فرکانس و ساختار صفات اختلال شخصیت DSM-IV در دانشجویان

Abstract A sample of French female college students (N = 201) completed the self-report Personality Disorder Questionnaire for DSM-IV (PDQ-4+; Hyler, 1994). Forty-two participants (21%) had a PDQ total score equal or greater to 30 suggesting a personality disturbance. The DSM-IV three-cluster classification of PDs was tested using confirmatory factor analysis and failed to produce an acceptable fit. An exploratory factorial analysis extracted a four-factor solution possessing both satisfactory fit and meaningful interpretations. This model was compared with the two other studies which have tested the three-cluster model and found that PDs grouped themselves into components differing from DSM-IV clusters. It may be unlikely to find an universal structure for DSM-IV PDs generalizable to all clinical and non-clinical populations.

. Introduction Personality disorder traits and diagnoses are frequent in community samples of young adults. For example, in a large epidemiological sample of young adults, Moran, Coffey, Mann, Carlin, and Patton (2006) found that the prevalence of personality disorders was 18.6%. Among young adults, college students constitute a vulnerable group at high risk for psychological morbidity (e.g., Harrison, Barrow, Gask, & Creed, 1999). In particular, Ekselius, Tillfors, Furmark, and Frederikson (2001) found that personality disorders (PDs) were significantly more often diagnosed in students than in younger subjects and older adults in a community sample. High rates of PDs have been found using both self-report questionnaires (Johnson & Bornstein, 1992; Sinha & Watson, 2004) or structured clinical interviews (Taylor, 2005). Moreover, Benton, Robertson, Tseng, Newton, and Benton (2003) found a growing number of college students with serious psychological problems, including PDs, among students attending university counselling centers. There are few studies of the structure of PDs among young adults. Further studies of PDs in young adults and college students are warranted. The most authoritative classifications of PDs is provided by the third and fourth editions of the Diagnostic and Statistical Manual of Mental Disorders (DSM-III, APA, 1980; DSM-III-R, APA, 1987; DSM-IV, APA, 1994) which present a three-cluster classification of personality disorders (PDs). Empirical testing of this three-cluster model using exploratory or confirmatory factorial analyses (EFA, CFA) have yielded mixed results. Some of them support the DSM-III and DSM-III-R three-factor structure of Axis II (e.g., Bagby, Joffe, Parker, & Schuller, 1993; Hyler & Lyons, 1988; Hyler et al., 1990) whereas others found that PDs show a poor fit to the clusters (Bell & Jackson, 1992; Deary, Peter, Austin, & Gibson, 1998). Others reported four (e.g., Kass, Skodol, Charles, Spitzer, & Williams, 1985) or five (Nestad et al., 1994) factor structures. As only 10 of the 93 DSM-III-R Axis II criteria were not changed in DSM-IV (Widiger, 2001), studies assessing DSM-IV structure of PDs are needed. Only three studies have been conducted all on psychiatric patient samples; they find little support for the DSM-IV three-cluster model. Yang, Bagby, Costa, Ryder, and Herbst (2002) used the Personality Disorder Questionnaire (PDQ-4+; Hyler, 1994) and a semi-structured interview for the assessment of PDs among Chinese psychiatric patients. Using CFA, the DSM-IV three-cluster model was compared to a one-factor model and a set of random three-factor models. Only the clinician-rated instrument was considered to support the DSM-IV three-cluster model although the fit indices were equivocal. Only two studies using principal component or cluster analyses proposed alternative structures for the DSM-IV PDs. In the study by Fossati et al. (2000) using a semi-structured interview for the assessment of PDs, DSM-IV PD cluster were not replicated in a sample of Italian psychiatric patients. Durrett and Westen (2005) using cluster analysis did not recover the DSM-IV three-cluster model in a sample of adolescent patients. There are no known studies of the DSM-IV three-cluster model among non-clinical or clinical young adults or college students. More studies are needed to explore the structure of PDs. Indeed, these studies are important as regard the debate about categorical versus dimensional classification systems. Whether PDs are accurately or optimally classified categorically or dimensionally is still a controversial topic (Widiger & Samuel, 2005). Assessment of clinical utility and user acceptability of both classification systems needs further empirical investigations (First, 2005). An interesting solution to the present state of knowledge may be the tandem use of categorical and dimensional assessments of PDs in research (Kessler, 2002). DSM classification of PDs offers this possibility. A dimensional approach of PDs based on criteria counts have been proposed by Kass et al. (1985) and been proved to be relevant. The dimensional approach considers PD traits as variant of basic personality traits continuously distributed in populations with indistinct boundaries between normal and abnormal personality. This dimensional conception provides a series of scales which could be used to give a profile of a person’s personality. Similarly, PD cluster could be converted into broad personality dimensions (Deary et al., 1998) which could be used to complement or compete with the many dimensional models of general personality functioning such as the three-factor models (Eysenck & Eysenck, 1975) or the ‘Big five’ five-factor model (Costa & McCrae, 1992). The objectives of the present study were to evaluate the prevalence of personality disorders and to further examine the structure of PD traits among college students; the DSM-IV three-cluster model and a one-factor model were tested using CFA. More relevant models were examined after using EFA and the fit of these models were compared. Following Fossati et al. and Yang et al., we used the PDQ-4+ to assess PD traits.

