Common and Unique Etiological Influences on Personality and Personality Disorders in Youth

Abstract

Background

Personality and personality disorders (PDs) are closely related constructs, but the nature of their relationship remains uncertain. Dimensional models suggest PDs are extreme manifestations of personality traits. We investigated the etiological links between broad personality factors and dimensional PD factors using quantitative analyses of common and unique genetic and environmental etiological factors affecting both normative and deviant personalities, emphasizing methods to minimize measurement bias.

Methods

Seven national twin cohorts participated (N = 2,353 twins from 1,188 families), assessing personality traits with questionnaires across three adolescent waves and PD traits via age-19 clinical interviews. Exploratory factor analyses of personality traits and the three-cluster ABC factors of PDs simplified the factor structure of the measures. Biometric twin analyses with Cholesky modeling identified genetic and environmental links among factors.

Results

Factor analysis of personality traits yielded non-resilient, disinhibited, and disaffected factors. PDs fit an emotional/erratic, anxious/fearful, and odd/eccentric factor structure. Phenotypic correlations varied, but the disinhibited factor had minimal relations to specific PD factors. Crucially, approximately one-third of the genetic variance was shared between personality and PDs. However, the majority of genetic variance and all non-shared environmental variance were trait-specific, suggesting predominantly distinct etiological pathways.

Conclusion

The significant genetic overlap supports the continuum model; results were bolstered by using broad factors and method triangulation. However, unique genetic and environmental influences point to distinct etiological pathways differentiating personality and PDs. These findings endorse the continuum model yet highlight the need for further research to grasp the complex relationships between personality and PDs.

Introduction

Personality and personality disorders (PD), although based within separate research traditions, are closely related constructs. Personality, understood as normative personality functioning, consists of stable patterns of thoughts, feelings, and behaviors that define individual uniqueness and adaptability. In contrast, PDs are conceptualized within psychiatric diagnostic systems as inflexible, enduring patterns of personality deviating from cultural norms, leading to significant distress or impairment. Recent advancements in psychopathology research increasingly favor dimensional over diagnostic/categorical approaches for understanding both symptom and PDs, promoting an integrated view of mental health [1–4]. The continuum model, which is rooted in the dimensional framework [5], posits that normative personality traits and PDs exist along a single spectrum [6]. In this model, typical broad traits, such as those outlined by the five-factor model, represent varying expressions of personality with PDs manifesting as extreme or maladaptive versions [7, 8]. This framework has implications for understanding the structural links between personality traits and disorders, as well as practical consequences; if they share common etiological pathways, changes in one may influence the other, offering valuable insights for prevention and intervention strategies [9, 10].

Evidence of associations in support of a continuum model of personality and PDs is mounting, as summarized in several recent reviews [11–16]. A recent metanalytic study of studies employing dimensional as well as categorical (taxonic) approaches to personality and psychopathology found very weak support for latent categorical structures in any of the included dimensions [17]. In fact, there was less support for categorical structure in PD than in any other psychopathology dimensions. However, although factor analytic approaches can map structural associations, they do not uncover etiological links between factors [5]. To further understand the etiological nature behind the phenotypical associations between personality and PDs, genetically informed designs are essential [18].

Personality traits and PDs are both strongly genetically influenced. Twin, family, and adoption studies have consistently demonstrated heritability estimates typically within the range from 40% to 60% for traits as defined by models like the five-factor model [19, 20]. The genetic influences on measures of general adaptive personality functioning, such as wellbeing [21], self-control [22], ego resilience [23], self-efficacy [24], and sense of coherence (SOC) [25], also yield moderate-to-high heritability estimates.

Likewise, genetically informative studies on PD show moderate-to-high heritability. Torgersen et al. [26] estimated heritability at 60% for PDs overall, with heritability estimates for specific PD within the range of 54–79% for narcissistic, borderline, histrionic, schizotypal, dependent, and self-defeating PD, and lowest at around 28–29% for schizoid, paranoid, and avoidant disorders. In a large register-based family study, Skoglund et al. [27] estimated the heritability of borderline PD of 46%, with the remaining variance due to non-shared environmental influences. The heritabilities of parent-evaluated PD in children and adolescents have been reported to be within the same range as those found in adult samples [28].

While genetic influences play a central role in the variance observed in personality traits and disorders, few studies have explored the common and unique genetic and environmental influences between these constructs, which are essential for understanding and acting on the implications of a continuity model. One reason for this scarcity of studies may be the challenges of obtaining statistically powered sample sizes and using clinically valid assessments of PDs in biometrical studies, which often rely on community-based samples [18]. Moreover, existing evidence regarding the extent of common etiology remains inconsistent. A recent twin study found considerable genetic overlap between the personality traits neuroticism and extraversion and avoidant PD [29]. High-to-moderate genetic correlations were found between borderline PD and the personality traits of neuroticism (positive), and conscientiousness and agreeableness (both negative) [30]. SOC was highly correlated with borderline PD in a study by [31]. Wright et al. [32] found that both maladaptive and normative traits showed similar heritability levels, with traits that correlated phenotypically also correlated genotypically, highlighting overlapping genetic components. However, there were also genetic and environmental contributors unique to maladaptive traits, distinct from normative traits.

As earlier noted, mixed results in psychiatric research may originate from diagnostic systems grounded in categorical clinical agreement on descriptive criteria. The persistent comorbidity observed across individual PDs [33–35] highlights the benefits of studying broader PD factors within a dimensional approach, which facilitates exploring empirically linked symptom clusters. Genetically informative studies have shown that phenotypic associations often mirror the underlying genetic structure of these relationships, revealing significant genetic comorbidity across individual PDs [36]. To further advance the etiological study of personality and PDs, research that focuses on empirically developed broader factors rather than isolated clinical categories is needed. Similarly, personality traits display both phenotypic and genetic correlations, with genetic influences accounting for covariance between these traits [37–39]. There is growing evidence of phenotypic comorbidity influenced by genetic pleiotropic effects that PDs can be organized in hierarchical structures alongside – or within – the general psychopathology structures of HiTOP. Livesley and Jong [40] found four general factors, while later developments have empirical data confirming a latent three-factor A-B-C cluster organization as well as one general PD factor [41]. PDs seem to be structured in the same hierarchical way as general psychopathology [42].

