Using Multiple Methods to Evaluate Associations Among Externalizing Psychopathology, Personality, and Relationship Quality: A Replication and Extension

Abstract

The current study evaluated associations among externalizing psychopathology, personality, and relationship quality in a sample of 794 couples. Personality and psychopathology were assessed using dimensional measures, and relationship attributes were assessed with both self-report and observer-reports of videotaped interactions. Results were consistent with prior work (i.e., Humbad et al., 2010) such that greater externalizing psychopathology remained a significant predictor of lower relationship adjustment, while controlling for personality traits. Importantly, dimensional measures of externalizing psychopathology showed stronger associations with relationship adjustment when compared to symptom count measures used in Humbad et al. (2010). These results highlight the importance of replication and extension studies, the usefulness of dimensional measures of psychopathology, and the value of multiple methods of assessment to increase confidence in the robustness of associations between pathological attributes of personality and features of romantic relationships.

Dissatisfying romantic relationships are associated with a number of negative outcomes, including poorer physical and mental health, diminished personal well-being, and lower overall life satisfaction (e.g., Heller, Watson, & Ilies, 2004; Proulx, Helms, & Buehler, 2007; Robles, Slatcher, Trombello, & McGinn, 2014; Whisman, 2013). Understanding the factors that contribute to relationship quality is therefore an important area of research. The vulnerability-stress-adaptation (VSA; Karney & Bradbury, 1995) model is a widely used framework for such studies. According to the VSA, individual characteristics like personality traits and psychopathology serve as distal risk factors for relationship distress by influencing interpersonal processes such as communication patterns and how couples adapt to stressful experiences. Extensive research has reported that both personality traits and psychopathology are related to relationship outcomes (see e.g., Dyrenforth, Kashy, Donnellan, & Lucas, 2010; Whisman, 2013). The current study aims to extend key findings in this area by examining the relationship between externalizing psychopathology, personality traits, and relationship quality in a large sample of married couples using a multimethod approach.

Enduring Vulnerabilities and Relationship Quality

Both psychopathology and personality traits are viewed as enduring vulnerabilities within the VSA model. Externalizing disorders, which include substance use disorders as well as antisocial personality disorder and psychopathy, can be particularly challenging for romantic relationships. The thoughts, feelings, and behaviors associated with externalizing problems are likely to be associated with relationship quality through the processes outlined by the VSA. First, externalizing problems increase the stressors the couples have to face. For instance, aggressive tendencies have been linked to future economic troubles (Caspi, Wright, Moffitt, & Silva, 1998; Conger, Martin, Masarik, Widaman, & Donnellan, 2015). Put simply, externalizing problems may increase the problems couples have to content with in their relationship. Second, a propensity toward externalizing problems undermines the ability of individuals to handle relationship conflicts when they do arise. Thus, it is no surprise that alcohol and drug use can create problems in romantic relationships. Substance use has been cited as a significant reason for why many married couples choose to divorce (e.g., Amato & Previti, 2003; Scott, Rhoades, Stanley, Allen, & Markman, 2013). In addition, antisocial/psychopathic personality traits, including aggressiveness, deceitfulness, manipulation, and impulsivity, are those that cause significant impairment in interpersonal relationships including romantic relationships. These traits have been associated with a number of negative outcomes in romantic relationships, including increased conflict and sexual infidelity (e.g., Capaldi & Owen, 2001; Jones & Weiser, 2014). Indeed, existing work has documented the negative association between the range of externalizing disorders and relationship distress (e.g., Cranford, Floyd, Schulenberg, & Zucker, 2011; Humbad, Donnellan, Iacono, & Burt, 2010; Marshal, 2003; Savard, Sabourin, and Lussier, 2006; Weiss, Lavner, & Miller, 2018; Whisman, Tolejko, & Chatev, 2007).

