Originally published by: The Journal of Nervous and Mental Disease
Written by: Benjamin Hummelen, MD, Theresa Wilberg, MD, PhD, Geir Pedersen, MA, Sigmund Karterud, MD, PhD

The study evaluated the quality of the DSM-IV obsessive-compulsive personality disorder (OCPD) construct as a prototype category. A sample of 2237 patients from the Norwegian Network of Psychotherapeutic Day Hospitals was examined by a variety of psychometric analyses. A high number of OCPD patients (77%) had co-occurrent PDs, but only the co-occurrence with paranoid was significantly higher than expected. Exploratory factor analysis of the PD criteria indicated that OCPD consists of 2 dimensions. The first dimension, perfectionism, was constituted by OCPD criteria only and was significantly related to obsessive-compulsive disorder. The second dimension, aggressiveness, included 2 OCPD criteria, reluctance to delegate and stubbornness, but was also defined by criteria from paranoid, antisocial, and borderline PD. Confirmatory factor analysis of the OCPD criteria indicated a poor fit of both a unitary model and a 3-dimensional model. Overall, the OCPD criteria had poor psychometric properties. Although it seems that the quality of the DSM-IV OCPD as a prototype construct is insufficient, it may be improved by deleting the criteria hoarding behavior and miserliness. Alternative criteria could be related to problems in close relationships involving the need for predictability. Such revisions may add a third dimension to the 2 dimensions of perfectionism and aggressiveness.

Obsessive-compulsive personality disorder (OCPD) received considerable clinical and empirical attention during the major part of the 20th century. However, interest in this category has waned during the last decades, despite its high frequency; prevalence estimates in clinical samples range from 5.1% to 16.4% (Alnaes and Torgersen, 1988; Fossati et al., 2000; Oldham et al., 1992; Stuart et al., 1998; Zanarini et al., 1998; Zimmerman et al., 2005) and from 0% to 9.3% in community surveys (Grant et al., 2005; Samuels et al., 2002; Torgersen et al., 2001).

According to DSM-IV (American Psychiatric Association, 1994), the essential feature of OCPD is a preoccupation with mental and interpersonal control, orderliness, and perfectionism, at the expense of flexibility, openness, and efficiency. The concept has its roots in psychoanalytical character typology as outlined by Freud (1908/1958), describing the “anal character” supposed to arise out of conflicts between parents and child over bowel training in the second to third year of life. According to Freud, the anal character consists of a particular constellation of traits, i.e., obstinacy, parsimony, and orderliness. A number of successive authors drew upon the work of Freud, contributing with more detailed descriptions of the anal character, e.g., Abraham (1923) and Rado (1959). Other authors used different terms but similar conceptualizations, e.g., the compulsive character of Reich (1945) and the obsessional character of Sandler and Hazari (1960).

In DSM-III (American Psychiatric Association, 1980), the descriptions of OCPD were highly influenced by the passive-ambivalent character portrayed by Millon (1981), resulting in the inclusion of the criteria of indecisiveness and restricted expression of affection. These criteria were retained in DSM-III-R (American Psychiatric Association, 1987) and finally deleted in DSM-IV (American Psychiatric Association, 1994). It is not clear on what grounds these criteria were deleted. A DSM-III-R study published in 1994 found only moderate support for these revisions (Nurnberg et al., 1994). Regarding DSM-IV (American Psychiatric Association, 1994), several studies have examined the psychometric properties of the OCPD criteria (Blais and Norman, 1997; Farmer and Chapman, 2002; Fossati et al., 2006; Grilo, 2004a,b; Grilo et al., 2004). All studies, except the study of Fossati et al. (2006), found that perfectionism was the criterion that performed best. With respect to the other criteria, however, the results of these studies were less consistent, though the 2 criteria that reflect parsimony (hoarding behavior and miserly) were likely to have the poorest psychometric properties. Chronbach’s
coefficients of the OCPD criteria ranged from 0.53 to 0.77 in the different studies, indicating insufficient to moderate internal consistency (Blais and Norman, 1997; Farmer and Chapman, 2002; Grilo, 2004b; Grilo and McGlashan, 2000; Grilo et al., 2001).

