The Impact of Personality Pathology Across Three Generations: Evidence from the St. Louis Personality and Intergenerational Network Study

Abstract

Personality disorder (PD) symptoms in a parent generation may confer risk for problems in future generations, but intergenerational transmission has not been studied beyond parent-child effects. We examined the generational transfer of risk associated with PDs using structural models of grandparent personality pathology and grandchild psychopathology among 180 adults (M_age=66.9), 218 of their children (M_age=41.2), and 337 of their grandchildren (M_age=10.5). We found evidence for general and heterotypic domain-specific transmission. Specifically, broad grandparent personality pathology was associated with broad grandchild psychopathology (B=.15, 95% CI [−.01, .31]); at the domain level, grandparent internalizing personality pathology was associated with grandchild externalizing psychopathology (B =.06, 95% CI [.01, .12]). Neither association was significantly mediated by parental personality pathology. These findings indicate that personality pathology in one generation confers risk for psychopathology across subsequent generations. Such intergenerational transmission operates across broad, rather than specific (i.e., individual disorder) psychopathology domains.

Pathological personality traits are associated with a variety of negative outcomes for individuals (e.g., poor social functioning, health problems, lower SES and education; Clark & Ro, 2014; Dixon-Gordon et al., 2015; Gleason et al., 2014; Iacovino, Bogdan, & Oltmanns, 2016; Jonason et al., 2017, Lazarus et al., 2014). These negative outcomes extend to social and familial relationships; for example, higher levels of personality pathology are associated with lower marital relationship quality and satisfaction, poorer parenting practices, and poorer offspring attachment (Adshead, 2015; Bouchard & Sabourin, 2009; Eyden et al., 2016; Oltmanns & Balsis, 2019; South, Boudreaux, & Oltmanns, 2020; Wilson et al., 2017). While emerging evidence suggests that personality pathology is associated with the intergenerational transmission of problems to a subsequent generation (e.g., childhood maltreatment; Paul et al., 2019; Stepp et al., 2011), we are not aware of any studies evaluating whether associations extend beyond the subsequent generation. Understanding the intergenerational transmission of problems related to personality pathology may facilitate the development of strategies and policies to mitigate its impact on subsequent generations.

Existing multigenerational studies have proved crucial in outlining the transmission of several problematic behavioral patterns associated with substantial public health burdens, such as depression (Gotlib et al., 2020; Olino et al., 2008; Pettit et al., 2008), aggressive parenting (Conger et al., 2003; Hops et al., 2003) antisocial behavior (Capaldi, Pears, et al., 2003; Thornberry et al., 2003), and other externalizing behaviors (Salvatore et al., 2015; Serbin et al., 2004). For example, findings of relatively modest intergenerational continuities in externalizing problems such as antisocial behaviors and substance use have led researchers to search for potential moderators of intergenerational associations, such as parental age or child gender (Capaldi et al., 2017). Similarly, findings of increased internalizing problems in the grandchildren of depressed grandparents, even in the absence of parental depression, has led researchers to speculate about mechanisms of this multigenerational persistence (Olino et al., 2008). Such investigations may help identify at-risk groups for preventative and intervention efforts.

Multigenerational studies of personality pathology offer a unique perspective on factors in the parent generation that may have ripple effects on the health, behavior, and socioemotional functioning of generations down the line, as well as factors that may exacerbate risk or, conversely, increase resiliency in high-risk individuals. Before this mechanistic work can be conducted, however, initial foundational research is required to examine whether effects of personality pathology in a parent generation persist several generations later, and if so, whether transmission is general, domain-level (i.e., occurs within certain related classes of disorders), or disorder-specific. PDs are differentially related to impaired interpersonal functioning in various relational domains; for example, in a meta-analysis examining PDs and interpersonal dysfunction, Wilson and colleagues (2017) found parental paranoid, schizotypal, avoidant, and borderline PDs to be associated with impaired parent-child relationships. It may be that certain broadband groupings of personality pathology in a parent generation (e.g., those associated with patterns of detachment) put future generations at risk for poor outcomes, but others (e.g., those associated with patterns of antagonism) do not. In the service of foundational work examining whether personality disorder symptomatology persists down the generational line, the present study aimed to investigate the effects of personality pathology in a sample of older adults on psychological problems in their children and grandchildren (i.e., across three generations).

A Case for Intergenerational Transmission of Personality Pathology

Personality pathology tends to complicate and degrade relationships over time (Hopwood et al., 2013); thus, it comes as no surprise that personality pathology in one individual has a detrimental impact on family members and relationships. In particular, pathological personality traits in parents have been tied to a wide array of negative outcomes in offspring. For example, Pearson and colleagues (2018) found high levels of maternal personality pathology to be associated with offspring self-harm, depression, and anxiety. Parental antisocial, narcissistic, and borderline PD are associated with offspring externalizing problems, and parental borderline PD is additionally associated with offspring internalizing problems, insecure attachment patterns, borderline symptoms, and emotion dysregulation (Bertino et al., 2012; Dutton et al., 2011; Eyden et al., 2016).

Furthermore, several individual personality disorders demonstrate intergenerational transmission across two generations, such that biological children of individuals diagnosed with personality disorders are themselves at increased risk for development of the disorder or traits related to it. Specifically, studies have identified intergenerational transmission effects across two generations for schizotypal, dependent, avoidant, borderline, and antisocial personality disorders (Baron et al., 1985; Di Giacomo & Clerici, 2017; Distel et al., 2009; Gjerde et al., 2012; Isomura et al., 2015; Loranger et al., 1982; Zanarini et al., 2009). Results of these studies suggest that transmission is not disorder-specific. For example, while borderline PD aggregates in families (including offspring), it also tends to co-aggregate with mood, anxiety, and substance use disorders, and histrionic, narcissistic, and antisocial PDs (Zanarini et al., 2009). Similarly, relatives (including children and grandchildren) of individuals with avoidant PD are themselves at increased risk of avoidant PD, but also for social anxiety disorder (Isomura et al., 2015). Thus, while intergenerational transmission of personality pathology has been investigated only at the disorder level, factors shared between different forms of personality pathology and psychopathology may better account for the familial links among individual PDs.

