Symptom-Level Analysis of DSM-IV/DSM-5 Personality Pathology in Later Life: Hierarchical Structure and Predictive Validity across Self-and Informant Ratings

Abstract

Dissatisfaction with the categorical model of personality disorder led to several investigations on alternative, dimensional systems. The majority of these studies were conducted at the syndrome-level where each diagnostic criterion is summed or averaged within each disorder. Studies at the symptom-level have identified symptom dimensions that define and cut across categories, but the number and nature of dimensions varies across studies. The purpose of the present study was to examine the hierarchical structure and impact of personality pathology at the symptom-level across self-and informant ratings in a large community sample of older adults (N = 1,630; ages 55 to 64). Results indicated that multiple structural patterns can be organized within a common hierarchical framework, with a general factor of maladjustment at the top, two broad dimensions of internalizing and externalizing pathology directly below, and progressively more specific symptom dimensions toward the bottom. Factors at each level of the hierarchy were similar across self-and informant ratings. The four-factor model showed significant incremental validity in predicting a range of life outcomes over simpler models, while increasingly complex models incrementally but modestly improved predictive power. Several consistencies emerged between the current findings and prior factor analytic studies. The most unexpected result was the conspicuous absence of a disinhibition factor reflecting antisocial and impulsivity-related problems. This anomaly may involve the older age of our sample and the changing expression of personality pathology in later life.

General Scientific Summary

This study examined the hierarchical structure of personality disorder (PD) symptoms and impact of the resulting symptom dimensions on health and well-being across self-and informant ratings. The PD symptom hierarchy includes eight levels, with a general factor at the top, two broad dimensions of internalizing and externalizing pathology directly below, and increasingly more specific symptom dimensions toward the bottom. Each factor at all hierarchical levels was interpretable and robust across self-and informant ratings; however, the fourth level may be fundamental to understanding outcomes associated with PD.

Despite efforts to provide an objective, categorically-based diagnostic and classification system of personality disorder (PD), several problems emerged and continue to endure in the American Psychiatric Association’s (APA) Diagnostic and Statistical Manual of Mental Disorders (DSM; APA, 2013). These problems have been reviewed extensively (e.g., Clark, 2007; Krueger & Eaton, 2010; Livesley, 2003; Widiger & Samuel, 2005) and indicate clearly that PD symptoms do not define categorical entities that are distinct from each other or from normality, but are best understood as existing on latent dimensions. Many critics have argued that a reconceptualization of PDs in terms of personality dimensions will lead to a more structurally valid model. While several studies have examined alternative models (e.g., Mulder, Newton-Howes, Crawford, & Tyrer, 2011; O’Connor & Dyce, 1998), relatively little consensus has emerged about the number and nature of latent dimensions. Moreover, most studies focused on only one of many possible factor analytic solutions (selected primarily on the basis of model fit criteria) while neglecting the predictive validity of the preferred model. The present paper explored a hierarchical model of DSM-IV/DSM-5 PD symptoms across self-and informant ratings. We add to the literature in several ways. First, we evaluate factor replicability across two informational sources at each level of the hierarchy. Second, as a means of comparing the validity of the resulting factor solutions, we assess each level’s ability to predict a number of self-and informant rated outcomes.

Higher-Order Factor Analytic Studies of Personality Pathology

Much of the factor analytic work focusing on DSM-based descriptions of PDs has used diagnoses (summed or averaged endorsements of the diagnostic criteria within each disorder) as the input indicators. In a quantitative synthesis of 33 studies, O’Connor (2005) reanalyzed the intercorrelations among the 10 DSM-IV PDs and found that they were best summarized by a four-component consensus structure. The mean congruence between the consensus structure and individual data sets was moderately strong, but there was considerable variation in the degree to which the structure could be replicated. This may indicate problems with not only PD assessment methods, but also with current conceptualizations of the PD categories themselves (Clark, Livesley, & Morey, 1997). Because the PDs reflect a heterogeneous combination of symptoms, important within-disorder variation is lost in syndrome-level analyses. The structure does not indicate which factors account for diagnostic covariation, but only whether the structure supports or does not support the DSM higher-order cluster organization.

Other work has focused explicitly on the higher-order structure of maladaptive personality scales. Studies using self-report measures have supported two-(e.g., Markon, Krueger, & Watson, 2005),¹ three-(e.g., Clark, 2005), four-(e.g., Livesley, Jang, & Vernon, 1998), and five-(e.g., Krueger, Derringer, Markon, Watson, & Skodol, 2012) factor models of maladaptive personality. Widiger and Simonsen (2005; see also Krueger et al., 2011) observed that several of these models converge on highly similar constructs and suggested that they could be integrated within a common hierarchical structure. The basic descriptive units analyzed in these studies, however, were symptom clusters or trait composite scales rather than individual symptoms per se.

Although the use of aggregate-level indicators often has theoretical and psychometric advantages, potential pitfalls exist, especially when the constructs of interest are multidimensional (Bagozzi & Heatherton, 1994; Little, Cunningham, Shahar, & Widaman, 2002; Smith, McCarthy, & Zapolski, 2009). This is clearly apparent for the PDs. Several studies have shown the multidimensional nature of many PDs, including borderline (Sanislow, Grilo, & McGlashan, 2000), schizotypal (Raine et al., 1994), and narcissistic (Blais, Hilsenroth, & Castlebury, 1997). Trait scales and symptom clusters often show improved psychometric properties over the PD categories, but the degree to which they represent unitary constructs is not always known. The use of multifaceted indicators in multivariate research can be problematic, causing distortions in the resulting structure and biases in model parameters (Bandalos & Finney, 2001).

In a review of factor analytic studies of DSM PDs, Sheets and Craighead (2007) wrote: “Future research that analyzes the criteria-level correlation matrices of multiple studies is likely to provide the most valid representation of the true structure of personality pathology” (p. 87). However, compared to studies at the syndrome-level, relatively few studies have been conducted at the criteria-level, making meta-analytic studies of the kind Sheets and Craighead advocated difficult to perform. Moreover, several existing criteria-level analyses have tended to focus on particular disorders, rather than the full range of PD symptomatology. Clark (1992) recommended pooling multiple data sets, but to our knowledge, such large-scale analyses of symptom-level data have not yet been published.

Criteria-Level Factor Analytic and Hierarchical Studies of Personality Pathology

Investigations at the criteria-level have taken a variety of approaches to examine the core dimensions that give rise to PD symptomatology. Several studies have failed to support the unidimensional structure of most PDs (e.g., South & Jarnecke, 2017). Studies that examined symptoms across multiple PDs have also failed to confirm the a priori structure of DSM, at both the cluster and PD levels (e.g., Huprich, Schmitt, Richard, Chelminski, & Zimmerman, 2010; Moldin, Rice, Erlenmeyer-Kimling, & Squires-Wheeler 1994). Rather, results from exploratory analyses show that items corresponding to each PD typically spread across two or more personality dimensions. Studies have reported 10-, 11-, and 12-factor models (Huprich et al., 2010; Hyler et al., 1990; Torgersen, Skre, Onstad, Edvardsen, & Kringlen, 1993), but these did not fully correspond to the DSM system.

Most studies investigating the factor structure of PD symptoms have focused on a single source of information (e.g., self-report or interview ratings). Only one factor analytic study of PD symptoms (that we are aware of) compared factor structures across multiple sources. Thomas, Turkheimer, and Oltmanns (2003) asked Air Force recruits and first-year college students to rate themselves and their peers on PD symptoms using an early version of the Multi-source Assessment of Personality Pathology (the instrument used in the present study). Parallel analyses (Horn, 1965) indicated seven and eight factors (in the Air Force and college student samples, respectively), with the seven-factor solution being the most interpretable. The seven factors were labeled: Histrionic/Narcissistic, Dependent/Avoidant, Detachment, Anger/Mistrust, Antisocial, Obsessive-Compulsive, and Schizotypal. Congruence coefficients between corresponding factors in the self-report data suggested moderate to strong similarity for the first three factors and fair similarity for Anger/Mistrust and Schizotypal, with Antisocial and Obsessive-Compulsive replicating least well.

An increasing number of investigations have converged in their support for a unified, tiered structure where PD trait or symptom dimensions can be conceptualized across multiple levels of abstraction (e.g., Gutiérrez, Vall, Peri, Gárriz, & Garrido, 2014; Markon et al, 2005; Van den Broeck et al., 2014; Wright & Simms, 2014). For example, Gutiérrez et al. (2014) conducted a joint factor analysis of scales from three instruments and found a replicable hierarchical structure across gender and methods from one to seven factors. Similarly, Wright and Simms (2014) reported a one-to five-factor structure of scales from three personality and PD measures. These studies elucidate the structure of PD across multiple levels of an integrative hierarchy and help alleviate concerns about the putative lack of agreement among alternative factor analytic solutions. However, very few studies have examined the hierarchical unfolding of factors at the level of PD symptoms, and no study has examined the similarity of hierarchical levels across rating sources.

Impact of PD Symptom Dimensions on Consequential Life Outcomes

As compared to the growing number of studies that examined the latent structure of personality pathology, few have addressed the external correlates of the resulting dimensions. Rather, most studies have tended to focus more on indices of model fit and patterns of associations with other personality constructs rather than the potential implications of the underlying traits. The clinical literature has documented a number of functional impairments in patients with a traditional PD diagnosis (e.g., Ansell, Sanislow, McGlashan, & Grilo, 2007; Skodol et al., 2002). However, as Vall et al. (2015) pointed out, because PD categories are composed of heterogeneous traits, it is unknown which aspect of disorder confers risk, or whether each constituent trait represents different risks for different outcomes. For example, in DSM-5’s Alternative Model for Personality Disorder, borderline PD is partially defined by the traits of negative affectivity (e.g., emotional lability), disinhibition (e.g., impulsivity), and antagonism (e.g., hostility). Borderline PD has been linked to a number of negative outcomes in several areas of functioning (e.g., Zweig-Frank & Paris, 2002), but it is unclear whether the affective instability, impulsive, or relational component (or a combination of these components) is driving these effects. As the field is moving toward dimensional models of personality pathology, more research is needed to quantify the impact of PD symptom dimensions on real-life outcomes.

