What Do Individual Differences in Asymmetric Prefrontal Activation Reflect?

This section will present a brief overview of recent work from my laboratory that was designed to examine individual differences in measures of prefrontal activation and their relation to different aspects of emotion, affective style, and related biological constructs. These findings will be used to address the question of what underlying constituents of affective style such individual differences in prefrontal activation actually reflect.

In both infants (Davidson & Fox, 1989) and adults (Davidson & Tomarken, 1989) we noticed that there were large individual differences in baseline electrophysiological measures of prefrontal activation and that such individual variation was associated with differences in aspects of affective reactivity. In infants, Davidson and Fox (1989) reported that 10-month-old babies who cried in response to maternal separation were more likely to have less left-sided and greater right-sided prefrontal activation during a preceding resting baseline compared with those infants who did not cry in response to this challenge. In adults, we first noted that the phasic influence of positive and negative emotion elicitors (e.g., film clips) on measures of prefrontal activation asymmetry appeared to be superimposed upon more tonic individual differences in the direction and absolute magnitude of asymmetry (Davidson & Tomarken, 1989).

During our initial explorations of this phenomenon, we needed to determine if baseline electrophysiological measures of prefrontal asymmetry were reliable and stable over time and thus could be used as a trait-like measure. Tomarken, Davidson, Wheeler, and Doss (1992) recorded baseline brain electrical activity from 90 normal subjects on two occasions separately by approximately 3 weeks. At each testing session, brain activity was recorded during eight 1-min trials, four trials with eyes open and four with eyes closed, presented in counterbalanced order. The data were visually scored to remove artifact and then Fourier-transformed. Our focus was on power in the alpha band (8-13 Hz), although we extracted power in all frequency bands (see Davidson, Chapman, Chapman, & Henriques, 1990, for a discussion of power in different frequency bands and their relation to activation). We computed coefficient alpha as a measure of internal consistency reliability from the data for each session. The coefficient alphas were quite high, with all values exceeding .85, indicating that the electrophysiological measures of asymmetric activation indeed showed excellent internal consistency reliability. The test-retest reliability was adequate with intraclass correlations ranging from .65 to .75, depending upon the specific sites and methods of analysis. The major finding of import from this study was the demonstration that measures of activation asymmetry based upon power in the alpha band from prefrontal scalp electrodes showed both high internal consistency reliability and acceptable test-retest reliability to be considered a trait-like index.

The large sample size in this reliability study enabled us to select a small

P.94

group of extreme left and extreme right-frontally activated subjects for magnetic resonance (MR) scans to determine if there existed any gross morphometric differences in anatomical structure between these subgroups. None of our measures of regional volumetric asymmetry revealed any difference between the groups (unpublished observations). These findings suggest that whatever differences exist between subjects with extreme left versus right prefrontal activation, those differences are likely functional and not structural.

On the basis of our prior data and theory, we reasoned that extreme left and extreme right frontally activated subjects would show systematic differences in dispositional positive and negative affect. We administered the trait version of the Positive and Negative Affect Scales (PANAS; Watson, Clark, & Tellegen, 1988) to examine this question and found that the left-frontally activated subjects reported more positive and less negative affect than their right-frontally activated counterparts (Tomarken, Davidson, Wheeler, & Doss, 1992; see figure 5-1). More recently with Sutton (Sutton & Davidson, 1997), we showed that scores on a self-report measure designed to operationalize Gray's concepts of Behavioral Inhibition and Behavioral Activation (the BIS/BAS scales; Carver & White, 1994) were even more strongly predicted by electrophysiological measures of prefrontal asymmetry than were scores on the PANAS scales (see figure 5-2). Subjects with greater left-sided prefrontal activation reported more relative B AS to BIS activity compared with subjects exhibiting more right-sided prefrontal activation.

P.95

We also hypothesized that our measures of prefrontal asymmetry would predict reactivity to experimental elicitors of emotion. The model we have developed over the past several years (see Davidson, 1992, 1994, 1995, for background) features individual differences in prefrontal activation asymmetry as a reflection of a diathesis that modulates reactivity to emotionally significant events. According to this model, individuals who differ in prefrontal asymmetry should respond differently to an elicitor of positive or negative emotion, even when baseline mood is partialed out. We (Wheeler, Davidson, & Tomarken, 1993) performed an experiment to examine this question. We presented short film clips designed to elicit positive or negative emotion. Brain electrical activity was recorded prior to the presentation of the film clips. Just after the clips

P.96

were presented, subjects were asked to rate their emotional experience during the preceding film clip. In addition, subjects completed scales that were designed to reflect their mood at baseline. We found that individual differences in prefrontal asymmetry predicted the emotional response to the films even after measures of baseline mood were statistically removed. Those individuals with more left-sided prefrontal activation at baseline reported more positive affect to the positive film clips, and those with more right-sided prefrontal activation reported more negative affect to the negative film clips. These findings support the idea that individual differences in electrophysiological measures of prefrontal activation asymmetry mark some aspect of vulnerability to positive and negative emotion elicitors. The fact that such relationships were obtained following the statistical removal of baseline mood indicates that any difference between left- and right-frontally activated in baseline mood cannot account for the prediction of film-elicited emotion effects that were observed.

In another study, we (Davidson, Dolski, et al., in preparation) examined relations between individual differences in prefrontal activation asymmetry and the emotion-modulated startle. In this study, we presented pictures from the International Affective Picture System (Lang et al., 1995) while acoustic startle probes were presented and the EMG-measured blink response from the orbicularis oculi muscle region was recorded (see Sutton, Davidson, Donzella, Irwin, & Dottl, 1997, for basic methods). Startle probes were presented both during the 6-s slide exposure as well as 500 ms following the offset of the pictures, on separate trials.¹ We interpreted startle magnitude during picture exposure as providing an index related to the peak of emotional response, while startle magnitude following the offset of the pictures was taken to reflect the recovery from emotional challenge. Used in this way, startle probe methods can potentially provide new information on the time course of emotional responding. We expected that individual differences during actual picture presentation would be less pronounced than individual differences following picture presentation, because an acute emotional stimulus is likely to pull for a normative response across subjects, yet individuals are likely to differ dramatically in the time to recover. Similarly, we predicted that individual differences in prefrontal asymmetry would account for more variance in predicting magnitude of recovery (i.e., startle magnitude post-stimulus) than in predicting startle magnitude during the stimulus. Our findings were consistent with our predictions and indicated that subjects with greater right-sided prefrontal activation show a larger blink magnitude following the offset of the negative stimuli, after the variance in blink magnitude during the negative stimulus was partialed out. Measures of prefrontal asymmetry did not reliably predict startle magnitude during picture presentation. The findings from this study are consistent with our hypothesis and indicate that individual differences in prefrontal asymmetry are associated with the time-course of affective responding, particularly the recovery following emotional challenge.

In addition to the studies previously described using self-report and psychophysiological measures of emotion, we have also examined relations between individual differences in electrophysiological measures of prefrontal asymmetry and other biological indices, which in turn have been related to differential reactivity to stressful events. Two recent examples from our laboratory include measures of immune function and cortisol. In the case of the former, we examined differences between left- and right-prefrontally activated subjects in

P.97

natural killer (NK) cell activity, because declines in NK activity have been reported in response to stressful, negative events (Kiecolt-Glaser & Glaser, 1991). We predicted that subjects with right prefrontal would exhibit lower NK activity compared with their left-activated counterparts because the former type of subject has been found to report more dispositional negative affect, to show higher relative BIS activity and to respond more intensely to negative emotional stimuli. We found that right-frontally activated subjects indeed had lower levels of NK activity compared to their left-frontally activated counterparts (Kang, Davidson, Coe, Wheeler, Tomarken, & Ershler, 1991).

Recently, in collaboration with Kalin, our laboratory has been studying similar individual differences in scalp-recorded measures of prefrontal activation asymmetry in rhesus monkeys (Davidson, Kalin, & Shelton, 1992, 1993). We (Kalin, Larson, Shelton, & Davidson, 1998) acquired measures of brain electrical activity from a large sample of rhesus monkeys (n = 50). EEC measures were obtained during periods of manual restraint. A subsample of 15 of these monkeys were tested on two occasions 4 months apart. We found that the test-retest correlation for measures of prefrontal asymmetry was .62, suggesting similar stability of this metric in monkey and man. In the group of 50 animals, we also obtained measures of plasma cortisol during the early morning. We hypothesized that if individual differences in prefrontal asymmetry were associated with dispositional affective style, such differences should be correlated with cortisol, since individual differences in baseline cortisol have been related to various aspects of trait-related stressful behavior and psychopathology (see e.g., Gold, Goodwin, & Chrousos, 1988). We found that animals with right-sided prefrontal activation had higher levels of baseline cortisol than their left-frontally activated counterparts (see figure 5-3). Moreover, when blood samples were collected 2 years following our initial testing, animals classified as showing

P.98

extreme right-sided prefrontal activation at age 1 year had significantly higher baseline cortisol levels when they were 3 years of age compared with animals who were classified at age 1 year as displaying extreme left-sided prefrontal activation. These findings indicate that individual differences in prefrontal asymmetry are present in nonhuman primates and that such differences predict biological measures that are related to affective style.

Affective Style and Psyche-pathology

Virtually all forms of psychopathology involve some abnormality in emotional processes, although the nature of these abnormalities is likely to differ among different disorders. The study of precisely what is abnormal in the emotional processing systems of individuals with different forms of psychopathology is very much in the earliest stages of investigation. We have used our findings in normal subjects as a foundation to probe the underlying neural substrates of affective and anxiety disorders with a major goal of understanding more precisely the nature of the abnormality in emotional processing in affective disorders.