In factor analytical studies including all PDs, OCPD had a tendency to come to the fore as a dimension that stands apart from the other personality dimensions (Fossati et al., 2006; Kass et al., 1985; Mulder and Joyce, 1997; Nestadt et al., 1994, 2006; Sanislow et al., 2002). However, some studies found a factor solution in which OCPD was not represented (Blais et al., 1997; Hyler et al., 1990; Moldin et al., 1994). To our knowledge, only one study has investigated the internal factor structure of the OCPD criteria. In an exploratory factor analysis of the OCPD criteria in a sample of patients with binge eating disorder, Grilo (2004b) found a 3-factor solution, i.e., rigidity, perfectionism, and miserliness—dimensions that bear resemblance to the triad of Freud, i.e., stubbornness, orderliness, and parsimony. We are not aware of studies that have confirmed this factor structure.

Freud (1908) was careful not to confuse the anal character with the obsessional neurosis, a diagnosis similar to the current concept of obsessive-compulsive disorder (OCD). Up to the present, the relationship between OCD and OCPD is controversial. Most studies have found a positive association between these disorders (Crino and Andrews, 1996; Diaferia et al., 1997; Eisen et al., 2006; Nestadt et al., 2006), but some have not (Baer and Jenike, 1992; Crino et al., 2005). As outlined by Krueger (2005), empirical research should not only focus on comorbidity, but also on the question why there is such an overlap between certain axis I and axis II disorders. There is a lack of studies examining which aspects of OCPD may account for the overlap with OCD.

Another issue that has received little attention is the quality of the OCPD construct as a prototype category. The prototype model is of special interest for personality pathology because it reflects the way humans construct mental representations of phenomena they encounter in the world. From DSM-III-R and onwards, the classification has adopted a prototype approach by defining the PD categories by sets of polythetic, descriptive, and nonetiological criteria. In this approach, any single criterion is neither necessary nor sufficient for a diagnosis to be set. Members of a prototype category can match the prototype to different degrees, and they may also have few or many features in common with other prototypes. Thus, a prototype model of PD is of particular interest as it allows heterogeneity within, and similarities across categories, and preserves both a categorical and a dimensional approach to personality pathology. The notion of prototypes as being mental representations is often misinterpreted as being unrelated to external reality. However, as maintained by Lilienfeld and Marino (1999), prototype concepts do mirror reality since they emerge from repeated experiences with real-world entities. Empirical research has an important role in the search for and validation of prototype categories.

It has been claimed that the validity of the different PDs as distinct categorical entities is not warranted because empirical research has failed to show a “point of rarity,” or at least a “zone of rarity,” in the distribution of the PD criteria (Kendell and Jablensky, 2003). However, according to Meehl (1999), this is not a good criterion to determine whether a construct has a dimensional or categorical structure because latent dimensions may present themselves as categories and the phenomenal expression of a category may be dimensional. According to Meehl (1999), a more appropriate way to evaluate the latent structure of a diagnostic category is by way of taxometric procedures. Such studies need large samples of patients, at least 300, according to Meehl (1995), and 600 if dichotomous variables are involved (Ruscio, 2000), which may be the reason why OCPD has not yet been subjected to taxometric evaluation (Haslam, 2003; Rothschild et al., 2003).

The sample in this study consists of a large number of patients, most of whom had personality disorders. The large sample size makes it possible to conduct a variety of psychometric analyses, like factor analysis and taxometric procedures. The analyses were organized according to 9 assumptions concerning the notion of OCPD as a prototype category:

1. The prototype should be heterogeneous at its boundaries, i.e., there should be multiple diagnostic co-occurrences. Ideally, the “distance” to the different PDs should be equal.

2. The OCPD construct should reveal a “true” personality construct, i.e., there should be no specific associations with any symptom disorder.

3. The OCPD criteria should be correlated more strongly with OCPD than with the other PDs.

4. The diagnostic efficiency of the OCPD criteria should be adequate.

5. Indicators of reliability should have acceptable values.

6. The OCPD criteria set should form a separate factor in an exploratory factor analysis including all the PD criteria. A confirmatory factor analysis (CFA) should demonstrate a unitary model as most parsimonious with better goodness-of-fit indices, compared with the 3-dimensional model identified by Grilo (2004b).

7. Category membership should be heterogeneous within its boundaries, i.e., there should be no privileged way of qualifying for the prototype (no “core OCPD”).