In a large-scale twin study, Czajkowski et al. [43] found that 40–60% of genetic influences on individual PDs, and all non-shared environmental influences, were unique to each PD. Further analyses by Kendler et al. [44] using the same dataset examined the genetic and environmental architecture of normative personality, abnormal personality traits, and PD symptoms. Their findings revealed a complex genetic and environmental etiological pattern. Their best-fitting model had three genetic factors and three unique environmental factors, of which two genetic and one environmental common factor were consistent with a continuity model while one genetic and two environmental factors were more supportive of a separate etiology model behind personality and PD. Finding mixed support for the continuum model of personality and PD relationships, the authors consequently advised caution regarding the adoption of a straightforward continuum model for adaptive and pathological personality.

Thus, the etiological evidence for placing personality and PDs on a single spectrum remains inconclusive, and more genetically informed studies are essential to clarify this relationship. Given the uncertainties in the field and the reliance on behavior measures – some of which are evolving with ongoing changes in perspectives within psychopathology – it is essential for these investigations to critically evaluate their methodological choices. To achieve robust and unbiased knowledge, it is likewise important to integrate diverse research methodologies [6].

This twin study aimed to explore the fundamental etiological connections between broad personality traits and the dimensional factors of PDs. Diverse methodological strategies were utilized to minimize common method bias and strengthen the identification of genuine associations. To facilitate method triangulation, personality was assessed through youth self-reports, while PDs were evaluated using clinical interviews. Additionally, personality was measured using a range of instruments drawn from diverse theoretical and conceptual models, whereas personality pathology was examined according to conventional clinical criteria for PDs. Exploratory factor analyses were conducted independently for the 14 personality measures and for 10 dimensional measures of PDs. Personality measurements were aggregated across three waves to enhance assessment validity. This approach minimizes random variations and transient influences, offering a more accurate representation of core traits. Aggregated measures, drawing on shared variance, are more reliable than time-specific measurements affected by temporary conditions, ensuring our findings reflect the stable nature of personality central to our research aims. By minimizing common method variance, the strength of observed associations may be reduced, but this approach improves the accuracy of interpretations by reflecting true relationships rather than measurement artifacts.

Based on the above methodological considerations and judging from results from earlier behavior genetic research, the main hypothesis was that the common genetic influences would be in the range of moderate to small. It was expected that most genetic and all environmental influences would be specific to each of the two constructs.

Methods

Sample and Procedure

Data come from the Oslo University Adolescent and Young Adult Twin Project [45, 46]. Twins were identified through the Norwegian Medical Birth Registry, and all twins born in Norway between 1988 and 1994 were invited to participate. A total of 4,668 twin pairs were alive and residing in Norway at the time of recruitment. Of these, 2,486 families (53.3%) had at least one consenting parent and one consenting twin. Full family consent – from both parents and both twins – was obtained in 1,976 families. Incomplete consent was primarily due to paternal death (98 families), maternal death (3 families), parental bereavement (1 family), paternal refusal (360 families), maternal refusal (40 families), or twin refusal (8 families). For a full overview of sample size by birth cohort, see online supplementary Table S1 (for all online suppl. material, see https://doi.org/10.1159/psp/aelag006). The project includes three waves of questionnaire data collected throughout adolescence and young adulthood, with 2 years between measurements (12–18 years at wave-1), and data from a face-to-face interview when the twins were around 19 years old. American Psychological Association ethical principles were followed in the conduct of the study. Attrition analyses indicated that female gender, but not zygosity, was associated with continued participation across waves. Personality characteristics reflecting higher conscientiousness, openness to experience, resilience, and structured study habits also predicted retention. To assess whether attrition biased heritability estimates, we compared monozygotic (MZ)/dizygotic (DZ) twin correlation ratios across participants completing one, two, or three waves. Ratios were highly similar across groups (MZ/DZ: 0.55/0.20, 0.59/0.12, and 0.59/0.18, respectively), suggesting that attrition did not meaningfully influence heritability estimates [45].

All twins who had participated in the interview and responded to at least one questionnaire wave were included in the analyses, resulting in an overall sample size of 2,353 twins (59% females) from 1,188 families, including 911 MZ and 1,442 DZ twins. The number of incomplete twin pairs was low (i.e., 1,165 complete and 23 incomplete). Of the 2,353 participants, 989 twins from 568 families participated in all three questionnaire waves and the interview, including 379 MZ and 610 DZ twins. Online supplementary Table S2 shows the total number and percentage of wave participation. The mean ages at wave-1, wave-2, and wave-3 were 15.3 (SD = 2.0), 17.0 (SD = 2.0), and 19.7 (SD = 2.0), respectively. The mean age at interview was 19.0 (SD = 1.1).

Zygosity was determined by a combination of a zygosity questionnaire [47] and genotyping, and the misclassification rate has been estimated to 0.45%. A detailed explanation of the procedure is provided in Skaug et al. [48].

Measures

Questionnaire Data

To avoid overly lengthy questionnaires and minimize dropouts and missing data, all scales in the questionnaires were abbreviated based on results from a pilot study [45]. Items with the highest item-to-trait correlations across different age groups and sexes were selected for inclusion in the study. The questionnaire scales are presented in online supplementary material S1. Cronbach’s alpha coefficients and test-retest correlations for all scales are presented in online supplementary Tables S3 and S4. Overall, the scales demonstrated good to acceptable internal consistency across all three waves (α = 0.70–0.86), whereas conduct problems and delinquency scales were more moderate (α = 0.33 and 0.67). Test-retest correlations across the three waves indicated substantial temporal stability for most constructs. Over the 2-year intervals (wave-1-Wave-2 and wave-2-wave-3), correlations were generally in the moderate-to-high range. The Big Five traits showed the highest stability, with 2-year correlations ranging from 0.61 to 0.76 and 4-year correlations from 0.55 to 0.61. The other constructs also showed meaningful stability, with 2-year correlations mostly between 0.50 and 0.68 and 4-year correlations typically between 0.37 and 0.55.

Big Five Personality. The Big Five personality traits were measured using the Hierarchical Personality Inventory for Children (HiPIC) [49]. The HiPIC assesses the domains of emotional stability (reversed neuroticism), extraversion, imagination (corresponding to openness), benevolence (corresponding to agreeableness), and conscientiousness. For this study, the full HiPIC (144 items) was abbreviated to a 40-item scale, with eight items per trait. Participants responded on a five-point scale ranging from 0 (not typical) to 4 (very typical), and mean scores for each of the five personality traits were calculated.