Vall et al. (2015) examined the ability of seven basic dimensions of personality pathology to predict a wide range of clinical outcomes, including dissatisfaction with a variety of life domains (e.g., job, health, social network), functional difficulties (e.g., housework, cognitive tasks, sleeping), and clinical severity (e.g., illegal drug use, depression, mental health service utilization). The seven dimensions together explained 17.6% of the variance, on average. Negative Emotionality accounted for the majority of this variance, whereas the remaining six dimensions explained considerably less. When the 7-factor model was compared to hierarchically superior solutions (i.e., the 1-to 6-factor models), gains in predictive power were modest. When the most complex 7-factor model was compared to the 1-factor model, increase in adjusted R² was 4.4%. These results indicate that Negative Emotionality had a disproportionality stronger effect on the prediction of several clinically relevant variables than the remaining six dimensions, and that complex models (i.e., 2-through 7-factor models) added incrementally but modestly over the most parsimonious 1-factor model.

Williams, Scalco, and Simms (2018) examined the criterion validity of general and specific factors derived from a confirmatory bifactor analysis of PD symptoms. The general factor correlated with several self-rated pathological personality traits, internalizing symptoms, and substance use, and it predicted interpersonal problems above and beyond these other forms of psychopathology. The incremental contribution of the general factor was not compared to the specific factors (or vice versa); thus, it is unclear if they might have added incremental predictive value beyond the general factor. Similarly, Conway, Hammen, and Brennan (2016) parsed general and specific factors from the 10 DSM-IV PD constructs and evaluated their ability to predict psychosocial and clinical variables in a sample of mother-offspring pairs. The general factor strongly predicted all outcomes. The specific factors together yielded only modest increases in explained variance in outcomes (e.g., social life, family, marital, work). Together, these studies provide initial construct validity for dimensional representations of PD by highlighting the relative impact these dimensions have on diverse life outcomes.

Overview of the Present Study

In the current analyses, we examined the dimensional structure and implications of personality pathology at the level of the DSM-IV/DSM-5 diagnostic criteria across self-and informant ratings. To delineate the structure of PD symptoms, we examined the hierarchical unfolding of factors at different levels of abstraction. Identifying links between factors at adjoining levels of the hierarchy can provide an integrative framework for evaluating relations between lower-and higher-order constructs. Finally, we evaluated the ability of the resulting factors across each level of the PD symptom hierarchy to predict a range of self-and informant rated outcomes. We chose four consequential life outcomes -mental and physical health, romantic relationship satisfaction, and social adjustment -that have been strongly linked with personality pathology (e.g., Oltmanns, Melley, & Turkheimer, 2002; South, Turkheimer, & Oltmanns, 2008). While we considered multiple decision rules to determine the “optimal” number of factors, we relied on these latter analyses to illuminate the level at which predictive power is maximized. We conducted these analyses using data from a sample of adults who were between the ages of 55 and 64 when they entered the study. Although the overall prevalence of personality pathology seems to remain relatively constant across the lifespan, specific features of some types of PD may increase in frequency as people get older (e.g., schizoid and obsessive-compulsive PDs) while others may decrease (e.g., antisocial and borderline PDs) (Balsis, Zweig, & Molinari, 2015; Oltmanns & Balsis, 2011). Thus, the resulting latent dimensions in our sample may differ as a function of items that may be more or less likely to be endorsed in this age group.

Method

Participants and Procedures

We analyzed data from the St. Louis Personality and Aging Network (SPAN), a longitudinal study of personality and health in later life (see Oltmanns, Rodrigues, Weinstein, & Gleason, 2014). The sample consisted of 1,630 middle-aged and older adults residing in the St. Louis area, with an average age of 59.5 (SD = 2.7; range = 55 to 64). Fifty-five percent of the sample was female. Reflecting the ethnic and racial composition of St. Louis, 65% identified as White, 32% as African American, and 2% as Hispanic or Latin American. The remaining 1% identified as a different race or did not respond to this question. The current analyses used data from 1,610 participants who completed a self-report measure of personality pathology (described below) at the baseline assessment. This assessment consisted of an informed consent process including a thorough explanation of the procedures, followed by a battery of interviews and questionnaires administered over a 3-hour session.

Based on a semi-structured diagnostic interview for PDs (the Structured Interview for DSM-IV Personality; Pfohl, Blum, & Zimmermann, 1997), 8% of the sample met DSM-IV criteria for at least one PD and 2% met criteria for PD Not Otherwise Specified (defined as the presence of 10 or more features that did not fit into any of the other existing PD categories) (Oltmanns et al., 2014). These numbers suggest that our sample exhibits rates of personality pathology that would be expected in a representative community sample. Participants were asked to indicate whether they have ever “received treatment for a mental disorder or advice from a mental health professional on problems in life.” Forty-four percent (N = 719) of the sample reported having received some form of treatment: 24% counseling, 8% medication, 9% counseling and medication, and 3% other form of treatment (e.g., inpatient hospitalization, convulsive therapy). Of those who received treatment, 96.7% (N = 695) specified a mental health problem: 47.8% mood disorder (e.g., depression, bipolar), 29.6% adjustment disorder (e.g., relationship problems, family issues), 11.8% anxiety disorder (e.g., generalized anxiety, post-traumatic stress), 5.3% substance disorder (e.g., alcoholism, addiction), and 5.5% other form of disorder (e.g., attention-deficit, sleep disorder). Four PD categories (antisocial, avoidant, borderline, and dependent) were more common among those participants who had received mental health treatment (Lawton & Oltmanns, 2013).

Participants were also asked to nominate an informant who knew them well and would be willing to complete a series of questionnaires about the participant. Overall, 90% (N = 1,467) of participants identified an informant who agreed to participate in the study. They were 68% female (N = 1,004) and between the ages of 18 and 92 (M=55.4, SD = 11.5). Sixty-seven percent identified as White, 31% as African American, and 1% as Hispanic or Latin American. The remaining 1% identified as a different race or did not respond to this question. Nearly half of the informants were a spouse or romantic partner of the participant (48%, N = 712). The remaining were typically either other family members (e.g., brother, sister, child; 28%, N = 404) or friends (22%, N = 328). On average, informants indicated that they had known their targets for 32 years (SD = 15.0) and knew them “better than anyone else” (M=1.5, SD = 0.6; scale is 1 to 5).

Personality Disorder Measure

Multi-source Assessment of Personality Pathology (MAPP; Oltmanns & Turkheimer, 2006).

The MAPP is an 80-item measure of personality pathology based on lay translations of the diagnostic criteria for the 10 PDs listed in DSM-IV. The items are listed in a quasi-random order to disguise the nature of the constructs assessed. All of the 79 DSM-IV PD symptoms were rewritten into words that avoided the use of technical terms and psychiatric jargon. One diagnostic criterion (Narcissistic PD criterion #8), “Is often envious of others or believes that others are envious of him or her,” was split into two separate items: (a) I am jealous of other people and (b) I think other people are jealous of me. Each item is rated on a 5-point, Likert-type scale ranging from 0 (I am never like this) to 4 (I am always like this). As noted above, one previous study reported the factor structure of the MAPP using self-and peer report data from younger adults (Thomas et al., 2003). Outside of this single study, the MAPP has primarily been used to provide a quantitative assessment of the 10 diagnostic categories. Convergent validity coefficients between the self-and informant MAPP varied between .13 (narcissistic PD) and .28 (paranoid and avoidant PDs; overall mean r =.23), indicating that some of the same symptoms are identified across sources, but that each source also provides unique information (Oltmanns et al., 2014).

Outcome Measures

Beck Depression Inventory (BDI-II; Beck, Steer, & Brown, 1996).

The BDI-II is a 21-item inventory that assesses the severity of depressive symptomatology. Each item is rated on a 4-point, Likert-type scale, with higher scores indicating greater depressive severity. The BDI-II has been found to possess adequate internal consistency and construct validity in samples of community-dwelling adults (e.g., Segal, Coolidge, Cahill, & O’Riley, 2008). Coefficient alpha for self-reports in the current sample was .89; informants did not complete the BDI-II.

Dyadic Adjustment Scale (DAS-4; Sabourin, Valois, & Lussier, 2005).

The DAS-4 is an abbreviated version of the DAS-32 (Spanier, 1976), a widely used measure of relationship satisfaction. The DAS-4 is a well-validated instrument that includes items about thoughts of relationship dissolution, how well the relationship is going, frequency of confiding in one another, and happiness in the relationship. A total score is computed by summing the items, with higher scores indicating higher relationship satisfaction. A total of 938 participants provided self-report data; 630 of 712 informants who were married to the participant also rated their own satisfaction in the relationship. Coefficient alphas were .78 for self-report and .81 for spouse-report.

RAND-36 Health Status Inventory (HSI; Hays & Morales, 2001).

The HSI is a self-report questionnaire consisting of 36 items assessing subjective physical and mental health. The present analyses used the general health composite scale, which is an aggregate of eight weighted physical and mental health subscales (i.e., physical functioning, role limitations due to physical problems, pain, general health perceptions, emotional well-being, role limitations due to emotional problems, social functioning, and energy/fatigue). Scores ranged from 31 to 83 (M = 68.2; SD = 9.8). Studies have supported the reliability and validity of the HSI in community adult samples (e.g., Vander Zee, Sanderman, Heyink, & de Haes, 1996). Response options vary by questions; however, all items are keyed in the direction of better health. Coefficient alphas for the subscales in the present older adult sample ranged from .75 (Pain) to .92 (Physical Functioning; Mdn. = .82).

We adapted 10 of the original 36 HSI items for informants to complete about the participant. Four items assessed general and physical health status (e.g., “during the past 4 weeks, to what extent have his/her physical health problems interfered with his/her ability to work or engage in physical activities?”); two assessed social functioning (e.g., “during the past 4 weeks, to what extent have his/her emotional problems interfered with his/her normal social activities with family, friends, neighbors, or groups?”); three assessed emotional well-being (e.g., “has s/he been a very nervous person”?); and one assessed energy/fatigue (“during the past 4 weeks, did s/he seem to be full of pep?”). Each item is rated on a 5-point, Likert-type scale with higher scores indicating better health. A total score was taken by averaging the items. Coefficient alpha was .87.

Social Adjustment Scale (SAS; Weissman, 1999).