One of the important sources of data on relations between brain function and emotion has come from studies of the affective styles of patients with localized brain lesions (see Robinson & Downhill, 1995, for a review). Robinson and his colleagues have reported that damage to the left frontal region is more likely to be associated with depression than damage to any other cortical region. Moreover, among patients with left hemisphere damage, more severe depressive symptomatology is present in those patients whose damage is closer to the frontal pole (Robinson, Kubos, Starr, Rao, & Price, 1984). Studies of regional brain function with neuroimaging of patients with psychiatric depressions have fairly consistently revealed a pattern of decreased blood flow or metabolism in left prefrontal regions at rest (Baxter et al., 1989; Bench, Friston, Brown, Scott, Frackowiak, & Dolan, 1992, 1993; Martinet et al., 1990; see George, Ketterer & Post, 1994, for review; see also Drevets et al., 1992, for a more complex pattern associated with pure familial depression).

We have conducted several studies examining regional brain electrical activity in depression. We hypothesized that most depression is fundamentally associated with a deficit in the approach/appetitive motivation system and should therefore be specifically accompanied by decreased activation in the left prefrontal region as measured by scalp electrophysiology. Henriques and Davidson (1991) obtained support for this hypothesis. Moreover, in another study, these authors demonstrated that the decrease in left prefrontal activation found among depressives was also present in recovered depressives who were currently euthymic, compared with never-depressed controls who were screened for lifetime history of psychopathology in both themselves as well as their first degree relatives (Henriques & Davidson, 1990). The findings from patients with localized unilateral brain damage, together with neuroimaging and electrophysiology studies in psychiatric patients without frank lesions, converge on the notion that depression is associated with a deficit in at least the prefrontal component of the approach system. We view this pattern of left prefrontal hypoactivation as a neural reflection of the decreased capacity for pleasure, loss of interest, and generalized decline in goal-related motivation and behavior.

Consistent with this notion are the data from another recent behavioral study

P.99

from my laboratory where we demonstrated using signal detection methods that depressed subjects were specifically hyporeactive to reward incentives (Henriques, Glowicki, & Davidson, 1994). In this study, we administered a verbal memory task under reward, punishment, and neutral incentive conditions. The rewards and punishments were monetary. Signal detection measures of sensitivity and response bias were computed. Nondepressed control subjects exhibited a more liberal response bias under both reward and punishment incentives. In other words, they were more likely to consider a stimulus as a signal if they were rewarded for correct hits or punished for misses. Depressed subjects showed a pattern quite similar to the controls in response to the punishment contingency. However, they failed to modify their response bias during the reward condition. In other words, the depressed subjects were less responsive to rewards compared with controls, while the groups showed no significant differences in response to punishment.

Based upon the evidence reviewed earlier, we hypothesized that, in contrast to depression, anxiety disorders would be associated with an increase in right-sided rather than a decrease in left-sided prefrontal activation, particularly during an acute episode of anxiety. To test this hypothesis, we (Davidson, Marshall, Tomarken & Henriques, in press) exposed social phobics who were particularly fearful of making public speeches to the threat of having to make a public speech. We recorded brain electrical activity during an anticipation phase where subjects were presented with an audiotaped countdown that noted how much more time there was until they were to make their speech. The taped recorded message was presented every 30 s for a total of 3 min. We found that the phobics showed a large and highly significant increase over baseline in right-sided prefrontal and right-sided parietal activation. During the same anticipation period, the controls showed a very different pattern of regional changes. The only change to reach significance was in the left posterior temporal region. We interpret this latter change as likely a consequence of verbal rehearsal in anticipation of making the public speech. No region in the right hemisphere exceeded an even liberal statistical threshold for increased activation relative to a baseline condition. The change in prefrontal activation among the phobics is consistent with our hypothesis of increased right-sided activation associated with an increase in anxiety. The increase in right parietal activation is consistent with Heller's (1990) hypothesis of increased right-sided activation associated with the arousal component of anxiety. Indeed, simultaneous measures of heart rate in this study indicated that the phobics had higher heart rate compared with the controls, particularly during the anticipation phase.

Research using self-report measures of positive and negative affect as well as experienced increases in autonomic arousal indicate that decreased positive affect is uniquely associated with depression, while increased autonomic arousal is uniquely associated with anxiety. However, reported negative affect is something that has been found to be common to both anxiety and depression (Watson, Clark, Weber, Assenheimer, Strauss, & McCormick, 1995). We have hypothesized that the decrease in left prefrontal activation may be specific to depression, while the increase in right-sided prefrontal activation (as well as right parietal activation) may be specific to certain components of anxiety. Considerably more research is required to understand the contribution being made by the activated right prefrontal region to negative affect. Other work (see Posner & Petersen, 1990, for review) indicates that portions of the right

P.100

prefrontal region are activated during certain types of vigilance and attention (e.g., Knight, 1991; see Posner & Petersen, 1990, for review). Anxiety-related negative affect is accompanied by heightened vigilance (e.g., McNally, 1998), which may be reflected in the right prefrontal increase.

One common region we believe to be associated with both anxiety and depression is the amygdala. While there is now a burgeoning literature on the anatomy and function of the amygdala (see Aggleton, 1993, for review), relatively little research has been conducted in intact humans, owing in large measure to the difficulty in imaging function in a structure that is relatively small (the adult human amygdala is not much more than 1 cm³ in volume). However, from what is known from both the animal and human studies, it appears that the amygdala plays an important role in assigning affective significance, particularly of negative valence, to both sensory as well as cognitive input (see LeDoux, 1992, for review). Using positron emission tomography (PET) to measure regional blood flow, several groups have reported increased blood flow in the amygdala in response to both behavioral (e.g., Schneider et al., 1995) and pharmacological (e.g., Ketterer et al., 1996) elicitors of negative affect. We have reported activation in the human amygdala using functional magnetic resonance imaging in response to aversive pictures (Irwin, Davidson, Lowe, Mock, Sorenson, & Turski, 1996). These studies suggest that activation in the human amygdala occurs in response to a broad range of elicitors of negative affect.

Both fMRI and O¹⁵ PET are ill-suited, for different reasons, for examining individual differences in resting or baseline levels of activation in the amygdala. As it is currently used, fMRI requires that at least two conditions be compared. What is measured is a relative difference in MR signal intensity between two or more conditions. Currently, fMRI is not calibrated in real physiological units. While O¹⁵ PET can be calibrated in real units, it reflects activity over a very short period of time (approximately 1 min) and thus, for psychometric reasons, is poorly suited to capture trait-like differences. It would be the equivalent of developing a single-item self-report instrument for assessing individual differences. PET used with flourodeoxyghicose (FDG) as a tracer, on the other hand, is well-suited to capture trait-like effects because the period of active uptake of tracer in the brain is approximately 30 min. Thus, it is inherently more reliable because the data reflect activity aggregated over this 30-min period. We have used resting FDG-PET to examine individual differences in glucose metabolic rate in the amygdala and its relation to dispositional negative affect in depressed subjects (Abercrombie, Larson, et al., 1998). We acquired a resting FDG PET scan as well as a structural MR scan for each subject. The structural MR scans are used for anatomical localization by coregistering the two image sets. Thus, for each subject, we used an automated algorithm to fit the MR scan to the PET image. Regions of interest (ROIs) were then drawn on each subject's MR scan to outline the amygdala in each hemisphere. These ROIs were drawn on coronal sections of subjects' MR images and the ROIs were then automatically transferred to the co-registered PET images. Glucose metabolism in the left and right amygdala ROIs were then extracted. The inter-rater reliability for the extracted glucose metabolic rate is highly significant with intraclass correlations between two independent raters ≥.97. We found that subjects with greater glucose metabolism in both the right and left amygdala report greater dispositional negative affect on the PAN AS scale (see figure 5-4). These findings indicate that individual differences in resting glucose metabolism in the amygdala are present

P.101

and that they predict dispositional negative affect among depressed subjects. Most nondepressed controls score so low on the PANAS trait negative scale that it is not possible to examine the same relation in this group because of the severe truncation of range for the PANAS scores.

The findings reviewed in this section indicate that the framework adopted for the study of individual differences in fundamental approach and with-drawal-related processes can be usefully applied in the study of psychopathology. A deficit in the approach system is viewed as a unique attribute of depressive disorders that is reflected in decreased left prefrontal activation. The acute symptoms of anxiety, as was described in our study with social phobics, was associated with a pronounced increase in both right-sided prefrontal and parietal activation. From research conducted in our laboratory as well as recent findings in the literature, it appears that amygdala activation may be a generic component of negative affect that is present in both anxiety and depression. Thus, differences between these disorders may be more pronounced for cortical systems that are critically involved in affect regulation and affect-cognition interaction, while subcortical contributions (in particular, the amygdala) may be common to both types of disorders and may in part be responsible for the substantial co-morbidity between these disorders (Watson et al., 1995).

Affective Style, Mood, and Anxiety Disorders: An Affective Neuroscience Approach

Among the most striking features of human emotion is the variability that is apparent across individuals in the quality and intensity of dispositional mood and emotional reactions to similar incentives and challenges. The broad ranges of differences in these varied affective phenomena has been referred to as “affective style” (Davidson, 1998). Differences among people in affective style appear to be associated with temperament (Kagan, Reznick, & Snidman, 1988), personality (Gross, Sutton & Ketelaar, 1998) and vulnerability to psychopathology (Meehl, 1975). Moreover, such differences are not a unique human attribute but appear to be present in a number of different species (see, e.g., Davidson, Kalin, & Shelton, 1993; Kalin, 1993).

The next section of this chapter will introduce conceptual distinctions among the various components of affective style and will highlight methodological challenges to their study. The third section will present a brief overview of the anatomy of two basic motivational/emotional systems—the approach and withdrawal systems. Then the fourth section will consider individual differences in these basic systems and indicate how such differences might be studied. The fifth section will address the relation between such individual differences and psychopathology. It is our intuition that some of the individual differences in basic processes of affective style are central to determining either resilience or vulnerability. Such differences can be conceptualized as diatheses that affect an individual's response to a stressful life event. Finally, the last section will consider some of the implications of this perspective for assessment, treatment, and plasticity.