8. PD prototypes are assumed to be dimensional constructs. Taxometric analyses should therefore indicate that OCPD has a latent dimensional structure.

9. Because the PD criteria reflect dysfunctional traits, the number of criteria met should be positively correlated with other measures of psychopathology, i.e., higher levels of symptoms and interpersonal problems, and lower levels of functional impairment. In other words, a full prototype should describe a person who is highly dysfunctional.

Aims of the Study

In summary, the aim of the study was to conduct a comprehensive evaluation of the quality of OCPD as a prototype category in a large sample of carefully assessed patients, most of whom had a PD diagnosis.

MATERIALS AND METHODS

Setting

This multisite study included patients from 12 units of the Norwegian Network of Psychotherapeutic Day Hospitals, specialized in the treatment of personality disorders (Karterud et al., 2003). All units adhered to the same treatment model consisting of short-term day treatment followed by long-term outpatient group therapy. The units complied with the diagnostic and data collection procedures required for membership in the Network (Karterud et al., 2003). The present study is based on data from the patient evaluation at admission.

Assessments

DSM-IV diagnoses were established according to the LEAD standard (“longitudinal, expert, all data;” Spitzer, 1983) in the following way: within 2 weeks after admission to day treatment, 1 staff member performed (a) the Mini-International Neuropsychiatric Interview for axis I diagnoses (Sheehan et al., 1994), and (b) the Structured Clinical Interview for DSM-IV Personality Disorders (SCID-II; First, 1994). The proposed diagnoses were then discussed at a staff conference and compared with other information, such as referral letters, the patients’ own written narratives, and several evaluation interviews. Two weeks after discharge the provisional symptom and PD diagnoses were discussed at a new case conference, taking into consideration clinical observations and additional information appearing during the 18-week treatment period. A final PD diagnosis required that the criteria from the original SCID-II protocol were confirmed by clinical observations. The SCID-II protocol was revised according to consensus among the staff members. There were no formal reliability tests of the diagnoses.

Functional impairment was assessed by global assessment of functioning (GAF; American Psychiatric Association, 1994), rated at admission, and subjected to consensus at staff conferences. A formal reliability study showed satisfactory reliability of GAF ratings (Pedersen et al., 2006). Self-reported symptoms and interpersonal problems at admission were assessed by the symptom check list (SCL-90-R; Derogatis, 1994) and circumplex of interpersonal problems (CIP; Pedersen, 2002). The CIP is a 48-item version of the inventory of interpersonal problems (IIP-C; Alden et al., 1990) and consists of 8 subscales. CIP sum score correlates 0.99 with the sum score of the circumplex version IIP-C (Pedersen, 2002). The items in SCL-90-R and CIP are rated on a 5-point Likert-scale from 0 to 4. Higher scores mean more symptomatic distress or interpersonal problems. Global Symptom Index (GSI) and CIP sum stand for the overall score of respectively subjective distress and interpersonal problems, assessed by SCL-90-R and CIP.

Subjects

From 1996 to 2005, 2367 patients were admitted to day treatment in one of the included day hospitals. From this sample, we successively excluded 57 patients because of missing diagnostic information or inconsistencies between the SCID protocol and final diagnosis, 36 patients because of missing data on GAF, GSI, or CIP sum, and 37 patients with more than 1 treatment period in the day hospital. Thus, the final sample consisted of 2237 patients.

Mean age was 35 years (SD = 9) and 71% of the patients was female. Twenty percent had no PD diagnosis, 54% had 1, 18% had 2, and 8% had 3 or more PD diagnoses. Patients with PD had an average of 1.5 PD diagnoses (SD = 0.8). The most frequent PD diagnosis was avoidant PD (39%), followed by borderline PD (24%) and PD not otherwise specified (18%, see also Table 1). One hundred and ninety-eight patients (9%) had OCPD. Mean number of symptom disorders was 2.2 (SD = 9.3). The majority of the patients (97%) had 1 or more symptom disorders: mood disorders, 74%; anxiety disorders, 68%; eating disorders 13%; substance-use disorders, 10%. The prevalence of OCD was 7%.

Participation in the study was voluntarily and all patients had given written consent before inclusion. The study was approved by the Regional Committee for Medical Research Ethics.