Self-Efficacy. Self-efficacy over the past year was assessed using a nine-item abbreviated version of the Children’s Perceived Self-Efficacy (CPSE) scales [50]. The scale included four items measuring perceived academic self-efficacy (e.g., finishing homework by deadlines, motivation to do schoolwork) and five items measuring perceived social self-efficacy (e.g., ability to make and keep friends, handling interpersonal conflicts). Participants responded on a five-point scale ranging from 0 (not easy at all) to 4 (very easy), and mean scores for the two self-efficacy domains were calculated.

Resilience. Resilience was measured using two scales: the Resilience Scale (RS) [51] and the Ego-Resiliency Scale (ER89) [52]. Each scale was abbreviated to include five items. The RS items assess the level of commitment to completing plans, interest in various subjects, determination, and self-belief. The ER89 items evaluate the extent to which a person finds joy in dealing with new situations, fills everyday life with interesting activities, and describes themselves as having a “strong” personality. Participants rated each item on a five-point scale ranging from 0 (not typical) to 4 (very typical), and mean scores were computed for each scale.

Loneliness. Feelings of loneliness over the past year were assessed using a five-item scale, which included the four-item survey version of the Revised University of California, Los Angeles (R-UCLA) Loneliness Scale [53] and one direct measure of loneliness (“I feel lonely”). Participants rated each item on a five-point scale, with responses ranging from 0 (not typical) to 4 (very typical). Mean scores were calculated, with higher scores indicating greater feelings of loneliness.

Sense of Coherence. SOC over the past year was measured using a shortened five-item version of the Sense of Coherence 13-item scale (SOC-13) [54]. The scale included items that assess the extent to which a person perceives oneself and the world as understandable, manageable, and meaningful. Participants responded on a seven-point scale ranging from 1 (very often) to 7 (rarely/never). Mean scores were calculated, with higher scores indicating stronger SOC.

Impulsivity. Impulsivity over the past year was assessed using seven items from the Control Scale (reversed impulsivity) within the brief form of the Multidimensional Personality Questionnaire (MPQ-BF) [55]. The scale measures tendencies to act spontaneously, the propensity to plan and organize, and the likelihood of making decisions through careful reasoning. Participants responded to each item on a three-point scale, ranging from 0 (false) to 2 (true). Mean scores were calculated, with higher scores indicating greater levels of impulsivity.

Conduct Problems. Conduct problems over the past year were assessed using the Conduct Problems Scale, a subscale in the Strengths and Difficulties Questionnaire (SDQ) [56, 57]. The scale comprises five items that address behavior related to temper, fights, lying, cheating, and stealing. Participants responded to each item on a three-point scale, ranging from 0 (not true) to 2 (certainly true). Mean scores were calculated, with higher scores indicating a greater frequency of conduct problems.

Delinquency. Delinquency over the past year was measured by asking participants about nine delinquent behaviors [58, 59], such as engaging in physical fights, stealing, carrying weapons, vandalism, and other rule breaking behaviors like staying out late at night when expected to be at home. Participants rated their responses on a four-point scale ranging from 0 (never) to 3 (very often). Mean scores were calculated, with higher scores indicating a higher frequency of delinquent behaviors.

Interview Data: PDs

PDs were assessed using a Norwegian version of the Structured Interview for DSM-IV Personality (SIDP-IV) [60]. The ratings in SIDP-IV are based on behaviors, cognitions, and feelings typical over the past 5 years, with each DSM-IV criterion being scored on a scale from zero to three: 0 = absent; 1 = subthreshold; 2 = present; 3 = strongly present. We adopted a dimensional approach, in which we analyzed PDs defined as counts of the number of endorsed criteria scored at the subclinical level or higher (i.e., SIDP-IV criteria scored 1 or higher). For ease of expression, we refer to these measures as PDs rather than the more cumbersome but accurate term dimensional representations of PD traits.

Inter-rater reliability was assessed based on two raters’ scoring of 55 audiotaped interviews. The intraclass correlation coefficient for the number of endorsed criteria scored ≥1 was as follows: paranoid PD 0.78; schizoid PD 0.44; schizotypal PD 0.76; histrionic PD 0.69; borderline PD 0.77; antisocial PD 0.83; narcissistic PD 0.56; avoidant PD 0.86; dependent PD 0.75; obsessive PD 0.75.

Statistical Analyses

Factor Analyses

All factor analyses were performed using the fa function from the psych package in R. For models with more than one factor, varimax rotation was applied to achieve a simpler and more interpretable factor structure. First, we conducted a series of factor analyses with one factor to create aggregated measures of each personality trait across the measurement waves. Missing questionnaire data were imputed using multiple imputation by fully conditional specification [61]. The mean scores based on 10 iterations were computed and then used as input variables in the factor analyses. Specifically, 14 separate factor analyses were performed, with the personality variables measured at the three measurement waves serving as input variables in each analysis (e.g., the factor analysis of neuroticism included neuroticism measured at wave-1, wave-2, and wave-3). These factor analyses allowed us to summarize multiple time point measurements into single factor scores. Because personality traits are, by definition, relatively stable over time, aggregated scores that capture the common variance shared by the measurements are considered more reliable and robust than any individual measurement at a single time point. Using aggregated scores eliminates the influence of random errors and transient conditions that might affect the personality measures at specific time points, thereby providing a more stable and consistent representation of the underlying personality trait. Next, we performed a factor analysis on the aggregated personality scores to identify higher order dimensions underlying the personality traits. Factors were extracted using minimum residual factor analysis with varimax rotation.

Finally, we conducted a three-factor analysis of the PDs. According to the Diagnostic and Statistical Manual of Mental Disorders (DSM), PDs are classified into three factors based on symptom similarities: the odd/eccentric factor A PDs (paranoid, schizoid, and schizotypal), the dramatic/emotional/erratic factor B PDs (antisocial, borderline, histrionic, and narcissistic), and the anxious/fearful factor C PDs (avoidant, dependent, obsessive-compulsive). Accordingly, we hypothesized that the PDs would group into three factors, mirroring the DSM cluster classification.

