The higher-order structure of common DSM mental disorders: internalization, externalization, and their connections to personality

https://doi.org/10.1016/S0191-8869(00)00106-9Get rights and content

Abstract

Comorbidity among mental disorders is commonly observed in clinical and epidemiological samples. Can comorbidity be understood as meaningful covariance, and is this covariance structure linked with personality? We addressed this question in a sample of 634 female and 549 male, middle-aged participants in the Minnesota Twin-Family Study (MTFS). Mental disorders were assessed using the Structured Clinical Interview for DSM-III-R, the Substance Abuse Module from the Composite International Diagnostic Interview, and a specially-designed interview for the assessment of antisocial personality disorder. Personality was assessed using the Multidimensional Personality Questionnaire. Relations among symptom scales for eight common DSM disorders were compatible with hypothesized underlying bivariate normal distributions. Polychoric correlations among these scales were well-fit by a two-factor model positing internalizing and externalizing factors, which, in turn, were correlated with broad personality dimensions. Internalizing was positively correlated with negative emotionality (and negatively with positive emotionality in women) and externalizing was negatively correlated with constraint.

These findings suggest that internalization, externalization, and their links to personality may provide a useful framework for understanding covariance among common adult mental disorders.

Introduction

In recent history, personality and mental disorders have been studied in distinct and non-overlapping literatures. This historical bifurcation, however, is clearly waning, as the field bears witness to a rapprochement between personality and psychopathology research. At one time, these fields were essentially united as topics of study under the broad rubric of human individual differences. Now that the age of doubt regarding the reality and utility of personality is essentially over, personality and psychopathology research are again showing signs of cross-fertilization (Watson & Clark, 1994).

In the research reported herein, we pursued the rapprochement between personality and psychopathology. We examined the higher-order structure of common mental disorders in a community-based sample of men and women, and we mapped this higher-order structure onto the higher-order structure of personality. Thus, our analyses addressed two basic issues. First, could “comorbidity” among mental disorders be understood as meaningful covariance, and hence, modeled as the result of basic underlying factors of psychopathological variation? Second, how did these basic underlying factors of psychopathological variation map onto higher-order personality traits? To set the stage for our analysis, we briefly outline the historical context of our work.

One key theme underlying the renewed interest of psychopathologists in personality has been the strong resurgence of personality psychology as a vital academic enterprise. This resurgence was based on a number of key research findings and closely-related conceptual developments. First, personality traits were found to be heritable. Evidence from twin studies, for example, is consistent in documenting broad heritabilities of 30–50% for higher-order dimensions of personality (Loehlin, 1992, Bouchard, 1994). In addition, molecular genetic work is beginning to bear fruit, as specific markers (e.g., DRD4) are being linked to personality variation (Plomin & Caspi, 1998). If personality traits are fictional, they are fictions with a notable genetic basis.

Second, personality traits were found to be stable in adulthood. Summarizing studies that varied markedly in instrumentation, sample size, gender composition, initial age of participants, and retest interval (two to thirty years), Costa and McCrae (1997) found that the median stability of different personality traits ranges from 0.34 to 0.77. As these authors point out, disattenuated for unreliability, these correlations would be even higher.

Finally, personality traits were found to be consequential. One could list a compendium of important social outcomes that are meaningfully correlated with personality traits, but such a “laundry list” does not do justice to the systematic nature of the relevant relations. Personality trait models provide the “glue” for a comprehensive psychological theory linking individual differences in domains such as affect and impulse control to consequential social outcomes (Watson, Clark & Harkness, 1994). For example, why do substance use disorders and antisocial behavior tend to occur in the same persons (Kessler et al., 1997, Morgenstern et al., 1997, Rounsaville et al., 1998)? Perhaps because low levels of constraint — a tendency toward impulsivity, sensation seeking, and non-traditional values — lead to behaviorally distinctive outcomes that, nevertheless, are psychologically coherent when reframed as indicative of impulsivity (Sher and Trull, 1994, Krueger, 1999b). In this way, personality trait models can predict and explain specific social behaviors, and the relations among such behaviors.

While these developments in personality psychology were underway, psychopathology research witnessed the rise of the modern Diagnostic and Statistical Manuals, i.e., DSM-III (American Psychiatric Association, 1980), DSM-III-R (American Psychiatric Association, 1987), and DSM-IV (American Psychiatric Association, 1994), and their operationalization in various structured interviews. One of the most interesting findings from research using structured interviews is the pervasiveness of “comorbidity” among DSM-defined mental disorders — the tendency for mental disorders to co-occur at greater than chance rates (Clark, Watson & Reynolds, 1995). This phenomenon is usually viewed through the lens of the categorical, “neo-Kraepelinian” measurement model underlying the modern DSMs. From this perspective, comorbidity is typically viewed as either a crisis or as a nuisance. Conceptually, comorbidity is a crisis in that it undermines the neo-Kraepelinian idea that there are various, separate, and discrete mental disorders. Operationally, comorbidity is a nuisance because various gymnastics are necessary to obtain samples for research, such as identifying rare “pure” individuals in highly comorbid treatment-seeking samples.