The Constituents of Affective Style

Many phenomena are subsumed under the rubric of affective style. A concept featured in many discussions of affective development, affective disorders, and personality is “emotion regulation” (Thompson, 1994). Emotion regulation

P.89

refers to a broad constellation of processes that serve to amplify, attenuate, or maintain the strength of emotional reactions. Included among these processes are certain features of attention that regulate the extent to which an organism can be distracted from a potentially aversive stimulus (Derrybeny & Reed, 1996) and the capacity for self-generated imagery to replace emotions that are unwanted with more desirable imagery scripts. Emotion regulation can be both automatic and controlled. Automatic emotion regulation may result from the progressive automization of processes that initially were voluntary and controlled and have evolved to become more automatic with practice. We hold the view that regulatory processes are an intrinsic part of emotional behavior and rarely does an emotion get generated in the absence of recruiting associated regulatory processes. For this reason, it is often conceptually difficult to distinguish sharply between where an emotion ends and regulation begins. Even more problematic is the methodological challenge of operationalizing these different components in the stream of affective behavior.

When considering the question of individual differences in affective behavior, one must specify the particular response systems in which the individual differences are being explored. It is not necessarily the case that the same pattern of individual differences would be found across response systems. Thus, for example, an individual may have a low threshold for the elicitation of the subjective experience (as reflected in self-reports) of a particular emotion but a relatively high threshold for the elicitation of a particular physiological change. It is important not to assume that individual differences in any parameter of affective responding will necessarily generalize across response systems,

P.90

within the same emotion. Equally important is the question of whether individual differences associated with the generation of a particular specific emotion will necessarily generalize to other emotions. For example, are those individuals who are behaviorally expressive in response to a fear challenge also likely to show comparably high levels of expressivity in response to positive incentives? While systematic research on this question is still required, initial evidence suggests that at least certain aspects of affective style may be emotion-specific, or at least valence specific (e.g., Wheeler, Davidson, & Tomarken, 1993).

In addition to emotion regulation, there are likely also intrinsic differences in certain components of emotional responding. For example, there may be individual differences in the threshold for eliciting components of a particular emotion, given a stimulus of a certain intensity. Thus, some individuals are likely to produce facial signs of disgust upon presentation of a particular intensity of noxious stimulus, whereas other individuals may require a more intense stimulus for the elicitation of the same response at a comparable intensity. This suggestion implies that dose-response functions may reliably differ across individuals. Unfortunately, systematic studies of this kind have not been performed, in part because of the difficulty of creating stimuli that are graded in intensity and designed to elicit the same emotion.

There are also likely to be individual differences in the peak or amplitude of the response. Upon presentation of a series of graded stimuli that differ in intensity, the maximum amplitude in a certain system (e.g., intensity of a facial contraction, change in heart rate, etc.) is likely to differ systematically across subjects. Some individuals will respond with a larger amplitude peak compared with others. Again, such individual differences may well be quite specific to particular systems and will not necessarily generalize across systems, even within the same emotion. Thus, the individual who is in the tail of the distribution in heart rate response to a fearful stimulus will not necessarily be in the tail of the distribution in facial response.

Another parameter that is likely to differ systematically across individuals is the rise time to peak. Some individuals will rise quickly in a certain response system, while others will rise more slowly. There may be an association between the peak of the response and the rise time to the peak within certain systems for particular emotions. Thus, it may be the case that for anger-related emotion, those individuals with higher peak vocal responses also show a faster rise time, but to the best of my knowledge, there are no systematic data related to such differences.

Finally, another component of intrinsic differences across individuals is the recovery time. Following perturbation in a particular system, some individuals recover quickly and others recover slowly. For example, following a fear-pro-voking encounter, some individuals show a persisting heart rate elevation that might last for minutes, while other individuals show a comparable peak and rise time, but recover much more quickly. As with other parameters, there are likely to be differences in recovery time across different response systems. Some individuals may recover rapidly in their expressive behavior, while recovering slowly in certain autonomic channels. As is noted in a later section, individual differences in recovery time may be particularly important for identifying individuals vulnerable to mood and anxiety disorders.

These specific parameters of individual differences describe affective chronometry—the temporal dynamics of affective responding. Very little is known about the factors that govern these individual differences and the extent to which such differences are specific to particular emotion response systems or generalize across emotions (e.g., is the heart rate recovery following fear similar to that following disgust?). Moreover, the general issue of the extent to which these different parameters that have been identified are orthogonal or correlated features of emotional responding is an empirical question that has yet to be answered. I hope to show that affective chronometry is a feature of affective style that is methodologically tractable and can yield to experimental study of its neural substrates.

We also hold that affective style is critical in understanding the continuity between normal and abnormal functioning and in the prediction of psychopathology and the delineation of vulnerability. On the opposite side of the spectrum, such individual differences in affective style will also feature centrally in any comprehensive theory of resilience. The fact that some individuals reside “off the diagonal” and appear to maintain very high levels of psychological wellbeing despite their exposure to objective life adversity is likely related to their affective style (Ryff & Singer, 1998). Some of these implications will be discussed at the end of this chapter.

We first consider some of the neural substrates of two fundamental emotion systems. This provides the foundation for a consideration of individual differences in these systems and the neural circuitry responsible for such differences.

The Circuitry of Approach-and Withdrawal-Related Emotion

Although the focus of my empirical research has been on measures of prefrontal brain activity, it must be emphasized at the outset that the circuit instantiating

P.91

emotion in the human brain is complex and involves a number of interrelated structures. Preciously few empirical studies using modern neuroimaging procedures that afford a high degree of spatial resolution have been performed (see George et al., 1995; Paradise et al., 1997, for examples). Therefore, hypotheses about the set of structures that participate in the production of emotion must necessarily be speculative and based to a large extent on the information available from the animal literature (e.g., LeDoux, 1987) and from theoretical accounts of the processes involved in human emotion.

Based upon the available strands of theory and evidence, numerous scientists have proposed two basic circuits each mediating different forms of motivation and emotion (see, e.g., Gray, 1994; Davidson, 1995; Lang, Bradley, & Cuthbert, 1990). The approach system facilitates appetitive behavior and generates certain types of positive affect that are approach related, e.g., enthusiasm, pride, and so on (see Depue & Collins, 1999, for review). This form of positive affect is usually generated in the context of moving toward a desired goal (see Lazarus, 1991, and Stein & Trabasso, 1992, for theoretical accounts of emotion that place a premium on goal states). The representation of a goal state in working memory is hypothesized to be implemented in dorsolateral prefrontal cortex. The medial prefrontal cortex seems to play an important role in maintaining representations of behavioral-reinforcement contingencies in working memory (Thorpe, Rolls & Maddison, 1983). In addition, output from the medial prefrontal cortex to nucleus accumbens (NA) neurons modulates the transfer of motivationally relevant information through the NA (Kalivas, Churchill, & Klitenick, 1993). The basal ganglia are hypothesized to be involved in the expression of the abstract goal in action plans and in the anticipation of reward (Schultz, Apicella, Romo, & Scarnati, 1995; Schultz, Romo, Ljungberg, Mirenowicz, Hollerman & Dickinson, 1995). The NA, particularly the caudomedial shell region of the NA, is a major convergence zone for motivationally relevant information from a myriad of limbic structures. Cells in this region of the NA increase their firing rate during reward expectation (see Schultz, Apicella, et al., 1995). There are likely other structures involved in this circuit which depend upon a number of factors including the nature of the stimuli signaling appetitive information, the extent to which the behavioral-reinforcement contingency is novel or overlearned, and the nature of the anticipated behavioral response.

It should be noted that the activation of this approach system is hypothesized to be associated with one particular form of positive affect and not all forms of such emotion. It is specifically predicted to be associated with pre-goal attain ment positive affect, the form of positive affect that is elicited as an organism moves closer toward an appetitive goal. Post-goal attainment positive affect represents another form of positive emotion that is not expected to be associated with activation of this circuit (see Davidson, 1994, for a more extended discussion of this distinction). This latter type of positive affect may be phenomenologically experienced as contentment and is expected to occur when the prefrontal cortex goes off-line after a desired goal has been achieved. Cells in the NA have also been shown to decrease their firing rate during post-goal consummatory behavior (e.g., Henriksen & Giacchino, 1993).

Lawful individual differences can enter into many different stages of the approach system. Such individual differences and their role in modulating vulnerability to psychopathology will be considered in detail later. For the moment, it is important to underscore two issues. One is that there are individual

P.92

differences in the tonic level of activation of the approach system which alters an individual's propensity to experience approach-related positive affect. Second, there are likely to be individual differences in the capacity to shift between pre- and post-goal attainment positive affect and in the ratio between these two forms of positive affect. Upon reaching a desired goal, some individuals will immediately replace the just-achieved goal with a new desired goal, and so will have little opportunity to experience post-goal attainment positive affect, or contentment. There may be an optimal balance between these two forms of positive affect, though this issue has never been studied.

There appears to be a second system concerned with the neural implementation of withdrawal. This system facilitates the withdrawal of an individual from sources of aversive stimulation and generates certain forms of negative affect that are withdrawal related. Both fear and disgust are associated with increasing the distance between the organism and a source of aversive stimulation. From invasive animal studies and human neuroimaging studies, it appears that the amygdala is critically involved in this system (e.g., LeDoux, 1987). Using functional magnetic resonance imaging (fMRI) we have recently demonstrated, for the first time, activation in the human amygdala in response to aversive pictures compared with neutral control pictures (Irwin et al., 1996). In addition, the temporal polar region also appears to be activated during with-drawal-related emotion (e.g., Reiman, Fusselman, Fox, & Raichle, 1989; but see Drevets, Videen, MacLeod, Haller, & Raichle, 1992). These effects, at least in humans, appear to be more pronounced on the right side of the brain (see Davidson, 1992, 1993, for reviews). In the human electrophysiological studies, the right frontal region is also activated during withdrawal-related negative affective states (e.g., Davidson, Ekman, Saron, Senulis & Friesen, 1990). At present it is not entirely clear whether this electroencephalogram (EEC) change reflects activation at a frontal site or whether the activity recorded from the frontal scalp region is volume-conducted from other cortical loci. The resolution of this uncertainty must await additional studies using positron emission tomography (PET) or fMRI, which have sufficient spatial resolution to differentiate among different anterior cortical regions. In addition to the temporal polar region, the amygdala and possibly the prefrontal cortex, it is also likely that the basal ganglia and hypothalamus are involved in the motor and autonomic components, respectively, of withdrawal-related negative affect (see Smith, DeVita, & Astley, 1990).