We were interested in examining how personality dimensions relate to PD dimensions. Therefore, we used the factor scores from the factor analyses mentioned above in subsequent analyses.

Associations between PDs and PD Dimensions

First, phenotypic associations between personality dimensions and PD dimensions were examined using correlation analyses. Next, the extent of shared and specific genetic and environmental variance underlying personality dimensions and PD dimensions was examined using a series of multivariate twin models. Twins reared together share the same family environment, MZ twins are genetically identical, whereas DZ twins share, on average, half of their segregating genes. Using data from twins allows us to partition the variance of an observed phenotype and the covariance between phenotypes into three sources of variance/covariance. Additive genetic influences (A; the effect of genes that together operate in an additive manner) are inferred by the extent to which the correlation between MZ twins is greater than the correlation between DZ twins. If influence of A is the only source to familial resemblance, the MZ correlation is expected to be twice as large as the DZ correlation. Shared environmental influences (C; environmental factors contributing to phenotypic similarity among family members) are inferred by the extent to which the DZ correlation is greater than half the size the MZ correlation. Any remaining variance/covariance not accounted for by A or C is attributed to non-shared environmental influences (E; any factors contributing to phenotypic dissimilarity among family members, including measurement error).

A series of four-variate Cholesky twin decompositions were fitted to determine specific and shared genetic variance underlying the personality dimensions and the PD dimensions. To examine the variance in a given PD dimension, we specified a model in which the three personality dimensions were entered as the first three variables, and PD dimension as the fourth variable. This allowed us to decompose the variance in the PD dimension into components shared with the personality dimensions and components unique to that PD dimension. Analogously, to examine the variance in a given personality dimension, we specified models in which the three PD dimensions were entered as the first three variables and the personality dimension as the fourth variable. This approach resulted in six separate four-variate models. Analyzing each outcome dimension in separate models, rather than combining all variables in a single six-variate model, was motivated by the factor structure of both the personality and PD dimensions. Both sets of dimensions were derived using varimax rotation, which by construction yields approximately orthogonal factors with minimal intercorrelations. Although a six-variate Cholesky could in principle capture any residual covariance among the dimensions, such covariance is expected to be minimal given the orthogonal factor structure produced by varimax rotation. However, combining all dimensions in a single model would conflate variance shared among the predictor dimensions with variance shared between predictors and the outcome, complicating interpretation for our specific research questions.

The Cholesky decomposition is among the most widely used twin analysis. This model provides a robust approach to estimate genetic and environmental sources of variance and covariance, with minimal theoretical assumptions [62]. In this model, one set of latent factors (A, C, and E) is specified for each of the variables, with the first set of factors loading on all measured variables, the second set loading on all variables except the first, and so forth. In this way, the final set of latent factors loads only on the last variable, accounting for the residual variance in the last variable not explained by influences from the preceding variables. Of note, because random measurement error is assumed to be uncorrelated across individuals and traits, it is absorbed into the trait-specific non-shared environmental (E) component in the multivariate classical twin/Cholesky framework [63]. For each personality and PD dimension, we report the total genetic variance, the proportion that is specific and the proportion that is shared with PD dimensions/personality dimensions. We also report genetic and environmental correlations between the dimensions.

All biometric models were conducted using the structural equation modeling package OpenMx [64]. Separate means were estimated for males and females to account for sex differences in mean levels. Competing models were compared using Akaike’s information criterion [65] and Bayesian information criterion [66], with lower values indicating better model fit. All models were fitted by using full-information maximum likelihood.

Results

Preparatory/Initial Analyses

First, a series of factor analyses with one factor were conducted on each of the personality trait measures across the three measurement waves. These analyses were conducted to create aggregated measures for each trait across the waves. The factor loadings and proportion of variance explained by each factor are presented in Table 1. The high factor loadings indicate that the measures of personality traits across the three measurement waves are well represented by a single factor.

Table 1.

Factor loadings and proportion of variance

​	N	E	O	A	C	SEF_A	SEF_S	RS	ER	LON	SOC	IMP	CON	DEL
Factor loadingsa
Wave-1	0.74	0.76	0.71	0.69	0.70	0.62	0.69	0.70	0.67	0.68	0.64	0.68	0.64	0.54
Wave-2	0.91	0.91	0.91	0.88	0.90	0.92	0.88	0.80	0.85	0.84	0.77	0.85	0.82	0.82
Wave-3	0.81	0.81	0.84	0.80	0.82	0.67	0.77	0.77	0.74	0.73	0.69	0.81	0.67	0.57
Proportion varianceb	0.67	0.68	0.68	0.63	0.66	0.56	0.61	0.57	0.57	0.57	0.49	0.61	0.51	0.43

N, neuroticism; E, extraversion; O, openness; A, agreeableness; C, conscientiousness; SEF_A, academic self-efficacy; SEF_S, social self-efficacy; RS, Resilience Scale; ER, ego resilience; LON, loneliness; SOC, sense of coherence; IMP, impulsivity; CON, conduct problems; DEL, delinquency.

^a The factor loading for the corresponding personality trait measure, measured at wave-1, wave-2, and wave-3.

^b Proportion of variance the factor accounts for out of the variables measured at all three waves.

Next, a factor analysis on the 14 aggregated personality factor scores was performed. Both the scree plot of eigenvalues and Kaiser’s criterion (eigenvalues >1) indicate the presence of three factors. Specifically, the scree plot showed a notable drop after the third factor, and the first four eigenvalues were 5.02, 2.30, 1.21, and 0.34, respectively. Therefore, we decided to retain three factors. The factor loadings are presented in Table 2. We labeled factor 1 as “non-resilient,” factor 2 as “disinhibited,” and factor 3 as “disaffected.”

Table 2.

Factor loadings of the personality factors

​	Non-resilient	Disinhibited	Disaffected
N	0.54	−0.10	0.54
E	−0.84	0.10	−0.13
O	−0.63	−0.16	0.09
A	0.03	−0.35	−0.66
C	−0.47	−0.82	−0.03
SEF_A	−0.33	−0.58	−0.28
SEF_S	−0.69	0.12	−0.21
RS	−0.76	−0.37	−0.12
ER	−0.89	−0.07	−0.08
LON	0.61	−0.01	0.49
SOC	−0.30	−0.05	−0.72
IMP	−0.20	0.70	0.04
CON	0.01	0.49	0.59
DEL	−0.09	0.47	0.40

Factor loadings greater than 0.40 appear in bold. The total variance explained by the 3-factor solution was 61%.

