- Email: [email protected]
The joint structure of normal and pathological personality: Further evidence for a dimensional model
Available online at www.sciencedirect.com
ScienceDirect Comprehensive Psychiatry 55 (2014) 667 – 674 www.elsevier.com/locate/comppsych
The joint structure of normal and pathological personality: Further evidence for a dimensional model Michael P. Hengartner a,⁎, Vladeta Ajdacic-Gross a , Stephanie Rodgers a , Mario Müller a , Wulf Rössler a, b a
b
University of Zurich, Department of Psychiatry, Psychotherapy and Psychosomatics Collegium Helveticum, a Joint Research Institute between the University of Zurich and the Swiss Federal Institute of Technology, Zurich, Switzerland
Abstract Objective: The literature proposes a joint structure of normal and pathological personality with higher-order factors mainly based on the fivefactor model of personality (FFM). The purpose of the present study was to examine the joint structure of the FFM and the DSM-IV personality disorders (PDs) and to discuss this structure with regard to higher-order domains commonly reported in the literature. Methods: We applied a canonical correlation analysis, a series of principal component analyses with oblique Promax rotation and a bi-factor analysis with Geomin rotation on 511 subjects of the general population of Zurich, Switzerland, using data from the ZInEP Epidemiology Survey. Results: The 5 FFM traits and the 10 DSM-IV PD dimensions shared 77% of total variance. Component extraction tests pointed towards a two- and three-component solution. The two-component solution comprised a first component with strong positive loadings on neuroticism and all 10 PD dimensions and a second component with strong negative loadings on extraversion and openness and positive loadings on schizoid and avoidant PDs. The three-component solution added a third component with strong positive loadings on conscientiousness and agreeableness and a negative loading on antisocial PD. The bi-factor model provided evidence for 1 general personality dysfunction factor related to neuroticism and 5 group factors, although the interpretability of the latter was limited. Conclusions: Normal and pathological personality domains are not isomorphic or superposable, although they share a substantial proportion of variance. The two and three higher-order domains extracted in the present study correspond well to equivalent factor-solutions reported in the literature. Moreover, these superordinate factors can consistently be integrated within a hierarchical structure of alternative four- and five-factor models. The top of the hierarchy presumably constitutes a general personality dysfunction factor which is closely related to neuroticism. © 2014 Elsevier Inc. All rights reserved.
1. Introduction Researchers and clinicians agree that personality disorders (PDs) need a conceptual redefinition. The existing definition and operationalization of PDs lack accuracy and adequacy [1–3]. The DSM-5 Research Planning Conference on Personality Disorders (held in December 2004, in Arlington, VA, USA) concluded that PDs seem not to be discrete clinical conditions with distinct aetiologies, but rather distinctions along dimensions of general personality ⁎ Corresponding author at: University of Zurich, Department of Psychiatry, Psychotherapy and Psychosomatics, PO Box 1930, CH-8021 Zurich, Switzerland. Tel.: +41 44 296 75 87; fax: +41 44 296 74 49. E-mail address: [email protected] (M.P. Hengartner). 0010-440X/$ – see front matter © 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.comppsych.2013.10.011
functioning, which suggest a redefinition of the categorical DSM-IV PDs [4]. The most widely established exponent of a dimensional approach is the PD conceptualization based on the fivefactor model of personality (FFM) [5,6]. These five broad domains of general personality functioning are neuroticism, extraversion, openness, agreeableness, and conscientiousness. Extensive research has indicated that there are mainly four higher-order personality domains that underlie PD constructs, representing neuroticism (i.e. emotional dysregulation and negative affectivity), introversion (i.e. social withdrawal and detachment), disagreeableness (i.e. hostility and antagonism), and conscientiousness (i.e. compulsivity and constraint) [3,5,7,8]. However, there is more to it than that. The literature additionally provides substantial evidence
668
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674
for widely acknowledged personality models consisting of two [9–11] or three [12–14] superordinate factors. Those alternative models are likewise nested within the FFM domains. For a thorough integration of alternative models within a common hierarchical structure see Widiger and Simonsen [15]. The DSM-5 personality and personality disorder work group (P&PDWG) proposed a dimensional PD model comprising the following five higher-order personality dimensions: 1) negative affectivity, 2) detachment, 3) antagonism, 4) disinhibition, and 5) psychoticism [16]. The five proposed higher-order personality domains therefore represent an integration of the well-described taxonomy of the four-factor models detailed above plus psychoticism. After subsequent modifications the DSM-5 P&PDWG conceived a hybrid model and proposed that the PD traits might only be utilized for the assessment of the residual diagnosis formerly known as PD not otherwise specified. At the end of 2012 the APA board of trustees ultimately dismissed the revised PD model and declared that it needed further research. As a consequence the widely criticized categorization according to DSM-IV will persist in DSM-5. Thus, the major objective of the present study was to examine the joint structure of normal and pathological personality with respect to the FFM and to compare those higher-order factors to domains commonly reported in the literature by analyzing data from a population-based community sample. 2. Methods 2.1. Study design and sampling This study was conducted within the scope of the Epidemiology Survey of the “Zurich Programme for Sustainable Development of Mental Health Services” (ZInEP; in German: Zürcher Impulsprogramm zur nachhaltigen Entwicklung der Psychiatrie), a research and health care programme involving several psychiatric research divisions and mental health services from the canton of Zurich, Switzerland. The Epidemiology Survey is one of the six ZInEP projects and consists of four components: 1) a short telephone screening, 2) a comprehensive semistructured face-to-face interview followed by self-report questionnaires, 3) tests in the sociophysiological laboratory, and 4) a longitudinal survey (see Fig. 1). Telephone screening and semi-structured interviews started in August 2010, the tests at the sociophysiological laboratory in February 2011, and the longitudinal survey in April 2011. The screening ended in May 2012 and all other components in September 2012. First, 9829 Swiss males and females aged 20–41 years at the onset of the survey and representative of the canton of Zurich, Switzerland, were screened by computer assisted telephone interview (CATI) using the SCL-27 [17]. All participants were randomly chosen through the residents’
Fig. 1. The sampling procedure of the ZInEP Epidemiology Survey.
registration offices of all municipalities of the canton of Zurich. Residents without Swiss nationality were excluded from the study. The CATI was conducted by GfK (Growth for Knowledge), a major market and field research institute, in accordance with instructions from the ZInEP research team. The overall response rate was 53.6%. Reasons for nonresponse were no response, only telephone responder, incorrect telephone number, communication impossible, unavailability during the study period, and refusal by a third person or the target person. In cases where potential subjects were available, the response rate was 73.9%. Second, 1500 subjects were randomly selected from the initial screening sample for subsequent face-to-face interviews (response rate: 65.2%). We applied a stratifying sampling procedure including 60% high-scorers (scoring above the 75th percentile of the global severity index of the SCL-27) and 40% low-scorers (scoring below the 75th percentile of the global severity index). The basic sampling design was adapted from the prospective Zurich cohort study [18] and was chosen to enrich the sample with subjects at high-risk of mental disorders. Such a two-phase procedure with initial screening and subsequent comprehensive interview with a stratified subsample is fairly common in epidemiological research [19]. Face-to-face interviews were carried out by experienced and extensively trained clinical psychologists. The interviews took place either at the participants' homes or at the Zurich University Hospital of Psychiatry. All participants who completed the semi-structured interview were additionally assigned to complete various questionnaires. For this purpose, the sample was divided into subsamples focusing
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674
either on psychosis (N = 820) or on personality disorders (N = 680). Out of totally 680 subjects in the personality disorder subsample, 169 (24.9%) refused to return or complete all of the questionnaires required for the present study, leading to a final sample size of N = 511. The ZInEP Epidemiology Survey was approved by the ethics committee of the canton of Zurich (KEK) to fulfil all legal and data privacy protection requirements and is in strict accordance with the declaration of Helsinki of the World Medical Association. All participants gave their written informed consent. 2.2. Instruments and measures Personality disorder dimensions were provided by the Assessment of DSM-IV Personality Disorders Questionnaire (ADP-IV) [20]. The ADP-IV design allows a dimensional trait-score and a categorical PD diagnosis for each of the DSM-IV PDs. The ADP-IV is a paper-pencil self-report instrument consisting of 94 items, which represent the 80 criteria of the 10 DSM-IV PDs and the 14 research criteria of the depressive and the passive–aggressive PDs. Each traitquestion is rated on a 7-point Likert scale, ranging from “totally disagree” to “totally agree”. For the present study we used the German translation by Doering et al. [21]. Internal consistency of the dimensional PD scales is good and test– retest reliability and concurrent validity of the dimensional trait-scores are also satisfactory [22]. Most importantly, the ADP-IV showed good concordance with the SCID-II interview [23] and may be considered an economical and valid alternative to semi-structured interviews. The Big Five Inventory short form (BFI-S) [24] is a German adaptation of the popular Big Five Inventory by John et al. [25]. The questionnaire consists of 15 items divided into the five broad domains of neuroticism, extraversion, openness, agreeableness, and conscientiousness. The items are rated on a 7-point Likert scale ranging from “don’t agree at all” to “completely agree”. The BFI-S has shown good reliability and validity [24].