Statistics

To evaluate the strength of association between OCPD and the other disorders, odds ratios were computed. Pearson’s correlation coefficients were applied to evaluate the association between the OCPD criteria and the other PDs, and between the number of OCPD criteria and GAF, GSI, and CIP. The diagnostic efficiency of the criteria was evaluated by sensitivity, specificity, positive, and negative predictive value. According to Streiner (2003), diagnostic efficiency should also be evaluated by indices that control for agreement that may occur by just chance, e.g., κ  efficient. Internal consistency was evaluated by “reliability statistics,” i.e., intercriterion correlations, corrected item total correlations, mean intercriterion correlation, and Chronbach’s ∝ efficient. Because of an increased risk of type I errors, because of the large sample size, we set an
level of 0.01 for all analyses.

To evaluate the factor structure of OCPD in relation to the other PD criteria, we performed an exploratory factor analysis of all PD criteria, except the juvenile criteria for antisocial PD, with principal component analysis (PCA) as the extraction method and Varimax with Kaiser Normalization as the rotation method. Selection for further examination of the factors was based on eigenvalues (they should be higher than 1.0), clinical coherence, and the place of the elbow in the screenplot. A limitation of PCA is that error and specific variances are not separated out. Therefore, a principal axis factoring was performed as a control analysis, with the inspection of the residual correlation matrix of the OCPD criteria.

To compare the 3-dimensional model of Grilo (2004b) with a unitary model, we performed a CFA according to Lisrel 8 (Jo ̈reskog and So ̈rbom, 1996). Patients who had fulfilled none of the 8 OCPD criteria were excluded from this analysis, leaving a sample of 1365 patients for the CFAs. The OCPD criteria were dichotomized and the CFAs were performed on an asymptotic covariance matrix with tetrachoric correlations. The normed fit index (NFI; Bentler and Bonett, 1980), the comparative fit index (CFI; Bentler, 1990), and the root mean square error of approximation (RMSEA; Steiger, 1990) were applied to fully evaluate the model. The NFI and the CFI both measure the fit of the model compared with the independence model. Both of them are derived from the chi square statistic, and are supposed to lie between 0 and 1. The CFI is somewhat less affected by sample size than the NFI. Values greater than 0.90 for these measures are normally required for good fit of a model. The use of chi square as a central statistic is based on the assumption that the model holds exactly in the population, an assumption that may be unreasonable in empirical research (Jo ̈reskog and So ̈rbom, 1996), because it may imply that models which hold approximately in the population will be rejected in large samples. Browne and Cudeck (1993) proposed a number of measures accounting for the error of approximation and for the precision of the measure itself. One of these population discrepancy functions is the root mean square error of approximation (RMSEA; Steiger, 1990), which measures the discrepancy per degree of freedom. A RMSEA of 0.05 or below indicates a good fit of the model, whereas values between 0.05 and 0.08 indicate a reasonable fit.

For all analyses, we dichotomized the gradient score of the SCID-II protocol; a score of 1 (criterion absent) or 2 (criterion partly fulfilled) was rescored as criterion absent, and a score of 3 kept its status as criterion present. Pearson’s correlation coefficient was used for all correlation statistics. Strictly, it is not optimal to use this coefficient on dichotomous variables. Therefore, we performed several control analyses using the original SCID-II scores (1 = criterion absent, 2 = criterion partly fulfilled, and 3 = criterion present): (1) PCA and reliability statistics with Pearson’s correlation coefficient; and (2) correlation statistics with Spearman’s rank correlation coefficient.