Results 3.1. Descriptive statistics The rate of endorsement of personality disorder traits was relatively high. The mean for PDQ total score was 29.4 (SD = 10.4, range 1–61). Forty-two participants (21%) had a PDQ total score equal or greater to 30. There were eight true responses to SQ item 64 and no true response to item 76. These eight participants were deleted from the analyses (n = 193). The Cronbach’s α averaged about .65 and ranged from a low .47 (obsessive-compulsive) to a high .83 (dependent) with the others PDs ranging as follows: avoidant (.80), borderline (.73), schizotypal (.70), narcissistic (.70), histrionic (.64), schizoid (.54), paranoid (.53), antisocial (.51). The frequencies of possible PD diagnoses were: Cluster A: paranoid, 30%; schizoid, 6%; schizotypal, 20%; Cluster B: antisocial, 9%; borderline, 30%; histrionic, 9%; narcissistic, 6%; Cluster C: avoidant, 27%; dependent, 6%; obsessive, 41%. The frequencies of possible PD diagnoses were 42% for Cluster A, 40% for Cluster B and 55% for Cluster C. All the correlations between cluster scores were significant and moderate: Pearson’s r coefficients ranged 0.42–0.45 (p < .001). Endorsement of PD traits was sufficient to permit factorial analyses and interpretation of the results. 3.2. Confirmatory and exploratory factorial analyses The correlated three-cluster model fit the data poorly (χ2/df = 112.7/35 = 3.2, GFI = 0.89, CFI = 0.35, RMR = 0.11, RMSEA = 0.09). EFA of the PD scores yielded four factors with eigenvalues exceeding 1.0. The eigenvalue curve suggested either a one-, two-, three-, four-factor solution. The three-factor solution which explained 46% of the variance was deleted because the borderline PD loaded on two factors (first factor: schizoptypal, paranoid, avoidant, obsessive, and borderline; second factor: antisocial and schizoid; third factor: histrionic, narcissistic, dependent, and borderline). The two-factor solution was deleted because it explained a too low proportion of the variance (35%). The first factor consisted in borderline, histrionic, schizotypal, avoidant, dependent, narcissistic, paranoid, obsessive PDs and the second factor consisted of antisocial and schizoid PDs. The four-factor solution which explained 58% of the variance was retained. The first factor consisted of borderline, schizotypal, obsessive, and avoidant PDs; the second factor of antisocial and schizoid PDs; the third of histrionic and narcissistic PDs; and the fourth of dependent and paranoid PDs. Table 1 presents this four-factor solution. The correlations between the four factors were weak (from 0.09 to 0.23). Only correlations between factor 1 and 3 (Pearson’s r = 0.23, p = .002) and between factor 2 and 4 (r = 0.14, p = .04) were significant. Three PDs had communality values lower than .2 indicating that they were not well accounted for by the four-factor model. The low to moderate communality values suggests that each PD is a clearly independent dimension. Table 1. Results of the factorial analysis of the PDQ-4+ PD scores conducted on a sample of 193 female college students: factor loadings, communalities (h2), eigenvalues, and explained variance PDs Factor 1 Factor 2 Factor 3 Factor 4 h2 Factor 1 Borderline 0.70 0.09 0.33 0.04 0.39 Schizotypal 0.70 0.24 0.03 0.12 0.33 Obsessive 0.64 −0.10 −0.01 −0.13 0.15 Avoidant 0.59 −0.06 −0.13 0.23 0.32 Factor 2 Schizoid 0.04 0.80 −0.13 0.07 0.13 Antisocial 0.00 0.68 0.22 −0.12 0.12 Factor 3 Histrionic 0.04 0.08 0.80 −0.20 0.33 Narcissistic 0.04 −0.01 0.68 0.40 0.28 Factor 4 Dependent 0.03 −0.21 −0.04 0.77 0.24 Paranoid 0.14 0.39 0.09 0.62 0.21 Eigenvalue 2.1 1.4 1.1 1.1 Explained variance 0.18 0.14 0.13 0.13 Table options The one-factor model and the four-factor model derived from the EFA were tested with CFAs. The one-factor model produced an almost acceptable fit (χ2/df = 78.7/35 = 2.3, GFI = 0.93, CFI =0.85, RMR = 0.06, RMSEA = 0.05) but all loadings were not significant. The uncorrelated four-factor model fit the data poorly (χ2/df = 175.9/38 = 4.6, GFI = 0.84, CFI = 0.53, RMR = 0.18, RMSEA = 0.12). A second confirmatory factorial analysis was performed with a correlated four-factor model which fit the data well (χ2/df = 56/29 = 1.9, GFI = 0.95, CFI = 0.91, RMR = 0.05, RMSEA = 0.04). All loadings were significant. The one-factor model and the correlated four-factor model were compared using the difference in χ2 values. The model χ2 of the one-factor model was 78.7 (df = 35) and the model χ2 of the four-factor model was 56 (df = 29). The difference in χ2 value (χ2 difference = 22.7, df difference = 6) exceeded the critical value of View the MathML sourceχ62 (12.6) and thus, was significant (p < .001) showing that the four-factor model was better able to reproduce the data than the one-factor model.