Likewise, personality traits of neuroticism and negative emotionality (i.e., the dispositional tendency to experience more negative emotions) and traits of disinhibition are also represented as enduring vulnerabilities within the VSA model, whereas personality traits of extraversion or positive emotionality (i.e., the tendency to experience positive emotions) typically represent “enduring strengths” within the VSA model. These “enduring strengths” are theorized to serve as protective factors for relationship quality. Indeed, the associations between personality traits and relationship outcomes have been well documented in existing work (e.g., Dyrenforth et al., 2010; Humbad et al., 2010; Karney & Bradbury, 1995). Consistent with the VSA model, this work has found that traits of positive emotionality are associated with greater marital adjustment in both individuals and their spouses, and traits of negative emotionality and disinhibition are associated with decreased marital adjustment.

The majority of the existing work examining associations between externalizing disorders, personality traits, and relationship outcomes has focused on specific disorders and personality traits in isolation, even though there is evidence that 1) a latent externalizing factor may underlie specific externalizing disorders (Krueger, Markon, Patrick, Benning, & Kramer, 2007; South, Krueger, & Iacono, 2011) and 2) personality traits have been consistently linked with psychopathology (e.g., Kotov, Gamez, Schmidt, & Watson, 2010; Krueger & Eaton, 2010; Widiger, 2011). For example, Miller and Lynam (2001) found that higher levels of negative emotionality and lower levels of constraint (i.e., greater disinhibition) were both associated with antisocial personality disorder. Such findings indicate a need to study both externalizing psychopathology and personality traits in the same analysis to disentangle whether the effects of externalizing psychopathology, for example, persist after controlling for personality disorders. Humbad et al. (2010) filled such a need using data from a sample of over 1,800 married couples. Externalizing psychopathology was defined as a composite of raw symptom counts of antisocial personality disorder, conduct disorder (retrospectively reported), and alcohol dependence as assessed through diagnostic interviews of the Diagnostic and Statistical Manual of Mental Disorders (DSM). The authors found evidence that both personality and psychopathology were associated with marital adjustment in separate analyses. Importantly, they found evidence that externalizing psychopathology remained associated with marital adjustment after controlling for normal personality traits, although the effect size was diminished. The current study aims to constructively replicate this prior work, and importantly, to extend our analyses to include observational measures of the romantic relationship.

The Current Study

The purpose of the current study is to constructively replicate and extend key findings from Humbad et al. (2010) to examine the relationship between externalizing psychopathology, personality traits, and relationship quality in a large sample of couples using both self-report measures and observational measures. Using the actor-partner interdependence model (Kenny, Kashy, & Cook, 2006), we plan to examine both actor effects (i.e., the associations between an individual’s psychopathology/personality and his or her own relationship quality) and partner effects (i.e., the associations between an individual’s personality/psychopathology and his or her partner’s relationship quality) for these relationships. We predict that greater externalizing psychopathology and greater negative emotionality will be associated with reduced relationship quality, and both of these associations will have significant actor and partner effects. We predict there will also be significant actor and partner effects for positive emotionality, such that greater positive emotionality will be associated with greater relationship quality for both individuals and their partners. For constraint, we predict that there will only be significant actor effects (as found in Humbad et al., 2010), such that greater constraint will be associated with greater relationship quality for individuals only (and not their spouses). We also expect to find that externalizing psychopathology will remain independently associated with relationship quality and have significant actor and partner effects, even after controlling for personality traits.

In addition, we will extend prior work with the addition of observational measures of the relationship. Observational research in the area of close relationships can provide important, objective information on dynamics occurring between partners that may not be entirely captured through self-report measures. Prior work has found that partners who communicate with greater levels of negative affect (e.g., hostility, contempt) and lower levels of positive affect (e.g., empathy, understanding) tend to be at higher risk for relationship dissatisfaction (e.g., Caughlin, Huson, & Houts, 2000; Johnson et al., 2005). Observational measures in conjunction with self-report measures can reduce concerns that shared-method variance inflates associations between individual differences and relationship processes, a common issue in the existing literature (e.g., Orth, 2013). Nonetheless, observational measures of the romantic relationship are often missing from studies with larger sample sizes given the data can be resource-intensive to collect.

