Personality traits and anxiety symptoms: The multilevel trait predictor model

Abstract

Investigation of relations between personality traits and mental disorders can inform key issues in psychopathology research. However, it has been hindered by extensive correlations among the traits. Building on studies of affect–psychopathology relations (e.g., the tripartite model), an organizational framework is proposed to solve this problem with respect to anxiety pathology. To test the resulting model, associations between four traits (negative emotionality, positive emotionality, anxiety sensitivity, and negative evaluation sensitivity) and four anxiety symptoms (chronic worry, obsessive-compulsive symptoms, panic, and social anxiety) were examined in an undergraduate sample (N=907). Confirmatory factor analyses supported operationalizations of the constructs in this study. Examination of the trait–symptom links using hierarchical multiple regression analyses supported most of the predicted relations. Specifically, negative emotionality emerged as a general predictor that was significantly related to all four symptom dimensions. In contrast, anxiety sensitivity was specific to panic and worry, whereas negative evaluation sensitivity was specific to social anxiety and worry. Finally, positive emotionality was uniquely related to social anxiety. The model accounted for a substantial amount of variance in the symptoms and almost all of the covariation among them.

Introduction

A large and flourishing area of research has sought to explicate the links between personality and psychopathology. This line of research is important for several reasons (for a discussion, see Watson & Clark, 1994). Most notably, it can play a key role in addressing basic diagnostic and taxonomic issues in psychopathology, such as explaining comorbidity and identifying its sources. Furthermore, investigations of the relations between personality and psychopathology can advance our understanding of etiology. For example, the broad trait of negative emotionality appears to contribute to all anxiety disorders and may account for a large portion of the comorbidity among these disorders (Mineka, Watson, & Clark, 1998; Watson, 1999). The basic goal of this paper is to articulate and test a structural framework for relations between anxiety disorders and relevant personality traits.

In following paragraphs, we will identify traits previously linked to anxiety disorders and discuss limitations of extant research, most notably the preponderance of correlated candidate trait predictors. We will propose a model—the multilevel trait predictor model—to overcome these problems. In developing this approach, we will consider principles that helped organize research on the relations between dimensions of mood and anxiety disorders, especially the integrative hierarchical model. Two types of traits will be considered for inclusion in the model: (1) traditional personality traits, such as the Big Five (Costa & McCrae, 1992), and (2) clinical traits that are dispositional constructs identified by psychopathology researchers as potential contributors to anxiety disorders.

Numerous studies have established that a basic personality dimension, termed neuroticism or negative emotionality (N/NE), has positive associations with all anxiety disorders, although its relation with generalized anxiety disorder (GAD) appears to be the strongest, and its link to specific phobia is the weakest (e.g., Bienvenu et al., 2001; Trull & Sher, 1994; Watson & Clark, 1995; Watson, Gamez, & Simms, 2005). Another general trait, extraversion or positive emotionality (E/PE), has shown consistent negative associations with agoraphobia, GAD, and especially social phobia (SP; Bienvenu et al., 2001; Gomez & Francis, 2003; Trull & Sher, 1994; Watson & Clark, 1995; Watson et al., 2005). There also is evidence that low conscientiousness is associated with obsessive-compulsive disorder (OCD; Rector, Hood, Richter & Bagby, 2002) and post-traumatic stress disorder (PTSD; Trull & Sher, 1994), while low Agreeableness has been linked only to PTSD (Trull & Sher, 1994). However, these associations are weaker and less consistent than those for N/NE and E/PE.

Psychopathology researchers have identified a number of trait-like constructs that are hypothesized to contribute to anxiety disorders. Anxiety sensitivity (AS) and negative evaluation sensitivity (NES) are among the best established clinical traits, and they both show fairly specific links to anxiety disorders (Fenigstein, Scheier, & Buss, 1975; Reiss & McNally, 1985). These clinical constructs were developed outside of the personality field, but they are conceptualized as dispositional characteristics, and the empirical evidence shows that their stability is comparable to that of typical personality dimensions (e.g., Fenigstein, 1987; Taylor, 1999). Accordingly, it is important to examine these clinical traits together with traditional personality traits.

Expectancy theory (Reiss, 1991) suggests that three fundamental dispositions are relevant to the generation of anxiety pathology, including AS and NES. AS refers to fears of physiological arousal symptoms. Individuals with high AS fear many strong arousal sensations, whereas those with low AS experience these sensations as unpleasant but nonthreatening. AS is believed to be a dispositional, trait-like characteristic that may precede the development of panic attacks or anxiety disorders (Taylor, 1999). AS was found to be highly elevated in patients with panic disorder and PTSD; lesser elevations were found in patients with GAD and SP (Rodriguez, Bruce, Pagano, Spencer, & Keller, 2004; Taylor, Koch, & McNally, 1992).