669
Velicer’s minimum average partial (MAP) test [27] and Horn’s parallel analysis (PA) test [28]. These two tests are widely acknowledged as providing the most accurate determinants for component retention [29,30]. Both MAP and PA were conducted with syntax programmes provided by O’Connor [30]. The component structure was inspected according to the guidelines provided by Costello and Osborne [29]. A clean component structure ideally comprises the following: that each component at least loads on three items higher than 0.50, that each item should exhibit a loading of at least 0.32, and that only one component should load higher than 0.32 on the same item (two or more components loading higher than 0.32 on the same item are referred to as cross-loadings). MVA, CCA and PCA were carried out with SPSS 20 for Macintosh. We additionally conducted a bi-factor exploratory factor analysis (EFA) using MPlus 7 [31]. A bi-factor EFA computes a general factor that loads on all items and that is orthogonal to the Geomin-rotated group factors [32]. The model fit was evaluated with the χ 2-test of model fit and the following approximate fit indices (AFI): the comparative fit index (CFI), the Tucker–Lewis index (TLI), the root mean square error of approximation (RMSEA), and the standardized root mean square residual (SRMR). According to the χ 2-test, a wellfitting model should provide an insignificant result (i.e., above the 0.05 threshold). However, with increasing sample sizes, a χ 2 value easily becomes significant and the test tends to reject also well-fitting models [33]. Recommended cut-off values of AFI for a good model fit are CFI N 0.95, TLI N 0.95, RMSEA b 0.06, and SRMR b 0.08 [34].
3. Results The CCA indicated that the 5 FFM traits and the 10 DSMIV PD dimensions shared 77.0% of total variance, which was of course statistically highly significant (λ = 0.2303, F(50, 2270.03) = 17.245, p b 0.001). The scree-test for the PCA is shown in Fig. 2. The eigenvalue and proportion of variance
2.3. Statistical analysis Values on all FFM and PD dimensions were missing completely at random (MCAR) according to Little’s MCAR test (χ 2 = 189.972, df = 182, p = 0.328). Therefore, to obtain complete data from all 511 participants included in the analysis we conducted a missing value analysis (MVA) as recommended by Schafer and Graham [26]. No variable was missing in more than maximally 9 cases (1.8%). MVA was carried out with the full information maximum likelihood estimation. A canonical correlation analysis (CCA) was conducted to estimate the proportion of total variance shared between all FFM traits and PD dimensions. Afterwards we conducted a series of principal component analyses (PCA) using oblique Promax rotation on a 15 × 15 item correlation matrix (comprising 10 ADP-IV scales and 5 BFI-S scales). We determined the best fitting component solution by inspecting
Fig. 2. The scree plot of a principal component analysis with 15 items.
670
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674
Table 1 Results of a principal component analysis with 15 items: initial eigenvalues and proportion of variance explained for the first 10 components.
Component 1 Component 2 Component 3 Component 4 Component 5 Component 6 Component 7 Component 8 Component 9 Component 10
Eigenvalue
% Variance
% Cumulative Variance
6.318 1.721 1.263 1.102 1.020 0.684 0.511 0.489 0.402 0.339
42.121 11.471 8.418 7.346 6.801 4.560 3.408 3.258 2.681 2.257
42.121 53.592 62.010 69.356 76.157 80.717 84.125 87.383 90.064 92.321
Components
explained for the first 10 components are shown in Table 1. The MAP pointed towards a two-component solution and the PA towards a three-component solution. The two-component solution is indicated in Table 2. The extraction of a structure with two components accounted for 53.6% of shared variance in normal and pathological personality. The first component loaded strongly on neuroticism and on all 10 PD dimensions. The second component showed strong negative loadings on extraversion and openness and strong positive loadings on schizoid and avoidant PDs. Conscientiousness and agreeableness showed no substantial loadings. Accordingly, the communality, i.e. the total variance explained by all extracted components, was low in those two traits (both h 2 b 0.120). The correlation between the two extracted components was r = 0.249. Although three cross-loadings emerged, the component structure was overall sufficiently clean. The three-component solution is indicated in Table 3. The extraction of three components explained 62.0% of variance shared between normal and pathological personality. The first component loaded again strongly on neuroticism and on
Table 2 Two-component solution of a principal component analysis: Promax rotated component-loadings and communalities (h 2) of the ADP-IV and BFI-S. Components
BFI-S Neuroticism BFI-S Extraversion BFI-S Openness BFI-S Conscientiousness BFI-S Agreeableness ADP-IV Paranoid ADP-IV Schizoid ADP-IV Schizotypal ADP-IV Antisocial ADP-IV Borderline ADP-IV Histrionic ADP-IV Narcissistic ADP-IV Avoidant ADP-IV Dependent ADP-IV Obsessive-Compulsive
1
2
h2
0.451 0.020 0.375 −0.096 −0.291 0.795 0.403 0.827 0.735 0.888 0.865 0.801 0.590 0.661 0.663
0.119 −0.846 −0.664 −0.296 −0.099 0.062 0.467 0.049 −0.199 −0.049 −0.219 −0.053 0.515 0.319 0.216
0.244 0.708 0.458 0.111 0.109 0.660 0.474 0.707 0.507 0.769 0.701 0.623 0.765 0.644 0.558
Loadings greater than 0.320 are indicated in bold.
Table 3 Three-component solution of a principal component analysis: Promax rotated component-loadings and communalities (h 2) of the ADP-IV and BFI-S.
BFI-S Neuroticism BFI-S Extraversion BFI-S Openness BFI-S Conscientiousness BFI-S Agreeableness ADP-IV Paranoid ADP-IV Schizoid ADP-IV Schizotypal ADP-IV Antisocial ADP-IV Borderline ADP-IV Histrionic ADP-IV Narcissistic ADP-IV Avoidant ADP-IV Dependent ADP-IV Obsessive-Compulsive
1
2
3
h2
0.493 −0.179 0.260 0.079 −0.078 0.796 0.509 0.821 0.566 0.863 0.782 0.772 0.760 0.784 0.800
−0.071 0.809 0.671 0.291 0.084 0.014 −0.408 0.029 0.250 0.129 0.287 0.123 −0.433 −0.239 −0.139
0.012 0.087 0.185 0.778 0.742 −0.063 −0.067 −0.075 −0.358 −0.053 −0.090 −0.062 0.087 0.102 0.230
0.248 0.709 0.472 0.616 0.574 0.660 0.475 0.707 0.606 0.770 0.704 0.623 0.806 0.682 0.651
Loadings greater than 0.320 are indicated in bold.
all PD dimensions. The second component was also consistent with the second component from the twocomponent solution delineated above, showing positive loadings on extraversion and openness and negative loadings on schizoid and avoidant PDs. The third component demonstrated two strongly positive loadings on conscientiousness and agreeableness and a moderate negative loading on antisocial PD. The correlation between components 1 and 2 as well as between 1 and 3 was r = −0.085 and r = −0.239, respectively. The correlation between components 2 and 3 was r = −0.114. Except for a few minor violations this component structure was likewise adequately clean. The high proportion of variance explained by the first component and its strong loadings on all PD dimensions point towards a general PD factor. Therefore we additionally conducted a series of bi-factor EFA with 2 factors (i.e. 1 general factor and 1 group factor) up to 8 factors (i.e. 1 general factor and 7 group factors). The 7- and 8factor solutions did not converge and the best fit to the data was found for the 6-factor solution. Although the χ 2 test was statistically highly significant (p b 0.001), the AFI indicated an acceptable to good model fit, depending on the AFI applied (CFI = 0.98; TLI = 0.94; RMSEA = 0.07; SRMR = 0.01). The 6-factor solution is indicated in Table 4. The general factor loaded strongly on neuroticism and on all PD trait-scores. The group factors depicting specific personality styles showed a rather intricate and messy factor structure owing to the limited number of items included in the analysis.
4. Discussion In this study we examined the joint structure of normal and pathological personality using the 5 FFM traits and the
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674
671
Table 4 Six-factor solution of a bi-factor exploratory factor analysis: factor-loadings and communalities (h 2) of the ADP-IV and BFI-S. General factor
BFI-S Neuroticism BFI-S Extraversion BFI-S Openness BFI-S Conscientiousness BFI-S Agreeableness ADP-IV Paranoid ADP-IV Schizoid ADP-IV Schizotypal ADP-IV Antisocial ADP-IV Borderline ADP-IV Histrionic ADP-IV Narcissistic ADP-IV Avoidant ADP-IV Dependent ADP-IV Obs.-Comp.