The taxonic structure of the OCPD criteria was evaluated by MAMBAC analyses (“mean above minus below a cut”), described by Meehl (1995, 1999). For each MAMBAC analysis, we put 1 criterion aside, resulting in 8 curves, 1 curve for each criterion. Taking the first OCPD criterion as an example, this criterion was taken aside and criteria 2–7 were used as “cutting indicator” (in this context, cutting refers to the diagnostic cutoff point based on the number of fulfilled criteria). Subsequently, we created fictive diagnoses of OCPD defined by 7 different cutoff points, based on the number of OCPD criteria fulfilled, from 1 criterion fulfilled to 7 criteria fulfilled. For instance, a cutoff point at 1 criterion means that only 1 criterion among criteria 2–7 has to be present to get the diagnosis. Now, for every cutoff point 2 groups were created, those above and those below the cutoff point (those with a fictive OCPD diagnosis and those without). Continuing the example of criterion 1, for each cutoff point, we calculated the mean value of this criterion in each group, and the differences in means between the 2 groups. It is the differences in these means, plotted in a graph and connected with each other that forms a MAMBA curve. This procedure was applied for all the OCPD criteria. If OCPD were a taxon, i.e., a nondimensional distribution of OCPD criteria, the graph should appear as a convex upward (a bump). However, in case of a low base rate, the figure may also appear as an increasing plot. A straight line or concave upward, resembling a dish, is indicative for a dimensional factor. It should be noted that MAMBAC requires 2 indicators of a hypothesized latent construct. In our analyses, we regarded the separate OCPD criterion as a dichotomous variable and the number of the remaining OCPD criteria as a continuous variable.

Results

Assumption 1: Multiple Co-occurrences With Other PDs

OCPD had a high overlap with other PDs; 77% of those with OCPD had also other PD diagnoses. OCPD co-occurred with all other PDs, indicating that OCPD is heterogeneous at its boundary (Table 1). However, OCPD was significantly closer associated with paranoid PD than with the other PDs. Though not significant, there was also a close association between OCPD and schizoid and narcissistic PD.

Assumption 2: No Specific Association With the Symptom Disorders

We assessed the co-occurrence patterns of OCPD with the most common symptom disorders, i.e., major depression, dysthymia, bipolar disorder type I/II, panic disorder, agoraphobia without panic disorder, social phobia, generalized anxiety disorder, posttraumatic stress disorder, eating disorders, and substance-use disorders. OCPD was significantly related to OCD (odds ratio = 3.7, p < 0.001, results not given in a table). Eighteen percent of the OCPD patients had OCD (35 of 198). Conversely, 24% of OCD patients had OCPD (35 of 146). There were no significant associations with other symptom disorders.

Assumption 3: The OCPD Criteria Should Correlate Highest With OCPD

As expected, the OCPD criteria correlated more strongly with OCPD than with the other PD categories (Table 2). However, paranoid PD correlated positively with several OCPD criteria, notably criterion 6 (reluctant to delegate) and criterion 8 (stubbornness). These criteria were also significantly correlated with borderline PD. Furthermore, most OCPD criteria were significantly associated with OCD, especially criterion 1, 2, and 6 (preoccupation with order, perfectionism, and reluctance to delegate).

Assumption 4: Diagnostic Efficiency Indices Should be Acceptable

Overall, the diagnostic efficiency of the OCPD criteria was moderate, and there was a large variability of the diagnostic efficiency indices (Table 3). The positive predictive values were modest for criteria 1–4 and 7 and poor for criteria 5, 6, and 8. Criterion 6, however, had the highest negative predictive value and sensitivity. Although criterion 7, miserliness, had the highest positive predictive values, the sensitivity was very low, i.e., only a few patients with OCPD had this criterion (17%). κ efficient, which can be considered as an overall test of diagnostic efficiency, was highest for criterion 1 and lowest for criterion 7. Taken as a whole, it can be stated that criteria 1– 4 and criterion 6 performed best, and criterion 5, 7, and 8 performed poorest.

Assumption 5: Indicators of Reliability Should be Acceptable

The OCPD criteria had moderate internal consistency as indicated by a Chronbach’s ∝ of .57, interitem correlations in the moderate to lower range, and a mean interitem correlation of 0.14. Especially criterion 7 had low interitem correlations. Criteria 1–4 and criterion 6 had the highest correlations with each other (Table 4).

Assumption 6: Coherent Factor Solutions

The first PCA, including all PD criteria, resulted in the extraction of 22 components with eigenvalues higher than 1.0, explaining 51% of the variance. The OCPD criteria loaded on 4 different components (component 10, 11, 14, and 19), explaining 6.7% of the variance. Hoarding behavior loaded on component 14 and miserly on component 19. However, these components explained only a very small amount of the variance, and they seemed to have insufficient clinical coherence. A control analysis, using principal axis factoring as the extraction method, yielded an identical factor solution. The full PCA of all PD criteria has not previously been reported, but the most relevant components are discussed separately in previous papers. In short, the first factor represented avoidant PD (Hummelen et al., 2006), the second borderline PD (Johansen et al., 2004), the third paranoid PD (not published), the fourth and 12th factor histrionic PD (not published), and the fifth factor dependent PD (Gude et al., unpublished data, 2005).