The current study will examine the role of observationally coded positive and negative affect in partners and their association with relationship quality, and it will examine how self-reported externalizing psychopathology and self-reported personality measures are associated with observer-rated relationship quality. We predict lower observer-rated positive affect and greater observer-rated negative affect will be associated with reduced relationship quality for both individuals and their partners. We also expect that there will continue to be significant actor and partner effects of externalizing psychopathology on relationship quality while controlling for observer-rated positive and negative affect. Finally, we expect to find consistency in the relationships between externalizing psychopathology, personality, and relationship quality when using both self-report and observational measures of relationship quality.

Method

Sample

Data for these analyses consisted of 794 couples (total N = 1,588 participants) with children recruited as part of the Twin Study of Behavioral and Emotional Development in Children (TBED-C), a study within the Michigan State University Twin Registry (MSUTR; Burt & Klump, 2013; Klump & Burt, 2006). Couples consisted of the biological mother of the twin and her current romantic partner, which was typically the biological father (n=764) but could have also been either a stepdad or current long-term dating partner (n=30). Participants ranged in age from 24–70 years, averaging 38.4 years for women (SD = 5.3) and 40.3 years for men (SD = 6.0). The majority of couples were married (94.7%) and the remainder were dating. Couples had been together for an average of 15.1 years (SD = 5.1).

The TBED-C families were recruited from the lower peninsula of Michigan. Of the 794 couples, 432 couples were part of a population-based sample of the TBED-C and 362 couples were part of an “at-risk” sample of families residing in modestly-to-severely disadvantaged neighborhoods. Families were recruited through birth records via anonymous recruitment mailings through the Michigan Department of Health and Human Services. Mailings in the at-risk sample were restricted to families residing in neighborhoods with census-level poverty data at or above the 2008 mean of 10.5%. The demographics of married/dating couples in the population-based sample were as follows: 93% White, 4% Black, 3% other ethnicities (across all, 2% also identified as Hispanic/Latino origin). The at-risk sample was slightly more ethnically diverse: 90% White, 7% Black, 3% other ethnicities (across all, 4% also identified as Hispanic/Latino). Full participant recruitment details and additional sample details can be found in Burt and Klump (2013).

Measures

Relationship adjustment.

The 32-item Dyadic Adjustment Scale (DAS; Spanier, 1976) was used to assess relationship adjustment in the current sample. To allow for direct comparisons with Humbad et al. (2010), we made use of the overall scale rather than the four highly correlated subscales of satisfaction, consensus, cohesion, and affective expression. Two additional items were also added to the original 32-item scale to assess the extent of agreement between partners regarding their parenting: how to raise the children and how to discipline the children. These questions were also added in the Humbad et al. (2010) sample given that conflicts over child rearing can play a significant role in perceptions of relationship adjustment for couples with children (e.g., Cui & Donnellan, 2009). The overall DAS score had excellent internal consistency for men and women in the current sample (α = .94). Approximately 8% of the total sample (n = 133) was missing data on this measure.

Externalizing Psychopathology.

The Adult Self-Report (ASR; Achenbach & Rescorla, 2003) is a self-report measure assessing a variety of clinical dimensions. The overall externalizing psychopathology scale was used (35 items; α = .85), which assesses multiple facets of externalizing, including aggression (i.e., getting into fights, threatening others, having a ‘hot’ temper), rule-breaking (i.e., drug use, excessive alcohol use, cheating, theft) and intrusiveness (i.e., teasing others, bragging, showing off). Analyses were conducted using the raw scale scores. Only 1.2% of the sample (N = 19) was missing data on this measure.

Personality.

The Multidimensional Personality Questionnaire – Brief (MPQ-BF; Patrick, Curtin, & Tellegen, 2002) was used to assess personality along three higher order factors: positive emotionality (the dispositional tendency to experience positive affect/emotions), negative emotionality (the dispositional tendency to experience negative affect/emotions), and constraint (self-control and behavioral restraint). We further separated positive emotionality into its agentic (high scorers are ambitious, socially dominant) and communal (high scorers report greater interpersonal connectedness and positive emotions from close relationships) subfactors given that communal positive emotionality is more closely related to relationship quality (e.g., Donnellan, Assad, Robins, & Conger, 2007; Humbad et al., 2010). Only 3.6% of the sample (N = 57) was missing data for personality. Reliabilities for the four factors ranged from α = .76-.87.