Despite these advances, this research has thus far focused on intergenerational transmission across only two generations, which limits our understanding of the persistence of risk associated with personality pathology over several generations. For example, it is important to examine whether associations between grandparent personality pathology and grandchild psychopathology can be accounted for by psychopathology in the intermediate (i.e., parent) generation. Borderline personality pathology presents as one relevant set of traits with which to index broad psychopathology. Borderline PD may be a marker of general personality pathology (Sharp et al., 2015), is associated with widespread psychiatric comorbidity (Grant et al., 2008), and maps onto both internalizing and externalizing dimensions of psychopathology (Eaton et al., 2011). Findings that parental borderline personality pathology, or a similar index of broad psychopathology, fully accounts for associations between grandparent and grandchild pathology would indicate that transfer of risk is perpetuated largely from parents to children, with no additional risk gained from greater personality pathology in grandparents.

Furthermore, the research examining intergenerational transmission almost exclusively relies on investigation of specific constructs (e.g., aggression; Conger et al., 2003) or disorders (e.g., borderline PD; Distel et al., 2009). Given that the constructs often studied (e.g., individual PDs, aggression, poor parenting practices, childhood maltreatment) tend to be interrelated, this approach may mask broader patterns of continuity among related bands of psychopathology. Transmission may occur at various levels of specificity, ranging from general (e.g., a broad domain of personality pathology conferring risk for psychological problems broadly in future generations) to domain-level (e.g., internalizing problems conferring risk for internalizing problems in future generations) to disorder-specific. Within domain-specific conferral, transmission may be homotypic, such that associations between the same domain of psychopathology are found across generations, or heterotypic, such that associations between different domains of psychopathology are found across generations. Research has not yet examined whether intergenerational transfer of risk associated with personality pathology may occur across general or domain-specific dimensions.

Models of Personality Pathology and Psychopathology

Psychopathology is often quantified in terms of a dimensional hierarchical structure. This empirically-based representation of pathology contrasts with the current categorical system of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5; American Psychiatric Association, 2013). Research on this hierarchical structure specifies latent factors that capture systematic covariation between disorders and symptoms. Higher-order latent factors are typically labeled Internalizing – common variance among disorders related to anxiety and depression – and Externalizing – common variance among disorders related to substance use, inattention, and antisocial behaviors (Achenbach, 1966; Krueger & Markon, 2006). The Internalizing-Externalizing model has been demonstrated across the lifespan and clinical status, establishing its utility as an empirically-based research framework for organizing psychopathology (Achenbach & Edelbrock, 1978; Eaton et al., 2015; Forbes et al., 2016). Furthermore, this model has been found to account for disorder persistence and predictive validity of future outcomes, such as suicide attempts, over and above disorder-specific variance (Eaton et al., 2013, 2015; Kessler et al., 2011; Kim & Eaton, 2015).

Problems with the DSM’s polythetic categories apply to PDs as well. PDs are highly comorbid with each other (Zimmerman et al., 2005) and with other psychological disorders (Lenzenweger et al., 2007), highlighting the need for examination of latent dimensional factors that may account for this comorbidity. Using factor analysis, several studies have identified relevant factors of DSM personality disorder criteria, with the number of factors ranging from two (Blackburn et al., 2005) to five (Trull & Widiger, 2013; Wright & Simms, 2015) to seven (Thomas et al., 2003; Trull et al., 2013). Hierarchical models of personality pathology further characterize personality pathology as including both general and specific factors (Boudreaux et al., 2019; Sharp et al., 2015; Williams et al., 2018; Wright et al., 2016). Because these factor structures are indicative of varying levels of specificity (with larger numbers of factors indicating more specificity in symptom clusters), examining intergenerational transmission via these various factor structures will help to elucidate the level of specificity at which intergenerational transmission may occur.

Using measures of common mental disorders, transdiagnostic psychopathology factors (i.e., Internalizing and Externalizing) have been found to account for transmission of specific psychological disorders from parents to offspring (Bornovalova et al., 2010; Hicks et al., 2004; Starr et al., 2014), suggesting that across two generations, transmission may occur across these broader pathways rather than across specific disorders. However, research has not examined whether presence of personality pathology in a parent generation may confer risk for psychological problems in future generations across these transdiagnostic pathways. In the personality pathology framework, the Internalizing factor captures comorbidity among PDs associated with negative affect and detachment (paranoid, schizoid, schizotypal, avoidant, dependent, obsessive-compulsive, and borderline), whereas the Externalizing factor captures comorbidity among PDs associated with antagonism and disinhibition (borderline, histrionic, narcissistic, antisocial; Blackburn et al., 2005; Boudreaux et al., 2019; Wright et al., 2012). The Internalizing and Externalizing factors may represent the highest or second-highest level of a more extensive hierarchical model of personality pathology, followed by maladaptive personality traits and facets at lower levels (e.g., Krueger et al., 2011; Widiger & Simonsen, 2005; Wright et al., 2012). In the present study, we examined single-factor and correlated two-factor (Internalizing-Externalizing) models of personality pathology in the interest of investigating general versus domain-specific transmission across three generations.