With regard to the CFA, only the 3-dimensional model according to Grilo is displayed (Fig. 1). Overall, the fit of 3-dimensional model (RMSEA = 0.042, NFI = .58, CFI = 0.63) was comparable with the fit of the unitary model (RMSEA = 0.035, NFI = 0.62, CFI = 0.70). For both models, the RMSEA can be considered NFI and CFI indicated a poor fit of both had moderate loadings on the perfectionism component and only the miserly criterion loaded on the miserliness component. The rigidity component had no substantial loadings—1 criterion (rigidity) even loaded negatively.

Assumption 7: Heterogeneity Within Its Boundaries

Theoretically, with 8 criteria and a threshold of 4 criteria, OCPD patients can have 163 possible combinations of any 4 OCPD criteria. We identified 80 different combinations (49%), which emerged in a frequency ranging from 1 to 13 times. Criterion 6 was particularly common in the different combinations. In former studies, concerning avoidant PD and borderline PD, we found a considerably larger percentage of different combinations, 62 of 64 and 136 of 256, respectively (Hummelen et al., 2006; Johansen et al., 2004). Thus, though OCPD is heterogeneous within its borders, the heterogeneity is less pronounced compared with avoidant and borderline PD.

Assumption 8: Latent Dimensional Structure

As can be seen in Figure 2, the MAMBAC curves were not centrally peaked. However, several criteria had slopes that tended to peak at the right side of the plot, especially criteria 4 and 6, and to a lesser degree, criteria 1, 2, and 3. Taking into account the low base rate of the OCPD criteria (in a taxon with a low base rate the peak is to be expected in the right sight of the plot), this might indicate that these criteria OCPD have a taxonic structure, i.e., the phenomenological expression of a latent category. It should be noted that in the exploratory factor analysis, these criteria constituted the first factor, and they were strongest associated with OCD.

Assumption 9: the More Criteria, the More Dysfunctional

The number of fulfilled OCPD criteria was negatively correlated with GAF (r = -0.09, p < 0.001) and positively correlated with GSI (r = 0.14, p < 0.001) and CIP (r = 0.13, p < 0.001). However, the correlation coefficients were small. Furthermore, after controlling for the total number of PD criteria by partial correlations, these associations lost their significance. Thus, the weak association between the number of OCPD criteria and measures of global functioning, symptomatic distress, and interpersonal problems is probably due to the presence of criteria other than the OCPD criteria.

Control Analyses

The control analyses, using the original (nominal) SCID-II scores (1 = criterion absent, 2 = criterion partly fulfilled, and 3 = criteria present), yielded an identical factor solution, and indices of internal consistency were somewhat higher (Chronbachs’s ∝ =0.65). Reliability analyses (assumption 3) with Pearson’s and Spearman’s rank correlation coefficients yielded results that were highly comparable with the original findings.

Discussion

In this study, we evaluated 9 assumptions pertaining to the construct validity and prototype nature of obsessive-compulsive personality disorder (OCPD). There was an extensive overlap with all other PDs; 77% of patients with OCPD had co-occurrent PDs, but only the association with paranoid PD was significantly higher than expected in comparison to the other PDs. However, OCPD was also closely associated with narcissistic and schizoid PD. Among the symptom disorders, only OCD was significantly related to OCPD. The OCPD criteria correlated more strongly with OCPD than with the other PD categories, but most criteria were also significantly correlated with paranoid PD.

In concordance with other studies on the psychometric properties of the OCPD criteria (Blais and Norman, 1997; Farmer and Chapman, 2002; Grilo, 2004a,b; Grilo et al., 2001; Sanislow et al., 2002), we found that orderliness and perfectionism (criterion 1and 2) were among the criteria that performed best, whereas hoarding behavior and miserly performed poorest. Inflexible morality performed somewhat better in our study than in most other studies and rigidity and stubbornness performed somewhat poorer in our study (i.e., Grilo et al., 2004). The internal consistency of the OCPD criteria in our study was substantial lower than in other studies (e.g., Grilo et al., 2001, 2003). A PCA of all PD criteria supported a 2-dimensional structure of the OCPD construct; the first component consisted of OCPD criteria only, the second component consisted of criteria from other PDs as well. CFA of the OCPD criteria, however, indicated a poor fit of both the unitary model and the 3-dimensional model of Grilo (2004b). Taxometric analyses suggested that the criteria constituting the first component had a taxonic structure—the phenomenological expression of a latent category. Finally, the associations between the number of OCPD criteria and general measures of psychopathology were weak when controlling for the presence of other PD criteria.