Observational Data.

Couples participated in a ten minute videotaped interaction that was later coded on eight dimensions using the Brief Romantic Relationship Interaction Coding Scheme (BRRICS; Humbad, Donnellan, Klump, & Burt, 2011). Prior to the interaction, couples were asked to select from a list three areas of common conflict in their relationship. They were then instructed to discuss their thoughts and feelings about a few commonly selected problem areas with the intention of trying to resolve the conflict while being videotaped. The straightforward BRRICS was specifically developed to code interaction data in large samples of couples because of its ease of use while still providing important observational information on the dyadic relationship (Humbad et al., 2011). The coding scheme provides eight codes, four at the individual level and four at the dyadic level. At the individual level, positive affect (e.g., displaying positive emotions, making positive/affiliative statements, providing respectful comments and feedback) and negative affect (e.g., use of harsh tones, criticism, hostility) were coded separately for men and women in each couple. At the dyadic level, positive reciprocity and negative reciprocity are dyadic versions of positive and negative affect (i.e., back and forth exchanges that are positive or negative in nature). Couples were also coded for presence/absence of the demand-withdraw pattern (i.e., when one partner approaches conflict while the other avoids conflict; see Christensen & Heavey, 1990), and a global rating of relationship satisfaction was made by coders (i.e., a 5-point rating of how satisfied the coder perceives the couple to be in their relationship ranging from “Extremely Low” to “Extremely High”). The current study made use of positive and negative affect for men and women to examine as predictors of relationship adjustment. The overall satisfaction code was examined as an outcome of personality and psychopathology predictors. Full details on coder training and reliability can be found elsewhere (Humbad et al., 2011), but reliabilities computed using intraclass correlations between all coders were adequate across the codes of interest in the current paper (ranging from 0.69–0.75). Approximately 7% of the sample (N = 110, or 55 couples) was missing observational data.

Data Analyses

Following Humbad et al. (2010), the actor-partner interdependence model (APIM; Kenny et al., 2006) was used to estimate associations between personality/psychopathology/observer-reported affect and relationship adjustment (see Figure 1 for the multivariate APIM). In this model, actor effects are denoted by a in Figure 1, whereas partner effects are denoted by p in Figure 1. As seen in Figure 1, for the multivariate model, all predictors were entered into the same model to examine unique actor and partner effects for each predictor. Prior to running this multivariate model, analyses were first conducted for each predictor in a separate APIM (e.g., externalizing psychopathology alone predicting relationship adjustment, negative emotionality alone predicting relationship adjustment). A multivariate model examining externalizing psychopathology and observer-reported positive and negative affect was also tested. To examine the association between personality, psychopathology, and observer-reported overall satisfaction, a regression analysis was conducted given each couple had only one score for observer-reported satisfaction. Mplus 7 was used to estimate all APIMs and it was used for the regression analysis.

Figure 1.

Multivariate Actor-Partner Interdependence Model (APIM; Kenny, Kashy, & Cook, 2006). a_w and a_h represent women’s and men’s actor effects, respectively (i.e., the association between one’s own personality/psychopathology and one’s own relationship adjustment); p_w and p_h represent women’s and men’s partner effects, respectively (i.e., the association between one’s own personality/psychopathology and one’s partner’s relationship adjustment). E₁ and E₂ respectively represent the residual variance in women’s and men’s relationship adjustment, after controlling for actor and partner effects. In this multivariate model, all predictors are entered at once, and thus, actor and partner effects for each variable represent associations between those variables and relationship adjustment after controlling for all other actor and partner effects.

Prior to all analyses, raw scores for all variables were standardized across the overall sample. This procedure followed recommendations by Kenny et al. (2006) to help generate standardized estimates which would facilitate comparisons to estimates reported in Humbad et al. (2010). Analyses were initially conducted separately for each predictor variable (i.e., externalizing psychopathology, personality traits, and observer-reported positive and negative affect and relationship adjustment). We then evaluated whether the effects of externalizing psychopathology persisted after controlling for personality.