The Present Study

Given the impact of personality pathology on familial relationships, and the importance of identifying pathways of intergenerational transmission for mental health more broadly, there is a clear need to determine whether transmission persists beyond parent-offspring effects. Doing so will lay the groundwork for investigation of genetic and/or environmental pathways via which transmission may operate, which will aid in identification of families who are most at risk for transmission and help to focus prevention and intervention efforts. The goal of the current study was to evaluate transmission across three generations via the broad domain of personality pathology. We examined evidence for intergenerational transmission of personality pathology in a community sample of older adults (N = 180, M_age = 66.9) and their children (N = 218, M_age = 41.2) and grandchildren (N = 337, M_age = 10.5). Further extending our understanding of personality pathology transmission to hierarchical structural models of psychopathology, we examined single factor and correlated (Internalizing – Externalizing) factor models of both personality pathology (in grandparents) and of broadband psychopathology (in grandchildren). In the case of evidence for the intergenerational transmission of personality pathology, comparison of these structural approaches to psychopathology allows for differentiation of general and specific transmission. When focusing on multiple generations, two types of transmission could be at play: broad risk conferral, wherein transmission occurs within general factors across generations, or domain-specific risk conferral, wherein transmission occurs within domain-specific pathways. Finally, we examined whether parental borderline personality pathology, a putative index of both internalizing and externalizing psychopathology (Eaton et al., 2011), accounted for variance shared between the grandparent and grandchild generations. We expected to find positive associations between grandparent personality pathology, parent borderline personality pathology, and grandchild psychopathology. However, given the novelty of the study, we viewed these analyses as primarily descriptive and did not advance specific hypotheses regarding expectations for general versus specific transmission.

Methods

Participants

The St. Louis Personality and Intergenerational Network (SPIN) study is designed to examine the intergenerational transmission of personality and health as well as related psychosocial and biological factors across three generations. Preliminary data were collected from the children (G2 N = 218; M_age = 41.2; SD_age = 4.8; age range = 28–51; 60.7% female; 69.3% White, 23.4% Black) and grandchildren (G3 N = 337; M_age = 10.5; SD_age = 3.6; age range = 4–18; 51.2% female; 71.3% White, 20.8% Black) of a subset of participants who completed the St. Louis Personality and Aging Network (SPAN) study (G1 N = 180; M_age = 66.9; SD_age = 2.82; age range = 61–73; 68.9% female; 70.6% White, 26.1% Black; Table 1; Oltmanns & Gleason, 2011). At baseline, SPAN included 1,630 adults recruited from St. Louis and the surrounding suburban areas via recruitment calls made by purchasing telephone records from a private sampling firm. Inclusion criteria for SPAN included baseline age between 55 and 64 and English fluency. Exclusion criteria included individuals with an advanced, life-threatening illness, imminent plans to move out of St. Louis, inability to read at a 6^th grade level, and the presence of current psychosis. People without a home address were also excluded because it would be extremely difficult to maintain contact with them over time. SPIN data collection is ongoing.

Table 1.

Demographic Characteristics and Descriptive Statistics

Variable	Demographic Characteristics
	G1 N = 180	G1 Informant N = 173		G2 N = 218	G3 N = 337
Age: M (SD)	66.9 (2.8)	52.6 (12.4)		41.2 (4.8)	10.5 (3.6)
Race
% White	70.9	71.7		71.6	71.3
% Black	26.3	26.0		24.2	20.8
% Other/not reported	2.8	2.3		4.3	7.9
% Female	68.9	64.2		60.7	51.2
	Descriptive Statistics
G1 PD Factor	M (SD)	G2 PD Factor*	M (SD)	G3 CBCL Scale*	M (SD)
Paranoid	−.21 (.74)	Borderline	60.45 (10.44)	Anxious-Depressed	15.83 (3.40)
Schizoid	−.21 (.77)			Withdrawn-Depressed	9.64 (2.36)
Schizotypal	−.10 (.41)			Somatic Complaints	12.22 (2.12)
Avoidant	−.22 (1.54)			Attention Problems	13.58 (3.72)
Dependent	.04 (.68)			Rule Breaking	18.42 (2.56)
Obsessive-Compulsive	−.28 (.98)			Aggression	21.72 (4.61)
Borderline	−.13 (1.01)
Histrionic	.09 (.65)
Narcissistic	−.19 (1.01)
Antisocial	−.02 (.27)

^*G2 FFI-Borderline PD scores and CBCL scales are computed as sums. For the CBCL, each scale is computed from a different number of items, which accounts for the differences in mean scores across the 6 scales.

Of the 733 original G1 participants who reported that they had children and grandchildren, 597 gave permission to contact all or some of their G2 adult children. Of 1,276 potential G2 participants, 773 reported having children. Of these, 325 consented to complete questionnaires and 218 completed questionnaires (online or through mail) about themselves and their G3 children.¹ G1 data for the present study were gathered from three in-person interviews conducted with participants: baseline, in-person follow-up 1 (IP-FU1) occurring 2–3 years after baseline, and in-person follow-up 2 (IP-FU2) occurring 2–3 years after IP-FU1. G1 participants received $60 for completing each in-person SPAN session, and G2 participants received $30 for completing questionnaires. The study was approved by the Institutional Review Board at Washington University in St. Louis.

Measures

Descriptive statistics for G1 PD factors, the G2 Borderline PD scale, and G3 CBCL scales can be found in Table 1.

G1 Personality Pathology

Personality pathology was assessed in G1 participants across three time points (baseline, IP-FU1, IP-FU2) using interviewer, self, and informant ratings. Descriptive statistics for G1 PD measures are in Table S1. Consistent with previous research with this sample, scores were treated continuously to preserve variation at diagnostically subthreshold levels (e.g., Paul et al., 2019). Informant reports were included given evidence that they add unique information about an individual’s personality that the participant may be unable or unwilling to provide (Oltmanns & Turkheimer, 2006).

Clinical Interview.

The SIDP-IV is a semi-structured clinical interview that assesses diagnostic criteria for DSM-IV personality disorders (Pfohl et al., 1997). Trained interviewers rated 79 items corresponding to criteria of 10 PDs on a scale from 0 (no pathology present) to 3 (pathology strongly present). SIDP-IV scores were treated continuously by summing responses across criteria at each assessment for the following PDs: Paranoid (PA), Schizoid (SZ), Schizotypal (ST), Antisocial (AS), Borderline (BD), Narcissistic (NA), Histrionic (HI), Avoidant (AV), Dependent (DE), and Obsessive-Compulsive (OC). The percentage of participants who met at least one criterion at one or more assessments ranged from 8.9% (Antisocial) to 52.8% (Obsessive-Compulsive). Inter-rater reliability ratings from a selected subsample of 265 video-recorded baseline interviews show good agreement (intraclass correlation coefficient: 0.77). Further details are provided in Oltmanns et al. (2014).