The SAS is a 56-item measure that assesses instrumental and expressive role performance in a variety of social contexts (e.g., work, social and leisure activities, relationships with extended family). For the current study, we adapted eight of the original 56 items for informants. Three items assessed social and leisure activities (e.g., “how many friends did s/he see [or has s/he been in contact with] in the last 2 weeks?”); three assessed quality of relationships (e.g., “how well has s/he been getting along with other people”); and two assessed work (e.g., “how well has s/he been able to do her/his work in the last 2 weeks?”). Each item is rated on a 5-point, Likert-type scale, scored in the direction of greater adjustment. A total score was computed by averaging the items. Coefficient alpha was .69.

Overview of Data Analytic Procedures

Number of Factors.

Three methods were used to determine the maximum number of factors to retain for extraction: parallel analysis (PA; Horn, 1965), the minimum average partial test (MAP; Velicer, 1976), and factor replicability. PA and the MAP test are widely recommended by statisticians (Velicer, Eaton, & Fava, 2000; Zwick & Velicer, 1986) and are increasingly being implemented with the availability of statistical software programs (Liu & Rijmen, 2008; O’Connor, 2000). Replicability of factor solutions is another highly recommended procedure (Fabrigar, Wegener, MacCullum, & Strahan, 1999) that has been adopted in a number of studies of normal and abnormal personality trait structure (e.g., Markon, Krueger, & Watson, 2005; McCrae, Zonderman, Costa, Bond, & Paunonen, 1996; O’Connor, 2002).

Exploratory Structural Equation Models of PD Symptoms.

Because previous studies have not supported the a priori DSM latent structure, we took an exploratory approach to modeling PD symptoms. Specifically, we performed an Exploratory Structural Equation Modeling (ESEM) analysis on the 80 items of the MAPP. ESEM is a powerful and flexible method for analyzing the latent structure of multidimensional constructs. It provides a compelling alternative to either confirmatory or exploratory factor analytic procedures (CFAs and EFAs) by integrating the many methodological advantages of both (Asparouhov & Muthén, 2009). Many writers have argued that CFAs may not be appropriate for modeling complex structures (such as personality traits), in part because of the overly restrictive CFA assumptions (Marsh et al., 2010). CFA requires each indicator to load on only one factor, which may diminish model fit due to potentially meaningful secondary loadings. Many of the most widely used measures of personality tend to perform poorly within a CFA context, but do quite well from an EFA perspective (Hopwood & Donnellan, 2010).

ESEM analyses were carried out using Mplus 7 statistical software (Muthén & Muthén, 1998–2012) using mean and variance-adjusted weighted least squares (WLSMV). WLSMV is a robust estimator for modeling ordinal data (Brown, 2006). Mplus estimates the polychoric correlations among the items, and then uses this correlation matrix as input for the factor analysis. The polychoric correlation is a measure of association between two ordinal variables that are theorized to have an underlying joint normal distribution, and helps to reduce the effect of statistical artifacts, such as artificial groupings of items based on similar distributional properties. The median response frequencies for responses of 0, 1, 2, 3, and 4 across all PD symptoms were 61.1%, 27.6%, 7.9%, 2.6%, and 0.8% for self-ratings, and 58.9%, 24.1%, 9.0%, 4.5%, and 1.8% for informant ratings. Descriptive statistics for each PD criterion are presented in Supplemental Table 1, as well as paired samples t tests and Cohen’s d values to index the size of the differences between self and informant ratings.

Goldberg’s (2006) top-down approach was used to examine the hierarchical structure of PD symptoms within an ESEM framework. This method involves estimating a series of factor models, beginning by extracting the most general factor and then successively extracting and rotating additional factors. One stops extracting factors when there is no variable with its highest loading on a new factor, or is lower than a user-defined cutoff. For each successive number of factors, factor scores are computed and saved. The across-model correlations between factor scores at each level and those below it serve to estimate the paths between levels of the hierarchy.² This procedure offers not only a useful method for elucidating the underlying dimensional structure of correlated variables, but also an adjunct for determining the number of factors to retain.

We examined the similarity of factors between self-and informant ratings across levels of the resulting PD symptom hierarchy. Specifically, we computed factor congruence coefficients (Φs; see Guadagnoli & Velicer, 1991) for parallel factors in each group following an orthogonal Procrustes rotation (Schönemann, 1966). We treated the Varimax rotated solution from the self-report data as the target matrix and rotated the loading matrix derived from the informant ratings to this target. Using conventional criteria, values in the range of .85-.94 suggest fair similarity, and values greater than .95 suggest strong similarity (Lorenzo-Seva & ten Berge, 2006). We examined multiple fit indices for each estimated model, including the comparative fit index (CFI), Tucker-Lewis incremental fit index (TLI), and root-mean-square error of approximation (RMSEA).

Prediction of Life Outcomes.

In order to compare the predictive validity of each hierarchical level, we used PD symptom dimensions that emerged across factor models to predict a number of self-(depressive symptoms, relationship satisfaction, general health) and informant (social adjustment, relationship satisfaction, general health) rated outcomes. We performed a series of hierarchical regressions where factors at each level of the hierarchy were entered as blocks (i.e., number of blocks equaled number of levels). Following Kotov et al. (2016), we tested the incremental validity of each level of the hierarchy above all preceding levels to ensure that the result for level N added new information not captured by general factors at higher-order levels. We report adjusted R² values to correct for the number of predictors in the model and tested for significance in change between blocks of predictors. Standardized regression coefficients are also presented. Regression analyses were conducted in SAS 9.4 (SAS Institute Inc., 2002–2012).

Results

Factor Structure

Number of Factors.

For the self-and informant report data, we performed separate PAs using a modified SAS program for variables with ordered categories (Liu & Rijmen, 2008). Eigenvalues from 50 data sets were computed, and the 95^th percentile of the distribution of random eigenvalues was generated and compared to the eigenvalues from exploratory factor analyses (eigenvalues are not available using ESEM). The first 10 factors in both the self-and informant report data had eigenvalues greater than those for parallel factors in the random data, suggesting that 10 factors should be retained. However, because PA has a tendency to overestimate (Zwick & Velicer, 1986), we also conducted the MAP test. In this procedure, the number of factors to retain is determined by examining the average squared partial correlation coefficients after partialing k (the number of variables) minus one factor out of the correlations between the variables. The number of factors that minimizes this value is the number that should be selected. Using a SAS program written by O’Connor (2000), the MAP test indicated that 6 factors should be retained in the self-report data, and 8 factors should be retained in the informant report data.

Model fit indices for the one-to ten-factor models (one represents the most parsimonious model and ten represents the upper bound suggested by the PA) in the self-and informant report data are presented in Supplemental Table 2. Using conventional criteria (Hu & Bentler, 1999), model fit reached acceptable levels in the self-report data for the four-factor solution (RMSEA = .028, CFI = .918, TLI = .909) and acceptable to excellent levels for the seven-factor solution (RMSEA = .023, CFI = .951, TLI = .941). In the informant report data, fit reached acceptable levels for the three-factor solution (RMSEA = .039, CFI = .914, TLI = .907) and excellent levels for the six-factor solution (RMSEA = .028, CFI = .958, TLI = .951).

Table 1 provides factor congruence coefficients for each of the ten models across self-and informant ratings. All factors in the first three models demonstrated strong similarity and continued to show strong consistency when additional factors were extracted. Factor II in the 4-factor solution was just below threshold for fair similarity, but was nearly equivalent across rating sources in the 6-factor solution. All factors reached fair to strong similarity in the six-factor solution. Replicability tended to improve for these six factors in more complex models (i.e., the 7-to 10-factor models). When 10 factors were extracted (as suggested by the PA), eight factors replicated (range = .89 to .98), but the remaining two showed no to fair similarity (Φs = .72 and .86).

Table 1.

Factor Congruence Coefficients for Ten Models of PD Symptoms across Self-and Informant Ratings on the Multi-Source Assessment of Personality Pathology

Number of Factors	Factor Congruence Coefficients
	I	II	III	IV	V	VI	VII	VIII	IX	X
1	.99
2	.98	.95
3	.98	.95	.96
4	.97	.84	.95	.96
5	.97	.84	.95	.96	.94
6	.97	.89	.95	.96	.88	.87
7	.99	.96	.94	.93	.97	.92	.77
8	.99	.91	.96	.94	.97	.96	.71	.92
9	.99	.87	.97	.93	.96	.90	.94	.95	.78
10	.98	.91	.93	.92	.97	.97	.72	.91	.89	.86

Note. Factor congruence coefficients are based on N = 1,610 self-reports and N = 1,464 informant reports. Numerical factor labels for factor models 1 through 8 correspond to Figure 1.

Hierarchical ESEM of PD Symptoms.

Congruence coefficients indicated fair to high similarity for at least six (or possibly eight) factors in the self-and informant ratings. We examined the interpretability of each factor across the various factor solutions. Factor scores were computed and saved, and then correlated across successive models to illustrate relations between factors at different levels of factor extraction. Figures 1 shows the hierarchical unfolding of PD symptom dimensions in the self-report data (see Supplemental Figure 1 for informant ratings).

Figure 1.

Hierarchical Representation of Self-Reported Personality Disorder Symptoms on the Multisource Assessment of Personality Pathology

Note. Correlations are presented for each factor at adjoining levels of the hierarchy. Numerical factor labels correspond to Table 1.

In the one-factor solution, 76 of 80 PD symptoms in the self-report data, and 74 of 80 in the informant report data, had loadings of at least .30, suggesting that this single factor represents General Maladjustment. Two factors emerged from this general factor, labeled Internalizing and Externalizing. Avoidant and dependent PD symptoms had the strongest loadings on the Internalizing factor, followed by items tapping emotional instability, withdrawal, paranoia, and behavioral rigidity. In contrast, the attention-seeking and grandiose features of histrionic and narcissistic PDs had the highest loadings on the Externalizing factor, followed by items tapping impulsiveness, anger proneness, manipulativeness, and deceitfulness. As shown in Table 1, factor congruence coefficients for Internalizing (i.e., Factor I) and Externalizing (i.e., Factor II) were .98 and .95, respectively. The odd or unusual features of schizotypal PD were equally spread across the Internalizing and Externalizing factors.