Tests for distinguishability following guidelines from Kenny et al. (2006) were first conducted to evaluate whether men and women were systematically different in terms of means, variances, and covariances for each variable across the full sample. We found evidence for distinguishability for all variables (i.e., chi-square values ranging from 15.2 to 98.3, df = 6, ps < .05) with the exception of observer-reported negative affect (i.e., chi-square = 10.4, df = 6, p = .11). These gender differences appeared to be driven by mean differences in predictor variables (e.g., gender differences in Communal Positive Emotionality; see Table 1). Given that we were primarily interested in whether gender moderates actor and partner effects, we relaxed constraints on means and variances for all subsequent models and tested whether actor and partner covariances could be constrained across men and women. The chi-square test, the comparative fit index (CFI), and the root-mean-square error of approximation (RMSEA) were used to assess model fit. We followed the convention that reasonable models should have values of CFI > .95, RMSEA < .06 given that the chi-square test alone is influenced by large sample sizes (see Bentler & Bonett, 1980; Hu & Bentler, 1999).

Table 1.

Descriptive Statistics For All Predictor and Outcome Variables

	Women Mean (SD)	Men Mean (SD)	Gender Difference (d)
Combined Sample (N = 794 couples):
Dyadic Adjustment Scale+	155.6 (18.0)	155.6 (16.6)	.00
Externalizing Psychopathology	6.8 (5.5)	8.1 (6.6)	−.22*
Communal Positive Emotionality	68.2 (14.3)	62.9 (16.2)	.35*
Agentic Positive Emotionality	59.6 (13.6)	63.1 (13.4)	−.26*
Negative Emotionality	29.8 (12.0)	30.0 (14.1)	−.01
Constraint	90.0 (11.7)	84.3 (12.1)	.48*
Observer-reported Positive Affect	3.9 (1.2)	3.6 (1.2)	.18*
Observer-reported Negative Affect	1.7 (1.0)	1.7 (1.7)	.01
Observer-reported Overall Satisfaction	3.7 (1.0)		N/A
Population-Based Sample (n = 432 couples):
Dyadic Adjustment Scale	155.5 (18.0)	156.1 (16.2)	−.04
Externalizing Psychopathology	6.7 (5.6)	7.7 (5.9)	−.18*
Communal Positive Emotionality	68.1 (14.2)	63.8 (15.9)	.29*
Agentic Positive Emotionality	59.1 (13.6)	63.4 (13.2)	−.32*
Negative Emotionality	29.5 (11.7)	29.2 (13.1)	.02
Constraint	89.9 (12.0)	84.5 (11.8)	.45*
Observer-reported Positive Affect	3.8 (1.1)	3.6 (1.1)	.20*
Observer-reported Negative Affect	1.6 (0.9)	1.6 (0.8)	.05
Observer-reported Overall Satisfaction	3.7 (1.0)		N/A
At-risk Sample (n = 362 couples)
Dyadic Adjustment Scale	155.7 (18.1)	154.9 (17.3)	.04
Externalizing Psychopathology	7.0 (5.5)	8.6 (7.3)	−.25*
Communal Positive Emotionality	68.3 (14.5)	61.7 (16.5)	.42*
Agentic Positive Emotionality	60.1 (13.6)	62.8 (13.6)	−.19*
Negative Emotionality	30.2 (12.3)	30.9 (15.1)	−.05
Constraint	90.2 (11.3)	84.0 (12.5)	.52*
Observer-reported Positive Affect	3.9 (1.3)	3.7 (1.3)	.16*
Observer-reported Negative Affect	1.9 (1.0)	1.9 (1.1)	−.02
Observer-reported Overall Satisfaction	3.8 (0.9)		N/A

Note. Cohens d provides gender difference effect sizes. Positive effect sizes indicate that women had higher means than men.

^*p < .05

⁺For reference purposes, means for women and men on the Dyadic Adjustment Scale based on the original 32-item scale are 145.7 (SD=17.6) and 145.6 (SD=16.0), respectively, in the overall sample.