Rather than viewing comorbidity as a crisis or a nuisance, it may be better viewed as an opportunity. Comorbidity suggests that something is awry with the conceptual measurement model underlying recent DSMs, and provides the impetus for improving that model. It is not, however, a threat to the key, seminal contribution of the neo-Kraepelinians: The notion that mental disorders are real, and can therefore be reliably measured. Indeed, if mental disorders were not reliably measurable, comorbidity would not be possible (i.e., unreliability of measurement sets a ceiling on covariance among disorders).

Comorbidity is, however, a problem that is difficult to cope with under the categorical, neo-Kraepelinian model. Nevertheless, no specific measurement model has made a strong bid to take the place of the categorical model. Psychometric theory may provide such a model, for at least some psychopathological conditions. For instance, in recent years, psychometric methods have begun to influence conceptualizations of the Axis II (personality) disorders (Cloninger, 1999, Costa and Widiger, 1994, Strack and Lorr, 1994). Although agreement in this literature is not universal, researchers studying personality disorders have begun to realize the advantages of the methods utilized by personality researchers studying normal-range personality variation. Specifically, personality psychologists have historically coped with phenotypic diversity and covariance among indicators by using tools developed by psychometricians, such as factor analysis. These tools have served personality research well. Although researchers may still disagree about the precise number and nature of the basic dimensions of personality (Eysenck, 1992, Livesley et al., 1998), the notion that a small number of basic higher-order dimensions are jointly necessary and sufficient to summarize personality variation is fairly incontrovertible (Watson et al., 1994).

Does this perspective also have utility in understanding Axis I (syndrome) mental disorders? Recently, Krueger applied the psychometric perspective to the comorbidity problem in research on common Axis I mental disorders in the Dunedin Multidisciplinary Health and Development Study (DMHDS; Krueger, Caspi, Moffitt & Silva, 1998) and in the National Comorbidity Survey (NCS; Krueger, 1999a). Epidemiological research shows that certain DSM-defined mental disorders occur with greater frequency in community samples. Specifically, prevalence rates for unipolar affective disorders, anxiety disorders, substance use disorders, and antisocial behavior disorders exceed prevalence rates for “severe” psychopathologies, such as mania and the non-affective psychoses (Kessler et al., 1994). Moreover, the common DSM disorders are systematically covariant. Krueger showed how two higher-order, psychologically-coherent dimensions of variation, internalization and externalization, were able to account for this systematic covariance. The factor connecting the unipolar mood and anxiety disorders was labeled internalization, to describe the propensity to express distress inwards that unites these disorders. The factor connecting the substance use and antisocial behavior disorders was labeled externalization, to describe the propensity to express distress outwards that unites these disorders. Importantly, internalization and externalization are conceived as separate dimensions, as opposed to opposite ends of the same dimension.

In the work reported herein, we attempted to replicate and extend this previous research on the structure of common mental disorders, and to connect this line of research to research on the structure of personality. Specifically, we employed data from middle-aged parent participants in the Minnesota Twin-Family Study (MTFS) to build upon Krueger’s previous work in a number of specific ways. First, participants in the DMHDS were interviewed using the Diagnostic Interview Schedule (DIS; Robins, Helzer, Cottler & Goldring, 1989), and participants in the NCS were interviewed using the University of Michigan version of the Composite International Diagnostic Interview (UM-CIDI; Wittchen, Kessler, Zhao & Abelson, 1995). In the MTFS, a different set of diagnostic instruments — the Structured Clinical Interview for DSM-III-R (SCID; Spitzer, Williams & Gibbon, 1987), the Substance Abuse Module from the Composite International Diagnostic Interview (SAM; Robins, Babor & Cottler, 1987) and a specially-designed interview for the assessment of antisocial personality disorder (Holdcraft, Iacono & McGue, 1998) were employed. In addition, interview data collected in the MTFS were submitted to a consensus process, whereas DMHDS and NCS diagnoses were made directly from interview data. Thus, we were able to determine the robustness of Krueger’s model to variation in diagnostic instrumentation and method.

Second, Krueger et al., 1998, Krueger, 1999a employed the tetrachoric correlation as an index of association between DSM diagnoses. The tetrachoric correlation operationalizes a liability-threshold model of disorder risk. This model states that a normally distributed continuum of risk underlies observed dichotomous mental disorder diagnoses. When only dichotomous diagnoses are used, there is no direct way to test the appropriateness of the liability-threshold model. Therefore, in the current research, we employed multi-point ordinal scales representing the number of criteria participants met toward a given DSM disorder. When more than two categories are available for at least one of the two variables being correlated, it is possible to test the fit of the underlying bivariate normal distribution to the frequency table formed by the cross-classification of participants on the two ordinal variables. Thus, the appropriateness of viewing bivariate associations among DSM variables (a.k.a. comorbidity) as jointly normal and continuous was submitted to statistical scrutiny.