N, neuroticism; E, extraversion; O, openness; A, agreeableness; C, conscientiousness; SEF_A, academic self-efficacy; SEF_S, social self-efficacy; RS, Resilience Scale; ER, ego resilience; LON, loneliness; SOC, sense of coherence; IMP, impulsivity; CON, conduct problems; DEL, delinquency.

Next, we performed a factor analysis on the PDs with three factors. Table 3 presents the factor loadings. The DSM categorizes the PDs into three clusters: the odd/eccentric cluster A PDs (paranoid, schizoid, and schizotypal), the dramatic/emotional/erratic cluster B PDs (antisocial, borderline, histrionic, and narcissistic), and the anxious/fearful cluster C PDs (avoidant, dependent, obsessive-compulsive). The diagnostic criteria for a given PD are multidimensional, and there is also considerable content overlap across PDs. Thus, completely “pure” factors were not expected. Two PDs showed notable cross-loadings that warrant brief discussion. Paranoid PD loaded on all three factors (0.47, 0.35, and 0.24). Although paranoid PD belongs to cluster A (odd/eccentric), its suspiciousness and interpersonal sensitivity share features with the dramatic and emotionally dysregulated presentations of cluster B (emotional/erratic), while its hypervigilance and rigidity resonate with the anxious and fearful trait characteristic of cluster C (anxious/fearful). Borderline PD loaded most strongly on the emotional/erratic factor (0.60) but also showed a moderate cross-loading on the anxious/fearful factor (0.32), which is not surprising given that anxiety and fear of abandonment are core features of BPD that overlap with cluster C pathology. Overall, these cross-loadings reflect the well-documented diagnostic co-occurrence and content overlap among PDs [67] and are consistent with the inherent complexity of PD classification. The primary loadings nonetheless remained interpretable and aligned with DSM cluster assignments, supporting the validity of the factor structure.

Table 3.

Factor loadings of the PD factors

​	Emotional/erratic	Anxious/fearful	Odd/eccentric
Paranoid	0.47	0.35	0.24
Schizoid	0.07	0.14	0.70
Schizotypal	0.35	0.26	0.55
Antisocial	0.36	0.02	0.17
Borderline	0.60	0.32	0.20
Histrionic	0.66	0.12	−0.04
Narcissistic	0.55	0.16	0.12
Avoidant	0.02	0.77	0.26
Dependent	0.25	0.59	0.06
Obsessive-compulsive	0.22	0.26	0.08

PDs corresponding to each factor a priori are highlighted in bold.

Phenotypic correlations between the personality dimensions and PD dimensions are provided in Table 4. The phenotypic correlations revealed significant associations between the personality and PD dimensions, although some were more pronounced. Specifically, the non-resilient personality dimension was most closely associated with the anxious/fearful PD dimension, the disaffected personality dimension had the strongest association with the emotional/erratic PD dimension, while the disinhibited personality dimension was generally weakly associated with all three PD dimensions. Before conducting the biometric analyses, a preliminary examination of the correlation matrices stratified by zygosity, as detailed in online supplementary Table S5 (within-trait correlations) and online supplementary Table S6 (cross-trait correlations), indicated that the DZ correlations were consistently less than half the size of the MZ correlations. This suggested an absence of shared environmental influences in the data.

Table 4.

Phenotypic correlations between personality dimensions and PD dimensions

Personality dimension	PD dimension
	emotional/erratic	anxious/fearful	odd/eccentric
Non-resilient	−0.08***	0.44***	0.25***
Disinhibited	0.18***	−0.10***	0.04*
Disaffected	0.36***	0.21***	0.10***

*p < 0.05. ***p < 0.001.

Biometrical Analyses

For all twin models, and consistent with the MZ/DZ phenotypic correlation patterns mentioned above, removing the shared environmental parameters (C) resulted in more parsimonious models. This was evidenced by lower Akaike’s information criterion and Bayesian information criterion values, as shown in online supplementary Table S7. Results from the Cholesky AE models examining the proportions of genetic and environmental variance in the PD dimensions that are specific and shared with the personality dimensions are summarized in Tables 5 and 6. Specifically, the proportion of additive genetic and non-shared environmental variances specific to each PD dimension and the portions shared with the personality dimensions are presented in Table 5.

Table 5.

Heritability (a²) and proportion of non-shared environmental variance (e²) within each PD dimension, together with the proportions of genetic and environmental variance explained by the personality dimensions

PD dimension	Genetic effects			Non-shared environmental effects
	a2	% shared with the personality dimensions	% specific	e2	% shared with the personality dimensions	% specific
Emotional/erratic	0.44	40.9	59.1	0.56	3.6	96.4
Anxious/fearful	0.49	41.7	58.3	0.51	9.8	90.2
Odd/eccentric	0.47	14.9	85.1	0.53	1.9	98.1

Table 6.

Heritability (a²) and proportion of non-shared environmental variance (e²) within each personality dimension, together with the proportions of genetic and environmental variance explained by the PD dimensions

Personality dimension	Genetic effects			Non-shared environmental effects
	a2	% shared with the PD dimensions	% specific	e2	% shared with the PD dimensions	% specific
Non-resilient	0.56	46.4	53.6	0.44	9.1	90.9
Disinhibited	0.53	15.1	84.9	0.47	2.1	97.9
Disaffected	0.51	32.0	68.0	0.49	6.1	93.9

PD, personality disorder.

The results in Table 5 showed that genetic and non-shared environmental influences each accounted for approximately half of the variance within all three dimensions of PDs. Approximately 40% of the genetic influences on the emotional/erratic and anxious/fearful PD dimensions overlapped with the personality dimensions. In contrast, genetic influences shared with personality on the odd/eccentric dimension were lower, accounting for only about 15% of the variance in that PD dimension. This pattern was even more pronounced for environmental influences. Less than 10% of the non-shared environmental factors (which also include measurement error) affecting each of the three PD dimensions were shared with those influencing personality.

Similarly, the proportion of variance specific to each personality dimension and the proportions shared with the PD dimensions are presented in Table 6.