Results

Descriptive Statistics

As seen in Table 1, means, standard deviations, and gender difference effect sizes are presented for each sample and for the combined sample. Positive effect sizes indicate that women reported higher mean levels of the variable than men. In the combined sample, men reported higher levels of externalizing psychopathology and agentic positive emotionality, whereas women reported higher levels of communal positive emotionality, constraint, and they were rated higher for positive affect by observers. There were no significant gender differences for reports of relationship adjustment (i.e., dyadic adjustment scale scores), negative emotionality, and observer-reported negative affect. Observer-reported satisfaction has only one score per couple (i.e., coders rated how satisfied they perceived each couple to be).

Although not reported in Table 1, effect sizes were also computed to compare across the samples (i.e., comparing men and women in the population-based sample to men and women in the at-risk sample). Both men and women in the at-risk sample had higher levels of observer-reported negative affect (i.e., ds = −.30 and −.23, respectively, ps < .05), and men in the at-risk sample were also more likely to report higher levels of externalizing psychopathology, though this effect was marginally significant (i.e., d = −.14, p = .054). The remaining sample differences were quite small and not statistically significant.

Correlations between all predictor variables are presented in Table 2. As seen there, greater externalizing psychopathology was associated with greater negative emotionality and observer-reported negative affect for both men and women. Greater communal positive emotionality, agentic positive emotionality, constraint, and observer-reported positive affect was associated with lower levels of externalizing psychopathology for men. For women, greater levels of constraint were associated with lower levels of externalizing psychopathology. These results suggest that externalizing psychopathology was related to self-reported personality traits as well as to observer reported indices of affect thereby underscoring the need to include personality and psychopathology in an overall model.

Table 2.

Correlations Between Externalizing Psychopathology, Personality, and Observer-Reported Positive and Negative Affect

	EXT	PEM-C	PEM-A	NEM	CON	PA	NA
Externalizing Psychopathology (EXT)	---	−.07	−.02	.56**	−.32**	−.07	.16**
Communal Positive Emotionality (PEM-C)	−.19**	---	.50**	−.19**	−.07*	.05	−.08*
Agentic Positive Emotionality (PEM-A)	−.13**	.54**	---	.00	−.04	.04	.01
Negative Emotionality (NEM)	.61**	−.22**	−.07	---	−.12**	−.09*	.17**
Constraint (CON)	−.29**	.03	−.01	−.13**	---	−.01	−.02
Observer-Reported Positive Affect (PA)	−.09*	.07	.01	−.15**	.00	---	−.35**
Observer-Reported Negative Affect (NA)	.21**	−.06	.03	.25**	−.06	−.35**	---

Note. N = 794 couples. Correlations for women are reported above the diagonal and correlations for men are reported below the diagonal.

^*p < .05

^**p < .01

Sample Differences

Prior to our primary analyses, we tested for possible differences between the population-based sample and the at-risk sample of the TBED-C via a series of nested APIMs using each predictor in a separate model (i.e., a simplified version of Figure 1 in which only one predictor was estimated at one time for men and women predicting relationship adjustment). Means and variances were first freely estimated across gender. We then tested whether means, variances, and actor and partner paths could be constrained to be equal across the two samples, with actor and partner effects also constrained across gender. We compared changes in CFI values across these models to determine the best fitting model for each predictor variable, using the suggestion the CFI change between models should not exceed .01 (Cheung & Rensvold, 2002). Although there was some evidence based on these tests that the two samples should not be combined for certain variables,¹ we ultimately chose to combine the two samples into one large sample for the current analyses in order to directly compare our results to those in Humbad et al. (2010). As is detailed below, the APIM for externalizing psychopathology, personality, and relationship adjustment across the combined sample fit the data quite well. Indeed, we examined externalizing psychopathology, personality, and relationship adjustment separately for each sample, and the estimates followed the same pattern for the combined sample (results available upon request). For consistency purposes, we also report APIM results using the combined sample for observer-reported positive affect and negative affect (separate sample results also available upon request).