Group factors
1
2
3
4
5
6
h2
0.496 −0.282 0.106 −0.158 −0.274 0.793 0.513 0.806 0.589 0.875 0.762 0.727 0.788 0.791 0.718
0.160 0.690 0.271 0.218 0.024 0.000 −0.521 −0.120 0.057 0.189 0.341 0.059 −0.439 −0.071 −0.048
−0.064 −0.044 0.150 0.103 0.759 −0.238 −0.016 −0.021 0.002 −0.018 0.055 −0.155 0.175 0.238 0.038
0.056 0.044 0.019 0.534 0.078 0.201 0.025 −0.003 −0.422 −0.036 −0.233 0.032 0.025 −0.123 0.155
−0.437 0.042 0.256 0.003 0.036 0.069 0.359 0.286 0.473 −0.060 −0.004 0.258 −0.044 −0.248 −0.004
−0.071 0.006 −0.195 0.062 −0.023 −0.038 0.052 −0.163 −0.012 −0.164 0.086 0.317 −0.040 0.085 0.364
0.426 0.579 0.265 0.415 0.686 0.714 0.565 0.746 0.710 0.835 0.738 0.745 0.855 0.774 0.702
Loadings greater than 0.320 are indicated in bold.
10 DSM-IV PD dimensions assessed in 511 subjects of the general population of the canton of Zurich, Switzerland. Normal and pathological personality traits shared altogether 77% of their total variance. It is therefore legitimate to conceive PDs as extreme variants along continuous common personality domains, although normal and pathological personality traits are neither completely isomorphic nor superposable. Thus, in agreement with previous reports [35,36], normal and pathological personality domains do not map 1:1, although they share a substantial proportion of variance. A joint structure of normal and pathological personality comprising two and three Promax-rotated components was determined by the two most reliable and valid component extraction tests. The two-component solution explained 54% of shared variance in FFM traits and DSM-IV PD dimensions. The first component loaded uniformly highly on neuroticism and on all 10 PD dimensions, whereas the second component loaded negatively on extraversion and openness and positively on schizoid and avoidant PDs. The first component mainly represents emotional instability, impulsivity and negative affectivity. These primary PD traits have previously been assigned to FFM neuroticism [37–40]. The proportion of total variance explained was distinctly higher for the first component. Neuroticism accordingly has a predominant effect on the joint structure of normal and pathological personality that exceeds the effects of other personality traits. In other words, neuroticism constitutes the basis of a general, unspecific personality dysfunction. This finding has consistently been reported in the literature [41,42]. The second component describes features of detachment, self-consciousness and anhedonia. Associations of these features with inverted extraversion (i.e. introversion) and low openness have also repeatedly been shown [38,43–45]. Moreover, a superordinate factor related to both extraversion and openness has consistently been replicated in the literature [9,10]. A
highly similar two-factor structure distinguishing a neuroticism factor from a detachment factor has additionally been proposed by Markon et al. [7] as well as by Watson et al. [46]. The three-component solution explained 62% of shared variance in FFM traits and DSM-IV PD dimensions. This common structure adds a third component to the two detailed above that loads positively on conscientiousness and agreeableness and negatively on antisocial PD. The third component thus typically describes disinhibition and aggressiveness, which have also been reported by others [40,46,47]. Altogether the three superordinate domains of this joint structure of normal and pathological personality closely resemble the three-factor solution identified by Markon et al. [7], which comprises negative emotionality, positive emotionality (i.e. sociability), and disinhibition. Other proponents of a similar structure with three higherorder domains are Clark’s [12] superordinate factors of negative temperament, positive temperament and disinhibition and Zuckerman and colleagues’ [14,48] superordinate factors of neuroticism–emotionality, extraversion–sociability and impulsive–unsocialized-sensation-seeking. Moreover, our three-component PCA structure also bears considerable resemblance to Eysenck’s [13,49] biological model of personality, which consists of neuroticism, extraversion and psychoticism. Our bi-factor EFA provides further evidence for the separation of a general personality dysfunction factor from specific PD styles and traits [50–53]. As detailed above, it appears that this general factor is closely linked to neuroticism, which is in line with the literature [41,42]. A factor of general personality dysfunction atop the hierarchy of pathological personality structure that loads on higherorder domains such as negative emotionality, detachment, and disinhibition has also been reported by the DSM-5 P&PDWG [54]. Unfortunately, the group factors that constitute the specific PD styles were rather inconclusive
672
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674
in our analysis, because they did not provide a clean factor structure. Specifically, most group factors were not reliable due to the absence of several high loadings. A stable factor should ideally comprise at least three loadings N0.500 [29]. We therefore abstain from a discussion of those group factors. However, that limitation should not weaken the main conclusion of the bi-factor EFA, which was conducted to provide further evidence for a general PD factor that loads on all DSM-IV PD dimensions. The common four- and five-factor models of normal and pathological personality have been discussed and reviewed by various authors [e.g. 3,5,8,55]. The DSM-5 P&PDWG conceived a five-factor model of pathological personality [16], which also closely converges with the FFM traits [40]. In this respect it is worth noting that Krueger et al. [16] did not follow the results of the MAP and PA tests, which recommended 3 and 6 factors, respectively. Interestingly, computing three instead of five superordinate factors based on the 25 primary traits of the DSM-5 PD model Wright et al. [54] found the following three higher-order domains: negative affectivity, detachment and externalizing. Those three factors correspond closely to the three components extracted in the present study and to the three-factor models detailed above. At the next level the externalizing factor split into what the authors referred to as antagonism and disinhibition. At the final level, the five-factor structure, the psychoticism factor ultimately emerged [54]. The DSM-5 P&PDWG narrowly defines psychoticism as schizotypy, which is not consistent with the conceptualization by Eysenck and Eysenck [13], which relates psychoticism more broadly to aggressiveness and impulsiveness. Nevertheless, on a less fragmented level with less factorial splitting, our two- and three-factor solutions are highly consistent with the dimensional PD model proposed by the DSM-5 P&PDWG. This integration of alternative factorial models within a common hierarchical structure is in accordance with the comprehensive review by Widiger and Simonsen [15]. 4.1. Limitations This study is subject to the following limitations: First, we used a short form of the FFM personality inventory. Although this instrument is validated, it does not assess all facets of the original form. This may have slightly influenced our results. Second, our PD measures relied on a self-report questionnaire, which is critically appraised by some researchers, although the ADP-IV showed good concordance with semi-structured interviews [23]. Third, the group factors of our bi-factor EFA were inconclusive because of insufficient factor loadings. Presumably an analysis on the primary facet level including various items would provide a cleaner factor structure and thus interpretable group factors. 4.2. Summary and conclusions With regard to the revisions of the DSM and ICD a dimensional PD concept is widely considered to be a marked
improvement, providing greater validity and reliability than the current dichotomous categorical definition [1–3,56,57]. Although the DSM-5 ultimately dismissed a re-conceptualized dimensional PD model and retained the 10 dichotomous DSM-IV PD categories it is important to provide further evidence for a joint higher-order structure of normal and pathological personality. Convergent integration of alternative dimensional models within a common hierarchical structure is needed in order that the dimensional trait-model of the DSM-5 P&PDWG might be moved from Section 3 into the main section of the manual, for instance in the subsequent edition DSM-5.1. In summary, we found that normal and pathological personality dimensions share a considerably large proportion of variance. We provide strong empirical support for a general personality dysfunction factor which loads strongly on neuroticism and on all DSM-IV PDs [42,54]. We also provide further evidence for a common two- and three-factor structure of normal and pathological personality as represented by associations between FFM traits and DSM-IV PD dimensions. Our extracted components show close correspondence with well-established and carefully validated personality models such as the SNAP [12] or the Big Three [14]. Moreover, our component-structure may be further split up in order to correspond to the intensively promoted four- and five-factor models [15]. Therefore we claim that this study makes an important contribution to a more valid and appropriate understanding and conceptualization of PDs.
Conflicts of interests None. Acknowledgment ZInEP was supported by a private donation. The donor had no further role in experimental design; the collection, analysis, and interpretation of data; the writing of this report; or the decision to submit this paper for publication. References [1] Clark LA. Assessment and diagnosis of personality disorder: perennial issues and an emerging reconceptualization. Annu Rev Psychol 2007;58:227-57. [2] Farmer RF. Issues in the assessment and conceptualization of personality disorders. Clin Psychol Rev 2000;20:823-51. [3] Trull TJ, Durrett CA. Categorical and dimensional models of personality disorder. Annu Rev Clin Psychol 2005;1:355-80. [4] Widiger TA, Simonsen E, Krueger R, Livesley WJ, Verheul R. Personality disorder research agenda for the DSM-V. J Pers Disord 2005;19:315-38. [5] Widiger TA, Livesley WJ, Clark LA. An integrative dimensional classification of personality disorder. Psychol Assess 2009;21:243-55. [6] Widiger TA, Mullins-Sweatt SN. Five-factor model of personality disorder: a proposal for DSM-V. Annu Rev Clin Psychol 2009;5:197-220.
