in contrast to the nonsignificant 13% decline for axis II comparison subjects.

Table 5 details Defense Style Questionnaire scores for the five image-distorting or borderline defenses assessed. Both groups experienced a significant decline over time in mean scores for each of these defenses except primitive idealization. These declines ranged from a low of 16% (splitting) to a high of 31% (projective identification), with devaluation (22%) and omnipotence (21%) occupying a midrange position. In addition, borderline patients reported significantly higher mean scores for the defenses of projective identification (20% higher) and splitting (23% higher).

These analyses were rerun after removing comparison subjects who had a diagnosis that would fit within the borderline personality organization construct (N=16). The results for these image-distorting/borderline defenses were basically the same as when these near-neighbor subjects were included in our comparison group.

While the relative differences described above are a form of effect size, we also calculated Cohen’s d (29) for the group effect (comparing the borderline patients with the axis II comparison subjects over time) for the four styles and 19 defenses studied. We found a large effect size for the maladaptive action style (0.80) and medium effect sizes for two defenses, acting out (0.64) and emotional hypochondriasis (0.58). The remaining effect sizes were small.

We next assessed significant multivariate time-varying predictors of time to recovery from borderline personality disorder—an outcome achieved by 60% of borderline patients by the time of the 16-year follow-up (24). We found that four time-varying defenses (of 14 that were significant in bivariate analyses [all but altruism, anticipation, sublimation, reaction formation, and omnipotence]) were significant multivariate predictors of time to recovery: humor (hazard ratio=1.18, SE=0.07; z-score=2.62, p=0.009, 95% CI=1.04–1.33), acting out (hazard ratio=0.81, SE=0.06; z-score=–2.90, p=0.004, 95% CI=0.71–0.94), emotional hypochondriasis (hazard ratio=0.82, SE=0.08; z-score=–2.01, p=0.044, 95% CI=0.68–0.99), and projection (hazard ratio=0.64, SE=0.10; z-score=–2.79, p=0.005, 95% CI=0.47–0.88).

Humor predicted a shorter time to recovery, with an 18% greater chance of recovery for each 1-point increase in score for humor. The three immature defenses predicted a longer time to recovery. For each 1-point increase in acting out, emotional hypochondriasis, and projection, the chances of recovery declined 19%, 18%, and 36%, respectively.

Three main findings emerge from this study. The first is that patients with borderline personality disorder had significantly higher scores over time than axis II comparison subjects on two lower-level defensive styles (image-distorting and maladaptive action) and seven specific defenses. One of these defenses (undoing) was neurotic according to Vaillant’s classification system, and four were immature: acting out, emotional hypochondriasis, passive aggression, and projection. All four of these defenses underlie clinical features (impulsivity, demandingness, masochism, and suspiciousness) that have been found to be extremely common among borderline patients (30). However, only demandingness has been found to be specific for the disorder (30).

Two image-distorting/borderline defenses were also found to discriminate borderline patients from axis II comparison subjects. Borderline patients had significantly higher mean scores on the defenses of projective identification and splitting than axis II comparison subjects. Of equal importance is that three other image-distorting/borderline defenses were not found to discriminate borderline patients from axis II comparison subjects: devaluation, omnipotence, and primitive idealization. Taken together, these results are consistent with the earlier findings of Perry and Cooper (6), who found that what they termed borderline defenses (projective identification and splitting) were strongly associated with borderline psychopathology, while what they termed narcissistic defenses (devaluation, omnipotence, and primitive idealization) were not. This finding held whether comparison subjects with a borderline personality organization diagnosis were included in or excluded from the analyses.

The second main finding is that borderline patients experienced significant improvement on three of the four styles and 13 of the 19 defenses studied. They had significantly higher scores over time on the adaptive style and one mature defense (anticipation) and significantly lower scores on two lower-level styles (image-distorting and maladaptive action) and two neurotic defenses (isolation and undoing). They also had significantly lower scores over time on all immature defenses and all image-distorting/borderline defenses except primitive idealization.

The significantly higher score on the mature defense of anticipation was relatively small (11%) and may not signify much clinically meaningful change. The significantly lower scores on the neurotic defenses of isolation and undoing were somewhat more robust (23% and 22%). In terms of significant improvement in immature defenses, denial saw a decline of only 5%. The other five immature defenses had larger declines: acting out (28%), fantasy (24%), emotional hypochondriasis (26%), passive aggression (15%), and projection (16%). Four defenses showed significant improvement (substantial declines) in image-distorting or borderline defenses: devaluation (22%), omnipotence (21%), projective identification (31%), and splitting (16%).

Looking at these data synthetically, it appears that the borderline patients were functioning in a more adaptive manner on all four levels of defense mechanisms studied. This improvement was the least robust in the mature defenses, which Vaillant (28) has described as often being mistaken for convenient virtues. Here, improvement for anticipation, while significant, was only 11%.

More robust change was seen in neurotic, immature, and image-distorting or borderline defenses. Eleven of the 14 defenses in these categories were found to have undergone a significant decline of 15% or more, seven a significant decline of 20% or more, and three a significant decline of 25% or more (acting out, emotional hypochondriasis, and projective identification).

The third main finding is that four time-varying defense mechanisms were found to be significant predictors of time to recovery from borderline personality disorder. It is not surprising that three of these defenses were immature according to Vaillant’s classification system: acting out, emotional hypochondriasis, and projection. Clearly, continued impulsivity, unremitting complaints of being misunderstood, and chronic distrust and suspiciousness would interfere with good social and vocational adjustment. However, the fact that humor predicts a shorter time to recovery is an unexpected finding. It may be that humor, which requires a well-functioning observing ego, paves the way for a more flexible and mature psychosocial adjustment.

Currently, there are six evidence-based treatments for borderline personality disorder (31–36). However, most clinicians do not use any of these treatments, because of their complexity and cost. The main goal of clinicians is to help their borderline patients move ahead in a more adaptive manner, and in this supportive effort, they will use both dynamically and behaviorally informed strategies, such as clarifications, appropriate confrontations, and skills coaching.

These clinicians could use the defensive functioning of their borderline patients to track their symptomatic and psychosocial progress over time. The advantage of such an approach is that tracking defensive functioning fits into a number of psychodynamic frameworks (i.e., ego psychology, object relations theory, self psychology) that can help guide treatment, while viewing each act of impulsivity or each insistent and persistent demand that attention be paid to one’s inner pain as a separate and somewhat surprising event can lead clinicians to feel unnecessarily discouraged or even nihilistic.

This study has several limitations. The most important of these is that using a self-report measure to assess defense mechanisms over time yields clinically less rich information on defensive functioning than Vaillant’s longitudinal vignette method or Perry and Cooper’s method of videotaped clinical encounters. In addition, participants may provide socially acceptable answers that are not consistent with their actual defensive functioning. For example, the highest mean baseline scores reported by both study groups were for the adaptive style. Another limitation is that because the participants were all inpatients at study entry, our results may not generalize to healthier outpatients or nonpatients with borderline personality disorder. In addition, a substantial percentage of our participants were in nonintensive outpatient treatment over time (37). Our results might be different from those for an untreated sample or a sample that had been treated with an empirically based treatment for borderline personality disorder rather than the treatment as usual received by the vast majority of our participants.

Taken together, the results of this study suggest that the longitudinal defensive functioning of borderline patients is distinct and improves substantially over time. They also suggest that immature defenses are the best predictor of time to recovery.