Moving down the hierarchy, the Internalizing factor at level 2 split to form two relatively broad factors labeled Negative Affectivity (Φ = .98) and Paranoid Withdrawal (Φ = .95), the latter of which shared a modest correlation with Externalizing. The Externalizing factor at level 3 split into two subfactors at level 4, labeled Emotion Dysregulation³ (Φ = .84) and Exhibitionism (Φ = .96). The Paranoid Withdrawal factor at level 4 split into two subfactors, Paranoid (Φ = .95) and Withdrawal (Φ = .94), at level 5. At level 6, the paranoid features of paranoid and schizotypal PDs and the odd or unusual features of schizotypal PD split to form two separable factors in the self-report data, the former labeled Paranoid Mistrust (Φ = .89) and the latter labeled Peculiarity (Φ = .87). In the informant data, Peculiarity (which included some impulsivity-related content) split from Exhibitionism at level 6. At level 7, obsessive-compulsive PD features, previously spread across Factors V and VI in the self-report data (at level 6), combined to form a single Rigidity factor, but failed to clearly emerge in the informant data as a separable factor until level 8. Instead, a bipolar factor (Dependency vs. Rigidity) separated from Social Inhibition and Paranoid/Rigidity at level 7 in the informant data. At level 8, Rigidity (Φ = .92) replicated across data sources, but Dependency (Φ = .71) did not.

When additional factors beyond eight were extracted, no items had loadings ≥ .30 and all had their highest loadings on other factors. The eight-factor solution therefore seemed to be a fair stopping point for modeling PD symptoms at the most differentiated level of representation in these data. Table 2 presents the loading matrix for the eight-factor model in the self-and informant ratings following an orthogonal Crawford-Ferguson Varimax rotation.⁴ Factor loadings are sorted by self-report, with corresponding values presented for informant report.

Table 2.

Eight-Factor Model of Self-and Informant Reported Personality Disorder Symptoms

Item	Description	Factor1	Factor2	Factor3	Factor4	Factor5	Factor6	Factor7	Factor8
DEP3	I don’t like to disagree with other people because I fear that they may reject me	.76/.67	.01/.07	.05/.01	.05/.00	.14/.14	.04/.07	.04/.23	−.03/.01
AVD4	I worry that other people will criticize or reject me	.75/.71	−.10/−.08	.09/.11	.15/.22	.19/.17	.09/.04	.04/.22	.08/.02
AVD5	I am reserved or shy when meeting new people because I worry that I might not measure up	.72/.72	.01/.05	.17/.16	.10/.09	−.11/−.07	.04/.07	−.05/.10	.07/−.01
AVD3	I am very controlled or inhibited with close friends because I am afraid people will make fun of me	.71/.69	.13/.18	.30/.19	.05/.12	.14/.17	.05/.01	.02/.19	.03/−.01
AVD1	I avoid working in teams because I am afraid someone will criticize or reject me	.67/.63	.12/.20	.33/.15	.12/.25	.15/.17	.04/.10	.01/.27	.00/.00
DEP1	I find it hard to make a simple decision without lots of advice from other people	.59/.38	.13/.05	.20/.16	.16/.08	.18/.11	.17/.17	.25/.58	.00/.06
AVD6	I am not as much fun or as attractive as other people	.58/.68	.26/.07	.15/.08	.14/.16	−.02/−.03	.08/.07	−.07/.22	.19/.16
DEP4	I am afraid to start or do things by myself	.57/.42	.07/.11	.12/.17	.24/.10	.11/.09	.07/.07	.31/.67	.00/.01
AVD2	I am unwilling to get involved with other people unless I am certain of being liked	.57/.58	.11/.13	.23/.20	.07/.23	.36/.37	.00/.02	.00/.16	.08/.06
HST7	I am easily influenced by other people	.56/.44	−.05/−.04	−.03/−.04	.15/.16	.35/.21	.08/.23	.04/.48	.02/.01
AVD7	I do not like to do or try new things because they might be embarrassing	.55/.59	.18/.27	.17/.09	.10/.13	.06/.06	−13/−.10	.09/.15	.21/.15
NAR82	I am jealous of other people;	.52/.45	−.02/.06	.04/.21	.24/.34	.41/.33	.02/.08	.00/.27	.05/−.10
BDL7	I feel emotionally unfulfilled or that life is meaningless	.49/.52	.13/.10	.21/.13	.30/.39	−.04/.00	.20/.19	.10/.18	−.02/−.08
DEP2	I depend on other people to take care of me	.45/.18	.03/.09	−.11/.01	.23/.27	.24/.13	.05/.04	.30/.65	.04/−.07
SZT1	When I see other people talking, I begin to think that they may be talking about me	.43/.56	−.02/.02	.41/.35	.17/.24	.19/.23	.14/.07	.1½2	−.03/−.06
SZD5	I am not interested in close relationships	.13/.22	.58/.60	.26/.22	.08/.20	−.03/.01	.07/.08	.02/−.03	.04/−.01
SZD1	Close relationships are not important to me (including being part of a family)	.10/.23	.50/.46	.35/.29	.06/.22	.04/.13	.04/.07	.03/.02	.01/−.09
SZT8	I have no close friends (other than family members)	.25/.33	.40/.44	.39/.20	.05/.26	.01/.12	.05/.06	−.11/.01	.06/−.07
SZD4	I don’t enjoy doing anything	.31/.37	.36/.40	.34/.18	.27/.31	.07/−.03	.09/.08	.13/.20	−.01/−.04
SZD7	I am not very good at showing my feelings	.33/.24	.35/.57	.26/.13	−.07/.04	.09/.07	−.03/.05	−.18/.08	.14/.05
SZD3	I have little interest in having a sexual relationship	.16/.19	.35/.30	.09/.09	.01/.15	−.14/−.02	.10/.02	.17/.06	.06/.08
SZD2	I prefer to do things alone	.13/.19	.25/.36	.22/.15	.09/.16	−.06/−.05	.15/.13	−.09/−.28	.17/.10
NAR7	It is not my job to listen to, or solve other people’s problems	.16/.10	.24/.38	.16/.08	.06/.25	.23/.19	.00/−.01	−.09/.14	.17/.10
SZT5	I am constantly on edge to make sure that other people don’t take advantage of me	.18/.26	.10/.09	.80/.76	.18/.33	.18/.18	.04/.08	.12/.17	.08/.03
PND1	I am constantly on the lookout to make sure that other people are not taking advantage, lying to, or harming me	.13/.16	.11/.11	.73/.74	.13/.25	.20/.19	.03/.04	.04/.17	.13/.08
PND2	I have a hard time trusting other people and I often wonder if I can trust my friends	.32/.35	.1½5	.67/.54	.22/.39	.06/.16	.09/.06	.01/.10	.06/.00
SZT9	I am nervous around other people because I don’t trust them	.39/.57	.14/.17	.57/.49	.19/.20	.16/.06	.07/.08	.04/.11	.03/.01
BDL9	When I am under stress, I may become paranoid or suspicious of people I usually trust, or have other strange experiences that are hard to explain	.33/.38	.04/.04	.55/.49	.29/.35	.13/.14	.26/.27	.06/.20	−.08/.00
PND7	I have concerns that my sexual partner is not being faithful to me	.20/.28	−.06/−.05	.47/.37	.10/.17	.14/.29	.28/.14	.10/.00	−.07/−.09
PND3	I do not want to share personal information with other people because I am afraid that it may get into the wrong hands	.21/.22	.07/.21	.43/.41	.01/.12	.06/.04	.07/.04	−.07/.04	.34/.18
PND4	Rather than taking what people say at face value, I try to read between the lines and figure out what they really mean	.20/. 26	−.01/.06	.42/.42	.08/.22	.13/.28	.20/.08	−.02/.08	.30/.20
HST8	I have gotten hurt in relationships because I thought that the relationship was closer thank the other person did	.28/.33	−.07/−.06	.40/.28	.09/.13	.08/.26	.28/.18	−.02/.03	.06/.04
NAR2	I find myself daydreaming about power, success and/or the perfect relationship that will be mine someday	.18/.18	−.05/−.10	.40/.25	.03/.13	.29/.36	.25/.25	.01/.11	−.03/−.02
OBC5	I can’t throw out old things even if they are of no use to me	.17/.14	.13/.14	.20/.16	.12/.12	.05/−.02	.17/.11	.05/.29	.11/.13
ATS4	I get mad easily and often get in fights	.25/.12	.03/.06	.22/.13	.76/.81	.19/.17	.03/.09	.07/.14	.00/−.05
BDL8	I have sudden, intense outbursts of anger	.25/.15	.02/.09	.18/.11	.67/.86	.18/.18	.11/.10	.09/.09	.08/.02
BDL6	I have strong mood swings in response to events; I have frequent periods of intense sadness, irritation or anxiety	.40/.40	.04/.00	.28/.21	.49/.63	.03/.08	.30/.24	.21/.17	.01/−.02
PND6	I become angry quickly when I am criticized	.40/.21	.02/.13	.14/.22	.46/.69	.27/.27	.02/−.02	.09/.15	.18/.03
ATS1	I repeatedly get in trouble with the police	.14/.05	.20/.19	.35/.17	.45/.23	.42/.32	.10/.31	.11/.31	−.23/−.34
OBC8	I can be rigid and stubborn	.15/.14	.07/.31	.22/.21	.43/.57	.09/.21	.10/.08	−.08/.11	.31/.08
BDL5	I have threatened to hurt, or kill myself	.32/.37	.03/.02	.22/.16	.40/.44	.07/−.07	.22/.38	.20/.21	−.08/−.16
BDL2	In close relationships (with friends and family members), I often switch back and forth between loving a person and hating him or her	.29/.27	.12/.10	.32/.27	.38/.54	.21/.28	.21/.15	.18/.19	.02/−.18
PND5	I hold grudges for a long time if I am insulted or injured	.27/.24	.03/.17	.31/.30	.35/.52	.18/.25	.00/−.05	.00/.18	.17/−.01
OBC2	I am a perfectionist and my perfectionism gets in the way of getting things done	.21/.26	.08/.11	.17/.20	.24/.17	.15/.18	.18/.11	.02/.05	.20/.18
HST1	I like being the center of attention and feel disappointed when I am not	.25/. 12	−.06/.01	.01/.03	.13/.27	.72/.73	.08/.11	.06/.30	.02/.02
NAR1	I think that I am much better than most other people	.10/.09	.07/.24	.11/.18	.07/.33	.66/.63	.06/.01	.04/.11	.24/.08
NAR6	I will do just about anything to get what I need or think I deserve even if it means having to ‘step on a few toes’	.10/.10	.14/.21	.40/.38	.32/.35	.65/.53	.00/.11	−.05/.29	.02/−.14
DEP5	I will do just about anything to get other people to take care of me	.43/.25	.11/.09	.30/.18	.21/.25	.58/.36	.07/.14	.24/.68	−14/−.13
HST4	I use physical appearance to draw attention to myself	.15/.14	−.15/−.08	.20/.14	−.05/.12	.57/.68	.18/.18	.09/−.01	.01/.06
NAR4	Being noticed and/or admired by others is important to me	.22/.18	−16/−.16	.02/.00	.02/.20	.53/.61	.10/.04	.14/.15	.11/.04
NAR81	I think other people are jealous of me	.17/.19	.00/.09	.23/.31	.00/.21	.52/.57	.14/.02	.05/.12	.13/−.02
NAR5B	I expect other people to do what I say	.19/.03	.05/.25	−.01/.13	.18/.41	.50/.44	−.04/−.07	.00/.24	.26/.21
HST2	I am more flirtatious than other people	−.06/−.01	−17/−.01	.22/.15	.14/.16	.50/.58	.27/.34	.00/.06	.02/−.03
NAR3	Because I am so unique, only other special people understand me	.06/.16	.18/.14	.29/.32	.10/.22	.49/.55	.36/.28	.16/.07	.09/.19
ATS5	I am adventurous; I like to do things even if it could be dangerous to me or others	.03/−.04	.02/.05	−.02/−.04	.22/.10	.47/.40	.38/.50	−.30/.07	−.10/.06
ATS2	I can be deceitful when I need to be	.15/.10	.08/.27	.21/.25	.33/.32	.38/.35	.06/.14	−.15/.23	.02/−.23
DEP7	After I break up with a girlfriend/boyfriend, I am likely to jump into another relationship	.13/.17	−.05/−.01	.17/.11	.14/.12	.35/.42	.10/.19	−.07/.17	.05/−.09
HST5	In conversations with other people (such as about my personal beliefs), I usually emphasize my personal feelings and impressions and am bored by details	.25/.22	.11/.16	.20/.12	.21/.41	.31/.32	.17/.17	.00/.21	.17/.10
NAR9	It is important to let other people know when they are incompetent and I don’t worry about whether they will like me	.00/−.08	.21/.22	.23/.28	.21/.42	.31/.35	.06/.03	−.04/.08	.18/.06
OBC3	My work is more important than spending time with friends and family, and/or having fun	.12/.21	.27/.31	.22/.16	.09/.06	.30/.27	.07/.08	−.10/−.05	.18/.27
BDL1	I will do almost anything to keep those that I love from leaving me	.25/.20	−.03/−.02	.21/.11	.11/.16	.30/.21	.09/.10	−.03/.27	.17/.23
ATS7	I seldom feel sorry or guilty of doing things that may have hurt others because I feel that my actions were justified	.16/.02	.14/.35	.24/.17	.24/.45	.27/.33	−.02/.09	−.12/.22	.16/.05
SZT3	I see, hear, or experience things differently from the way other people do	.10/.13	.15/.16	.35/.37	.05/.27	.18/.19	.51/.48	.13/.16	.20/.25
SZT4	Things make sense to me in a way that they may not for other people	.09/.09	.16/.19	.24/.24	.04/.22	.21/.18	.46/.40	.04/.10	.33/.34
BDL4	I am impulsive and have done things that could be dangerous to me	.15/.15	−.02/.06	.21/.10	.33/.33	.35/.33	.45/.65	−.22/.19	−.05/−.13
HST6	My expressions of emotion are stronger than most others’	.11/.11	−.07/−.22	.10/.16	.30/.57	.25/.17	.41/.29E	.19/.07	.24/.22
HST3	I am not afraid to show my emotions, and my emotions can change quickly	.12/.13	−.11/−.18	.09/.12	.34/.55	.17/.12	.38/.24	.15/.05	.11/.15
SZT7	I act or dress in an eccentric (or odd) manner	.19/.18	.23/.12	.10/.12	.17/. 17	.29/.20	.38/.40	.05/.20	−.04/.03
SZT6	I find myself laughing or crying when those around me are not laughing or crying	.17/.15	.06/−.13	.12/.09	.29/.22	.17/.09	.37/.32	.20/.18	.00/.18
ATS3	I like to do things on the fly without planning ahead	.12/.07	.04/.05	−14/−.16	.15/.19	.31/.21	.32/.40	−.05/.31	−.03/−.08
SZT2	I am superstitious or believe in mind-reading	.17/.19	−.07/−.09	.14/.24	.04/.06	.16/.14	.30/.33	.10/.18	−.03/−.04
ATS6	I have failed to do what was expected of me, such as completing my work or paying bills (Not due to circumstances that I could not control)	.24/.19	.18/.14	.25/.07	.24/.31	.17/.13	.28/.31	.13/.40	−10/−.15
BDL3	Compared to others, my opinions and preferences change more frequently	.27/.29	.13/.05	.13/.22	.20/.24	.26/.15	.27/.33	−.03/.24	.02/−.04
DEP6	I feel scared or uncomfortable when left alone to care for myself	.46/.40	.02/−.09	.22/.20	.31/.18	.14/.12	.08/.07	.62/.65	.00/.08
DEP8	I am afraid of being left alone to care for myself	.43/.41	−.08/−.14	.18/.15	.20/.18	.11/.10	.06/.04	.46/.59	−.01/.04
OBC4	Compared to others, I have very high standards when it comes to morals and ethics	.02/.04	.04/−.07	.18/.05	−.04/−.02	.23/.06	.16/−.06	.13/−.08	.44/.47
OBC7	I am more concerned with saving money than my peers are	.10/.09	.06/.14	.19/.27	.07/−.05	.17/.04	.04/−.02	−.09/−.09	.39/.42
OBC6	I need to do everything myself because no one else will do them right	.25/.22	.10/.18	.33/.28	.18/.26	.14/.27	.11/.10	−.07/−.13	.37/.27
OBC1	I am very concerned with details, rules, lists and schedules; I spend a great deal of time getting organized	.14/.07	.02/.00	.31/.29	.06/.00	.10/.15	.09/−.08	.08/−.09	.35/.39
SZD6	I don’t care whether other people praise or criticize me	.01/−.11	.16/.30	.08/.12	.06/.12	.17/.10	.14/.09	−.01/.01	.22/.24