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674 [7] Markon KE, Krueger RF, Watson D. Delineating the structure of normal and abnormal personality: an integrative hierarchical approach. J Pers Soc Psychol 2005;88:139-57. [8] Watson D, Clark LA, Harkness AR. Structures of personality and their relevance to psychopathology. J Abnorm Psychol 1994;103:18-31. [9] DeYoung CG. Higher-order factors of the Big Five in a multiinformant sample. J Pers Soc Psychol 2006;91:1138-51. [10] Digman JM. Higher-order factors of the Big Five. J Pers Soc Psychol 1997 Dec;73(6):1246-56. [11] Krueger RF, Tackett JL. Personality and psychopathology: working toward the bigger picture. J Pers Disord 2003;17:109-28. [12] Clark LA. Manual for the Schedule for Nonadaptive and Adaptive Personality. Minneapolis: University of Minnesota Press; 1993. [13] Eysenck SBG, Eysenck HJ. Manual of the EPQ (personality questionnaire). San Diego, CA: Educational and Industrial Testing Service; 1975. [14] Zuckerman M, Kuhlman DM, Joireman J, Teta P, Kraft M. A comparison of three structural models for personality: the big three, the big five, and the alternative five. J Pers Soc Psychol 1993;65:757-68. [15] Widiger TA, Simonsen E. Alternative dimensional models of personality disorder: finding a common ground. J Pers Disord 2005; 19:110-30. [16] Krueger RF, Derringer J, Markon KE, Watson D, Skodol AE. Initial construction of a maladaptive personality trait model and inventory for DSM-5. Psychol Med 2012;42:1879-90. [17] Hardt J, Egle UT, Kappis B, Hessel A, Brähler E. Symptom Checklist SCL-27: results of a representative German survey. Psychother Psych Med 2004;54:214-23. [18] Angst J, Gamma A, Neuenschwander M, Ajdacic-Gross V, Eich D, Rössler W, et al. Prevalence of mental disorders in the Zurich cohort study: a twenty year prospective study. Epidemiol Psichiatr Soc 2005; 14:68-76. [19] Dunn G, Pickles A, Tansella M, Vazquez-Barquero JL. Two-phase epidemiological surveys in psychiatric research. Br J Psychiatry 1999; 174:95-100. [20] Schotte CKW, de Doncker D. ADP-IV questionnaire. Antwerp, Belgium: University Hospital Antwerp; 1994. [21] Doering S, Renn D, Hofer S, Rumpold G, Smrekar U, Janecke N, et al. Validation of the German version of the Assessment of DSM-IV Personality Disorders (ADP-IV) questionnaire. Z Psychosom Med Psychother 2007;53:111-28. [22] Schotte CKW, de Doncker D, Vankerckhoven C, Vertommen H, Cosyns P. Self-report assessment of the DSM-IV personality disorders. Measurement of trait and distress characteristics: the ADP-IV. Psychol Med 1998;28:1179-88. [23] Schotte CKW, de Doncker DA, Dmitruk D, Van Mulders I, D'Haenen H, Cosyns P. The ADP-IV questionnaire: differential validity and concordance with the semi-structured interview. J Pers Disord 2004; 18:405-19. [24] Schupp J, Gerlitz J-Y. BFI-S: Big Five Inventory-SOEP. In: & Glöckner-Rist A, editor. Zusammenstellung sozialwissenschaftlicher Items und Skalen (ZIS Version 1200). Bonn: GESIS; 2008. [25] John OP, Donahue EM, Kentle RL. The Big Five Inventory — versions 4a and 54. Berkeley, CA: University of California, Berkeley, Institute of Personality and Social Research; 1991. [26] Schafer JL, Graham JW. Missing data: our view of the state of the art. Psychol Methods 2002;7:147-77. [27] Velicer WF. Determining the number of components from the matrix of partial correlations. Psychometrika 1976;41:321-7. [28] Horn JL. A rationale and test for the number of factors in factor analysis. Psychometrika 1965;30:179-85. [29] Costello AB, Osborne JW. Best practices in exploratory factor analysis: four recommendations for getting the most from your analysis. Pract Assess Res Eval 2005;10(7) [Available online: http:// pareonline.net/getvn.asp?v=10&n=7]. [30] O'Connor BP. SPSS and SAS programs for determining the number of components using parallel analysis and velicer's MAP test. Behav Res Methods Instrum Comput 2000;32:396-402.
673
[31] Muthén LK, Muthén BO. Mplus user's guide. Seventh ed. Los Angeles, CA: Muthén & Muthén; 1998–2012. [32] Jennrich RI, Bentler PM. Exploratory bi-factor analysis. Psychometrika 2011;76:537-49. [33] Steiger JH. Understanding the limitations of global fit assessment in structural equation modeling. Pers Individ Differ 2007;42:893-8. [34] Schreiber JB, Stage FK, King J, Nora A, Barlow EA. Reporting structural equation modeling and confirmatory factor analysis results: a review. J Educ Res 2006;99:323-37. [35] Samuel DB, Simms LJ, Clark LA, Livesley WJ, Widiger TA. An item response theory integration of normal and abnormal personality scales. Pers Disord 2010;1:5-21. [36] Stepp SD, Yu L, Miller JD, Hallquist MN, Trull TJ, Pilkonis PA. Integrating competing dimensional models of personality: linking the SNAP, TCI, and NEO using Item Response Theory. Pers Disord 2012; 3:107-26. [37] Blais MA. Clinician ratings of the five-factor model of personality and the DSM-IV personality disorders. J Nerv Ment Dis 1997;185:388-93. [38] Leibing E, Jamrozinski K, Vormfelde SV, Stahl J, Doering S. Dimensions of personality—relationship between DSM-IV personality disorder symptoms, the five-factor model, and the biosocial model of personality. J Pers Disord 2008;22:101-8. [39] Livesley WJ, Jang KL, Vernon PA. Phenotypic and genetic structure of traits delineating personality disorder. Arch Gen Psychiatry 1998; 55:941-8. [40] Thomas KM, Yalch MM, Krueger RF, Wright AG, Markon KE, Hopwood CJ. The convergent structure of DSM-5 personality trait facets and five-factor model trait domains. Assessment 2013;20:308-11. [41] Krueger RF, Eaton NR, Clark LA, Watson D, Markon KE, Derringer J, et al. Deriving an empirical structure of personality pathology for DSM-5. J Pers Disord 2011;25:170-91. [42] Saulsman LM, Page AC. The five-factor model and personality disorder empirical literature: a meta-analytic review. Clin Psychol Rev 2004;23:1055-85. [43] Piedmont RL, Sherman MF, Sherman NC, Dy-Liacco GS, Williams JE. Using the five-factor model to identify a new personality disorder domain: the case for experiential permeability. J Pers Soc Psychol 2009;96:1245-58. [44] Reynolds SK, Clark LA. Predicting dimensions of personality disorder from domains and facets of the Five-Factor Model. J Pers 2001;69:199-222. [45] Tackett JL, Silberschmidt AL, Krueger RF, Sponheim SR. A dimensional model of personality disorder: incorporating DSM cluster A characteristics. J Abnorm Psychol 2008;117:454-9. [46] Watson D, Clark LA, Chmielewski M. Structures of personality and their relevance to psychopathology: II. Further articulation of a comprehensive unified trait structure. J Pers 2008;76:1545-86. [47] Nestadt G, Costa Jr PT, Hsu FC, Samuels J, Bienvenu OJ, Eaton WW. The relationship between the five-factor model and latent Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition personality disorder dimensions. Compr Psychiatry 2008;49:98-105. [48] Zuckerman M, Kuhlman DM, Camac C. What lies beyond E and N? Factor analyses of scales believed to measure basic dimensions of personality. J Pers Soc Psychol 1988;54:96-107. [49] Eysenck HJ. Biological basis of personality. Nature 1963;199:1031-4. [50] Bornstein RF, Huprich SK. Beyond dysfunction and threshold-based classification: a multidimensional model of personality disorder diagnosis. J Pers Disord 2011;25:331-7. [51] Hopwood CJ, Malone JC, Ansell EB, Sanislow CA, Grilo CM, McGlashan TH, et al. Personality assessment in DSM-5: empirical support for rating severity, style, and traits. J Pers Disord 2011;25: 305-20. [52] Livesley WJ. An empirically-based classification of personality disorder. J Pers Disord 2011;25:397-420. [53] Morey LC, Berghuis H, Bender DS, Verheul R, Krueger RF, Skodol AE. Toward a model for assessing level of personality functioning in
674
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674
DSM-5, part II: empirical articulation of a core dimension of personality pathology. J Pers Assess 2011;93:347-53. [54] Wright AG, Thomas KM, Hopwood CJ, Markon KE, Pincus AL, Krueger RF. The hierarchical structure of DSM-5 pathological personality traits. J Abnorm Psychol 2012;121:951-7. [55] Samuel DB, Widiger TA. A meta-analytic review of the relationships between the five-factor model and DSM-IV-TR personality
disorders: a facet level analysis. Clin Psychol Rev 2008;28: 1326-42. [56] Verheul R. Clinical utility of dimensional models for personality pathology. J Pers Disord 2005;19:283-302. [57] Widiger TA, Lowe JR. A dimensional model of personality disorder: proposal for DSM-V. Psychiatr Clin North Am 2008;31: 363-78.