Self and Informant Report.

The SPAN study acquired self- and informant-report of G1 personality. Most (91%) participants had an associated informant consent to the SPAN study protocol and report on the participant’s personality at the baseline assessment (Table 1). Informants completed mailed or online questionnaires about their associated participant and received $30 remuneration at each time point (i.e., baseline, IP-FU1, IP-FU2).

The Revised NEO Personality Inventory (NEO PI-R; Form S for self; Form R for informant; Costa & McCrae, 1992) consists of 240 items designed to assess the Big Five personality domains of neuroticism, extraversion, openness, agreeableness, and conscientiousness, as well as 30 lower-order facets. NEO PI-R personality pathology scores were generated independently for self and informant report by summing facet scale scores mapping onto each PD within each assessment (e.g., scores on the facets of anxiousness, angry hostility, depressiveness, impulsiveness, vulnerability, feelings, actions, and (low) deliberation are summed together to form the borderline NEO-PI-R scale; Lynam & Widiger, 2001) and had excellent internal consistency across all three time points (all Cronbach’s alphas > 0.87; Table S2). NEO-PI-R PD scales had moderate inter-rater agreement between self- and informant-reports within each study wave (median r = .495 [range, .39-.61]; Table S3) and excellent test-retest reliability (self-report: median r = .86 [range, .76 to .90]; informant-report: median r = 0.815 ([range, .74-.88]; Table S4).

The Multisource Assessment of Personality Pathology (MAPP; Okada & Oltmanns, 2009; Oltmanns & Strauss, 1998) is an 80-item measure of personality pathology based on lay translations of DSM-IV PD diagnostic criteria. Self- and informant MAPP PD scores were calculated by summing responses across the PD items. The percent of participants who met at least one PD criterion at one or more assessments ranged from 18.3% (Dependent PD) to 75.6% (Schizoid PD) per self-report, and from 27.2% (Dependent PD) to 88.9% (Obsessive-Compulsive PD) per informant-report. The MAPP scales had variable internal consistency across all three time points (self-report: median α = .66 [range, .37-.84]; informant-report: median α = 0.77 [range, .49-.87]; Table S2).² MAPP scales had relatively poor inter-rater (i.e., self- and informant-report) consistency across the three time points (median r = .24 [range, −.02-.38]; Table S2) and moderate to excellent test-retest reliability (self-report: median r = .58 [range, .29-.75]; informant-report: median r = .70 [range, .47-.82]; Table S4).

G1 PD Factor Score Generation.

We computed factor scores for each of the 10 DSM PDs. Exploratory structural equation modeling in Mplus 7.3 (Muthén & Muthén, 1998-2020) was used to derive PD factor scores from clinical interviews (SIDP-IV) and self and informant report (NEO PI-R and MAPP) data across in-person assessments (baseline, IPFU1, IPFU2). We hypothesized one-factor models to account for the correlations among the measures allocated to each PD factor. Model fits for all PD models were adequate to good, with factor loadings ranging from .24 to .98 (Table S5). Estimated factor scores for each PD were used in subsequent nalyses.

G1 Personality Pathology Factor Estimates.

Regression-based factor score estimates were computed from one- and two-factor models of personality pathology using the 10 PD factor scores. The one-factor model reflected loadings of all estimated PD factor scores. Consistent with previous research (Blackburn et al., 2005; Wright et al., 2012), in the G1 two-factor model, the G1 Internalizing (INT) latent factor reflected loadings of factor scores for paranoid, schizoid, schizotypal, avoidant, dependent, obsessive-compulsive, and borderline PDs. The G1 Externalizing (EXT) latent factor reflected loadings of factor scores for narcissistic, histrionic, antisocial, and borderline PDs³.

G2 Borderline Personality Pathology

G2 borderline personality pathology (BPP) was assessed using a 24-item Five Factory Inventory-Borderline Personality Disorder (FFI-BPD; Few et al., 2015) composite sum score generated from the NEO Five-Factor Inventory (Costa & McCrae, 1992). This FFI-BPD composite: (1) converges with explicit BPD assessments (r values = 0.35–0.72) and correlates highly with a time-invariant component of borderline pathology (r = 0.81); (2) is heritable (40%) with a large genetic correlation with explicit measures of BPD (r_g = 0.84); and (3) shows a highly similar profile of correlations with clinical criterion variables (e.g., childhood maltreatment, depression, alcohol use disorder; Few et al., 2015, Baranger et al., 2020). These data suggest that the FFI-BPD index is a valid indicator of borderline personality pathology. In the present sample, FFI-BPD had acceptable-good internal consistency (α=0.79, 95% CI [0.76, 0.83]). Other data indexing psychopathology and personality pathology were not available for G2 participants, which precluded us from examining broad Internalizing and Externalizing domains in this generation.

G3 Psychopathology

G3 psychopathology was assessed using the 113-item parent-report Child Behavior Checklist (CBCL; Achenbach & Rescorla, 2001), which measures past 6-month psychopathology characteristics in children aged 6 to 18.⁴ In the present study, dimensional composite sum scores were computed for the following scales: anxious/depressed, withdrawn/depressed, somatic complaints, attention problems, rule breaking behavior, and aggressive behavior. Coefficient alphas ranged from 0.73 (rule breaking behavior) to 0.87 (aggressive behavior; Table S6).

G3 Psychopathology Factor Estimates.

Regression-based factor score estimates were computed from one- and two-factor models of psychopathology using the 6 CBCL scale scores. The one-factor model reflected loadings of all estimated CBCL scale scores. Consistent with previous research (Achenbach, 1966; Laceulle et al., 2015), the G3 Internalizing (INT) latent factor reflected loadings of CBCL scales anxious/depressed, withdrawn/depressed, and somatic complaints; the Externalizing (EXT) latent factor reflected loadings of CBCL scales rule breaking, aggression, and attention problems.