NES is typically defined as exaggerated fears relating to negative evaluation. According to expectancy theory, individuals high in NES are prone to experiencing discomfort in social situations and tend to avoid them. Consistent with Reiss’ model, NES shows clear specificity to SP, as patients with SP have elevated scores on measures of NES relative to normal controls and to patients with panic disorder or OCD (Ball, Otto, Pollack, Uccello, & Rosenbaum, 1995; Jostes, Pook, & Florin, 1999; Saboonchi, Lundh, & Ost, 1999).

In sum, previous research has linked anxiety disorders to a number of different traits, including broad personality dimensions and clinical traits. Importantly, many of these constructs are at least moderately related with one another, which makes it difficult to determine what traits are ultimately responsible for the observed correlations with anxiety disorders. Some studies have attempted to address this question by controlling for the contributions of certain traits (e.g., Gamez, Watson & Doebbeling, in press; Norton, Cox, Hewitt, & McLeod, 1997; Schmidt, Lerew, & Jackson, 1999). However, the picture is far from complete, as no study has included measures of all of these constructs. A comprehensive organizational framework is needed, and we believe that such framework can be adapted from a related area: research on links between the anxiety disorders and basic dimensions of affect. Organizational models have been successfully applied to the study of anxiety–affect relations, including the most recent proposal, the integrative hierarchical model of anxiety and depression (Mineka et al., 1998). Although, this model does not explicitly consider traits, its basic logic can be extended to personality–psychopathology research.

The integrative hierarchical model evolved from two influential models of relations between affect, anxiety and depression: the tripartite model (Clark & Watson, 1991) and the hierarchical model of the anxiety disorders (Barlow, 1991; Zinbarg & Barlow, 1996). The tripartite model attempted to explain the comorbidity of anxiety and depression by linking these phenomena to three dimensions of mood: negative affect, positive affect, and anxious arousal. According to the model, (low) positive affect and anxious arousal are unique features of depression and anxiety, respectively, while negative affect is the shared feature that primarily is responsible for the observed comorbidity among these disorders.

The hierarchical model also posited a shared factor of anxious apprehension—which is very similar to the negative affect dimension of the tripartite model (Brown, Chorpita, & Barlow, 1998; Zinbarg & Barlow, 1996)—but emphasized distinctions between individual anxiety disorders and assigned a unique contributor to each disorder. For instance, Brown and colleagues (1998) found that anxious arousal shows a unique association with panic disorder rather than being linked to all anxiety disorders.

The integrative hierarchical model has built on the hierarchical model and provided a further level of differentiation. The integrative model postulated that mood correlates of depressive and anxiety disorders can be classified in three categories: (1) a general factor, which is shared by all disorders; (2) a specific factor, which is linked to some disorders but not others; and (3) a unique factor, which is not shared with any other disorder.¹ Thus, there is one general factor, several specific factors, and one unique factor per disorder. Importantly, multiple degrees of specificity are possible (some constructs may contribute to two disorders, others to three, etc). Furthermore, the model allows multiple specific factors to contribute to a given disorder. In the integrative hierarchical model, the comorbidity between anxiety disorders is thought to result from the influence of both the general factor and these specific factors; the unique factors are uncorrelated by definition and, therefore, do not contribute to comorbidity.

This basic scheme of organizing correlates according to generality of their contributions can be applied to traits and facilitate the search for core dispositions associated with anxiety disorders. In other words, one trait (or a few traits) would be considered the general predictor contributing to all disorders, several other traits would be specific predictors linked to a few disorders, and a number of other traits would be uniquely associated with just one syndrome. Importantly, this framework can guide the analytic plan for investigations in this area. As noted earlier, a serious problem in personality-psychopathology research is that the traits are correlated; that is, they are not independent contributors to psychopathology. The logic of the integrative hierarchical model specifies that the contributions of the traits should be assessed in order of decreasing generality: contributions of the general trait should be assessed first, contributions of specific traits estimated next (also in order, according to how many other syndromes they are linked to), and the unique traits examined last. In other words, each new level needs to show incremental validity over more general traits. This multilevel analysis neatly partitions the variance in correlated traits, so that each factor makes an independent contribution in a meaningful order.

What traits should be placed at each level of this model? Current knowledge about the components of the integrative hierarchical model, however limited, provides a reasonable starting point. The model relates anxiety disorders to dimensions of affect: negative affect is the general factor that is common to all disorders, positive affect is a specific factor associated with depression and social phobia (and possibly a few other disorders), and anxious arousal is a unique factor associated with panic (Mineka et al., 1998). Notably, these transient affective states are closely linked with certain affective traits, especially N/NE and E/PE (Watson, 2000). Thus, it is possible to generate hypotheses about potential trait contributors to a disorder on the basis of relevant features identified in the integrative hierarchical model, and vice versa.