Note. N = 1,610 and 1,464 for self-and informant reports, respectively. Items correspond to DSM-IV diagnostic criteria. Factors correspond to the eight-factor model in Table 1 and Figure 1. Loadings ≥ .30 in bold. Values before the slash are based on self-report data; values after slash are based on informant data. Factor1 = Social Inhibition; Factor2 = Detachment; Factor3 = Paranoid Mistrust; Factor4 = Angry Hostility; Factor5 = Exhibitionism; Factor6 = Peculiarity; Factor7 = Dependency; Factor8 = Rigidity.

Impact of PD Symptom Dimensions on Consequential Life Outcomes

Up to this point, we used multiple methods to determine the optimal number of factors to retain, including PA, the MAP test, model fit indices, and factor replicability. These procedures failed to converge on a single, consistent number of factors (e.g., PA and the MAP test suggested between 6 and 10 factors, while fit indices suggested fewer). When viewed from a hierarchical perspective, however, multiple factor models could be tenable. Thus, perhaps a more crucial consideration is whether different levels of the hierarchy can capture meaningful predictive variance. To address this question, we sought to predict a number of self-and informant rated outcomes from the symptom dimensions that emerged across the various factor solutions.

Focusing first on the prediction of self-reported outcomes using self-rated PD symptom dimensions as predictors (see blue [black] lines in Figure 2), the 2-, 3-, and 4-factor models showed significant incremental validity in predicting depressive symptoms, general health ratings, and relationship satisfaction.⁵ Note that, with the exception of relationship satisfaction, R² (solid lines) and adjusted R² values (dotted lines) were very similar, indicating little shrinkage due to the number of predictors in each model. Across all three outcomes, the largest increase in the proportion of variance explained was at level 4. For example, adjusted R² reached .32 in predicting depressive symptoms, representing an 8.8% increase when compared to levels 1, 2, and 3 combined. Similarly, level 4 contributed an additional 10.7% and 2.4% explained variance to the prediction of general health and relationship satisfaction, above and beyond all simpler levels. In contrast, all subsequent levels (i.e., levels 5 through 8) added only a cumulative 1.1%, 1.7%, and 4.4% increase to each outcome, respectively. A similar trend emerged when using informant rated PD symptom dimensions as predictors (see orange [gray] lines in Figure 2), though the total variance explained was much smaller than those based on self-ratings. However, across all three outcomes, the 4-factor model consistently showed significant incremental validity. Subsequent levels did add significantly to each outcome, but the greatest increase in explained variance was again at level 4, with the exception of relationship satisfaction (where level 8 contributed an increase of 1.7%, and level 4 contributed an increase of 1.3%).

Figure 2.

Prediction of Self-Reported Outcomes from Self-and Informant Rated PD Symptom Factors.

Note. Solid blue [black] lines show R² values for factor scores based on self-report data (dotted blue [black] lines show adjusted R² values); solid orange [gray] lines show R² values for factor scores based on informant report data (dotted orange [gray] lines show adjusted R² values). R² values are presented for each level of the hierarchy, including all preceding levels. Asterisks indicate significant ΔR² values when comparing a given hierarchical level against all simpler structures combined.

* p < .05. ^** p < .01. ^*** p < .001; two-tailed. The colored version of this figure is available in the online article.

Focusing next on the prediction of informant reported outcomes using informant rated PD symptom dimensions as predictors (see orange [gray] lines in Figure 3), the 4-, 6-, 7-, and 8-factor models showed significant incremental validity in predicting social adjustment, general health, and relationship satisfaction. While there appears to be a roughly linear trend between adjusted R² values and number of factors extracted, there is a clear shift in the lines at level 4 corresponding to the greatest increase in explained variance. For each outcome, adjusted R² reached .27, .30, and .32 at level 4, representing increases of 7.0%, 8.1%, and 6.7% over all preceding levels combined. In contrast, all subsequent levels (i.e., levels 5 through 8) added a cumulative 4.7%, 8.5%, 3.4%. With the exception of relationship satisfaction, a similar trend emerged when using self-rated PD symptom dimensions as predictors (see blue [black] lines in Figure 3). For social adjustment and general health, the 4-factor model again demonstrated the greatest increase in explained variance. While the 2-factor model showed significant incremental validity over level 1 in predicting social adjustment and general health, the increase in adjusted R² was much lower than that of the 4-factor model (2.0% vs. 4.6% for social adjustment and 2.5% vs. 8.6% for general health).