Demographic Covariates

G1 and G2 participants self-reported their age, race, ethnicity, and gender. Due to limited endorsement of races other than White and Black, in the present study, race was coded as one dichotomous variable representing White and non-White. G1 and G2 participants indicated their highest level of education on a 9-point scale ranging from 1 (less than high school) to 9 (professional degree; Table S7). G1 and G2 participants reported their annual household income on an 8-point scale ranging from 1 ($20,000 or less) to 8 ($140,000 or more; Table S7). G1 socioeconomic status (SES) was computed by standardizing responses to each variable and then summing the standardized values. In the present study, scores on this SES metric ranged from −3.28 to 3.64 (M_SES = −0.02, SD_SES = 1.74). G2 participants reported on G3 age and gender.

Statistical Analyses

First, we estimated univariate associations between G1 PD factor scores (n=10), the G2 borderline scale score (n=1), and G3 psychopathology subscales (n=6). Second, we estimated relationships between broad measures (i.e., latent general, internalizing, and externalizing factors) of personality pathology (G1) and psychopathology (G3) across generations and their relation to G2 BPP. More specifically, we estimated the following associations: 1) G1 general personality pathology factor, G3 general psychopathology factor, and G2 BPP; and 2) G1 INT and EXT personality pathology factors, G3 INT and EXT psychopathology factors, and G2 BPP, while accounting for INT and EXT overlap within G1 and G3 generations. Third, and finally, we examined whether G2 BPP indirectly linked any significant associations between G1 personality pathology and G3 psychopathology using a 3–2-1 multilevel mediation model clustered by family.

Because the sample contains related participants, we accounted for the nonindependence of data by using linear mixed-effect models and nesting data according to this familial structure (i.e., G3 individuals nested in G2 families, which were nested within G1s) in all models. Analyses were conducted with and without the following fixed-effects covariates in applicable models: G1 and G2 age, gender, race (White/not-White), and SES, and G3 age and gender. Scale scores above the 99^th percentile and below the 1^st percentile were winsorized (i.e., replaced with those respective percentile values prior to analyses). Personality pathology and psychopathology scores were standardized before being entered in linear mixed-effect models, so while estimates are reported as unstandardized (B), they can be interpreted in terms of standard deviation units. Analyses were conducted using the psych, lavaan, and nlme packages of R statistical software (Pinheiro et al., 2020; R Core Team, 2020; Revelle, 2019; Rosseel, 2012) and Mplus v. 8.4 software (Muthén & Muthén, 1998–2020).

Results

Associations between G1 personality disorder symptoms, G2 borderline personality pathology, and G3 CBCL subscales were generally small when not including fixed-effects covariates (i.e., ranging from |B| <0.01 - .23) with the largest association between G2 BPP and G3 aggressive behavior (Table 2). Including covariates, the G1 personality pathology factor was positively, but not significantly, associated with G2 BPP (B=.13, 95 % CI [−.01, .27], p = .061; Figure 1A) and the G3 psychopathology factor (B=0.15 [95% C.I. −0.01, 0.31], p =0.067; Figure 1B). A test of equality indicated that the regression coefficients for the relationships between G1 broad personality pathology and G3 broad psychopathology, and between G2 borderline personality pathology and G3 broad psychopathology, did not differ significantly (z = 0.43, p = .67).

Table 2.

Associations (Standardized Betas) Between G1 Personality Pathology Factor Scores, G2 Borderline PD Symptoms, and G3 Psychopathology Scales

Predictor Variable	Outcome Variable
	G2 Borderline	G3 Anxious-Depressed	G3 Withdrawn-Depressed	G3 Somatic	G3 Attention Problems	G3 Rule Breaking	G3 Aggressive
G1 Paranoid	.15*	.02	.01	.14	.05	.17**	.11
G1 Schizoid	.10	.02	.04	.14	.14*	.16**	.09
G1 Schizotypal	.17*	.02	.04	.14*	.11	.15*	.16**
G1 Avoidant	.12	.08	.02	.11	.17**	.13*	.16*
G1 Dependent	.06	.08	.01	.08	.17**	.10	.14*
G1 Obsessive-Compulsive	.07	−.01	−.03	.09	.03	.07	.03
G1 Borderline	.21**	.02	.04	.17*	.07	.16**	.15*
G1 Histrionic	−.02	.03	<.01	.04	<.01	−.08	−.01
G1 Narcissistic	.11	−.08	−.03	.02	−.16*	.02	−.03
G1 Antisocial	.11	−.05	<.01	.05	−.16*	.01	.01
G2 Borderline	--	.15*	.19**	.18*	.13	.18*	.23**

Note. Beta estimates are estimated from models nested by family to account for data nonindependence.

^*p < .05

^**p < .01.

Figure 1.

Relationship Between Single Factors and Internalizing and Externalizing Factors of G1 Personality Pathology and G3 Psychopathology and G2 Borderline Personality Pathology

Note. FFI-BPD, Five Factor Index Borderline Personality Disorder scale. Solid lines indicate statistically significant effects. 95% confidence intervals are in brackets. G1 age, sex, race (White/Non-White), and socioeconomic status, and G3 age and sex, were included as covariates. G1 Internalizing was included as a covariate when using G1 Externalizing as a predictor and vice versa. In (b), G1 Internalizing was included as a covariate when using G1 Externalizing as a predictor and vice versa, and G3 Externalizing was included as a covariate when using G3 Internalizing as an outcome and vice versa. Scores were standardized before being entered in linear mixed-effect models, so B values can be interpreted in standard deviation units.

When evaluating G1 and G3 internalizing and externalizing factors, G1 INT was uniquely associated with greater G3 EXT (B=0.06 [0.01, 0.12], p =0.033); however, there were no significant homotypic (i.e., G1 INT – G3 INT, G1 EXT – G3 EXT) associations and G1 EXT was not associated with G3 INT (all |B|s<0.03, both ps>0.49; Figure 1C). G1 INT (B=.09, 95% CI [−.05, .23], p = .203) and G1 EXT (B=.12, 95% CI [−.01, .25], p=.069; Figure 1D) were positively, though not significantly associated with G2 BPP. G2 BPP was associated with higher G3 broad psychopathology (B=.20, 95% CI [.05, .34], p=.009; Figure 1E). At the G3 domain level, G2 BPP was not significantly associated with G3 INT (B=.05, 95% CI [−.01, .11], p=.083) or EXT (B=.04, 95% CI [−.01, .09], p=.135; Figure 1F).