ScienceDirect Comprehensive Psychiatry 55 (2014) 667 – 674 www.elsevier.com/locate/comppsych
The joint structure of normal and pathological personality: Further evidence for a dimensional model Michael P. Hengartner a,⁎, Vladeta Ajdacic-Gross a , Stephanie Rodgers a , Mario Müller a , Wulf Rössler a, b a
b
University of Zurich, Department of Psychiatry, Psychotherapy and Psychosomatics Collegium Helveticum, a Joint Research Institute between the University of Zurich and the Swiss Federal Institute of Technology, Zurich, Switzerland
Abstract Objective: The literature proposes a joint structure of normal and pathological personality with higher-order factors mainly based on the fivefactor model of personality (FFM). The purpose of the present study was to examine the joint structure of the FFM and the DSM-IV personality disorders (PDs) and to discuss this structure with regard to higher-order domains commonly reported in the literature. Methods: We applied a canonical correlation analysis, a series of principal component analyses with oblique Promax rotation and a bi-factor analysis with Geomin rotation on 511 subjects of the general population of Zurich, Switzerland, using data from the ZInEP Epidemiology Survey. Results: The 5 FFM traits and the 10 DSM-IV PD dimensions shared 77% of total variance. Component extraction tests pointed towards a two- and three-component solution. The two-component solution comprised a first component with strong positive loadings on neuroticism and all 10 PD dimensions and a second component with strong negative loadings on extraversion and openness and positive loadings on schizoid and avoidant PDs. The three-component solution added a third component with strong positive loadings on conscientiousness and agreeableness and a negative loading on antisocial PD. The bi-factor model provided evidence for 1 general personality dysfunction factor related to neuroticism and 5 group factors, although the interpretability of the latter was limited. Conclusions: Normal and pathological personality domains are not isomorphic or superposable, although they share a substantial proportion of variance. The two and three higher-order domains extracted in the present study correspond well to equivalent factor-solutions reported in the literature. Moreover, these superordinate factors can consistently be integrated within a hierarchical structure of alternative four- and five-factor models. The top of the hierarchy presumably constitutes a general personality dysfunction factor which is closely related to neuroticism. © 2014 Elsevier Inc. All rights reserved.
1. Introduction Researchers and clinicians agree that personality disorders (PDs) need a conceptual redefinition. The existing definition and operationalization of PDs lack accuracy and adequacy [1–3]. The DSM-5 Research Planning Conference on Personality Disorders (held in December 2004, in Arlington, VA, USA) concluded that PDs seem not to be discrete clinical conditions with distinct aetiologies, but rather distinctions along dimensions of general personality ⁎ Corresponding author at: University of Zurich, Department of Psychiatry, Psychotherapy and Psychosomatics, PO Box 1930, CH-8021 Zurich, Switzerland. Tel.: +41 44 296 75 87; fax: +41 44 296 74 49. E-mail address: [email protected] (M.P. Hengartner). 0010-440X/$ – see front matter © 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.comppsych.2013.10.011
functioning, which suggest a redefinition of the categorical DSM-IV PDs [4]. The most widely established exponent of a dimensional approach is the PD conceptualization based on the fivefactor model of personality (FFM) [5,6]. These five broad domains of general personality functioning are neuroticism, extraversion, openness, agreeableness, and conscientiousness. Extensive research has indicated that there are mainly four higher-order personality domains that underlie PD constructs, representing neuroticism (i.e. emotional dysregulation and negative affectivity), introversion (i.e. social withdrawal and detachment), disagreeableness (i.e. hostility and antagonism), and conscientiousness (i.e. compulsivity and constraint) [3,5,7,8]. However, there is more to it than that. The literature additionally provides substantial evidence
668
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674
for widely acknowledged personality models consisting of two [9–11] or three [12–14] superordinate factors. Those alternative models are likewise nested within the FFM domains. For a thorough integration of alternative models within a common hierarchical structure see Widiger and Simonsen [15]. The DSM-5 personality and personality disorder work group (P&PDWG) proposed a dimensional PD model comprising the following five higher-order personality dimensions: 1) negative affectivity, 2) detachment, 3) antagonism, 4) disinhibition, and 5) psychoticism [16]. The five proposed higher-order personality domains therefore represent an integration of the well-described taxonomy of the four-factor models detailed above plus psychoticism. After subsequent modifications the DSM-5 P&PDWG conceived a hybrid model and proposed that the PD traits might only be utilized for the assessment of the residual diagnosis formerly known as PD not otherwise specified. At the end of 2012 the APA board of trustees ultimately dismissed the revised PD model and declared that it needed further research. As a consequence the widely criticized categorization according to DSM-IV will persist in DSM-5. Thus, the major objective of the present study was to examine the joint structure of normal and pathological personality with respect to the FFM and to compare those higher-order factors to domains commonly reported in the literature by analyzing data from a population-based community sample. 2. Methods 2.1. Study design and sampling This study was conducted within the scope of the Epidemiology Survey of the “Zurich Programme for Sustainable Development of Mental Health Services” (ZInEP; in German: Zürcher Impulsprogramm zur nachhaltigen Entwicklung der Psychiatrie), a research and health care programme involving several psychiatric research divisions and mental health services from the canton of Zurich, Switzerland. The Epidemiology Survey is one of the six ZInEP projects and consists of four components: 1) a short telephone screening, 2) a comprehensive semistructured face-to-face interview followed by self-report questionnaires, 3) tests in the sociophysiological laboratory, and 4) a longitudinal survey (see Fig. 1). Telephone screening and semi-structured interviews started in August 2010, the tests at the sociophysiological laboratory in February 2011, and the longitudinal survey in April 2011. The screening ended in May 2012 and all other components in September 2012. First, 9829 Swiss males and females aged 20–41 years at the onset of the survey and representative of the canton of Zurich, Switzerland, were screened by computer assisted telephone interview (CATI) using the SCL-27 [17]. All participants were randomly chosen through the residents’
Fig. 1. The sampling procedure of the ZInEP Epidemiology Survey.
registration offices of all municipalities of the canton of Zurich. Residents without Swiss nationality were excluded from the study. The CATI was conducted by GfK (Growth for Knowledge), a major market and field research institute, in accordance with instructions from the ZInEP research team. The overall response rate was 53.6%. Reasons for nonresponse were no response, only telephone responder, incorrect telephone number, communication impossible, unavailability during the study period, and refusal by a third person or the target person. In cases where potential subjects were available, the response rate was 73.9%. Second, 1500 subjects were randomly selected from the initial screening sample for subsequent face-to-face interviews (response rate: 65.2%). We applied a stratifying sampling procedure including 60% high-scorers (scoring above the 75th percentile of the global severity index of the SCL-27) and 40% low-scorers (scoring below the 75th percentile of the global severity index). The basic sampling design was adapted from the prospective Zurich cohort study [18] and was chosen to enrich the sample with subjects at high-risk of mental disorders. Such a two-phase procedure with initial screening and subsequent comprehensive interview with a stratified subsample is fairly common in epidemiological research [19]. Face-to-face interviews were carried out by experienced and extensively trained clinical psychologists. The interviews took place either at the participants' homes or at the Zurich University Hospital of Psychiatry. All participants who completed the semi-structured interview were additionally assigned to complete various questionnaires. For this purpose, the sample was divided into subsamples focusing
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674
either on psychosis (N = 820) or on personality disorders (N = 680). Out of totally 680 subjects in the personality disorder subsample, 169 (24.9%) refused to return or complete all of the questionnaires required for the present study, leading to a final sample size of N = 511. The ZInEP Epidemiology Survey was approved by the ethics committee of the canton of Zurich (KEK) to fulfil all legal and data privacy protection requirements and is in strict accordance with the declaration of Helsinki of the World Medical Association. All participants gave their written informed consent. 2.2. Instruments and measures Personality disorder dimensions were provided by the Assessment of DSM-IV Personality Disorders Questionnaire (ADP-IV) [20]. The ADP-IV design allows a dimensional trait-score and a categorical PD diagnosis for each of the DSM-IV PDs. The ADP-IV is a paper-pencil self-report instrument consisting of 94 items, which represent the 80 criteria of the 10 DSM-IV PDs and the 14 research criteria of the depressive and the passive–aggressive PDs. Each traitquestion is rated on a 7-point Likert scale, ranging from “totally disagree” to “totally agree”. For the present study we used the German translation by Doering et al. [21]. Internal consistency of the dimensional PD scales is good and test– retest reliability and concurrent validity of the dimensional trait-scores are also satisfactory [22]. Most importantly, the ADP-IV showed good concordance with the SCID-II interview [23] and may be considered an economical and valid alternative to semi-structured interviews. The Big Five Inventory short form (BFI-S) [24] is a German adaptation of the popular Big Five Inventory by John et al. [25]. The questionnaire consists of 15 items divided into the five broad domains of neuroticism, extraversion, openness, agreeableness, and conscientiousness. The items are rated on a 7-point Likert scale ranging from “don’t agree at all” to “completely agree”. The BFI-S has shown good reliability and validity [24].