Figure 3.

Prediction of Informant-Reported Outcomes from Self-and Informant Rated PD Symptom Factors.

Note. Solid orange [gray] lines show R² values for factor scores based on informant report data (dotted orange [gray] lines show adjusted R² values); solid blue [black] lines show R² values for factor scores based on self-report data (dotted blue [black] lines show adjusted R² values). R² values are presented for each level of the hierarchy, including all preceding levels. Asterisks indicate significant ΔR² values when comparing a given hierarchical level against all simpler structures combined.

* p < .05. ** p < .01. *** p < .001; two-tailed. The colored version of this figure is available in the online article.

The preponderance of evidence suggests that level 4 is the most informative level for predicting these particular outcomes. Table 3 presents the factor loading matrix for the 4-factor model across self-and informant ratings. To evaluate the unique effects of each factor, we entered self-and informant factor scores (separately) in a regression model to predict depressive symptoms, social and dyadic adjustment, and general health (see Table 4).⁶ Focusing first on the coefficients that share the same source of measurement (i.e., self-rated predictors and self-rated outcomes, informant rated predictors and informant rated outcomes), Emotion Dysregulation had the strongest effects on all outcomes (average standardized b is .31 for self, .40 for informant), except for informant rated general health. Social Inhibition (.18 for self, .28 for informant) and Paranoid Withdrawal (.22 for self, .13 for informant) had small to medium effects, whereas Exhibitionism had the smallest effects (.13 for self, .07 for informants) and were in the opposite direction than the other three predictors, indicating a small but protective effect on outcomes. Focusing next on the coefficients that do not share the same source of measurement (i.e., informant rated predictors and self-rated outcomes, self-rated predictors and informant rated outcomes), we see a similar trend, but the effects are lower in magnitude. One striking difference, however, is that self-rated Social Inhibition in not associated with informant rated social or dyadic adjustment, whereas informant ratings are moderately associated with these outcomes.

Table 3.

Four-Factor Model of Self-and Informant Reported Personality Disorder Symptoms

Item	Description	Factorl	Factor2	Factor3	Factor4
DEP3	I don’t like to disagree with other people because I fear that they may reject me	.76/.67	.06/.01	.03/.16	.12/.06
AVD4	I worry that other people will criticize or reject me	.74/.72	.15/.14	.05/.16	.21/.13
AVD3	I am very controlled or inhibited with close friends because I am afraid people will make fun of me	.71/.68	.08/.14	.31/.33	.11/.05
AVD5	I am reserved or shy when meeting new people because I worry that I might not measure up	.69/.67	.08/.05	.19/.26	−.08/−.11
AVD1	I avoid working in teams because I am afraid someone will criticize or reject me	.68/.67	.13/.29	.32/.26	.12/.10
DEP4	I am afraid to start or do things by myself	.62/.73	.31/.20	.09/.01	.07/.15
DEP1	I find it hard to make a simple decision without lots of advice from other people	.61/.66	.30/.15	.19/.02	.15/.21
AVD2	I am unwilling to get involved with other people unless I am certain of being liked	.59/.56	.03/.23	.25/.33	.34/.27
AVD7	I do not like to do or try new things because they might be embarrassing	.59/.53	−.03/.14	.26/.36	.04/−.07
HST7	I am easily influenced by other people	.57/.64	.13/.18	−.05/−.10	.35/.27
AVD6	I am not as much fun or as attractive as other people	.56/.67	.04/.10	.33/.24	.02/−.04
DEP6	I feel scared or uncomfortable when left alone to care for myself	.56/.72	.53/.18	.08/−.03	.07/.25
NAR82	I am jealous of other people;	.55/.54	.14/.37	.03/.16	.40/.28
DEP5	I will do just about anything to get other people to take care of me	.51/.61	.30/.38	.19/−.07	.49/.40
DEP2	I depend on other people to take care of me	.50/.52	.24/.34	−.11/−.17	.20/.18
BDL7	I feel emotionally unfulfilled or that life is meaningless	.50/.56	.35/.40	.24/.14	−.02/.00
DEP8	I am afraid of being left alone to care for myself	.49/.69	.40/.15	.03/−.07	.06/.21
PND6	I become angry quickly when I am criticized	.46/.26	.31/.69	.17/.21	.31/.22
SZT1	When I see other people talking, I begin to think that they may be talking about me	.44/.61	.30/.25	.29/.29	.18/.21
BDL6	I have strong mood swings in response to events; I have frequent periods of intense sadness, irritation or anxiety	.42/.47	.59/.60	.24/.13	.08/.16
ATS4	I get mad easily and often get in fights	.34/.19	.57/.82	.18/.05	.21/. 17
BDL8	I have sudden, intense outbursts of anger	.33/.18	.53/.84	.17/.10	.23/.16
BDL5	I have threatened to hurt, or kill myself	.35/.48	.51/.48	.16/.01	.07/.06
HST3	I am not afraid to show my emotions, and my emotions can change quickly	.11/.17	.48/.44	.07/.05	.27/.27
SZT6	I find myself laughing or crying when those around me are not laughing or crying	.17/.25	.47/.18	.13/.01	.21/.28
ATS1	I repeatedly get in trouble with the police	.23/.25	.47/.42	.26/−.05	.35/.35
BDL9	When I am under stress, I may become paranoid or suspicious of people I usually trust, or have other strange experiences that are hard to explain	.33/.47	.46/.38	.44/.34	.14/.26
BDL2	In close relationships (with friends and family members), I often switch back and forth between loving a person and hating him or her	.33/.37	.45/.59	.30/.15	.22/.26
HST6	My expressions of emotion are stronger than most others’	.11/.17	.45/.46	.14/.06	.37/.36
ATS6	I have failed to do what was expected of me, such as completing my work or paying bills (Not due to circumstances that I could not control)	.25/.41	.39/.43	.25/−.07	.16/.20
PND7	I have concerns that my sexual partner is not being faithful to me	.18/.28	.36/.18	.33/.26	.16/.31
SZT7	I act or dress in an eccentric (or odd) manner	.18/.29	.31/.26	.19/.09	.30/.31
SZT2	I am superstitious or believe in mind-reading	.14/.31	.26/.11	.07/.05	.20/.28
SZT5	I am constantly on edge to make sure that other people don’t take advantage of me	.22/.36	.30/.36	.71/.60	.15/.29
PND1	I am constantly on the lookout to make sure that other people are not taking advantage, lying to, or harming me	.16/.28	.20/.29	.68/.58	.18/.29
PND2	I have a hard time trusting other people and I often wonder if I can trust my friends	.34/.37	.28/.45	.61/.52	.07/.15
SZT8	I have no close friends (other than family members)	.26/.27	.01/.39	.54/.38	−.01/−.03
SZT9	I am nervous around other people because I don’t trust them	.41/.56	.25/.23	.53/.50	.13/.06
SZD1	Close relationships are not important to me (including being part of a family)	.14/.20	.05/.39	.51/.41	−.03/−.01
PND3	I do not want to share personal information with other people because I am afraid that it may get into the wrong hands	.19/.21	.00/.14	.50/.47	.15/.07
SZD5	I am not interested in close relationships	.16/.14	.04/.39	.50/.47	−.07/−.14
PND4	Rather than taking what people say at face value, I try to read between the lines and figure out what they really mean	.18/.27	.16/.20	.45/.42	.23/.33
OBC6	I need to do everything myself because no one else will do them right	.25/.11	.09/.24	.45/.47	.24/.26
SZT3	I see, hear, or experience things differently from the way other people do	.05/.24	.36/.34	.43/.34	.29/.40
SZD7	I am not very good at showing my feelings	.32/.20	−.17/.24	.43/.39	.08/−.08
SZD4	I don’t enjoy doing anything	.35/.41	.28/.42	.42/.30	.03/−.10
SZT4	Things make sense to me in a way that they may not for other people	.04/.15	.24/.27	.41/.33	.34/.34
SZD2	I prefer to do things alone	.11/.00	.07/.21	.39/.42	−01/−.12
OBC1	I am very concerned with details, rules, lists and schedules; I spend a great deal of time getting organized	14/−.01	.08/−.08	.37/.41	.18/.19
OBC3	My work is more important than spending time with friends and family, and/or having fun	.14/.11	.00/.12	.37/.44	.32/.20
NAR9	It is important to let other people know when they are incompetent and I don’t worry about whether they will like me	.05/−.02	.10/.48	.34/.27	.33/.33
OBC8	I can be rigid and stubborn	.19/.17	.23/.63	.33/.30	.20/.17
HST8	I have gotten hurt in relationships because I thought that the relationship was closer thank the other person did	.24/.32	.27/.12	.33/.24	.15/.30
OBC7	I am more concerned with saving money than my peers are	.10/.00	−.05/−.09	.33/.47	.26/.09
ATS7	I seldom feel sorry or guilty of doing things that may have hurt others because I feel that my actions were justified	.20/.12	.07/.57	.32/.23	.30/.29
PND5	I hold grudges for a long time if I am insulted or injured	.31/.30	.23/.55	.32/.28	.22/.20
OBC4	Compared to others, I have very high standards when it comes to morals and ethics	.02/−.04	03/−.15	.30/.24	.32/.13
NAR7	It is not my job to listen to, or solve other people’s problems	.19/.12	−.06/.35	.30/.26	.23/.09
OBC5	I can’t throw out old things even if they are of no use to me	.17/.28	.17/. 17	.27/.14	.08/.06
SZD3	I have little interest in having a sexual relationship	.16/.17	.08/.22	.25/.25	−.16/−.09
OBC2	I am a perfectionist and my perfectionism gets in the way of getting things done	.22/.24	.21/.17	.25/.30	.22/.19
HST1	I like being the center of attention and feel disappointed when I am not	.29/.25	.11/.32	−.05/.01	.70/.69
NAR1	I think that I am much better than most other people	.15/.11	.00/.41	.17/.27	.67/.52
NAR6	I will do just about anything to get what I need or think I deserve even if it means having to ‘step on a few toes’	.20/.27	.20/.50	.36/.23	.61/.50
HST4	I use physical appearance to draw attention to myself	.15/.12	.14/.12	.06/.15	.57/.67
HST2	I am more flirtatious than other people	−.05/.06	.28/.24	.09/.06	.55/.63
NAR4	Being noticed and/or admired by others is important to me	.24/.22	.09/.15	−.06/−.02	.54/.58
ATS5	I am adventurous; I like to do things even if it could be dangerous to me or others	−.01/.04	.24/.19	.02/−.04	.53/.51
NAR81	I think other people are jealous of me	.18/.23	.08/.27	.21/.28	.53/.50
NAR5B	I expect other people to do what I say	.24/.11	−.01/.46	.09/.24	.52/.37
NAR3	Because I am so unique, only other special people understand me	.08/.18	.31/.28	.33/.37	.50/.59
BDL4	I am impulsive and have done things that could be dangerous to me	.10/.30	.41/.44	.20/−.03	.45/.50
ATS2	I can be deceitful when I need to be	.19/.23	.18/.48	.22/.15	.39/.29
DEP7	After I break up with a girlfriend/boyfriend, I am likely to jump into another relationship	.14/.26	.12/.18	.13/.03	.38/.42
HST5	In conversations with other people (such as about my personal beliefs), I usually emphasize my personal feelings and impressions and am bored by details	.27/.29	.19/.45	.27/.17	.35/.33
ATS3	I like to do things on the fly without planning ahead	.10/.25	.22/.28	−.08/−.21	.35/.30
BDL1	I will do almost anything to keep those that I love from leaving me	.25/.31	.09/.14	.21/.12	.34/.29
NAR2	I find myself daydreaming about power, success and/or the perfect relationship that will be mine someday	.16/.25	.24/.15	.29/.12	.31/.44
BDL3	Compared to others, my opinions and preferences change more frequently	.26/.41	.25/.30	.18/.11	.29/.26
SZD6	I don’t care whether other people praise or criticize me	01/−.11	.04/.19	.22/.28	.23/.12