The nature of the relation between these two hypothesized affect systems also remains to be delineated. The emotion literature is replete with different proposals regarding the interrelations among different forms of positive and negative affect. Some theorists have proposed a single bivalent dimension that ranges from unpleasant to pleasant affect, with a second dimension that reflects arousal (e.g., Russell, 1980). Other theorists have suggested that affect space is best described by two orthogonal positive and negative dimensions (e.g., Watson & Tellegen, 1985). Still other workers have suggested that the degree of orthogonality between positive and negative affect depends upon the temporal frame of analysis (Diener & Emmons, 1984). This formulation holds that when assessed in the moment, positive and negative affect are reciprocally related, but when examined over a longer time frame (e.g., dispositional affect) they are orthogonal. It must be emphasized that these analyses of the relation between positive and negative affect are all based exclusively upon measures of self

P.93

report and therefore their generalizability to other measures of affect are uncertain. However, based upon new data to be described here, we believe that a growing corpus of data does indeed indicate that one function of positive affect is to inhibit concurrent negative affect.

Physiology

Another issue of interest to temperament researchers is the underlying physiological markers of individual differences in emotional reactivity. Findings from hemispheric lateralization studies are of special relevance here in understanding the links between emotion and pathology. These studies have shown differential hemispheric activation for the underlying action tendencies, such as approach and withdrawal, rather than differential activation for specific discrete emotions (Davidson, Ekman, Saron, Senulis, & Friesen, 1990; Fox & Davidson, 1987, 1988). These studies also point to important parallels between different levels of analysis—from neural activation to behavioral indices—in approaching emotion and pathology links. For example, several of the behavioral explanations Cole and Zahn-Waxler (1994) offer to account for the emotional profile of children with disruptive behavior disorders seem to be organized around action tendencies rather than specific discrete emotions. We will try to illustrate the parallels that emerge when we consider action tendencies rather than specific discrete emotions.

Both the clinical and empirical observations of children with disruptive behavior disorders show these children to be low in sadness and fear and high in anger and inappropriate joy expressions (Cole & Zahn-Waxler, 1994). Cole and Zahn-Waxler suggest that the low reactivity in sadness and fear may be a function of atypical or nonnormative coping mechanisms. This approach, therefore, clusters low reactivity in sadness and fear together through their common underlying action tendency: withdrawal. Cole and Zahn-Waxler propose that children with disruptive behavior disorders may also cope by masking sadness and fear with anger and joy, the latter two being associated with the underlying action tendency of approach. Alternatively, they propose that these children may also cope by inhibiting the expression of sadness and fear, that is, not showing any withdrawal tendencies, but rather communicating only indifference. Both of these proposals are based on behavioral observation from the clinical and empirical literature, but both emphasize action tendencies rather than discrete emotions.

As already noted, hemispheric lateralization studies with both normative and nonnormative populations show differential activation for approach and withdrawal tendencies rather than the observed discrete emotions (Davidson et al., 1990; Fox & Davidson, 1987, 1988). Most of these studies link relative left frontal activation with approach, and relative right frontal activation with withdrawal tendencies (Fox, 1994). Studies have also suggested that the dynamic balance between the two hemispheres is altered in disordered populations. For example, studies have suggested that individuals with impulsive and hyperactive

P.275

temperaments tend to show an inhibition of withdrawal (i.e., disinhibition) through relative lower right frontal activation. In fact, individuals at risk for externalizing disorders tend to exhibit this pattern of relative lower right frontal activation. In contrast, individuals with depressed mood show relative lack of activation in the left frontal region suggesting a deficiency in approach and positive affect rather than an excess in withdrawal tendencies (Henriques & Davidson, 1990; see also Davidson, Chapter 5 of this volume). This deficiency of depressed individuals in showing approach-related behaviors converges with Zahn-Waxier et al.'s (1984) finding of deficient increases in joy reactions of toddlers with bipolar parents. Zahn-Waxler et al. found suggestive evidence that when stressful background stimulation was removed, toddlers with bipolar parents failed to show increased joy reactions in comparison to control children.

Such convergence in findings from hemispheric lateralization studies and behavioral differences found for at-risk children illustrate and emphasize the importance of indices of approach and withdrawal in emotional reactivity. More rigorous parallels at different levels of analysis are clearly important for our understanding of the relationship between emotion and psychopathology.

Direct Processes of Transmission

Also critical to questions of emotion and the development of psychopathology are studies that concentrate on the mechanisms by which parental psychopathology gives rise to nonnormative patterns in children's emotional responses. As Cummings and Davies (1994) argue in their review on the effects of maternal depression on child outcomes, a large body of research has concentrated on indirect mechanisms of influence, such as parenting and attachment. Zahn-Waxler's work has also been a part of this trend that has emphasized indirect mechanisms of influence, especially in delineating aspects of parenting practices associated with risk status (Zahn-Waxler et al., 1990). However, commentaries by Cummings (1995) and Seifer (1995) in the special section of Developmental Psychology on maternal depression, note that concentration on such indirect mechanisms is an overall limitation of the field. They suggest that this limitation is a function of our failure to generate a theoretical framework that organizes and informs our inquiry on the nature of the influence of maternal depression on developmental outcomes.

Studies conducted by Tronick (Tronick, 1989; Tronick, Als, & Brazelton, 1977; Tronick & Giannino, 1987) and Field (1984, 1994) are rare examples of efforts in the literature that have examined the direct mechanisms through which maternal depression can influence child outcomes. Most of these studies have been conducted with infants and their depressed caregivers using face-to-face interaction paradigms. These studies have shown that infant-depressed mother interactions are characterized by nonsynchronous affective expression (i.e., nonmatching affective states in valence) and an overall higher degree of negative affective exchanges compared to controls. Furthermore, experimental studies have demonstrated that similar patterns in emotional responding can be reliably elicited in infants with nondepressed caregivers, when the caregiver simulates depressed mood. However, these studies are often limited to infancy; furthermore, there are virtually no studies that examine the change or the persistence in such emotional responding beyond infancy. Neither are there any

P.276

studies that have examined the extent to which these infants generalize non-normative patterns in emotional responding to their interactions with nondepressed adults.

Nevertheless, these findings are important to questions regarding the relations between emotion and pathology in several ways. First, because the experimental studies have utilized random assignment, their pattern of findings is not saturated by the large individual difference variation in affective styles, especially in control mother-infant dyads. Although we can certainly expect that both maternal and infant individual differences in emotionality likely account for a significant proportion of the observed variation in the frequency of positive and negative affective states, the synchronous pattern cannot solely be a function of these individual differences. Second, because infants' emotional reactions can be reliably altered with simple simulation of depressed mood by otherwise nondepressed caregivers, the findings suggest that adults' affective reactions have an asymmetrical influence in the maintenance of synchronous affective exchanges.

This asymmetry, however, does not imply that infant emotional characteristics do not influence mothers' reactions. There are numerous studies showing that difficult temperament in infancy is a correlate of maternal depression (Crockenberg & Smith, 1982). However, the interesting question this asymmetry gives rise to is how similar processes operate beyond infancy in mother-child interactions. If we were to interpret this asymmetry as contagion of mother's affective state to the infant we may speculate that such asymmetry is not likely to maintain its influence, or operate in the same fashion beyond the first year of the infant's life. The reasoning behind such a claim is that infants have limited capacity in modulating motoric approach and withdrawal reactions. However, once upward locomotion emerges there are likely to be significant opportunities for the toddler to execute withdrawal and approach reactions in multiple situations. Thus, opportunities for continued contagion of mothers' affective state are likely to be significantly altered. Perhaps, child characteristics are likely to exert stronger influences on the dyadic nature of the affective exchanges in mother-child interactions beyond the first year with their emerging motoric abilities.

Another interesting issue these findings raise is the source of this asymmetry. As already reviewed, hemispheric lateralization studies suggest that depression may be a reflection of the inability to express positive affect and execute approach behaviors, rather than an increased tendency to express negative affect and execute withdrawal behaviors. Thus, it may be that a mother's overall capacity to express approach-positive affect states is a critical element in the generation of synchronous affective exchanges. This suggests the importance of an issue raised earlier about the role individual differences in emotional reactivity may play in dyadic affective exchanges. There are no studies, however, that exemplify the extent to which individual differences in both maternal and infant emotional reactivity contribute to the variation in these dyadic exchanges. Despite the long-standing appreciation of the need to incorporate the temperamental differences that children bring to their environments (Maccoby & Martin, 1983), researchers insist on ignoring these very important sources of variation. These findings thus alert us to the heretofore largely ignored influence of maternal personality as well.

P.277

At this juncture we must acknowledge that these rich, although ignored, sources of variation in individual differences in emotionality are also genetically based. Numerous studies have shown that a large percentage of the variation in individual differences in both infant-child (Goldsmith & Campos, 1986; Matheny, 1980; Torgersen & Kringlen, 1978) and adult emotionality (Loehlin, 1992; McGue, Bacon, & Lykken, 1993; Plomin, Chipuer, & Loehlin, 1990) is genetically influenced.