G2 BPP did not indirectly link the G1 broad personality pathology factor to the G3 broad psychopathology factor (indirect effect: B = .03, 95% CI [−.02, .09], p = .246; Figure 2A; Table S8). The association between the G1 general personality pathology factor and the G3 general psychopathology factor (path c: B = .15, 95% CI [−.01, .31], p = .067) was slightly attenuated and was less precise with the inclusion of G2 borderline personality pathology as a mediator (path c’: B = .13, 95% CI [−.24, .49], p = .495). Similarly, G2 BPP did not indirectly link G1 INT personality pathology to G3 EXT psychopathology (indirect effect: B = .004, 95% CI [−.004, .01], p = .326). G1 INT and G3 EXT (path c: B = .06, 95% CI [.01, .12], p = .033) remained significantly associated when including G2 borderline personality pathology as a mediator (path c’: B = .05, 95% CI [.001, .10], p = .044; Figure 2B, Table S8).

Figure 2.

G2 Borderline Personality Pathology Does Not Indirectly Link G1 Personality Pathology and G3 Psychopathology at General or Domain Levels

Note. Scores were standardized before being entered in multilevel mediation models, so B reflects the standardized estimate. 95% confidence intervals are in brackets. G1 age, sex, race (white vs. non-white), and socioeconomic status, G2 age and sex, and G3 age and sex were entered as fixed-effects covariates. In (b), G1 Externalizing and G3 Internalizing were additionally entered as fixed-effects covariates. b₁ reflects the b path estimate at the G2 level whereas b₂ reflects the b path estimate at the G1 level.

Discussion

Our study examining personality pathology (G1: grandparents, G2: children) and psychopathology (G3: grandchildren) across 3 generations yielded 3 primary findings. First, we find evidence that personality pathology among grandparents is associated with greater borderline personality pathology in their children, and greater psychopathology in their grandchildren at small effects (B values ranging from .13 to .15; Table 2; Figure 1). Second, the intergenerational association between G1 personality pathology and G3 psychopathology was largely attributable to general factors (i.e., global personality pathology, Figure 1B; INT and EXT factors, Figure 1C) as opposed to disorder-specific indicators (Table 2). Moreover, this transmission appeared to be domain-specific but in a heterotypic (rather than homotypic) pattern, such that, on average, the grandchildren of individuals with high INT personality pathology showed substantially higher levels of EXT psychopathology. Third, although G2 BPP was consistently associated with elevated G1 personality pathology and G3 psychopathology, many of these findings did not reach nominal levels of significance and G2 BPP did not mediate any G1-G3 associations (Figure 2). Collectively, these findings suggest that personality pathology reverberates across generations and that this is most visible with broad spectrum indicators of pathology, as opposed to unique disorder-specific presentations.

Grandparent Personality Pathology is Associated with Grandchild Psychopathology

In our approximation of bivariate associations at the disorder-specific level, we found striking, albeit small, associations between individual PD factor scores in G1 and psychopathology scales in G3. Specifically, we found positive associations between G1 paranoid, schizoid, schizotypal, avoidant, dependent, and borderline PD factor scores and G3 somatic complaints, attention problems, rule-breaking behavior, and aggressive behavior. Surprisingly, effect sizes for associations between scale scores for consecutive generations were similar in magnitude to those for associations between G1 and G3 scores (i.e., highest |B|s = .21, .23, and .17 for associations between G1 and G2 scores, G2 and G3 scores, and G1 and G3 scores, respectively), indicating that grandparent and parent personality pathology accounts for similar amounts of variance in grandchild psychopathology.

Non-Specific Pathology Contributes to Intergenerational Transmission.

The evidence for general transmission suggested that higher levels of the broad domain of personality pathology conferred risk for psychological problems broadly two generations later. These findings are consistent with existing two-generation research pointing to parental psychopathology as a robust but non-specific predictor of offspring psychological disorders (McLaughlin et al., 2012). This indicates that, to some extent, there may be common underlying mechanisms/correlates conferring risk across various forms of personality pathology. Alternatively, there may be a common consequence or level of impairment that is the product of heterogeneous problems that are perpetuated across generations. While our data cannot speak to what this mechanism or pattern might be, the finding of general transmission is consistent with theoretical conceptualizations of personality disorders as sharing a core feature of interpersonal dysfunction (Hopwood et al., 2013), which may result in broad-reaching negative consequences for an individual’s family both within and across generations. An alternative perspective on the general factor of psychopathology is also consistent with our results. That interpretation views the general factor as reflecting general impairment that is nonspecific and secondary to the variables that load on it (Oltmanns et al., 2018; Smith et al., 2020).

G1 INT and G3 EXT Heterotypic Associations.

Beyond this general pattern attributed to the general factor of psychopathology, our results also point to heterotypic (G1 internalizing to G3 externalizing) transmission at the more narrowly defined domain level. The effect size for this association was small (B = .06), and was smaller than univariate effect sizes between individual G1 PD factor scores and G3 psychopathology scales, likely because the heterotypic transmission effect size is the most stringent test of specificity (i.e., it controlled for both G1 externalizing and G3 internalizing). This finding suggests that broadband groupings of personality pathology associated with patterns of negative affect and detachment do seem to confer specific risk for problems related to rule-breaking and aggression two generations later. This observed heterotypic pattern supports previous findings of associations between parental anxiety and externalizing problems in young offspring, and between grandparent depression and externalizing problems in young grandchildren (Olino et al., 2008), though in the same study, grandparent anxiety was associated with decreased grandchild externalizing.