669
Velicer’s minimum average partial (MAP) test [27] and Horn’s parallel analysis (PA) test [28]. These two tests are widely acknowledged as providing the most accurate determinants for component retention [29,30]. Both MAP and PA were conducted with syntax programmes provided by O’Connor [30]. The component structure was inspected according to the guidelines provided by Costello and Osborne [29]. A clean component structure ideally comprises the following: that each component at least loads on three items higher than 0.50, that each item should exhibit a loading of at least 0.32, and that only one component should load higher than 0.32 on the same item (two or more components loading higher than 0.32 on the same item are referred to as cross-loadings). MVA, CCA and PCA were carried out with SPSS 20 for Macintosh. We additionally conducted a bi-factor exploratory factor analysis (EFA) using MPlus 7 [31]. A bi-factor EFA computes a general factor that loads on all items and that is orthogonal to the Geomin-rotated group factors [32]. The model fit was evaluated with the χ 2-test of model fit and the following approximate fit indices (AFI): the comparative fit index (CFI), the Tucker–Lewis index (TLI), the root mean square error of approximation (RMSEA), and the standardized root mean square residual (SRMR). According to the χ 2-test, a wellfitting model should provide an insignificant result (i.e., above the 0.05 threshold). However, with increasing sample sizes, a χ 2 value easily becomes significant and the test tends to reject also well-fitting models [33]. Recommended cut-off values of AFI for a good model fit are CFI N 0.95, TLI N 0.95, RMSEA b 0.06, and SRMR b 0.08 [34].
3. Results The CCA indicated that the 5 FFM traits and the 10 DSMIV PD dimensions shared 77.0% of total variance, which was of course statistically highly significant (λ = 0.2303, F(50, 2270.03) = 17.245, p b 0.001). The scree-test for the PCA is shown in Fig. 2. The eigenvalue and proportion of variance
2.3. Statistical analysis Values on all FFM and PD dimensions were missing completely at random (MCAR) according to Little’s MCAR test (χ 2 = 189.972, df = 182, p = 0.328). Therefore, to obtain complete data from all 511 participants included in the analysis we conducted a missing value analysis (MVA) as recommended by Schafer and Graham [26]. No variable was missing in more than maximally 9 cases (1.8%). MVA was carried out with the full information maximum likelihood estimation. A canonical correlation analysis (CCA) was conducted to estimate the proportion of total variance shared between all FFM traits and PD dimensions. Afterwards we conducted a series of principal component analyses (PCA) using oblique Promax rotation on a 15 × 15 item correlation matrix (comprising 10 ADP-IV scales and 5 BFI-S scales). We determined the best fitting component solution by inspecting
Fig. 2. The scree plot of a principal component analysis with 15 items.
670
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674
Table 1 Results of a principal component analysis with 15 items: initial eigenvalues and proportion of variance explained for the first 10 components.
Component 1 Component 2 Component 3 Component 4 Component 5 Component 6 Component 7 Component 8 Component 9 Component 10
Eigenvalue
% Variance
% Cumulative Variance
6.318 1.721 1.263 1.102 1.020 0.684 0.511 0.489 0.402 0.339
42.121 11.471 8.418 7.346 6.801 4.560 3.408 3.258 2.681 2.257
42.121 53.592 62.010 69.356 76.157 80.717 84.125 87.383 90.064 92.321
Components
explained for the first 10 components are shown in Table 1. The MAP pointed towards a two-component solution and the PA towards a three-component solution. The two-component solution is indicated in Table 2. The extraction of a structure with two components accounted for 53.6% of shared variance in normal and pathological personality. The first component loaded strongly on neuroticism and on all 10 PD dimensions. The second component showed strong negative loadings on extraversion and openness and strong positive loadings on schizoid and avoidant PDs. Conscientiousness and agreeableness showed no substantial loadings. Accordingly, the communality, i.e. the total variance explained by all extracted components, was low in those two traits (both h 2 b 0.120). The correlation between the two extracted components was r = 0.249. Although three cross-loadings emerged, the component structure was overall sufficiently clean. The three-component solution is indicated in Table 3. The extraction of three components explained 62.0% of variance shared between normal and pathological personality. The first component loaded again strongly on neuroticism and on
Table 2 Two-component solution of a principal component analysis: Promax rotated component-loadings and communalities (h 2) of the ADP-IV and BFI-S. Components
BFI-S Neuroticism BFI-S Extraversion BFI-S Openness BFI-S Conscientiousness BFI-S Agreeableness ADP-IV Paranoid ADP-IV Schizoid ADP-IV Schizotypal ADP-IV Antisocial ADP-IV Borderline ADP-IV Histrionic ADP-IV Narcissistic ADP-IV Avoidant ADP-IV Dependent ADP-IV Obsessive-Compulsive
1
2
h2
0.451 0.020 0.375 −0.096 −0.291 0.795 0.403 0.827 0.735 0.888 0.865 0.801 0.590 0.661 0.663
0.119 −0.846 −0.664 −0.296 −0.099 0.062 0.467 0.049 −0.199 −0.049 −0.219 −0.053 0.515 0.319 0.216
0.244 0.708 0.458 0.111 0.109 0.660 0.474 0.707 0.507 0.769 0.701 0.623 0.765 0.644 0.558
Loadings greater than 0.320 are indicated in bold.
Table 3 Three-component solution of a principal component analysis: Promax rotated component-loadings and communalities (h 2) of the ADP-IV and BFI-S.
BFI-S Neuroticism BFI-S Extraversion BFI-S Openness BFI-S Conscientiousness BFI-S Agreeableness ADP-IV Paranoid ADP-IV Schizoid ADP-IV Schizotypal ADP-IV Antisocial ADP-IV Borderline ADP-IV Histrionic ADP-IV Narcissistic ADP-IV Avoidant ADP-IV Dependent ADP-IV Obsessive-Compulsive
1
2
3
h2
0.493 −0.179 0.260 0.079 −0.078 0.796 0.509 0.821 0.566 0.863 0.782 0.772 0.760 0.784 0.800
−0.071 0.809 0.671 0.291 0.084 0.014 −0.408 0.029 0.250 0.129 0.287 0.123 −0.433 −0.239 −0.139
0.012 0.087 0.185 0.778 0.742 −0.063 −0.067 −0.075 −0.358 −0.053 −0.090 −0.062 0.087 0.102 0.230
0.248 0.709 0.472 0.616 0.574 0.660 0.475 0.707 0.606 0.770 0.704 0.623 0.806 0.682 0.651
Loadings greater than 0.320 are indicated in bold.
all PD dimensions. The second component was also consistent with the second component from the twocomponent solution delineated above, showing positive loadings on extraversion and openness and negative loadings on schizoid and avoidant PDs. The third component demonstrated two strongly positive loadings on conscientiousness and agreeableness and a moderate negative loading on antisocial PD. The correlation between components 1 and 2 as well as between 1 and 3 was r = −0.085 and r = −0.239, respectively. The correlation between components 2 and 3 was r = −0.114. Except for a few minor violations this component structure was likewise adequately clean. The high proportion of variance explained by the first component and its strong loadings on all PD dimensions point towards a general PD factor. Therefore we additionally conducted a series of bi-factor EFA with 2 factors (i.e. 1 general factor and 1 group factor) up to 8 factors (i.e. 1 general factor and 7 group factors). The 7- and 8factor solutions did not converge and the best fit to the data was found for the 6-factor solution. Although the χ 2 test was statistically highly significant (p b 0.001), the AFI indicated an acceptable to good model fit, depending on the AFI applied (CFI = 0.98; TLI = 0.94; RMSEA = 0.07; SRMR = 0.01). The 6-factor solution is indicated in Table 4. The general factor loaded strongly on neuroticism and on all PD trait-scores. The group factors depicting specific personality styles showed a rather intricate and messy factor structure owing to the limited number of items included in the analysis.
4. Discussion In this study we examined the joint structure of normal and pathological personality using the 5 FFM traits and the
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674
671
Table 4 Six-factor solution of a bi-factor exploratory factor analysis: factor-loadings and communalities (h 2) of the ADP-IV and BFI-S. General factor
BFI-S Neuroticism BFI-S Extraversion BFI-S Openness BFI-S Conscientiousness BFI-S Agreeableness ADP-IV Paranoid ADP-IV Schizoid ADP-IV Schizotypal ADP-IV Antisocial ADP-IV Borderline ADP-IV Histrionic ADP-IV Narcissistic ADP-IV Avoidant ADP-IV Dependent ADP-IV Obs.-Comp.