As Scarr and McCartney (1990) point out, however, the environmental variation we have discussed up to this point may be a reflection of the correlated genotypes among family members. More specifically, a child's genotype is correlated with the environment he/she is embedded in. In other words, children's genotypes are correlated with those of their parents, who in turn create and seek the environments that support those genetic tendencies. Thus, there is considerable genetic variation in “environmental measures.” For example, many studies indicate that around 50% of the variation in both infant negative affectivity (Goldsmith & Campos, 1986; Matheny, 1980; Torgersen & Kringlen, 1978) and adult neuroticism (Loehlin, 1992; McGue et al, 1993; Plomin et al., 1990) is heritable and most of the remaining variation is due to nonshared environmental influences. Thus, negative affect exchanges observed in infant-de-pressed mother dyads are likely to be largely mediated by the correlated genotypes of infant and mother. On the other hand, findings on infant positive affectivity from behavior genetic studies reveal that in addition to a heritable component, infant positive affectivity is significantly influenced by shared environmental factors (Goldsmith, Buss, & Lemery, 1997). Such shared environmental factors for positive affectivity suggest that nonsynchronous affective exchanges with a depressed mother, who has a decreased ability to execute approach and express positive affect, may be a function of concurrent depressed mood of the mother. The implication would be that infants may not generalize their responding to nondepressed caregivers. In any case, this argument shows that behavior genetic studies on trait measures of emotionality are important in teasing apart sources of influence for negative child outcomes. In other words, these studies are relevant to links between emotion and pathology because they point to sources of influence that have implications for treatment and preventive measures.

Conclusion

In this commentary we drew heavily on some of Zahn-Waxler's work with at-risk populations in addressing issues about emotion and the development of childhood psychopathology. We illustrated that her work has been central to formulations of various questions in understanding the relations between emotion and pathology. We argued that fuzzy concepts such as dysregulation need not be evoked when establishing links between normative and nonnormative patterns of emotional responding. In addition, we showed that issues of interest to temperament researchers—such as genetic and environmental origins, physiological underpinnings, stability and instability, and measurement and data-analytic techniques—can go a long way in generating explanatory power in linking emotion and psychopathology.

Stability

One issue of interest to temperament researchers has been the extent of stability in various temperament dimensions. Many studies have delineated patterns of continuity and discontinuity in trait measures of emotionality (Hagekull, 1989; Huttunen & Nyman, 1982). These studies suggest that instability is as much a part of development as stability is. There is evidence from personality development literature both in adulthood and childhood (Tellegen, 1988) which suggests that stability in any given individual difference dimension may also be construed as a meta-characteristic that cuts across content dimensions of social behavior. Thus, there are some individuals who appear to be stable in a number of personality characteristics, while there are others who appear to show fluctuations or discontinuities in a number of personality characteristics.

Persistence in the expression of “inappropriate behaviors,” construed in very general terms, is likely to be associated with continued risk for pathology. For example, persistence in aggressive behavior throughout childhood is associated with externalizing tendencies (Cole & Zahn-Waxier, 1994). Thus, our ability to model stability or change in traits over time would also increase our ability to predict and differentiate the dynamics in risk status. Such use of the concept of stability in trait measures has been examined in adult personality studies. Tellegen (1988) has suggested a regression-based approach to quantifying consistency in personality assessments in adult populations. This approach involves partialling the mean and the variability of an individual's score on a given trait measure, collected at multiple time points. Examining the interaction of these main effects (i.e., the mean and variability) is one way in which we can capture both change and stability as a meta-dimension.

The use of such approaches with measures of temperamental characteristics of at-risk children can allow us to capture such a meta-dimension in various

P.274

emotional reactions. Examining stability and change would also help explain why some children's anger propensities in toddlerhood, for example, predict a later increase in externalizing tendencies for some individuals but not for others. In other words, stability as a meta-dimension on individual differences in emotionality may help show how context-inappropriate expressions of certain emotions can evolve into behaviors that meet diagnostic criteria for a variety of disorders. Similarly, instability in measures of emotionality may account for our decreased ability to predict the externalizing and/or internalizing tendencies of some children (Zahn-Waxler et al., 1990).

Development of Child Psychopathology

The field of developmental psychopathology allows researchers the unique opportunity to probe issues pertaining to both developmental course and etiology of psychopathology. It also provides the opportunity to understand the complexities and establish boundaries for normative development. The role of emotion in childhood psychopathology has not been systematically investigated. Carolyn Zahn-Waxler is one of the leading developmentalists who stresses the role emotions play in childhood pathology, and she examines both the externalizing and the internalizing disorders through a common emotional framework. Zahn-Waxler's work on parental depression as a risk factor for childhood pathology has been central to formulations of fundamental questions regarding the role of emotion in childhood psychopathology as well as in the development of both observational and experimental methodology.

In this chapter, we will examine the implications of Zahn-Waxler's theoretical and empirical work for our understanding of the links between emotion and psychopathology. First, we will examine the framework she has utilized to investigate the role of emotions in various normative and nonnormative developmental phenomena. Second, we will examine how the concept of emotion dysregulation comes to play a central role in linking emotion processes to the development of pathology. We will try to show how the use of this concept of emotion dysregulation fails to live up to its promise of linking emotion and pathology in several ways. Third, we will offer several ways in which the use of concepts from the temperament domain can reduce our reliance on problematic explanatory constructs such as dysregulation. Finally, we will examine how studies on the mechanisms of transmission of parental psychopathology may have ignored some central questions about emotion-related processes.

Various researchers who study the role of emotions in psychopathology, either at the symptom or syndrome level, adopt a particular scheme in conceptualizing emotions (Clark, in press). These conceptualizations are not only central to the measures and assessments subsequently generated, but they also have

P.267

implications for the questions formulated on the role of emotions in psychopathology. We will first introduce the theoretical components of Zahn-Waxler's emotion framework and then examine how they have been reflected in her empirical work with at-risk populations. We will draw heavily on some of the findings from her empirical work with at risk populations to illustrate several points on both the nature of her contribution to the field, and the questions it raises about our current understanding of emotion and psychopathology relations.

Emotion Scheme

In Cole and Zahn-Waxler (1994), we find that discrete emotion theory constitutes the core of the framework Zahn-Waxler has adopted in investigating emotions and their role in childhood pathology (Ekman, 1984; Izard, 1977; Tomkins, 1984). The following four discrete emotions take the center stage in this framework: anger, sadness, fear, and joy. Embedded in this framework are further distinctions within anger that are based on its contextual elicitors. Among these contextual elicitors are goal frustration, self-assertion, and rage, which refers to anger expression in the absence of clear elicitors or provocation. Similar distinctions within anger have been discussed by other researchers as well (Kagan, 1981; Tomkins, 1963).

P.268

In contrast, there are no contextually based distinctions within the sadness and fear systems, although many other researchers have distinguished between fear of novelty and social fear/shyness (Campos, Barrett, Lamb, Goldsmith, & Sternberg, 1983; Goldsmith & Campos, 1982, 1990; Goldsmith & Rothbart, 1991; Kagan, 1981). Thus, in this form the scheme primarily adheres to formulations of the discrete emotion theorists.

In examining how this framework has come to play a role in Zahn-Waxler's empirical work, we find that some components of the model have weighed more heavily than others. For example, distinctions drawn within anger have figured consistently in her studies with at-risk populations. The value of these distinctions within anger become apparent as we examine how they have been operationalized and have helped differentiate later behavior problems. Four forms of anger expressions have been proposed as reflecting both rage and anger in contexts of goal frustration and self-assertion. For example, we find that interpersonal physical aggression and object struggles with peers have been viewed as tapping anger expression in response to goal frustration and/or self-assertion (Zahn-Waxler, lannotti, Cummings, & Denham, 1990). On the other hand, physical aggression toward an unfamiliar adult and undirected or out-of-control aggression are proposed to tap the third form of anger expression, namely rage (Zahn-Waxler, Cummings, McKnew, & Radke-Yarrow, 1984; Zahn-Waxler et al., 1990). Thus, we find that what is empirically retained are normative and non-normative expressions of anger. Out-of-control aggression and aggression toward an unfamiliar adult are viewed as examples of nonnormative expressions of anger. In fact, when the factor structure of these four distinct forms of aggression is examined, the distinction between normative and nonnormative expressions of anger empirically holds up (Zahn-Waxler et al., 1990). Furthermore, these nonnormative forms of anger expression were shown to be correlates (Zahn-Waxler et al., 1984, 1990) and predictors of externalizing problems 3–4 years later, above and beyond the maternal diagnosis (i.e., depression) and sex of the child (Zahn-Waxler et al., 1990).

In contrast, when we examine how other components of the model have figured in Zahn-Waxier's empirical work, we find that differences in sadness, fear, and joy have been examined as a function of risk status in only one study of children with bipolar parents (Zahn-Waxler et al., 1984). The findings from this small sample suggest, however, that there may be no significant differences in overall mean levels of joy, sadness, and fear expressions as a function of risk status. One fruitful endeavor that would further probe differences between sadness and fear would be to capitalize on distinctions within contextual elicitors of sadness and fear similar to those made for anger. For example, novelty- or object-based fear versus social fear toward mildly friendly though unfamiliar adults or peers have both been empirically shown to be independent dimensions in the general population (Campos et al., 1983; Goldsmith & Campos, 1990; Kagan, 1981; Kagan, Reznick, & Gibbons, 1989; Kochanska, 1991). There is also some speculation and some suggestive evidence that sadness is often a co-occurring response in situations typically associated with anger. Goldsmith (personal communication, 1993) has suggested that this may be a reflection of attributional biases in the development of a sense of self (e.g., low self-esteem), where some children exhibit sadness rather than anger when the integrity of the self is socially challenged.