Clark, Watson, and Mineka (1994) attempted to do just that. On the basis of data linking negative affect to N/NE, positive affect to E/PE, and anxious arousal to AS, they proposed that (a) N/NE is a general trait associated with anxiety disorders and depression, (b) E/PE is a specific contributor to depression and SP, and (c) AS is a specific contributor to panic disorder and perhaps to other anxiety disorders, with the exception of specific phobia. Unfortunately, there has been little research in this direction since. A primary aim of this paper is to build on the proposal of Clark and colleagues (1994) and use key features of the integrative hierarchical model to organize findings on relations between traits and anxiety disorders. The second aim is to test the resulting multilevel trait predictor model of anxiety disorders or MTPM.

Our review of the relations between personality and psychopathology led us to propose the following model. N/NE clearly is a general contributor to anxiety disorders. (Low) E/PE appears to be a specific factor with a particularly strong link to SP as well as a possible lesser connection to GAD. AS is a specific contributor to panic disorder with possible links to PTSD, SP and GAD. NES appears to be a unique factor associated with SP. There is some evidence that conscientiousness and agreeableness are specific contributors to OCD and PTSD, but the data currently are too limited for these traits to be specified in the model. Importantly, the status of a trait in the MTPM is not based on its standing in personality taxonomy (i.e., N/NE and E/PE are broad dimensions, whereas AS and NES are relatively narrow traits), but rather on the generality of its contributions to psychopathology. This model clearly is incomplete—for instance, it lacks unique factors for many of the disorders—but we can expect it to account for a substantial proportion of the variance in each anxiety disorder as well as the covariances (comorbidity) between the disorders.

The idea of organizing trait contributors into specific and general predictors has already been voiced in the anxiety literature. Sexton, Norton, Walker, and Norton (2003) evaluated contributions of one general trait (N/NE) and two specific predispositions (AS and intolerance of uncertainty) to four symptom dimensions: worry, panic, OCD, and health anxiety. Norton, Sexton, Walker, and Norton (2005) extended this model by including another general trait (E/PE), an additional syndrome (dysphoric mood), and assessing unique contributions of each vulnerability factor in a series of hierarchical multiple regressions. These investigations were not based on the fully elaborated multilevel approach of the MTPM, but similar to the MTPM they incorporated clinical traits as well as personality dimensions, and distinguished between general and specific predictors. One important difference between the two approaches is that Norton et al. (2005) considered E/PE a general contributor, on par with N/NE, whereas E/PE is a specific predictor in the MTPM.

In this first test of the MTPM, we restricted our focus to four anxiety disorders—panic disorder, SP, OCD, and GAD—as they currently are best described by the model. We elected to operationalize psychopathology as continuous symptom dimensions rather than as dichotomous diagnoses (see Watson, 2005, for a discussion of the advantages of dimensional versus dichotomous assessment). It should be acknowledged that this approach only attempts to capture the central feature of a disorder, and other important characteristics (e.g., course) may be left out. Specifically, we identified panic attack symptoms as the core feature of panic disorder, social anxiety as the central symptom of SP, and pervasive, chronic worry as the primary component of GAD. OCD was operationalized using three correlated symptom dimensions—checking, cleaning, and ordering/rituals—in accordance with recent structural evidence (see Watson & Wu, 2005). We also assessed four key dispositional components of the MTPM: N/NE, E/PE, AS and NES.

In analyzing our data, we began by examining the relations among the trait and symptom variables to determine whether they represent valid indicators of the target constructs. Next, we performed a series of hierarchical regression analyses to evaluate the links between these traits and symptoms. A hierarchical approach (as opposed to simultaneous entry of the predictors) was necessary to establish incremental validity of specific traits over more general traits, following the logic outlined earlier. We decided, a priori, to consider not only statistical significance, but also practical significance, as criteria for interpreting an association as meaningful. Specifically, only regression coefficients that were statistically significant and at least moderate in size (β⩾.20) were interpreted.

We expected that N/NE would contribute significantly to each symptom dimension but would have an especially strong association with worry (Mineka et al., 1998; Watson et al., 2005). E/PE was hypothesized to account for additional variance in social anxiety and, to a lesser extent, worry. We predicted that AS would contribute incremental variance to panic, worry, and social anxiety, although we expected the association with panic to be the strongest. Finally, we hypothesized that NES would correlate only with social anxiety, after N/NE is controlled. In other words, we expected N/NE to emerge as a general predictor, AS and E/PE as specific predictors (associated with three and two syndromes, respectively), and NES as a unique contributor to social anxiety. The hypotheses are summarized in the first section of Table 1.

We further hypothesized that these traits jointly would account for a substantial amount of variance in all four types of symptoms; however, in light of previous research (Brown et al., 1998; Watson et al., 2005) we expected that they would explain the most variance in worry and the least variance in obsessive-compulsive symptoms. Finally, we anticipated that controlling for general and specific—but not unique—trait contributors would produce a noticeable decrease in the correlations among symptoms. In fact, if all of the general and specific trait contributors to the four disorders have been identified and properly controlled, the resulting partial correlations should be trivial and non-significant.