Group factors
1
2
3
4
5
6
h2
0.496 −0.282 0.106 −0.158 −0.274 0.793 0.513 0.806 0.589 0.875 0.762 0.727 0.788 0.791 0.718
0.160 0.690 0.271 0.218 0.024 0.000 −0.521 −0.120 0.057 0.189 0.341 0.059 −0.439 −0.071 −0.048
−0.064 −0.044 0.150 0.103 0.759 −0.238 −0.016 −0.021 0.002 −0.018 0.055 −0.155 0.175 0.238 0.038
0.056 0.044 0.019 0.534 0.078 0.201 0.025 −0.003 −0.422 −0.036 −0.233 0.032 0.025 −0.123 0.155
−0.437 0.042 0.256 0.003 0.036 0.069 0.359 0.286 0.473 −0.060 −0.004 0.258 −0.044 −0.248 −0.004
−0.071 0.006 −0.195 0.062 −0.023 −0.038 0.052 −0.163 −0.012 −0.164 0.086 0.317 −0.040 0.085 0.364
0.426 0.579 0.265 0.415 0.686 0.714 0.565 0.746 0.710 0.835 0.738 0.745 0.855 0.774 0.702
Loadings greater than 0.320 are indicated in bold.
10 DSM-IV PD dimensions assessed in 511 subjects of the general population of the canton of Zurich, Switzerland. Normal and pathological personality traits shared altogether 77% of their total variance. It is therefore legitimate to conceive PDs as extreme variants along continuous common personality domains, although normal and pathological personality traits are neither completely isomorphic nor superposable. Thus, in agreement with previous reports [35,36], normal and pathological personality domains do not map 1:1, although they share a substantial proportion of variance. A joint structure of normal and pathological personality comprising two and three Promax-rotated components was determined by the two most reliable and valid component extraction tests. The two-component solution explained 54% of shared variance in FFM traits and DSM-IV PD dimensions. The first component loaded uniformly highly on neuroticism and on all 10 PD dimensions, whereas the second component loaded negatively on extraversion and openness and positively on schizoid and avoidant PDs. The first component mainly represents emotional instability, impulsivity and negative affectivity. These primary PD traits have previously been assigned to FFM neuroticism [37–40]. The proportion of total variance explained was distinctly higher for the first component. Neuroticism accordingly has a predominant effect on the joint structure of normal and pathological personality that exceeds the effects of other personality traits. In other words, neuroticism constitutes the basis of a general, unspecific personality dysfunction. This finding has consistently been reported in the literature [41,42]. The second component describes features of detachment, self-consciousness and anhedonia. Associations of these features with inverted extraversion (i.e. introversion) and low openness have also repeatedly been shown [38,43–45]. Moreover, a superordinate factor related to both extraversion and openness has consistently been replicated in the literature [9,10]. A
highly similar two-factor structure distinguishing a neuroticism factor from a detachment factor has additionally been proposed by Markon et al. [7] as well as by Watson et al. [46]. The three-component solution explained 62% of shared variance in FFM traits and DSM-IV PD dimensions. This common structure adds a third component to the two detailed above that loads positively on conscientiousness and agreeableness and negatively on antisocial PD. The third component thus typically describes disinhibition and aggressiveness, which have also been reported by others [40,46,47]. Altogether the three superordinate domains of this joint structure of normal and pathological personality closely resemble the three-factor solution identified by Markon et al. [7], which comprises negative emotionality, positive emotionality (i.e. sociability), and disinhibition. Other proponents of a similar structure with three higherorder domains are Clark’s [12] superordinate factors of negative temperament, positive temperament and disinhibition and Zuckerman and colleagues’ [14,48] superordinate factors of neuroticism–emotionality, extraversion–sociability and impulsive–unsocialized-sensation-seeking. Moreover, our three-component PCA structure also bears considerable resemblance to Eysenck’s [13,49] biological model of personality, which consists of neuroticism, extraversion and psychoticism. Our bi-factor EFA provides further evidence for the separation of a general personality dysfunction factor from specific PD styles and traits [50–53]. As detailed above, it appears that this general factor is closely linked to neuroticism, which is in line with the literature [41,42]. A factor of general personality dysfunction atop the hierarchy of pathological personality structure that loads on higherorder domains such as negative emotionality, detachment, and disinhibition has also been reported by the DSM-5 P&PDWG [54]. Unfortunately, the group factors that constitute the specific PD styles were rather inconclusive
672
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674
in our analysis, because they did not provide a clean factor structure. Specifically, most group factors were not reliable due to the absence of several high loadings. A stable factor should ideally comprise at least three loadings N0.500 [29]. We therefore abstain from a discussion of those group factors. However, that limitation should not weaken the main conclusion of the bi-factor EFA, which was conducted to provide further evidence for a general PD factor that loads on all DSM-IV PD dimensions. The common four- and five-factor models of normal and pathological personality have been discussed and reviewed by various authors [e.g. 3,5,8,55]. The DSM-5 P&PDWG conceived a five-factor model of pathological personality [16], which also closely converges with the FFM traits [40]. In this respect it is worth noting that Krueger et al. [16] did not follow the results of the MAP and PA tests, which recommended 3 and 6 factors, respectively. Interestingly, computing three instead of five superordinate factors based on the 25 primary traits of the DSM-5 PD model Wright et al. [54] found the following three higher-order domains: negative affectivity, detachment and externalizing. Those three factors correspond closely to the three components extracted in the present study and to the three-factor models detailed above. At the next level the externalizing factor split into what the authors referred to as antagonism and disinhibition. At the final level, the five-factor structure, the psychoticism factor ultimately emerged [54]. The DSM-5 P&PDWG narrowly defines psychoticism as schizotypy, which is not consistent with the conceptualization by Eysenck and Eysenck [13], which relates psychoticism more broadly to aggressiveness and impulsiveness. Nevertheless, on a less fragmented level with less factorial splitting, our two- and three-factor solutions are highly consistent with the dimensional PD model proposed by the DSM-5 P&PDWG. This integration of alternative factorial models within a common hierarchical structure is in accordance with the comprehensive review by Widiger and Simonsen [15]. 4.1. Limitations This study is subject to the following limitations: First, we used a short form of the FFM personality inventory. Although this instrument is validated, it does not assess all facets of the original form. This may have slightly influenced our results. Second, our PD measures relied on a self-report questionnaire, which is critically appraised by some researchers, although the ADP-IV showed good concordance with semi-structured interviews [23]. Third, the group factors of our bi-factor EFA were inconclusive because of insufficient factor loadings. Presumably an analysis on the primary facet level including various items would provide a cleaner factor structure and thus interpretable group factors. 4.2. Summary and conclusions With regard to the revisions of the DSM and ICD a dimensional PD concept is widely considered to be a marked
improvement, providing greater validity and reliability than the current dichotomous categorical definition [1–3,56,57]. Although the DSM-5 ultimately dismissed a re-conceptualized dimensional PD model and retained the 10 dichotomous DSM-IV PD categories it is important to provide further evidence for a joint higher-order structure of normal and pathological personality. Convergent integration of alternative dimensional models within a common hierarchical structure is needed in order that the dimensional trait-model of the DSM-5 P&PDWG might be moved from Section 3 into the main section of the manual, for instance in the subsequent edition DSM-5.1. In summary, we found that normal and pathological personality dimensions share a considerably large proportion of variance. We provide strong empirical support for a general personality dysfunction factor which loads strongly on neuroticism and on all DSM-IV PDs [42,54]. We also provide further evidence for a common two- and three-factor structure of normal and pathological personality as represented by associations between FFM traits and DSM-IV PD dimensions. Our extracted components show close correspondence with well-established and carefully validated personality models such as the SNAP [12] or the Big Three [14]. Moreover, our component-structure may be further split up in order to correspond to the intensively promoted four- and five-factor models [15]. Therefore we claim that this study makes an important contribution to a more valid and appropriate understanding and conceptualization of PDs.
Conflicts of interests None. Acknowledgment ZInEP was supported by a private donation. The donor had no further role in experimental design; the collection, analysis, and interpretation of data; the writing of this report; or the decision to submit this paper for publication. References [1] Clark LA. Assessment and diagnosis of personality disorder: perennial issues and an emerging reconceptualization. Annu Rev Psychol 2007;58:227-57. [2] Farmer RF. Issues in the assessment and conceptualization of personality disorders. Clin Psychol Rev 2000;20:823-51. [3] Trull TJ, Durrett CA. Categorical and dimensional models of personality disorder. Annu Rev Clin Psychol 2005;1:355-80. [4] Widiger TA, Simonsen E, Krueger R, Livesley WJ, Verheul R. Personality disorder research agenda for the DSM-V. J Pers Disord 2005;19:315-38. [5] Widiger TA, Livesley WJ, Clark LA. An integrative dimensional classification of personality disorder. Psychol Assess 2009;21:243-55. [6] Widiger TA, Mullins-Sweatt SN. Five-factor model of personality disorder: a proposal for DSM-V. Annu Rev Clin Psychol 2009;5:197-220.