These studies show that the application of this framework has proven useful in understanding emotion-based predictors of externalizing tendencies (Zahn-Waxler et al., 1990) with at-risk populations. It is also the case, as these findings illustrate, that the distinctions within anger are largely responsible for this improvement in prediction. Another central element of these studies is the concept of emotion dysregulation. Zahn-Waxler evokes this concept as an explanatory construct in linking emotion to psychopathology. However, its utility is questionable in further elucidating the relationship between emotion and risk for childhood pathology. Thus we will next examine in detail the different uses of this concept in an effort to show its limited utility.

Dysregulation Concept

The concept of emotion dysregulation has been evoked in research on developmental psychopathology by numerous researchers and also by those who study normative developmental processes (Calkins, 1994; Cole, Michel, & Teti, 1994). As Thompson (1994) has entitled his recent chapter, the concept of dysregulation continues to be a theme in search of a definition. The intuitive appeal of the term “dysregulation” is partly to blame for its frequent use. It has been evoked to discuss ineffective coping strategies as well as to explain deviations from the average or established norms for emotion-laden behavior in a variety of domains. Thus it has become a central construct in discussions regarding links between emotion and the development of psychopathology. This concept has also been evoked in several empirical and theoretical contexts in Zahn-Waxler's work. The two distinct uses of this concept by Zahn-Waxler are in many ways exhaustive of how many other researchers view and define emotion dysregulation.

According to Zahn-Waxler, context-inappropriate expressions of affective behavior can be viewed as emotion dysregulation. For instance, aggression toward an unfamiliar adult and out-of-control aggression are viewed as

P.269

dysregulation in the anger system. The concept of dysregulation is also applied to con-text-inappropriate expressions of joy. Here, examples are drawn from the literature on disruptive behavior disorders where joy responses are observed, when empathic concern at another's distress or guilt at wrongdoing is called for (Cole & Zahn-Waxier, 1994). Thus, dysregulation in this use of the term applies to activation of a particular emotion in inappropriate situations.

Emotion dysregulation has also been used to refer to nonnormative patterns in the modulation of emotional responses with changing environmental demands. This latter use of the term is distinct from dysregulation as context-in-appropriate expressions of emotions in many ways. To illustrate these distinctions it is important to elaborate upon a paradigm that has been used to observe dynamic modulation in multiple emotional reactions. This paradigm has been frequently used in studies of children's affective responses to interadult anger, where the interest has been to extrapolate children's emotional reactions to marital conflict (Cummings, 1987; Cummings, Zahn-Waxler, & Radke-Yarrow, 1984; Cummings, Pellegrini, Notarius, & Cummings, 1989). Typically, children are exposed to a series of stressful and neutral background events, while the changes in their affective reactions are observed. Zahn-Waxler has utilized this paradigm to compare differences in emotional reactions of toddlers from bipolar and depressed families, and their peers from control families (Zahn-Waxler et al., 1984, 1990).

These designs are advantageous to quantifying multiple parameters in emotional reactions. For example, one approach may be to aggregate the frequency with which given discrete emotions are expressed without respect to the nature of the background events. A second approach is to select particular background situations in which to examine differences in the intensity or frequency of particular kinds of emotional responses. For example, aggression toward an unfamiliar adult is often observed in such selected circumstances. This latter kind of approach thus yields differences in the extent to which children express normative versus nonnormative anger expressions as a function of their risk status (Zahn-Waxler et al., 1984, 1990).

A third approach is to partition differences in emotional reactions as a function of the changes in the affective tone of background events. This kind of approach gives rise to indices where modulation in emotional responding in relation to changing situational demands can be examined. In fact, Zahn-Waxler et al. (1984) found that children from control families showed an increased tendency to express joy following shifts in background events from stressful to neutral situations. On the other hand, these same control children did not show any comparable decrease or increase in the level of anger, fear, or sadness expressions. In contrast, children with bipolar parents did not show any increased tendency to express joy in

P.270

response to shifts from stressful to neutral background situations. They were, however, similar in their patterns of sadness, anger, and fear expressions, that is, there was no increase or decrease in mean levels of these negatively valenced emotions. Thus, dysregulation in this latter case refers to inadequate modulation in affective responses given changing contextual demands. In contrast, dysregulation in its former use, as context-inappro-priate expressions of a given discrete emotion, refers to overall mean level differences across a variety of situations.

The use of the term dysregulation, both as context-inappropriate expressions of discrete emotions and as an inability to modulate emotional reactions in response to changes in environmental demands, summarizes many developmentalists' views on what emotion regulation and dysregulation are. For example, Calkins (1994) defines regulations as “…processes and strategies which are used to manage arousal so that successful interpersonal functioning is possible” (p. 53). Similarly, Cole et al. (1994) define regulation as “an ongoing process of individual's emotion patterns in relation to moment-by-moment contextual demands” (p. 74). These definitions describe the function of regulation and suggest consequences for dysregulation. However, they fall short of elucidating what may go awry in the generation of the observed emotional response. For example, context-inappropriate expressions of anger, such as aggression toward an unfamiliar adult, refer to the occurrence of a discrete emotion in “unlikely” circumstances. Similarly, there may be multiple factors in a child's inadequate modulation of emotional responding with changing situational demands. For example, dysregulation may arise from failure to appraise changes in the affective tone of a number of situations in general, or it may be limited to a difficulty in appraising shifts only from stressful to neutral or positive affective backgrounds. It may also have very little to do with appraisals; rather, it may reflect individual differences in the propensity to express and experience positive affect. In all of these putative situations as well as in the definitions, the use of the term dysregulation fails to specify what it may be that gets dysregulated in the generation of an emotional response.

The objective here is not to underestimate the value of these findings with respect to our understanding of the ways in which emotional behavior is associated with risk status for later psychopathology. In fact, these dysregulation studies have important descriptive value in elucidating aspects of emotional behavior that seem to be associated with risk status. However, they have very limited explanatory power for our understanding of the ways in which emotion may be a factor in the development of behavior problems. To lend specificity to this point, let us take as an example the additional unique and predictive variance explained by nonnormative patterns in anger expressions after sex and parental psychopathology have been partialled out. This unique variance points to context-inappropriate forms of toddler anger expression as a predictor of externalizing problems. But such unique variance does not identify what parameters in the generation of such emotional responses are indicative of dysregulation. Thus, dysregulation fails to become distinct from the individual differences in the propensity to express these parameters of emotionality.

There are, however, multiple ways in which we can relate emotion to risk for childhood psychopathology without relying on the concept of dysregulation. We can gain both specificity and explanatory power if we adopt parameters that are used to characterize individual differences in emotionality. Individual differences in emotionality represent the primary topic of interest for temperament researchers. Despite the variability in the current conceptualizations of temperament, most agree that temperament refers to biologically based propensities in the expression of emotion and activity level (Campos et al., 1983; Goldsmith, Buss, Plomin, Rothbart, Thomas, Chess, Hinde, & McCall, 1987). Temperament researchers use a variety of parameters in characterizing individual differences in emotionality. Among the parameters used are the following: latency to, intensity of, and recovery to a neutral baseline state (Campos et al., 1983; Rothbart, 1989b). These parameters have also been adopted by researchers to focus on the processes of emotion regulation and provide a sound link among

P.271

concepts in the emotion nomological network. The use of these parameters can help reduce our reliance on fuzzy concepts such as dysregulation, because they can elucidate which aspects of the observed variability in emotion expression are associated with risk status.

These parameters are often utilized in the item pool of temperament assessment instruments such as the Infant Behavior Questionnaire (IBQ), Toddler Behavior Assessment Questionnaire (TBAQ), and Children's Behavior Questionnaire (CBQ) (Campos et al., 1983; Goldsmith & Rothbart, 1991; Goldsmith, 1996; Rothbart, 1981, 1989a, 1989b; Rothbart & Ahadi, 1994; Rothbart, Ahadi, & Hershey, 1994). For example, recovery to a neutral baseline state is construed as a separate dimension of individual differences and forms a separate scale on Rothbart's IBQ and CBQ. Another important component of the item pool in temperament assessment instruments has to do with the variety of situations in which a given emotion is expressed. For example, a child's propensity for anger expression is sampled in situations that involve both goal frustration in nonsocial situations and self-assertive behaviors in more social situations. Temperament researchers often refer to situational consistency in emotional responding as reactivity within a given discrete emotion (Bates, 1989; Rothbart, 1989b). As already noted, research on childhood inhibition has also shown that forming further distinctions in situational consistency is a meaningful enterprise, where consistency or reactivity in fearful behavior from social to nonsocial or novel situations appears distinct. Situational consistency in a given discrete emotion may also have little to do with other parameters such as average peak intensity in expression. For example, children who frequently get angry in a variety of limited situations may not, on average, show higher intensity of anger expressions compared to those who tend to get angry in a significantly fewer number of contexts. Thus, parameters such as latency to, reactivity, recovery from, and intensity are partially nonredundant parameters along which one may examine individual differences in discrete emotionality (Goldsmith, personal communication; Losoya, Lemery, Bowden, & Goldsmith, 1992).

The argument here is that the use of these parameters may be sufficient in elucidating what may be normative and nonnormative (or “dysregulated”) in the generation of discrete emotions. Thus, specificity in measurement of emotional reactions along these parameters can describe and help disentangle what may go awry with the emotional responding of at-risk children. We would like to illustrate the consequences of unspecificity in measurement in attempting to explain the links between emotion and pathology. We will take as an example the association between the anger reactions of toddlers to unfamiliar adults during brief encounters and risk status (i.e., depressed mother). This association points to anger reactions in “unlikely” circumstances. We would like to suggest three possible but distinct underlying processes that may give rise to such reactions.