Surprisingly, we found no evidence for within-domain transmission – greater externalizing-type personality pathology in grandparents was not associated with greater externalizing psychopathology in grandchildren, on average (controlling for grandparent and grandchild internalizing domains). Similarly, greater internalizing-type personality pathology in grandparents was not associated with greater internalizing psychopathology in grandchildren, on average (controlling for grandparent and grandchild externalizing domains). Further, we found evidence of heterotypic transmission only for the internalizing to externalizing pathway; higher levels of externalizing-type personality pathology in grandparents were not associated with internalizing problems in grandchildren. Although these findings suggest an intriguing pattern of G1 INT and G3 EXT heterotypic association that has been previously observed (e.g., Olino et al., 2008) it is notable that a test of equality of regression coefficients indicated that domain-level G1 to G3 associations did not significantly differ from each other (all ps > .07). Although larger samples are needed to ascertain whether our observed heterotypic association (i.e., G1 INT & G3 EXT) differs from homotypic (e.g., G1 INT & G3 INT) associations, we speculate about factors that may contribute to this heterotypic association below.

Heterotypic transmission may be explained via three mechanisms: chance, shared vulnerability factors, or causal factors (Costello et al., 2003; Lahey et al., 2011; Shevlin et al., 2017). In the case of shared vulnerability factors, the same underlying etiological influences would be responsible for higher levels of diverse types of pathology in both grandparents and grandchildren. In the case of causal factors, internalizing personality pathology in grandparents would cause externalizing psychopathology in grandchildren, potentially through a variety of mediated pathways – some of which we discuss below.

A case could be made for shared genetic factors underlying both general transmission and domain-specific transmission. Personality pathology is broadly heritable, whether measured categorically or dimensionally. Using family, twin, and adoption studies, modest to moderate heritability estimates have been found for all ten DSM personality disorders (see Reichborn-Kjennerud, 2010 for a review). However, it is likely that genetic etiological factors operate across broader domains than individual PDs. For example, in a multivariate twin study, Kendler and colleagues (2003) found that genetic associations between individual PDs were better accounted for by three genetic factors – 1) broad vulnerability to personality pathology or negative emotionality, 2) high impulsivity and low agreeableness, and 3) introversion – indicating that genetic risk for personality pathology operates via vulnerability to one of these three factors, rather than vulnerability to a specific PD. However, in the present study, our finding of heterotypic domain-level transmission suggests that something specific to internalizing-type personality pathology puts future generations at higher risk for externalizing problems.

Little research has examined characteristics unique to internalizing-type personality pathology which may contribute to this specific risk conferral. As noted previously, internalizing personality pathology is associated with patterns of negative affect and detachment (Wright et al., 2012). Similarly, in a meta-analysis of associations between personality disorders and dysfunctional interpersonal styles, Wilson, Stroud, and Durbin (2017) found that internalizing-type PDs were associated with socially avoidant traits – related to difficulty expressing feelings and feeling anxious around others – whereas externalizing traits were not. Interestingly, in the same meta-analysis, internalizing-type PDs (paranoid, schizotypal, avoidant) were associated with impairments in parent-child relationships, whereas externalizing-type PDs were not. In tandem with the findings of the present study, these results raise a testable hypothesis – that traits related to social avoidance, detachment, and negative affect may set the stage for problematic parent-child relationships and transmission of problematic symptoms across generations. Importantly, we may be able to test this hypothesis in future studies using SPIN data. We are currently expanding our G2 and G3 samples and are collecting more well-rounded data on both personality pathology and other forms of psychopathology in G2 participants. With this expanded sample, we hope to have more power to examine pathways of general and domain-specific transmission across all three generations.

Finally, it is important to consider more mundane factors that may have contributed to us only finding a domain-level heterotypic association between G1 INT and G3 EXT. Broadly, there was little evidence that any G1 variable, be it broad (i.e., G1 INT or G1 EXT; Figure 1C) or specific (i.e., specific G1 PDs; Table 2) was associated with G3 INT within our sample. Indeed, we found few associations between any of the G1 PD scores and G3 Internalizing. The overall lack of association between G3 INT and broad or specific G3 personality pathology suggests that grandparent personality pathology may not confer risk for G3 INT or that we were generally underpowered to detect G3 INT associations. Relatedly, the heterotypic transmission found in the present study may be a function of the ages at which we assessed psychopathology in grandparents and grandchildren. Psychopathology persistence across an individual’s own lifespan may be heterotypic, such that cross-disorder and cross-domain (i.e., externalizing to internalizing persistence) is not unusual (Kessler et al., 2011; Reef et al., 2010). In particular, behaviors related to externalizing psychopathology, such as antisocial behaviors and conduct problems, are strongly predicted by age, with many of these behaviors peaking in adolescence and young adulthood and then decreasing substantially across adulthood (Capaldi et al., 2016). It is possible that if we were to assess psychopathology later in the life of G3s, we would not find a heterotypic transmission pattern due to developmental differences in the presence of externalizing and internalizing psychopathology.

Borderline Personality Pathology in Parents Does not Mediate Grandparent-Grandchild Associations

Although G2 BPP was significantly associated with broad G3 psychopathology, the associations with any G1 variable and with G3 domain-level phenotypes (i.e., INT, EXT) only approached significance. Similarly, although G2 BPP attenuated associations between G1 personality pathology and G3 psychopathology, there was no evidence of significant mediation (Figure 2). While these results suggest that the associations between G1 and G3 that we observed may be independent of G2 BPP, it is important to acknowledge limitations of this analysis. First, our preliminary intergenerational data acquisition did not comprehensively assess personality, personality pathology, or psychopathology among the G2 generation.⁵ As a result, we had to rely on a more specific index of borderline personality traits among G2 individuals. However, while we are unable to generate broad domains of personality pathology or psychopathology among G2 participants, evidence suggests that borderline personality pathology is an indicator of broad psychopathology (Eaton et al., 2011; Sharp et al., 2015). Second, we had a relatively small sample size with which to detect small multilevel mediation effects. Nevertheless, we hope to be able to conduct more stringent tests of multigenerational intergenerational transmission in future studies as data continue to be acquired.