Our factor analyses correspond poorly with the factor solution of Grilo (2004b), who found 3 components; rigidity, perfectionism, and miserliness. In fact, only the perfectionism component was replicated in our study (our first component), with preoccupation with order and perfectionism as the most significant features. Our second component—the rigidity component of Grilo—comprises mostly criteria from other PDs, i.e., paranoid, borderline, and antisocial PD, all criteria reflecting problems with aggression. It would therefore be better to conceptualize this factor as aggressiveness. This factor may explain the high overlap between OCPD and paranoid PD, which also was found in recent DSM-IV cooccurrence studies of comparable clinical populations (Farmer and Chapman, 2002; Fossati et al., 2000; Zimmerman et al., 2005). Grilo’s third component, miserliness, was not replicated by our factor analyses. Thus, the 2 dimensions of OCPD may best be conceptualized as perfectionism and aggressiveness.

The link between perfectionism and aggressiveness is in line with classical and contemporary psychoanalytical thinking. Freud (1908) attributed aggressiveness to the anal character, e.g. anal sadism. The psychoanalytic tradition conceptualised the motivational drive in the anal phase as striving for autonomy. When autonomy conflicts remained unresolved, one assumed that the aggressive drive became fixated in the character traits of (over-) control, stubbornness and rigidity. We will suggest a somewhat different etiology, linking the 2OCPD factors to evolution and narcissism. According to Kohut (1977), perfectionism and aggressiveness are aspects of the grandiose self. We will suggest that perfectionism is an aspect of the broader systemizing mechanism (SM), which Baron-Cohen has conceptualized based upon his studies of autism spectrum conditions (Baron-Cohen, 2006). SM is the evolutionary solution to the problem of predicting lawful (nonintentional) events. In normal development individuals balance their SM with their empathizing mechanism, which makes it possible to predict human agents. SM is a dimensional tendency (low – high) and it is gender biased (men being more influenced than women). Abramson et al. (2005) found an association between systemizing (restrictive/repetitive) behavior in autistic probands and obsessive-compulsive features in their parents. We suggest that OCPD individuals have both a temperamental and a systemizing genetic predisposition. Being high on systemizing means an inborn tendency to interpret the world according to mechanical concepts. Being low on empathizing mechanism means problems with intersubjective sharing, cocreation of meaning, and flexible cooperativeness. We suggest that OCPD develops out of an intersubjective matrix where children with a moderate to high inborn tendency of systemizing mechanism and thus displaying more rigidity, stubbornness, and perfectionism than average, are met by rigid and inflexible countermeasures by parents who may share the same genetic disposition. With regard to temperament, if there is a predisposition toward rage reactivity, this intersubjective matrix may handle the rage less than optimally and fails in containing and transforming the rage into healthy assertiveness as well as failing to modify the archaic grandiose self. It should also be noted that the link between rigidity and aggression has been highlighted in recent research on OCPD (Villemarette-Pittman et al., 2004).

The notion of perfectionism as an inborn tendency is supported by our taxometric analyses, indicating that perfectionism may be an expression of a latent categorical variable, assumably a genetic vulnerability (Shafran and Mansell, 2001). The overlap between OCPD and OCD is probably due to this shared genetic vulnerability. However, it may also be due to the inclusion of OCPD criteria that belongs more to within the rubric of OCD. This was supported by a recent study of Nestadt et al. (2006), who suggested that hoarding behavior is better conceptualized as an OCD feature than as a PD trait. Furthermore, a recent report from the collaborative study found that hoarding behavior, together with perfectionism and preoccupation with order, were strongest associated with OCD (Eisen et al., 2006). However, in our study, we did not find a significant association between the hoarding criterion and OCD. It has recently been inferred that hoarding behavior is not just a simple trait but reflects a multifaceted problem involving at least 4 aspects: (1) information processing deficits; (2) emotional attachments; (3) behavioral avoidance (avoidance of decision making); and (4) erroneous beliefs about the nature of possessions (Frost and Hartl, 1996). Thus, the poor psychometric properties of the hoarding criterion may not only be due to its low relevance in patients with OCPD, but also to the failure of 1 single criterion to catch such a complex phenomenon. The same may be true for the criterion of miserliness, which had even poorer psychometric properties.