The results reveal that genetic and non-shared environmental influences accounted for approximately half of the variance within each personality dimension. The genetic influences shared with PD ranged from moderate (non-resilient and disaffected) to small (disinhibited). Environmental influences on the personality dimensions were predominantly specific to each dimension, with those shared with PD explaining less than 10% of the variance within each personality dimension. As stated earlier, the non-shared environmental component also captures measurement error, and these estimates should therefore be interpreted with appropriate caution.

Genetic and environmental correlations between each of the personality dimensions and PD dimensions are presented in Table 7. The biometric results in Table 7 revealed that the strongest etiological associations (both genetic and non-shared environmental) were between the non-resilient personality dimension and the anxious/fearful and the odd/eccentric PD dimensions. Closely reflecting the pattern of phenotypic correlations in Table 4, the disaffected personality dimension demonstrated a similar pattern, albeit with associations of slightly lower magnitude. Genetic correlations for these personality dimensions with PD were generally moderate, while environmental correlations were generally low. The weakest etiological associations with any PD factor were found in the disinhibited personality dimension, which displayed generally nonsignificant associations, except for a moderate genetic correlation with emotional/erratic PD and a weak negative environmental correlation with anxious/fearful PD.

Table 7.

Genetic and environmental correlations between the personality dimensions and the PD dimensions

​	Emotional or erratic PDs		Anxious or fearful PDs		Odd or eccentric PDs
	r A	r E	r A	r E	r A	r E
Non-resilient	−0.09 [−0.20, 0.02]	−0.09 [−0.01, −0.17]	0.61 [0.53, 0.69]	0.23 [0.16, 0.31]	0.37 [0.27, 0.47]	0.14 [0.06, 0.22]
Disinhibited	0.37 [0.26, 0.47]	0.03 [−0.05, 0.11]	−0.06 [−0.16, 0.04]	−0.12 [−0.20, −0.04]	0.06 [−0.05, 0.17]	0.00 [−0.08, 0.09]
Disaffected	0.55 [0.45, 0.65]	0.19 [0.11, 0.26]	0.22 [0.12, 0.33]	0.18 [0.10, 0.26]	0.15 [0.04, 0.27]	0.06 [−0.03, 0.14]

95% confidence intervals in brackets. Statistically significant correlations in bold. r_A = additive genetic correlation; r_E = non-shared environmental correlation.

Discussion

The aim of the study was to investigate the etiological relationship between broad personality factors and associated PD dimensional factors, based on longitudinal data from seven national cohorts of adolescent twins. The results showed that approximately one-third of the genetic variation was shared between personality traits and PD dimensions. This suggests a moderate etiological association in support of a dimensional relationship, as proposed by several researchers [6, 67–69]. Given the study’s efforts to enhance measurement diversity and minimize common method bias, the association appears robust. However, importantly, the majority of genetic variance was specific to each dimension, indicating strong etiological distinctiveness of personality and PDs – consistent with prior research [32, 70]. All non-shared environmental variance was likewise trait-specific, though these estimates should be interpreted with care because the E component necessarily includes random measurement error. Overall, while these findings demonstrate a partial alignment with the notion that PDs are extreme variations of personality traits, the large amount of specific etiological variance clearly underscores the necessity for more complex dimensional models to capture the interplay between shared and distinct etiological influences in personality and PDs [5].

To explore the etiological connections between personality dimensions and PDs in greater detail, the study findings reveal several key associations. The anxious/fearful PDs showed the strongest etiological connection with the non-resilient personality factor, while the emotional/erratic PDs were primarily associated with the disaffected personality dimension. The emotional/erratic PDs also demonstrated a genetic link with the disinhibited personality factor, while the odd/eccentric PDs showed moderate etiological associations with the non-resilient personality factor. This pattern of etiological overlap suggests a dimensional relationship between specific personality dimensions and PD constellations.

While the disinhibited personality factor showed genetic overlap specifically with the emotional/erratic PDs, both the non-resilient and disaffected personality factors were linked to multiple PD factors, both phenotypically (Table 4) and etiologically (Table 5). The non-resilient personality factor was strongly genetically related to the anxious/fearful PDs and moderately to the odd/eccentric PD dimension. The disaffected personality dimension had the strongest genetic relation to the emotional/erratic factor and a moderate connection to the anxious/fearful PD dimension. These cross-dimensional etiological relationships echo recent findings by Kendler et al. [44]. Based on a different factor analytical approach, these researchers identified three genetic factors linking specific combinations of personality traits and PDs. Their genetic factor 1, which shows parallels to the anxious/fearful × non-resilient genetic covariation in our data, was characterized by high loadings on numerous PDs alongside the personality trait of neuroticism. Two further genetic factors emerged in the Kendler et al. [44] data. One named antisocial/antagonism factor bearing resemblance to a combination of the emotional/erratic × disaffected (high neuroticism and high externalizing) and the emotional/erratic × disinhibited factor (low neuroticism and high externalizing); and one named schizoid/introversion factor, parallel to the odd/eccentric × non-resilient constellation in our data.

Interestingly, an interpretation proposed by the researchers of their findings was that of a hierarchical structure with the genetic factor 1 as a genetic “general criteria for PDs” factor, with trait neuroticism or negative emotionality playing significant roles [44] across all disorder dimensions. Several researchers [71–73] investigating a hierarchical structure for PDs have proposed a common factor potentially formed by neuroticism or negative emotionality, often in interaction with severity or general impairment. In the present study, as shown in Table 2, trait neuroticism along with different constellations of general mental health measures (SOC, resilience, self-efficacy) loaded heavily on both the non-resilient and the disaffected personality factor in the factor analysis, which again as shown in Table 5 exhibited a relatively strong common genetic influence with the anxious/fearful and the odd/eccentric PD factors. However, the two personality factors in the present study demonstrated different patterns of strength in their genetic cross-correlations with the various PD dimensions. The non-resilient was more strongly genetically associated with what one could call the internalizing PD factors (anxious/fearful and odd/eccentric), and the disaffected more with the externalizing PD factors (strongly to the emotional/erratic and moderately to anxious/fearful), each with very weak or no etiological relationship with the personality factor that was on the opposite end of the internalizing/externalizing PD range. Moreover, the disinhibited factor, characterized by low conscientiousness and high impulsivity, showed an even more narrow genetic association specifically with the externalizing emotional/erratic PDs, supporting its links to antisocial behavior and “antisocial/antagonistic” factors also reported by Kendler et al. [44] and others [44, 74]. Disinhibition has also been proposed as a personality trait underlying general psychopathology [75], but this was not supported in the present study. Thus, despite the significant etiological cross-correlations identified, there was no clear candidate within the present data for any of the personality factors having etiological influences across all PDs. Others have also found limited or conditional support for the contribution of a general PD factor above and across the individual PD clusters [73, 76].