Note. N = 1,610 and 1,464 for self-and informant reports, respectively. Items correspond to DSM-IV diagnostic criteria. Factors correspond to the one-factor model in Table 1 and Figure 1. Loadings ≥ .30 in bold. Values before the slash are based on self-report data; values after slash are based on informant data. Factor1 = Social Inhibition; Factor2 = Emotion Dysregulation; Factor3 = Paranoid Withdrawal; Factor4 = Exhibitionism.

Table 4.

Regression Coefficients of the Four PD Symptom Dimensions in Predicting Life Outcomes

	Self-Rated Outcomes						Informant Rated Outcomes
	Depression		Dyadic Adjustment		General health		Social Adjustment		Dyadic Adjustment		General health
Predictor	b (SE)	Std. b	b (SE)	Std. b	b (SE)	Std. b	b (SE)	Std. b	b (SE)	Std. b	b (SE)	Std. b
Self-Report
Social Inhibition	1.91 (.14)	.28***	−.23 (.11)	−.06*	−2.13 (.24)	−.19***	−.02 (.02)	−.02	.07 (.14)	.02	−.08 (.02)	−.10***
Emotion Dysregulation	2.85 (.16)	.37***	−.81 (.13)	−.20***	−4.62 (.27)	−.37***	−.14 (.02)	−.20***	−.47 (.16)	−.12**	−.26 (.02)	−.30***
Paranoid Withdrawal	1.67 (.14)	.25***	−.62 (.11)	−.17***	−2.76 (.24)	−.25***	−.12 (.02)	−.20***	−.51 (.14)	−.14***	−.10 (.02)	−.13***
Exhibitionism	−.90 (.14)	−.13***	.29 (.11)	.08*	1.83 (.24)	.17***	.10 (.02)	.17***	.04 (.15)	.01	.13 (.02)	.17***
Adjusted R2	.31		.08		.27		.10		.03		.14
Informant Report
Social Inhibition	1.07 (.18)	.16***	−.26 (.12)	−.07*	−1.97 (.29)	−.18***	−.15 (.02)	−.25***	−.73 (.12)	−.20***	−.31 (.02)	−.39***
Emotion Dysregulation	1.02 (.18)	.14***	−.70 (.13)	−.19***	−1.38 (.30)	−.12***	−.24 (.02)	−.38***	−1.82 (.13)	−.48***	−.27 (.02)	−.34***
Paranoid Withdrawal	.43 (.18)	.06*	−.30 (.13)	−.08*	−.76 (.30)	−.07*	−.09 (.02)	−.15***	−.58 (.13)	−.15***	−.07 (.02)	−.08**
Exhibitionism	−.76 (.18)	−.11***	.15 (.13)	.04	.83 (.29)	.07**	.05 (.02)	.08**	.23 (.13)	.06	.07 (.02)	.08**
Adjusted R2	.06		.05		.06		.25		.31		.30

Note. Ns for self-reported depressive symptoms, dyadic adjustment, and general health were 1,594, 939, and 1,577 for self-rated predictors and 1,444, 872, and 1,427 for informant rated predictors. Ns for informant reported social adjustment, dyadic adjustment, and general health were 963, 627, and 1,197 for self-rated predictors and 966, 630, and 1,200 for informant rated predictors. b = unstandardized coefficient; Std. b = standardized coefficient; SE = standard error.

^***p < .001,

^**p < .01,

^*p < .05; two-tailed.

Discussion

The interlocking goals of the current study were to conduct a symptom-level factor analysis of DSM-IV/DSM-5 PDs and to validate the resulting solutions using important life outcomes as assessed by the target and an informant. Our results generally fit well within the context of similar structural models of normal personality and psychopathology, with some unique differences that may be due to the age of our sample. We discuss our findings with regard to each goal in turn.

Factor Modeling of PD Symptoms

The questionnaire items analyzed in the present study were based on the diagnostic criteria for each PD listed in DSM-IV, representing decades of clinical experience and understanding regarding the nature of PD symptomatology. Our results indicated that PD symptoms are hierarchically organized with a general factor of maladjustment at the top, two broad dimensions of internalizing and externalizing pathology directly below, and increasingly more specific symptom dimensions toward the bottom. With a few exceptions, factors at each level of the hierarchy were similar across self-and informant ratings. When a single factor was fit to the data, all but four items in the self-report data and six in the informant data had positive loadings greater than .30. This is consistent with other studies that have identified a general factor underlying personality pathology (Sharp et al., 2015; Wright, Hopwood, Skodol, & Morey, 2016). The strongest loading items appear to broadly reflect a distress or severity factor. Sharp et al. (2015) speculated that the nature of this factor may involve features closely related to borderline PD. While borderline features were indeed among the strongest loading items, the composition of this factor in our sample was very broad with a majority of strongly loading items representing symptoms of avoidant and dependent PDs. This finding fits more closely with the suggestion that the general PD factor reflects maladaptivity rather than BPD more specifically (Oltmanns, Smith, Oltmanns, & Widiger, in press). In other words, people who endorse these items are experiencing impairment or dysfunction in their lives, and those problems may stem from any of several more specific deficits (e.g., in emotional regulation, interpersonal functioning, etc.).

The internalizing and externalizing factors identified in the two-factor model of our data are consistent with higher-order models of psychopathology (e.g., Caspi et al., 2014; Krueger, Caspi, Moffitt, & Silva, 1998). The internalizing dimension consists of mood and anxiety disorders (e.g., major depression, generalized anxiety, panic disorder), whereas the externalizing dimension represents antagonistic and impulsivity-related problems (e.g., substance use, conduct, and oppositional defiant disorders). When PDs are also included in factor analyses, avoidant, borderline, dependent, paranoid, and obsessive-compulsive PDs evidence strong loadings on an internalizing factor; antisocial and, sometimes, borderline PDs load strongly on an externalizing factor (e.g., Kotov et al., 2011; Røysamb et al., 2011). The current study broadens the externalizing dimension by including histrionic and narcissistic traits, consistent with other symptom-level analyses that identified qualities such as pathological jealousy, egocentricity, and vain and demanding behaviors (Nestadt et al., 1994).

Our three-factor solution departs from Big Three models of personality (e.g., Eysenck, Eysenck, & Barrett, 1985; Tellegen & Waller, 2008) by the lack of a unique disinhibition or psychoticism dimension. Rather, a broad factor we labeled Paranoid Withdrawal, which split away from the Internalizing and Externalizing factors, includes symptoms of paranoid, schizoid, and schizotypal PDs. We relabeled Internalizing as Negative Affectivity and Externalizing as Exhibitionism in order capture the affective instability and histrionic/narcissistic traits that mainly defined each respective factor. The composition of Paranoid Withdrawal agrees well with previous research that identified a factor consisting of Cluster A PDs (e.g., Fossati et al., 2000; O’Connor, 2005). Interestingly, Markon (2010) reported a strong association between “pathological introversion” and “thought disorder” factors, the latter of which included paranoia, eccentricity, and schizoid characteristics. Similarly, Kotov et al. (2011) found that a thought disorder dimension (which comprised symptoms of mania and psychosis in addition to Cluster A PDs) separates from the internalizing and externalizing spectra. Although symptoms of disordered thinking are not exhaustively sampled in the current PD criterion sets, our results indicate that the aberrant cognitions and eccentric behaviors of schizotypal PD covary with symptoms of paranoia and social withdrawal at a very broad level of analysis.

At the fourth level of the hierarchy, a cohesive Social Inhibition dimension, defined by avoidant and dependent PD symptoms, separated from Negative Affectivity and remained virtually unchanged when additional factors were extracted (cf. Fossati et al., 2006; Mulder & Joyce, 1997). Exhibitionism also remained intact at the fourth and succeeding levels.

At lower-order levels, Paranoid Mistrust and Detachment split from Paranoid Withdrawal, and a circumscribed Angry Hostility factor split from Emotion Dysregulation. Features of schizotypal and obsessive-compulsive PDs formed narrowly-defined factors in the self-report data. Interestingly, Dependency was a somewhat larger factor (as defined by the number of indicators) in the informant rather than self-report data at the lowest-order level (i.e., level 8).⁷ This suggests that, in comparison to the target persons themselves, informants were more likely to perceive and rate similarly dependency-related behaviors. It would be worthwhile to investigate the potential mechanisms associated with this discrepancy, if it is shown to replicate. It may be the case that dependency has a greater impact on persons who are responsible for meeting the person’s needs than on the person who is viewed as being dependent.