First, it may reflect high reactivity in the anger system. For example, the toddler may show activation in the anger system in a variety of situations both context inappropriate and appropriate. Second, it may reflect an alternative, although distinct, kind of reactivity in the anger system. For example, the toddler may not be particularly likely to express anger during object struggles and goal frustrative contexts; rather the child may show a tendency to express anger during brief encounters with unfamiliar adults, when the more likely response in the general population may be to show inhibition. Third, these observed mean

P.272

differences in anger expressions during brief encounters with adults may not have much to do with individual differences in the anger system at all. Rather, these differences may reflect low reactivity in the fear system, especially in social situations, which hinders the more normative inhibitory response to novel situations. Alternatively, these differences may reflect low reactivity in the joy system, which is partially supported by the inadequate increases in joy expressions with the removal of stressful stimuli found with at-risk children. Clearly, this example elucidates the necessity of specificity of measurement when examining normative and nonnormative processes in the emotion systems. Specifically, both situational consistency of emotional reactions and parame-ters—such as latency to and recovery from—must be measured in order to disentangle what appears nonnormative or “dysregulated.”

This example also underscores the importance of certain unconventional data-analytic strategies in approaching links between emotion and pathology. Variable-centered approaches to understanding links between individual differences in emotion and risk for pathology—such as regression—are the first to present themselves to researchers. However, such regression-based approaches may be limited in their ability to empirically separate the three distinctions just drawn. Regression-based data-analytic approaches emphasize the relative standing of individuals on multiple variables taken one at a time. Thus, these approaches are most fruitful when there is a relatively “significant” or large degree of nonredundancy among variables that are to be evaluated for their unique contribution to the outcome variability. For example, maternal reports of temperament from instruments that contain largely independent scales, such as Goldsmith's TBAQ (Goldsmith, 1996), Rothbart's IBQ (Rothbart, 1981), andCBQ (Rothbart & Ahadi, 1994; Rothbart et al., 1994), are ideal for these variable-cen-tered approaches. These instruments capitalize on specific sets of situations in which proneness to a variety of emotional reactions is likely to be expressed rather than the overall impressions of the caregiver. Hence, proneness to angry arousal and fearful arousal scales are largely independent of each other and can be distinguished from a general distress proneness dimension.

However, such independence is often meaningless at the level of the individual. In other words, such independence points to the fact that for the items that comprise these instruments there are equal numbers of individuals who are likely to be high in both fearful and angry arousal, and those who are likely to be high in one and not the other. Such distinctions are crucial to our understanding of the links between emotion and pathology. Person-centered data-an-alytic techniques can overcome this shortcoming by empirically examining and forming clusters of individuals in the multidimensional space across a variety of emotional domains. Thus, each cluster is composed of individuals who are similar in their standing on various trait measures of emotionality. For example, children who are both prone to angry and fearful arousal and low in positive affectivity form a distinct cluster from those who are prone to both angry arousal and positive affectivity, but low in tearfulness. Thus, individuals who are at risk for psychopathology can be associated with particular emotion profiles. The rigorous use of such cluster-analytic techniques may be especially relevant to our understanding of the relations between emotional precursors and later diagnostic status, which is also categorical in nature.

We tried to emphasize two very important points in understanding emotion and pathology links without evoking a broad construct such as dysregulation.

P.273

The first and foremost among these is the necessity of acquiring specificity in measurement of both the situational context and the emotional reactions. Parameters such as latency to, reactivity across situations, and recovery from or intensity should prove useful in distinguishing among emotional reactions. The second point was that we should consider more unconventional data-analytic techniques such as cluster analysis to render our findings and inferences both clearer and more relevant to understanding the totality of an individual's emotional profile. These techniques can also help bridge the gap between the categorical nature of clinical assessment and the continuous measures typically generated in research settings.

Thus, movement away from general, nonspecific concepts is necessary if we are to explain, rather than simply describe, differences that appear to be associated with risk status. The use of concepts from the emotion process—tem-perament—help lend validity and specificity to our theoretical discussions about links between emotion and pathology. Other issues of interest to temperament researchers may also serve to alert us to questions relevant to the relationship between emotion and psychopathology. We will touch upon two issues of interest to temperament researchers and try to elucidate ways in which these issues may help our discussion in linking emotion and pathology without reliance on fuzzy concepts such as dysregulation.

Affective Instability

It again seems relatively straightforward to match Siever and Davis's (1991) dimension of affective instability to dysregulation in Gray's BAS or Depue's behavioral facilitation system (BFS), which in turn have been linked to extraversion/positive emotionality. Siever and Davis relate this dimension to the mood disorders on Axis I and to the “dramatic” Cluster B (especially borderline and histrionic personality disorders) on Axis II. As one example of parallel biological findings between these two types of disorders, they cite research documenting similar abnormalities in brain functioning during sleep (e.g., shorter and more variable times between falling asleep and the onset of rapid eye movements associated with dreaming) in both mood disorders and affectively unstable personality disrders.

In contrast to the almost universal association of personality disorder features with the high end of the neuroticism dimension, the two ends of this dimension are predicted to be associated with different disorders (Widiger, 1993). Specifically, histrionic personality disorder is predicted to be most strongly associated with the extraverted end of the dimension and avoidant and schizoid personality disorders with the introverted end. Similarly, when measures of personality and personality disorders were factored together, histrionic and schizoid personality disorders marked opposite ends of an extraversion-intro-version dimension (Wiggins & Pincus, 1994). As noted earlier, avoidant personality disorder loaded mostly strongly on the neuroticism dimension, but it also had a strong secondary loading on introversion.

As mentioned earlier, it is important to distinguish between the level or strength of the positive emotionality/BAS dimension and its variability. Its close association with mood disorder may stem from the fact that both level and variability are important in the manifestation of manic and depressive episodes. These two parameters, however, may be differentially important in personality disorders. When histrionic and schizoid personality disorders mark an extra-version—introversion dimension, the assumption is that the major parameter is level. However, association of this dimension with borderline personality disorder may be based more in the parameter of variability.

In a study of mood variability in nonpatients and patients with major depression, borderline personality disorder, or premenstrual syndrome (PMS), participants completed mood ratings in the morning and evening for 2 weeks (Cowdry, Gardner, O'Leary, Leibenluft, & Rubinow, 1991). Depressed patients had the lowest overall mood, followed by borderline patients; PMS patients did not differ from nonpatients on overall mood level. The greatest overall variability (largest SDs) was seen in PMS patients; depressed patients had the smallest SDs, while those of the borderline patients and nonpatients were between the two extremes. However, on a measure of the degree of randomness of the day-to-day variability, the borderline patients were higher than either depressed or PMS patients, but only the PMS patients differed from nonpatients on this index. Finally, borderline patients also showed the greatest average mood

P.190

change from morning to evening; again the depressed patients showed the least variability, with nonpatients and PMS patients falling between the two.

Slavney and Rich (1980) conducted a similar study comparing mood variability in patients with “hysterical personality disorder” versus a control group of “other personality and neurotic diagnoses” (p. 402). Participants completed mood ratings four times a day for 5 days. There was no difference in the overall mood level between the groups, but the patients with hysterical personality disorder showed significantly greater rating-to-rating and day-to-day variability than the control group. Regrettably, both these studies measured mood with a simple “worst mood ever” to “best mood ever” visual analog scale, so the extent to which variation in positive versus negative affect contributed to the overall variability is unknown. When these two affects were measured separately, greater variability in positive than negative affect was found in both a large undergraduate sample and a smaller sample of community-dwelling adult men (Watson & Clark, 1994), but whether these same results would be obtained using psychiatric patients is unknown as well. Nevertheless, it seems possible that poor regulation of the BAS may underlie the hyperreactive moodiness of those with dramatic cluster personality disorders.

Akiskal (1991) has taken the idea of a link between mood and these personality disorders a step further by proposing that many so-called personality disorders are, in fact, unrecognized manifestations of mood disorders. He describes “irritable-cyclothymic,” and “hyperthymic” temperaments that have close parallels in the DSM-IV descriptions of borderline and narcissistic personality disorders, respectively. He also describes a “depressive” temperament that is quite similar to the appended “depressive personality disorder.” The following is taken from Akiskal's description of the irritable cyclothymic patient: “Minor provocation resulted in angry outbursts [and] the emotional storm would not abate for hours or days…. Interpersonal crises are further amplified by their pouting, obtrusive, dysphoric, restless, and impulsive behavior…. A tempestuous life-style that creates interpersonal havoc… [largely due to] the volatile nature of the moods, and the erratic and high-risk behaviors” (pp. 47–48). It seems likely that most clinicians would consider this a description of borderline personality disorder.

Although Akiskal focuses on temperament, and “emphasizes disposition[s] that are closest to the biological underpinnings of drive, affect, and emotion” (p. 43), his full view is that adult personality represents individuals' adaptation to ongoing environmental experiences, given biological predispositions. Indeed, twin studies (and to a lesser extent family and adoption data) have indicated that this dimension not only has a substantial genetic component, but is also—perhaps more than the other dimensions—subject to environmental influences (Nigg & Goldsmith, 1994; Tellegen et al., 1988). Most likely, a complex biopsychosocial model will be needed to account for all of the data.

Impulsivity/Aggression

As in the attempt to link Gray's fight-flight system with the temperamental dimensions discussed earlier, associating Siever and Davis's (1991) dimension of impulsivity/aggression with these dimensions is less clear-cut than for the BIS and BAS. However, as suggested earlier, a reasonable case may be made for linking impulsivity/aggression with the fight-flight system, the personality dimensions of disinhibition, psychoticism, low conscientiousness and low agreeableness, somatic anxiety, alarm reactions, and

P.191

the specific negative affect of hostility. It is noteworthy that Zuckerman's (1991) “impulsive unsocialized sensation seeking” dimension also has subfactors of impulsivity and aggression. Finally, Fowles (1993) has noted that recent work in “affective neuroscience” (Panksepp, 1992) suggests the existence of a fourth affective system, in which rage is separated off from panic. Thus, it remains unclear whether one or two dimensions/systems are needed to account for this general domain and, if two, what is the nature of their interrelation, for it seems unlikely that they would be completely independent of each other.