Although the data did not support a single universal PD genetic factor connecting normative personality and PDs, the complex pattern of cross-dimensional genetic overlap and specific etiological influences may still align with dimensional and hierarchical models. Findings from biogenetic studies have shown that personality and mental illness, including PDs, are inherently polygenic quantitative traits [77], with genetic influences operating in probabilistic rather than deterministic ways [78–80]. Numerous genetic effects are broad and nonspecific, influencing multiple functional domains, thus accounting for much of the prevalent comorbidity observed across psychiatric diagnoses [81]. Conversely, certain genes are specific to particular domains and diagnoses. This genetic influence, where higher order factors are broadly impacted by pleiotropic (multifunctional) effects, affecting several elements within and beneath those factors without influencing parallel hierarchical elements, is conceptually consistent with hierarchical approaches such as the HiTOP initiative [5].

The HiTOP model posits broad spectra (e.g., internalizing, externalizing, thought disorder) spanning multiple disorders [5]. The disinhibited × emotional/erratic factor reported here shows parallels with the disinhibited externalizing spectrum of the HiTOP [82]. At the same time, the substantial etiological variance found in the present study unique to each of the personality and PD dimensions indicates that hierarchical models capture only part of the underlying structure. While higher order genetic influences appear to contribute to shared liability across traits and PDs, much of the genetic variance and all non-shared environmental variance remain trait-specific, with the latter reflecting both genuine non-shared environmental effects and an inseparable measurement error component. Thus, hierarchical models provide a useful context for interpreting our pattern of partial genetic overlap alongside pronounced specificity. However, the etiological architecture found here should not be taken as a direct test of a hierarchical structure. It is also important to note that, in the present study, factors were derived empirically from PD criteria rather than through explicit hierarchical modeling, and the genetic correlations observed were moderate rather than strong.

While our finding of approximately one-third genetic overlap supports a continuum model linking normative personality to PD liability, it also suggests a degree of etiological distinctiveness between normative personality and PDs. In comparison, Reichborn-Kjennerud et al. [83] reported essentially complete genetic overlap between several DSM-5 section II PD diagnoses and their corresponding maladaptive section III trait domains. This pattern suggests that, whereas normative traits share meaningful genetic roots with PDs, maladaptive trait dimensions that are explicitly designed to index pathology appear to capture these shared genetic risk factors more fully. Thus, genetic overlap between normative and disordered personality traits clearly exists, but the degree and specificity of this overlap may increase as one moves from broad, normative trait measures to maladaptive trait constructs that more directly reflect disordered personality functioning.

All environmental influences in our study were non-shared and specific to each personality and PD dimension. Although our measures showed adequate reliability, measurement error could not be fully separated from true environmental effects in these estimates. Moreover, unique experiences such as distinct life events or peer influences typically have modest, transient effects in longitudinal studies [84, 85]. Accordingly, while firm conclusions about environmental influences are not possible, our findings support the prevailing view that genetic factors – likely mediated by complex neurobiological and neurobehavioral mechanisms [86] and interacting with unique environmental contexts [87, 88] – predispose individuals to disorders characterized by both common and disorder-specific etiological vulnerabilities, most of which are distinct from normative personality traits.

Overall, these results, based on a broad dimensional factor approach, are consistent with prior genetically informed studies, indicating an intricate etiological relationship between personality and PDs. This multifaceted nature challenges the notion of simply placing normative personality and pathological personality along a single continuum, highlighting the need to explore more complex models to understand their relationship.

Strengths and Limitations

A major strength of this study lies in its behaviorally genetically informative twin design combined with a multi-method/multi-wave approach, enabling a unique examination of the etiological association between PDs and personality while reducing common method bias. Specifically, personality factors were derived from exploratory factor analyses of multiple state-of-the-art scales measured across three waves, while PD factors were developed from factor analyses of dimensional scores from clinical interviews at age 19. This approach is designed to enhance the reliability and validity of each factor [18, 89, 90], while minimizing common method error between variables. Furthermore, the large population-based sample, encompassing seven national twin cohorts, enhances the generalizability of the findings. While diagnostic criteria have made it rare to study PDs before the age of 18, they do not suddenly emerge in young adulthood [91], PDs often emerge and tend to be most severe during adolescence [92]. Furthermore, empirical evidence suggests that factor structures in adolescents in most criteria resemble those of adult samples [93, 94]. Likewise, some personality traits especially related to the developmental tasks of adolescents, i.e., reward sensitivity (extraversion) and inhibitory control (conscientiousness), are under environmental pressure [95], and hierarchical organizations of personality psychopathology in adults have been indicated in adolescent samples [96], making this phase especially interesting for studying etiological sources of normative and pathological personality variation. Although attrition typically threatens generalizability in longitudinal studies, previous analyses based on the material revealed that it did not affect heritability estimates [45]. Simulation studies have suggested that attrition mainly affects prevalence estimates, not the validity of associations between variables of interest [97, 98]. Additionally, self-reports may be more effective for capturing affective traits and norm- or law-breaking behaviors, which parents and teachers of adolescents might not fully perceive or access.

Certain aspects of our methodological approach warrant clarification. In this study, PD was assessed dimensionally, using symptom-count scores that combined subclinical and clinical levels into a single continuous index of PD traits. This approach is commonly used in biometric PD research [99] because it better captures the extended phenotype, covers the full range of etiological risk [83], and increases power for low-prevalence diagnoses in community samples. It is also consistent with dimensional diagnostic frameworks such as RDoC [100] and DSM-5 AMPD [99], which emphasize underlying genetic and neurobiological mechanisms. However, because our PD indices include subclinical as well as clinical symptom levels, the findings primarily concern dimensional PD liability, and generalization to categorical clinical diagnoses should be made with caution. Results should also be corroborated with molecular genetic studies.