Predictive Validity of the PD Symptom Hierarchy

Analyses of life outcome data suggested that four dimensions were needed to maximize the predictive validity of PD symptoms. The four-factor model significantly contributed to the prediction of all outcomes over simpler models, while increasingly complex models improved predictive power only modestly. The four dimensions were Social Inhibition, Emotion Dysregulation, Paranoid Withdrawal, and Exhibitionism. Of these four dimensions, Emotion Dysregulation was the strongest predictor of the outcomes we chose to examine, regardless of who (i.e., self or informant) generated the ratings. This finding is consistent with numerous studies that have documented the social and personal costs of Neuroticism as rated by self and informants (e.g., Balsis, Cooper, & Oltmanns, 2015; Connelly & Ones, 2010). On the other hand, Exhibitionism had the smallest effects, and they were in the opposite direction of the other three dimensions. Although this finding seems somewhat counterintuitive, it agrees well with other research that identified an adaptive or high-functioning subtype of narcissistic PD (e.g., Raskin & Terry, 1988; Russ, Shedler, Bradley, & Westen, 2008) as well as a previous report that informant ratings of histrionic PD features were associated with better self-reported social functioning (Oltmanns, Melley, & Turkheimer, 2002). For example, Russ et al. (2008) reported that patients in an exhibitionistic subtype (defined by an exaggerated sense of self-importance, but also by energetic, outgoing, attention-seeking, and competitive behaviors) had higher adaptive functioning and more stable employment than those in other narcissistic subgroups. While the magnitude of associations in our study was quite small, individuals with high scores on Exhibitionism did tend to score in the direction of better adjustment and health as compared to those with low scores.

Although the four-factor model demonstrated the greatest incremental validity, both lower-and higher-order levels added significant predictive variance over preceding levels. With regard to self-reported PD symptoms, the construct validity of the two-and three-factor models was generally supported. In the two-factor model, Internalizing (but not Externalizing) was a moderate to strong predictor of depressive symptoms, relationship satisfaction, and general health. Paranoid Withdrawal, in the three-factor model, also emerged as a significant predictor of these outcomes. However, when this factor split into Paranoid Mistrust and Detachment, only the former maintained significant relations with all outcomes, suggesting that the withdrawal component (representing mostly schizoid characteristics) had negligible impact on life outcomes (see also Soeteman, Hakkaart-van Roijen, Verheul, & Busschbach, 2008). Lower-order models inconsistently contributed additional explained variance, but factors within these models were significant predictors in some cases. For example, in the eight-factor model, Peculiarity and Dependency were predictive of depressive symptoms and poor health. Interestingly, Rigidity had a protective effect on these outcomes. This result is similar to other studies that found less functional impairment and greater work success in patients diagnosed with obsessive-compulsive PD than other types of PD (e.g., Skodol et al., 2002; Ullrich, Farrington, & Coid, 2007).

With regard to informant rated PD symptoms, the more complex six-, seven-, and eight-factor models helped improve the predictive utility of PD symptoms. Patterns of associations with external criteria were generally similar to those based on self-report data, but with at least two notable exceptions. First, whereas self-reported Detachment was minimally predictive of life outcomes, informant ratings had stronger effects. This is particularly evident in how the informants rated the social adjustment of the target and their satisfaction with regard to being in a relationship with that person. This result suggests that informants recognize certain qualities or impairments that the targets themselves may either be unaware of or do not consider problematic. This potential lack of insight may be especially harmful to those (informants) who are in close, intimate relationships with detached partners (targets), who may be unaware of their own issues but also less able to identify the negative impact that they are having on their significant other. Second, and relatedly, whereas self-reported PD symptoms had little effect on how they rated the quality of their intimate relationship, how their partners rated them did have a strong effect on how the partners rated their own satisfaction in the relationship. For example, spouse ratings of their partner’s personality (i.e., the target) explained nearly 40% of their satisfaction in their relationship, whereas self-ratings explained approximately 7% of their partner’s ratings (and 12% of their own ratings). This finding indicates that it is not just how the target sees him or herself, but how the target’s partner sees him or her that is consequential to their feelings about and adjustment to the relationship (see also Brock, Dindo, Simms, & Clark, 2016).

Disinhibition and Later Life

Absent from our hierarchical model of PD symptoms is a distinct, separable factor reflecting disinhibited, impulsive, and antisocial characteristics. While it is possible that symptoms or traits related to disinhibition may be underrepresented in the PD diagnostic criteria, some previous studies have reported a unique antisocial factor (e.g., Sharp et al., 2015). Notably, Thomas et al. (2003), using an earlier version of the MAPP, did find a distinct antisocial factor in samples of young adults (military recruits and first-year college students).⁸ Therefore, it seems unlikely that the number of disinhibition items in the MAPP is responsible for this result.

It is probably most important to consider the older age of the current sample when trying to understand the absence of a disinhibition factor. Antisocial PD is less common in older age groups (Balsis, Zweig, & Molinari, 2015), and some of the criteria may be less relevant for older adults. Balsis, Gleason, Woods, and Oltmanns (2007) used item response theory to examine the item-level functioning of DSM-IV criteria across younger (18–34) and older (65–98) age groups. Four antisocial criteria were analyzed, and two of them showed measurement bias. The item “failure to conform to norms regarding legal behavior” had a significantly higher difficulty parameter in the older age group, indicating that older adults (with similar levels of pathology relative to younger adults) are less likely to endorse this criterion. Interestingly, the criterion “deceitfulness, as indicated by repeated lying, use of aliases, or conning others for personal profit or pleasure” had a lower difficulty parameter, suggesting that older adults may use this strategy more readily than younger adults. Consistent with these findings, the item “I repeatedly get in trouble with the police” had a very low endorsement rate (M = .03, SD = .26) and “I can be deceitful when I need to be” was relatively higher (M = .57, SD = .77). The failure of a disinhibition factor to emerge in our analyses indicates that these items do not appear to define a distinct factor that is separable from other forms of personality pathology in later life.

To our knowledge, this study is the only investigation that has examined the factor structure of PD symptoms in a representative sample of older adults. Most other similar studies have been based either on samples of convenience (e.g., college students) or patient groups being treated for various kinds of mental health problems. The upper age limit in these studies rarely exceeds middle to late 40s. The lack of PD research in older age groups leaves an open question of whether PD symptomatology, or more specifically, groupings of PD symptoms, change in middle or later life. Some features of disinhibited-related disorders manifest differently in later life and in ways that are not fully captured by the existing diagnostic criteria -a phenomenon that has been described as heterotypic continuity. For example, antisocial older adults may be more willing to conform their behavior to societal pressures than younger adults, but they might express their ambivalence toward rules and expectations at an interpersonal-level, such as through manipulativeness and deceitfulness. They may no longer act physically aggressive and intimidating toward others, but rely on other covert, nonaggressive methods for getting what they want. The changing face of antisocial pathology in later life is an intriguing area for future research to address.

Limitations and Future Directions

Despite minor differences across two information sources, these initial results support the idea of a common hierarchical structure underlying PD symptoms by showing how alternative factor solutions can be fit within a unified framework. The inclusion of both self-and informant ratings is a notable strength of this study. However, our reliance on a single measure of PD (the MAPP) is an important limitation that should be recognized. Whether these results hold across alternative measures is an empirical question worthy of addressing. Future studies that consider different measures and assessment methods will help further elucidate the structure of personality pathology and inform the generalizability of the present study’s findings.

The restricted age range of our sample can be viewed as both a limitation and a strength. A preponderance of the factor modeling studies of personality disorders and personality disorder symptoms has relied on younger samples (cf. South & Jarnecke, 2017). Thus, understanding the factor structure of PD symptoms in an older sample is novel and important. Because personality traits change across the lifespan, it seems reasonable to expect that the expression of personality pathology will also vary with time. This may be especially true with regard to externalizing factors. Yet, what we do not know is, in the same sample and cohort of individuals, if the structure of PD symptoms would change over time. This type of intensive longitudinal design would be ideal but difficult to implement. Future studies that include both younger and older adults would be one way to determine if the empirically derived factors found here are robust across age groups.

A thorough exploration of real life outcomes and the validity of PD factors will obviously require consideration of a much wider range of variables than we were able to explore in this paper. The analyses reported here follow examples from previous studies (e.g., Kotov et al., 2016) that have suggested how this process may continue to unfold. Our findings are offered as tentative rather than definitive solutions to important questions being asked regarding the structure of personality pathology. It will be critically important for future studies to include measures that cut across many levels of functioning, addressing the widest possible range of outcomes that are known to be influenced by personality, i.e., from social and occupational functioning to physical health and mortality (Ozer & Benet-Martinez, 2006). In each domain, investigators should employ informant rated outcomes as well as self-report measures of adjustment and performance. Biological indices of health, such as inflammation markers regarding immune system functioning and abnormalities in neurotransmission, will also be invaluable in this search (e.g., Beauchaine, Klein, Crowell, Derbridge, & Gatzke-Kopp, 2009; Dilorio et al., under review). The work will require replication of findings in large samples of participants, using both representative community samples as well as clinical patients of all ages and backgrounds.

Conclusion

The item-level analyses reported here support a hierarchical representation of personality pathology. By detailing the structure of PD symptoms at different levels of analysis, seemingly dissimilar results across studies can be accommodated within a unified model. Although factors at each of eight hierarchical levels were interpretable and robust across self-and informant ratings, the four-factor model may be fundamental to understanding outcomes associated with PD. Our results did not support the structure of the DSM PD categories or the DSM PD cluster system. Rather, they corresponded more closely to the trait dimensions in the DSM-5 and ICD-11 models of personality, with the exception of a disinhibition factor. The absence of this factor might reflect a lack of relevant diagnostic criteria for antisocial PD in later life as well as the possibility that its expression changes in older adults. Another difference between the models is the presence of a strong Exhibitionism dimension in the diagnostic criteria (reflecting intense attachments, overly dramatic displays of emotion, and need for admiration), which appears to have no clear equivalent in either the DSM-5 or ICD-11 trait models. Although the DSM-5 model does include subordinate traits within the antagonism domain that capture features of histrionic (i.e., attention seeking) and narcissistic (i.e., grandiosity) PDs, it will be important for future research studies to address the clinical impact these traits have on personality functioning and how these effects might compare to other antagonistic personality features. As more attention is given to hierarchical representations of PD symptoms and traits, conceptual relations between lower-and higher-order constructs will become clearer and lead to a better understanding of how each level of the hierarchy can explain different forms of psychopathology and related outcomes.