Tellegen (1985) has suggested that the dimension he calls Constraint reflects individual differences in a “‘preparedness’ to respond to a range of emotion-re-lated circumstances … with either caution … or with recklessness” (p. 697). This is consistent with Siever and Davis's (1991) characterization of impulsive/ aggressive individuals as action-oriented and as likely to have “difficulty anticipating the effects of their behavior, learning from undesirable consequences of their previous behaviors, and inhibiting or delaying action appropriately” (p. 1650). These characteristics are descriptive of certain personality disorders, especially antisocial and borderline, and an extraordinary amount of relevant research has been conducted on psychopathy. For example, impulsive behavior has been linked with serious delinquency that is stable over time (White et al., 1994). Similarly, men who score high on the Psychopathy Check List (Hare, 1980)—which assesses such characteristics as lack of empathy, shallow affect, and impulsivity—spend more time in prison than nonpsychopathic criminals, at least until about age 40. Most noteworthy in this regard is the dramatic increase in the criminal activity of psychopaths from the early to the late 20s, suggesting marked failure to adapt their behavior following release from incarceration (Hare, McPherson, & Forth, 1988).

There is some indication that cognitive factors play an important role in this domain. For example, research has demonstrated that psychopathic or antisocial individuals have impaired cognitive abilities (Smith, Arnett, & Newman, 1992), fail to leam from negative feedback (Patterson & Newman, 1993), and have difficulty delaying gratification (Sher & Trull, 1994). Antisocial behavior in adolescents—a precursor to adult personality disorder—is strongly predicted by neuropsychological deficits (Moffitt, 1993), especially in higher order “executive” cognitive functions. Neuropsychological dysfunction also has been found in individuals with borderline personality disorder (Judd & Ruff, 1993). Moreover, attention deficit disorder in childhood has been linked to both adolescent conduct disorder (Lilienfeld & Waldman, 1990) and adult antisocial personality disorder (Mannuzza, Klein, Bessler, & Malloy, 1993).

A study of 283 male adoptees revealed that having a delinquent or criminal biologic parent was associated with increased attention deficit disorder, aggressivity, and antisocial personality disorder in the adopted away sons, suggesting a genetic basis for the observed relations (Cadoret & Stewart, 1991). Taken together these data describe a pattern of genetically based neurophysiological and neuropsychological abnormalities that are linked with attentional deficits and poor ability to monitor and self-regulate behavior. The difficulties in self-regu-lation encompass both deficits in inhibiting inappropriate behavior (impulsive behaviors) and in exhibiting strongly active responses (aggressive behaviors). The resulting impulsive/aggressive behavior pattern has been studied primarily in psychopathic or antisocial individuals, but this pattern can be observed in individuals with other personality disorders as well, especially those in the

P.192

“dramatic” cluster (borderline, histrionic, and narcissistic). Especially intriguing is the question of why more males than females are diagnosed with antisocial and narcissistic personality disorder, whereas more females than males are diagnosed with borderline and histrionic personality disorder. Perhaps the different socialization experiences of men and women with impulsive/aggressive styles lead to different behavioral expressions of this trait (Lilienfeld, 1992). In any case, it is unlikely that biological factors alone will be able to account for the observed differences.

Empirical Tests of the Proposed Integrated Model

Linking these three sets of dimensions with the emotion-related criteria of the Axis II personality disorders, several specific hypotheses—shown in table 9-6— can be made about how they should be associated. Specifically, because they contain specific anxiety-related criteria (see table 9-4), schizotypal, avoidant, and dependent personality disorders were hypothesized to correlate with the anxious-inhibited/negative emotionality/BIS dimension. Because affective instability and excessive emotionality are defining characteristics of borderline and histrionic personality disorder, they were hypothesized to correlate with the affective instability/positive emotionality/BAS dimension. Finally, because they contained criteria specifically related to anger, aggression, and impulsivity, paranoid, antisocial, and borderline personality disorders were hypothesized to correlate with the third impulsive-aggressive/disinhibited/fight-flight dimension. These are not, by any means, the only hypotheses that one could make regarding relations between personality disorders and these dimensions (e.g., Widiger's 1993 summary of relations between the personality disorders and the dimensions of the five-factor approach includes many more possibilities), but these seem a priori the most clearly substantiable.

Table 6 Relations Between Personality Disorder and Three Psychobiological Dimensions of Temperament

Empirical Correlations Dimension Trait Hypothesized Relations Sample 1 Sample 2 Anxiety/Inhibition (BIS) NT Schizotypal .16 .18 Avoidant .34** .35* Dependent .43** .57** Affective Instability (BAS) PT Borderline Histrionic .12 .25* .03 .28+ Inpulsivity/Aggression (fight/flight) DIS Paranoid .41** .42** Antisocial .66** .59** Borderline .46** .41** Note: BIS = Behavioral Inhibition System; BAS = Behavior Activation System; NT = Negative Temperament; PT = Positive Temperament; DIS = Disinhibition. Sample 1 N = 88; Sample 2 N = 40. ** p <.01; * p <.05;fp <.10.

P.193

It is interesting to consider what existing empirical data might be used to test these hypothesized relations. One relevant type of data is the observed pattern of comorbidity among personality disorders: Disorders that are hypothesized to relate to the same dimension should co-occur more frequently than those that are not. An examination of Widiger and Rogers's (1989) review of the comorbidity data provides some support for the stated hypotheses. First, examining the co-occurrence of schizotypal, avoidant, and dependent personality disorders, which were all hypothesized to relate to the anxiety/inhibition dimension, avoidant personality disorder was found to co-occur most frequently with schizotypal (26%) and dependent (20%) personality disorders; however, schizotypal and dependent personality disorders were not highly co-occurent (5%). Second, borderline and histrionic personality disorders—which were hypothesized to share the affective instability dimension—each co-occurred most strongly with the other (46%). Finally, paranoid, antisocial, and borderline personality disorders all were hypothesized to score high on the impulsivity/ag-gression dimension. There was a high degree of overlap between antisocial and borderline personality disorders (26%, which was the greatest overlap for antisocial and second greatest for borderline). However, neither of these disorders overlapped with paranoid personality disorder (1% and 5%, respectively), which was likely due in large part to its very low prevalence (7%) in these data. (By contrast, all of the other disorders compared here were two to five times more prevalent.)

In addition to these tests based on reported comorbidity patterns, data also were available from two independent samples of patients to test these hypotheses. Participants in both samples completed the Structured Interview for DSM-III-R Personality (SIDP-R; Pfohl, Blum, Zimmerman, & Stangl, 1989), a semistructured interview for personality disorders used to rate each DSM-III-R criteria on a four-point scale ranging from “not present” to “prominent symptom.” Ratings for each criteria were summed to form a total score for each Axis II diagnosis. Participants also completed the Schedule for Nonadaptive and Adaptive Personality (SNAP; Clark, 1993), a true-false format, 375-item self-report measure of 15 trait dimensions relevant to personality disorder. Incorporated into the SNAP are the three scales of the GTS (Clark & Watson, 1990) described earlier. table 9-6 also presents the correlations in each sample for each of the hypothesized dimension-disorder relations.

These data confirmed most of the hypothesized relations. Clearly, the anxi-ety/inhibition component of schizotypal personality disorder was not supported, but both avoidant and dependent personality disorder, as hypothesized, were related to negative emotionality. Contrary to hypothesis, positive emotionality was not related to borderline personality disorder and was only weakly related to histrionic personality disorder. Most likely this reflects the fact that the GTS scale taps typical positive mood level, whereas the hypothesized relation is with dysregulation in this mood dimension. Therefore, a measure of mood variability such as Depue's General Behavior Inventory (1987) would probably provide a better test of this hypothesis. The third hypothesis—that the dimension of impulsivity/aggression, operationalized here using the GTS Disinhibition scale, would be related to paranoid, antisocial, and borderline personality disorders—was strongly confirmed.

It is important to emphasize the replicability of the supported relationships. As mentioned, these data sets were collected entirely independently; the only

P.194

thing they have in common is their use of the same measures. The larger data set was collected in Texas and represents a heterogeneous patient sample drawn from both inpatient and outpatient settings, including a state hospital, a college counseling center, a community mental health center, a private practice, and a hospital-based family practice clinic. The smaller data set was collected by Pfohl and colleagues in the Department of Psychiatry at the University of Iowa Hospitals and Clinics. Therefore, it is reasonable to conclude that these findings are robust across diverse samples, although whether they would be replicated with different measures of the dimensions and the disorders remains to be tested.

In addition to hypotheses based on the DSM criteria, other hypotheses formulated from the literature may also be examined using Widiger and Rogers (1989) comorbidity data. For example, both borderline and antisocial personality disorders may be characterized as high on the impulsive/aggressive, fight-flight/disinhibition dimension, whereas they should differ on anxiety/inhibi-tion, BIS, negative affect, neuroticism, with antisocial individuals scoring low and borderline individuals scoring high. By contrast, those in the “anxious” Cluster C (avoidant, dependent, and obsessive-compulsive personality disorders) would appear to anchor the high end of the anxiety/inhibition dimension and the constrained end of disinhibition. This analysis suggests there should be the lowest comorbidities between antisocial personality disorder and the Cluster C diagnoses, intermediate comorbidity between borderline personality disorder and the Cluster C diagnoses, and high comorbidities among the Cluster C diagnoses.

Partial confirmation of these hypotheses is again found in Widiger and Rogers's (1989) review, which reported only a 2% overlap of avoidant and dependent personality disorders with antisocial personality disorder, and no overlap at all between antisocial and obsessive-compulsive personality disorder in any of the four studies. Borderline personality disorder showed intermediate level overlap with avoidant and dependent personality disorder (19% each), which were themselves also moderately comorbid (20%). However, none of the diagnoses overlapped with obsessive-compulsive personality disorder, which again was likely due in large part to its very low prevalence (6%). Thus, from a different viewpoint, we again find reasonable confirmation of the hypothesized phenomenological relations. Of course, these analyses do not test the proposed links with the underlying biological dimensions.