Regarding the 3-cluster solution of the DSM system (American Psychiatric Association, 1994), we contend that OCPD does not conceptually belong to cluster C. These patients are not predominately anxious and fearful and they do not share the coping strategy of avoidance. Moreover, we would suggest to be careful not to overemphasize the positive aspects of OCPD, like conscientiousness and responsibility, as was done for instance in the obsessive prototype coined by Westen and Shedler (2000). Rather, it is the association between perfectionism and traits like excessive morality, stubbornness, and aggression that forms the core pathology of OCPD. Generally, in describing a comprehensive PD construct, one should also include problems in close relationships. It has been suggested that the difficulties OCPD persons experience within intimate human interactions is related to the fear of loosing control over their affects (Young et al., 2003, p. 261). The systemizing mechanism theory suggests that the reduced ability to predict human behavior may be related to these difficulties. If an individual with OCPD experience the partner as unpredictable and cannot explain the partner’s behavior by logical deduction, they may experience irritability, anger, or even rage, from which follows a disturbing feeling of not being able to control, and predict, one’s own inner experiences and behavioral acts. The SM theory may also imply that the current criterion overconscientiousness and inflexible morality may be a less than optimal description of the obsessive-compulsive person. Rather, the proneness to impose one’s own rules on others may appear as excessive morality.

Some caution should be shown with regard to the interpretation of the results. First, it should be noted that most of the patients in this sample had a low level of functioning. The OCPD criteria may function differently in samples of patients with a higher level of functioning. Second, there was no formal reliability check of the diagnoses. However, there are reasons to believe that the LEAD procedure contributed to a high validity of the diagnoses. Currently, empirical data are lacking, but a reliability study of the present LEAD procedure is under way. Third, the prevalence of the different PDs varied substantially, from 1% to 39%. This may have resulted in type II errors regarding some infrequent PDs (e.g., schizoid PD), and some symptom disorders (e.g., anorexia nervosa). Fourth, it is important to take into account that the internal consistency may be artificially high because of the “halo-effect.” Because the criteria for the different PDs were assessed sequentially, in accordance with the structure of the SCID-II, clinicians might be inclined to rate criteria similarly within a given disorder, leading to an artificial high coherence of the APD criteria. The application of the LEAD procedure may have minimized the halo-effect, but empirical evidence is lacking. Fifth, it should also be realized that the OCPD components only explained a small amount of the variance in the PCA. However, the link between perfectionism and aggression has theoretical coherence and recent research supported the importance of aggression in persons with OCPD (Light et al., 2006; Villemarette-Pittman et al., 2004). Finally, with regard to the taxometric analyses, there were 3 limitations. First, we used dichotomous indicators. However, this problem should be minimized because of the large sample; according to Ruscio (2000), samples of at least 600 should be used with dichotomous indicators. Second, we only conducted 1 type of taxometric analysis while it is recommended to use several types, e.g., both MAMBAC and MAXCOV. Third, overall, the OCPD criteria had poor psychometric properties. Thus, the findings concerning the taxometric analyses should be considered as preliminary. We are planning a more comprehensive study on this subject.

In conclusion, our findings support a 2-dimensional structure of OCPD, consisting of perfectionism and aggressiveness. The problems with aggression underscore the importance of poor affect regulation in persons with personality pathology. The quality of OCPD as a PD prototype category may be improved by deleting 2 criteria, i.e., hoarding behavior and miserliness, which also may reduce the overlap with OCD. These criteria can be replaced by criteria emphasizing the need for predictability in a variety of situations, notably within close relationships. Such revisions may result in a comprehensive PD construct including at least 3 aspects, i.e., perfectionism, aggressiveness, and interpersonal difficulties.

References available in original article in The Journal of Nervous and Mental Disease.