Finally, Krueger et al., 1998, Krueger, 1999a speculated about relations between internalization, externalization, and broad traits identified in the personality literature, but presented no empirical data with regard to this issue. Specifically, Krueger et al. (1998) speculated that internalization might be linked with negative emotionality (a tendency to experience anxiety, anger, and alienation), and externalization with a lack of constraint (a tendency to engage in risky behavior, to act on impulse, and to endorse non-traditional values). Negative emotionality and constraint are constructs measured by Tellegen’s Multidimensional Personality Questionnaire (Tellegen, 1985, Tellegen, 2000, Tellegen and Waller, 2000), which was completed by the MTFS parents. Thus, in the current research, we were able to directly examine relations between internalization, externalization, and the higher-order personality traits measured by the MPQ.

Section snippets

Participants

Participants were parents of adolescent Minnesota-born twins enrolled in the Minnesota Twin-Family Study (MTFS), an epidemiological investigation of the development of substance abuse and related mental disorders. Although parents of both male and female twins are enrolled in the MTFS, for this report, data from the parents of female twins were used because, at intake, these parents received a more extensive mental disorder assessment than parents of male twins. Twin families were ascertained

Appropriateness of the liability-threshold model

Does the liability-threshold model hold for pairwise relations among the DSM symptom-count variables? The computer program PRELIS 2.2 (Jöreskog & Sörbom, 1996b) was used to create contingency tables for all possible pairings of the eight DSM variables. That is, for each pairing of variables (e.g., A with x scale points, and B with y scale points), PRELIS creates a table with dimensions x by y showing the number of persons who fell in each cell of the table, due to their combination of scores on

Discussion

We investigated the latent structure of variables representing the criteria for eight common DSM mental disorders in a large, representative sample of middle-aged men and women. We found no reason to reject a liability-threshold model to account for the bivariate relations among these variables. The hypothesis that normally-distributed liabilities underlie observed patterns of co-risk for these eight DSM variables could not be rejected. This was a key finding because it allowed us to compare

Acknowledgements

The Minnesota Twin Family Study is supported in part by USPHS grants AA00175, AA09367, and DA05147. We thank Auke Tellegen, Kristen Kling, and the anonymous reviewers for their helpful comments.

References (52)

  • P.T Costa et al.

    Longitudinal stability of adult personality

  • H.J Eysenck

    Four ways five factors are not basic

    Personality and Individual Differences

    (1992)
  • H.U Wittchen et al.

    Reliability and clinical validity of UM-CIDI DSM-III-R generalized anxiety disorder

    Journal of Psychiatric Research

    (1995)
  • T.M Achenbach

    Empirically based assessment and taxonomy: applications to clinical research

    Psychological Assessment

    (1995)
  • T.M Achenbach et al.

    The classification of child psychopathology: a review and analysis of empirical efforts

    Psychological Bulletin

    (1978)
  • T.M Achenbach et al.

    Psychopathology of childhood

    Annual Review of Psychology

    (1984)
  • H Akaike

    Factor analysis and AIC

    Psychometrika

    (1987)
  • H.S Akiskal et al.

    The relationship of personality to affective disorders

    Archives of General Psychiatry

    (1983)
  • Diagnostic and statistical manual of mental disorders

    (1980)
  • Diagnostic and statistical manual of mental disorders

    (1987)
  • Diagnostic and statistical manual of mental disorders

    (1994)
  • T.J Bouchard

    Genes, environment and personality

    Science

    (1994)
  • A.T Church et al.

    Exploratory and confirmatory tests of the big five and Tellegen’s three- and four-dimensional models

    Journal of Personality and Social Psychology

    (1994)
  • L.A Clark et al.

    Temperament, personality, and the mood and anxiety disorders

    Journal of Abnormal Psychology

    (1994)
  • L.A Clark et al.

    Diagnosis and classification of psychopathology: challenges to the current system and future directions

    Annual Review of Psychology

    (1995)
  • D.L Dunner

    The issue of comorbidity in the treatment of panic

    International Clinical Psychopharmacology

    (1998)
  • H.J Eysenck

    Normality-abnormality and the three-factor model of personality

  • J.R Graham

    MMPI-2: assessing personality and psychopathology

    (2000)
  • L.C Holdcraft et al.

    Antisocial Personality Disorder and depression in relation to alcoholism: a community-based sample

    Journal of Studies on Alcohol

    (1998)
  • W.G Iacono et al.

    Psychophysiological prediction of substance abuse

  • K Jöreskog et al.

    LISREL 8: user’s reference guide

    (1996)
  • K Jöreskog et al.

    PRELIS 2: user’s reference guide

    (1996)
  • R.C Kessler et al.

    Lifetime co-occurrence of DSM-III-R alcohol abuse and dependence with other psychiatric disorders in the National Comorbidity Survey

    Archives of General Psychiatry

    (1997)
  • R.C Kessler et al.

    Lifetime and 12-month prevalence of DSM-III-R psychiatric disorders in the United States: results from the National Comorbidity Survey

    Archives of General Psychiatry

    (1994)
  • Cited by (318)

    • Psychological Assessment in Public Safety Personnel Screening

      2022, Comprehensive Clinical Psychology, Second Edition
    View all citing articles on Scopus
    View full text