In sum, results of the present study indicated that risk associated with personality pathology persists across three generations, but further research is required to elucidate mechanisms of this transmission. Intergenerational transmission of personality pathology is likely explained by an additive or interactive combination of genetic and environmental factors. Genetic vulnerabilities to PD symptoms and related problems likely interact with adverse environments, and adverse environments are more likely to occur in families of individuals with high levels of personality pathology (Fatimah et al., 2019; Laulik et al., 2013; Wilson & Durbin, 2012). At the family level, parental personality pathology may also increase risk for child psychopathology via perpetration of abuse and/or neglect, exposure to relationship violence, and creation of coercive or invalidating family environments that maladaptively reinforce children’s emotional and behavioral problems (Beauchaine et al., 2009; Johnson et al., 2006; Laulik et al., 2013, 2016; Paul et al., 2019; Stepp et al., 2011). Genetic risk may be exacerbated by poor parenting practices, which are often included in theoretical and statistical models of familial transmission of personality pathology and related constructs (Adshead, 2015; Oliver et al., 2009; Shaw & Starr, 2019), and which are commonly implicated in the development of internalizing and externalizing psychopathology in children (Pinquart, 2017a, 2017b). However, research focusing on genetic and environmental contributions to familial similarity in personality pathology has historically focused on non-twin family studies or classical twin studies, but extended twin-family designs are lacking (see Distel et al., 2009, for an exception focusing on borderline personality features). Extended twin-family studies incorporate siblings, parents, and spouses of twins and allow for increased power to distinguish between different types of genetic and environmental effects (Keller et al., 2010), and may be key in elucidating genetic and environmental transmission patterns across multiple generations.

Limitations and Future Directions

Given the methodological and practical difficulties in collecting data across several generations in a single family, multigenerational family data are rare but have the ability to answer unique research questions (Serbin & Stack, 1998). Utilizing data from grandparents and grandchildren allowed us to examine the persistence of personality pathology across three generations, representing a novel contribution to the literature on intergenerational transmission of psychopathology. However, multigenerational studies should be interpreted in the context of their inherent limitations. For example, it is difficult to fully correct for the clustered nature of these data, which include siblings and cousins at the G3 level, particularly when sufficient statistical power to detect small intergenerational effects necessitates large sample sizes at each generational level (Snijders, 2005). Additionally, personality pathology was assessed in only one grandparent and one parent for each G3 participant. Efforts to more fully model intergenerational transmission should include assessment in multiple people at each generational level. Further, whereas G1 data were collected via multiple informants and methods, only G2 participants reported on G2 and G3 psychopathology. Because parents have less access to their children’s internal thoughts and feelings, parents may be limited in their ability to report on their children’s internalizing symptoms (De Los Reyes & Kazdin, 2005). Additionally, parents with mental health problems may over-report child behavior problems compared to other informants (Najman et al., 2000), suggesting that future studies should strive to incorporate multiple informants at all generational levels.

Moreover, the study design would benefit from access to larger samples, which would allow researchers to use more complex statistical techniques to model family data and provide more power to detect small effects. As SPIN data collection is ongoing, it may be feasible to implement such techniques using this dataset in the future. In the context of increased power, we recommend three primary directions for future study.

First, mechanisms that may play a role in intergenerational transmission should be investigated. As described above, intergenerational transmission is likely a product of genetic and environmental influences, and elucidating these mechanisms should be prioritized in future research. For example, further study of environmental factors that may exacerbate or protect against the transfer of risk associated with personality pathology, such as childhood trauma or parenting practices, has clear consequences for the development of prevention and treatment interventions. Secondly, and along the same lines, future research should incorporate information from all relevant generations in a multigenerational study. As discussed previously, given our relatively small sample size for three-level mixed-effects models and our imperfect measurement of personality pathology in G2 participants, we were hesitant to include G2 data in the present study. We recognize the importance of fully modeling personality pathology or psychopathology in the G2 generation in future SPIN studies, where possible. In particular, it would be interesting to simultaneously analyze transmission patterns from G1s to G2s and from G2s to G3s to determine whether the G1 Internalizing to G3 Externalizing transmission pattern found in the present study is also found in these parent-child subsets. Additionally, certain G2 factors, such as assortative mating and parenting behaviors, likely predict variance in offspring (G3) psychopathology above and beyond the influence of genetic and environmental transmission accounted for by G1. Lastly, future studies should strive to employ modeling approaches better suited to larger datasets, though complementary to those used in the present study. For example, advanced bifactor modeling could be used to simultaneously model general and specific transmission across multiple generations in studies with access to larger samples.

Each of these recommendations would be best employed with substantially larger samples and the use of complex multigenerational twin-family designs. However, as noted before, these studies are practically, methodologically, and empirically difficult to implement. We argue that it is important for researchers with unique multigenerational data to explore research questions to the extent that they can, despite lacking access to optimal study designs, while being appropriately cautious in the interpretation of such findings.

Conclusions

In a community sample of grandparents (G1) and their children (G2) and grandchildren (G3), the present study tested the extent to which associations between grandparent personality pathology and grandchild psychopathology were general or specific, and whether they were mediated by parental borderline personality pathology. Results indicated that associations between general factors of G1 personality pathology and G3 psychopathology were better accounted for via specific transmission; grandparents with higher levels of internalizing personality pathology tended to have grandchildren with higher levels of externalizing psychopathology. Further, parental borderline personality pathology did not substantially mediate the association between general factors of G1 personality pathology and G3 psychopathology or the unique association between G1 internalizing personality pathology and G3 externalizing psychopathology, though these preliminary findings should be replicated with more comprehensive measures of parental personality pathology or psychopathology. These results demonstrate how using dimensional models enhances our understanding of the transfer of risk associated with personality pathology across generations. Specifically, while focusing on intergenerational transmission of specific disorders or constructs may be an overly narrow approach, researchers should also be cautious in utilizing only the broadest levels of analysis. Finally, these results underscore the need for increased multigenerational research to examine potential risk and protective factors associated with the intergenerational familial impact of personality pathology.