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674 [7] Markon KE, Krueger RF, Watson D. Delineating the structure of normal and abnormal personality: an integrative hierarchical approach. J Pers Soc Psychol 2005;88:139-57. [8] Watson D, Clark LA, Harkness AR. Structures of personality and their relevance to psychopathology. J Abnorm Psychol 1994;103:18-31. [9] DeYoung CG. Higher-order factors of the Big Five in a multiinformant sample. J Pers Soc Psychol 2006;91:1138-51. [10] Digman JM. Higher-order factors of the Big Five. J Pers Soc Psychol 1997 Dec;73(6):1246-56. [11] Krueger RF, Tackett JL. Personality and psychopathology: working toward the bigger picture. J Pers Disord 2003;17:109-28. [12] Clark LA. Manual for the Schedule for Nonadaptive and Adaptive Personality. Minneapolis: University of Minnesota Press; 1993. [13] Eysenck SBG, Eysenck HJ. Manual of the EPQ (personality questionnaire). San Diego, CA: Educational and Industrial Testing Service; 1975. [14] Zuckerman M, Kuhlman DM, Joireman J, Teta P, Kraft M. A comparison of three structural models for personality: the big three, the big five, and the alternative five. J Pers Soc Psychol 1993;65:757-68. [15] Widiger TA, Simonsen E. Alternative dimensional models of personality disorder: finding a common ground. J Pers Disord 2005; 19:110-30. [16] Krueger RF, Derringer J, Markon KE, Watson D, Skodol AE. Initial construction of a maladaptive personality trait model and inventory for DSM-5. Psychol Med 2012;42:1879-90. [17] Hardt J, Egle UT, Kappis B, Hessel A, Brähler E. Symptom Checklist SCL-27: results of a representative German survey. Psychother Psych Med 2004;54:214-23. [18] Angst J, Gamma A, Neuenschwander M, Ajdacic-Gross V, Eich D, Rössler W, et al. Prevalence of mental disorders in the Zurich cohort study: a twenty year prospective study. Epidemiol Psichiatr Soc 2005; 14:68-76. [19] Dunn G, Pickles A, Tansella M, Vazquez-Barquero JL. Two-phase epidemiological surveys in psychiatric research. Br J Psychiatry 1999; 174:95-100. [20] Schotte CKW, de Doncker D. ADP-IV questionnaire. Antwerp, Belgium: University Hospital Antwerp; 1994. [21] Doering S, Renn D, Hofer S, Rumpold G, Smrekar U, Janecke N, et al. Validation of the German version of the Assessment of DSM-IV Personality Disorders (ADP-IV) questionnaire. Z Psychosom Med Psychother 2007;53:111-28. [22] Schotte CKW, de Doncker D, Vankerckhoven C, Vertommen H, Cosyns P. Self-report assessment of the DSM-IV personality disorders. Measurement of trait and distress characteristics: the ADP-IV. Psychol Med 1998;28:1179-88. [23] Schotte CKW, de Doncker DA, Dmitruk D, Van Mulders I, D'Haenen H, Cosyns P. The ADP-IV questionnaire: differential validity and concordance with the semi-structured interview. J Pers Disord 2004; 18:405-19. [24] Schupp J, Gerlitz J-Y. BFI-S: Big Five Inventory-SOEP. In: & Glöckner-Rist A, editor. Zusammenstellung sozialwissenschaftlicher Items und Skalen (ZIS Version 1200). Bonn: GESIS; 2008. [25] John OP, Donahue EM, Kentle RL. The Big Five Inventory — versions 4a and 54. Berkeley, CA: University of California, Berkeley, Institute of Personality and Social Research; 1991. [26] Schafer JL, Graham JW. Missing data: our view of the state of the art. Psychol Methods 2002;7:147-77. [27] Velicer WF. Determining the number of components from the matrix of partial correlations. Psychometrika 1976;41:321-7. [28] Horn JL. A rationale and test for the number of factors in factor analysis. Psychometrika 1965;30:179-85. [29] Costello AB, Osborne JW. Best practices in exploratory factor analysis: four recommendations for getting the most from your analysis. Pract Assess Res Eval 2005;10(7) [Available online: http:// pareonline.net/getvn.asp?v=10&n=7]. [30] O'Connor BP. SPSS and SAS programs for determining the number of components using parallel analysis and velicer's MAP test. Behav Res Methods Instrum Comput 2000;32:396-402.
673
[31] Muthén LK, Muthén BO. Mplus user's guide. Seventh ed. Los Angeles, CA: Muthén & Muthén; 1998–2012. [32] Jennrich RI, Bentler PM. Exploratory bi-factor analysis. Psychometrika 2011;76:537-49. [33] Steiger JH. Understanding the limitations of global fit assessment in structural equation modeling. Pers Individ Differ 2007;42:893-8. [34] Schreiber JB, Stage FK, King J, Nora A, Barlow EA. Reporting structural equation modeling and confirmatory factor analysis results: a review. J Educ Res 2006;99:323-37. [35] Samuel DB, Simms LJ, Clark LA, Livesley WJ, Widiger TA. An item response theory integration of normal and abnormal personality scales. Pers Disord 2010;1:5-21. [36] Stepp SD, Yu L, Miller JD, Hallquist MN, Trull TJ, Pilkonis PA. Integrating competing dimensional models of personality: linking the SNAP, TCI, and NEO using Item Response Theory. Pers Disord 2012; 3:107-26. [37] Blais MA. Clinician ratings of the five-factor model of personality and the DSM-IV personality disorders. J Nerv Ment Dis 1997;185:388-93. [38] Leibing E, Jamrozinski K, Vormfelde SV, Stahl J, Doering S. Dimensions of personality—relationship between DSM-IV personality disorder symptoms, the five-factor model, and the biosocial model of personality. J Pers Disord 2008;22:101-8. [39] Livesley WJ, Jang KL, Vernon PA. Phenotypic and genetic structure of traits delineating personality disorder. Arch Gen Psychiatry 1998; 55:941-8. [40] Thomas KM, Yalch MM, Krueger RF, Wright AG, Markon KE, Hopwood CJ. The convergent structure of DSM-5 personality trait facets and five-factor model trait domains. Assessment 2013;20:308-11. [41] Krueger RF, Eaton NR, Clark LA, Watson D, Markon KE, Derringer J, et al. Deriving an empirical structure of personality pathology for DSM-5. J Pers Disord 2011;25:170-91. [42] Saulsman LM, Page AC. The five-factor model and personality disorder empirical literature: a meta-analytic review. Clin Psychol Rev 2004;23:1055-85. [43] Piedmont RL, Sherman MF, Sherman NC, Dy-Liacco GS, Williams JE. Using the five-factor model to identify a new personality disorder domain: the case for experiential permeability. J Pers Soc Psychol 2009;96:1245-58. [44] Reynolds SK, Clark LA. Predicting dimensions of personality disorder from domains and facets of the Five-Factor Model. J Pers 2001;69:199-222. [45] Tackett JL, Silberschmidt AL, Krueger RF, Sponheim SR. A dimensional model of personality disorder: incorporating DSM cluster A characteristics. J Abnorm Psychol 2008;117:454-9. [46] Watson D, Clark LA, Chmielewski M. Structures of personality and their relevance to psychopathology: II. Further articulation of a comprehensive unified trait structure. J Pers 2008;76:1545-86. [47] Nestadt G, Costa Jr PT, Hsu FC, Samuels J, Bienvenu OJ, Eaton WW. The relationship between the five-factor model and latent Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition personality disorder dimensions. Compr Psychiatry 2008;49:98-105. [48] Zuckerman M, Kuhlman DM, Camac C. What lies beyond E and N? Factor analyses of scales believed to measure basic dimensions of personality. J Pers Soc Psychol 1988;54:96-107. [49] Eysenck HJ. Biological basis of personality. Nature 1963;199:1031-4. [50] Bornstein RF, Huprich SK. Beyond dysfunction and threshold-based classification: a multidimensional model of personality disorder diagnosis. J Pers Disord 2011;25:331-7. [51] Hopwood CJ, Malone JC, Ansell EB, Sanislow CA, Grilo CM, McGlashan TH, et al. Personality assessment in DSM-5: empirical support for rating severity, style, and traits. J Pers Disord 2011;25: 305-20. [52] Livesley WJ. An empirically-based classification of personality disorder. J Pers Disord 2011;25:397-420. [53] Morey LC, Berghuis H, Bender DS, Verheul R, Krueger RF, Skodol AE. Toward a model for assessing level of personality functioning in
674
M.P. Hengartner et al. / Comprehensive Psychiatry 55 (2014) 667–674
DSM-5, part II: empirical articulation of a core dimension of personality pathology. J Pers Assess 2011;93:347-53. [54] Wright AG, Thomas KM, Hopwood CJ, Markon KE, Pincus AL, Krueger RF. The hierarchical structure of DSM-5 pathological personality traits. J Abnorm Psychol 2012;121:951-7. [55] Samuel DB, Widiger TA. A meta-analytic review of the relationships between the five-factor model and DSM-IV-TR personality
disorders: a facet level analysis. Clin Psychol Rev 2008;28: 1326-42. [56] Verheul R. Clinical utility of dimensional models for personality pathology. J Pers Disord 2005;19:283-302. [57] Widiger TA, Lowe JR. A dimensional model of personality disorder: proposal for DSM-V. Psychiatr Clin North Am 2008;31: 363-78.