Next, although attrition analyses suggested that dropout did not bias heritability estimates, selective participation with respect to family composition, ethnicity, and socioeconomic background cannot be fully ruled out, and findings should be interpreted with this caveat in mind. Another methodological consideration concerns the inclusion criterion, which required data from at least one of the three measurement waves. Although missing data were handled through multiple imputation by fully conditional specification, the missing-at-random assumption underlying this approach cannot be formally verified. It is possible that participants who dropped out after one wave differ systematically from those who completed all three waves in ways not captured by the observed data or the imputation model. This may limit the precision of longitudinal estimates and the generalizability of findings.

Our study, employing a factor analytical approach on the spectrum level of normative and psychopathological personality, yielded results akin to earlier research utilizing a single factor/disorder measurement approach [44]. However, behavior genetic studies at even lower hierarchical levels, such as nuanced trait or symptom-level analyses, would help further evaluate these findings. Particularly, recent studies have found that genetic effects are smaller compared to non-shared environment when moving down in the factor hierarchy of personality, indicating that environmental effects in a possibly polyfactorial way [88] act to differentiate on more specific levels of specificity [101, 102].

While internal consistency and inter-rater reliability were generally moderate to high, they nonetheless indicate some degree of measurement error. Notably, the low inter-rater reliability for schizoid PD (intraclass correlation coefficient = 0.44) suggests that much of the disorder-specific environmental variance likely reflects measurement error and rater bias. As noted, true non-shared environmental effects cannot be separated from measurement error in the model, so related findings have been interpreted with caution.

Given the developmental stage of our emerging adult sample, specific considerations regarding measurement and diagnosis are needed. A limitation is the DSM-IV 5-year symptom duration applied to a sample averaging 19 years old, anchoring symptom onset in mid-adolescence. While this criterion may enhance diagnostic rigor, its validation in adults calls for caution when applied to emerging adults. As a further point of note, the present study includes measures of conduct problems and delinquent behavior within the normative personality factor, even though such indicators are also commonly used to assess externalizing psychopathology within the domain of mental disorders. As noted by Moffitt et al. [103, 104] and others [105–108], adolescence is a life phase where engagement in externalizing behaviors is so prevalent as to be considered a developmentally normative aspect of this period. Personality measures are typically developed on adult personality or childhood temperament and not particularly aimed at measuring normative adolescent personality. Thus, to address the absence of stage-appropriate personality measures and ensure the coverage of relevant traits that suitably represent behaviors prevalent among young people, we included measures of externalizing behaviors that could reasonably serve as normative variants for the numerous PDs within the externalizing domain of mental disorders.

Finally, it is important to note that estimates from genetically informed twin designs are inherently conditional on the societal and cultural contexts in which they are obtained. This study used seven national twin cohorts from Norway, which, although large and population-based, come from a society with relatively high socioeconomic and ethnic homogeneity. Shared environmental influences (C) were nonsignificant, indicating that macro-environmental factors such as family composition and parental education did not contribute to within-family similarity in this context. However, this homogeneity constrains the environmental variation captured in our data, and such factors may play a more pronounced role in more heterogeneous settings. Studies in more socioeconomically and ethnically diverse populations are needed to evaluate the generalizability of the etiological architecture observed here.

Practical Implications and Further Research

Our findings suggest that prevention and treatment efforts should neither rely on overly general notions of common risk nor focus exclusively on disorder-specific mechanisms. The substantial shared genetic variance supports transdiagnostic interventions targeting broad vulnerability factors (e.g., neuroticism/emotional dysregulation), while the considerable disorder-specific variance indicates a continued need for tailored interventions addressing the distinct features of each PD. For dimensional PD diagnosis, the results support models that integrate broad trait dimensions with disorder-specific configurations rather than assuming a single underlying continuum. For early intervention, they point to the value of identifying and modifying both general temperament-based vulnerabilities and more specific risk profiles associated with emerging PD features. Future research should examine how such transdiagnostic and disorder-specific approaches can be optimally integrated.

Conclusion

The study reveals a moderate yet significant genetic overlap between personality and PD factors, alongside substantial trait-specific influences suggesting distinct etiological pathways. While the common genetic effect was modest in size, the thorough methodological steps taken to mitigate common method bias reinforce the credibility of the findings, indicating that genetic influences appear to contribute to shared liability across traits and PDs. Yet much of the genetic variance and all non-shared environmental variance remained trait-specific, with the latter reflecting both genuine non-shared environmental effects and an inseparable measurement error component. This complexity challenges the idea that normative and pathological personality can be located on a single continuum, even at the level of broad factors. It instead points to the need to consider additional or alternative models for their relationship, including dimensional diagnostic frameworks that accommodate both shared and disorder-specific liabilities and that can inform early, tailored interventions.

Acknowledgment

Center for Child and Adolescent Mental Health, Eastern and Southern Norway, contributed to data collection.

Statement of Ethics

The study was approved by the Norwegian Data Inspectorate (06/00308-2/CAO) and the Regional Committees for Medical and Health Research Ethics, Norway (S-06026). The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation in this study was provided by the participants’ legal guardians/next of kin.

Conflict of Interest Statement

The authors have no conflicts of interest to declare.

Funding Sources

This study was supported by the Department of Psychology, University of Oslo, and by the Research Council of Norway (Grants 170089, 213722, and 213760). Eirunn Skaug was funded by Foundation Dam (Grant No. 426701).

Author Contributions

S.O.T. and T.W.: conceptualization, formal analysis, funding acquisition, investigation, methodology, project administration, resources, supervision, validation, and writing – original draft, review, and editing. E.S.: conceptualization, data curation, formal analysis, investigation, methodology, project administration, validation, visualization, and writing – original draft, review, and editing.

Funding Statement

This study was supported by the Department of Psychology, University of Oslo, and by the Research Council of Norway (Grants 170089, 213722, and 213760). Eirunn Skaug was funded by Foundation Dam (Grant No. 426701).

Data Availability Statement

The datasets presented in this article are not readily available because data collection received preapproval in 2005 from the Norwegian Data Protection Authority, mandating a 20-year period of individual data protection followed by either data deletion or anonymization. This approval was extended in 2025; anonymized data will be available upon request after 2028. Requests to access the datasets should be directed to trine.waaktaar@psykologi.uio.no.