Univariate Actor-Partner Interdependence Model Results

Results of the univariate APIM analyses can be found in Table 3. For each model, actor and partner paths were first constrained to be equal across gender (i.e., the actor effect was the same for men and women as was the partner effect), and means and variances for men and women were freely estimated. Fit statistics and path coefficients are displayed under “constrained model” in Table 3. As seen in Table 3, the constrained model generally provided a good fit to the data, as indicated by CFI values > .99 and RMSEA values < .06, and the majority of chi-square values were not statistically significant. However, the constrained model for observer-reported positive affect was a poor fit based on a significant chi-square value and RMSEA value > .06, and therefore path estimates from a fully saturated model (i.e., where actor and partner paths vary for men and women, 0 df) are reported. Results from the fully saturated model are not presented in instances where the constrained model was a good fit to the data as indicated by changes in CFI of less than .01 across the constrained and fully saturated models. Finally, the last two columns of Table 3 present constrained path estimates from the same univariate analyses conducted in Humbad et al. (2010) for comparison purposes.

Table 3.

Univariate Results: Individual APIM Results for all Predictors and Relationship Adjustment

Predictor	Fit Indices For Constrained Model			Path Estimates For Constrained Model		Path Estimates For Unconstrained Model				Humbad et al. (2010) Path Estimates
	χ2 on 2 df	CFI	RMSEA	a (aw= am)	p (pw= pm)	aw	pw	am	pm	a (aw= am)	p (pw= pm)
Externalizing	3.3	.997	.028	−.29**	−.17**	---	---	---	---	−.07**	−.04**
Communal PEM	.4	1	0	.27**	.10**	---	---	---	---	.30**	.15**
Agentic PEM	3.2	.996	.028	.15**	.07**	---	---	---	---	.06**	.03
Constraint	.6	1	0	.11**	.06*	---	---	---	---	.08**	.03
Negative Emotionality	.02	1	0	−.28**	−.12**	---	---	---	---	−.24**	−.10**
Positive Affect	12.2*	.974	.08	.22**	.06*	.38**	.17**	.08	−.06	N/A	N/A
Negative Affect	6.7*	.99	.054	−.28**	−.12**	---	---	---	---	N/A	N/A

Note. N = 794 couples. Models were first fit constraining actor (a) and partner (p) effects to be equal across women (w) and men (m; i.e., a_w= a_m; p_w= p_m). Nearly all models provided an excellent fit to the data, as indicated by the confirmatory fit index (CFI) values > .99 and root-mean-square error of approximation (RMSEA) values < .06. The model fit for Positive Affect was not adequate, and thus path estimates for the unconstrained or fully saturated model (i.e., a and p paths varied by gender) are also presented. Path estimates of the constrained model from Humbad et al. (2010) are provided in the last two columns for comparison purposes. Standardized variables were used, thus path coefficients can be interpreted as standardized estimates. APIM = actor-partner interdependence model; PEM = Positive Emotionality; N/A = not applicable.

^*p < .05

^**p < .01

Consistent with our hypotheses, greater levels of communal and agentic positive emotionality, and lower levels of negative emotionality and externalizing psychopathology were all associated with greater levels of relationship adjustment for individuals and their partners. Although we predicted to only find actor effects for constraint, we found greater levels of constraint were associated with greater levels of relationship adjustment for individuals and their partners. Similar to the results with self-reported personality traits and consistent with our hypothesis, greater observer-reported negative affect was associated with lower self-reported relationship adjustment for individuals and their spouses. Interestingly and inconsistent with our hypothesis, greater observer-reported positive affect in women was related to their own self-reported relationship adjustment and that of their partners, but there were no significant actor or partner effects of men’s observer-reported positive affect on relationship adjustment.

Multivariate Actor-Partner Interdependence Model Results

We then conducted a multivariate APIM in which we simultaneously estimated separate actor and partner effects for each of the predictor variables to evaluate their independent associations with relationship adjustment. In this model, agentic positive emotionality was excluded because it is less closely related to relationship dynamics (e.g., Donnellan et al., 2007) than communal positive emotionality, and to allow for direct comparisons with results from the multivariate model in Humbad et al. (2010), in which agentic positive emotionality was also excluded. Similar to the univariate model, we initially constrained actor and partner effects to be equal across gender while means and variances were freely estimated for men and women. Results are presented in Table 4. We also include the multivariate results of the same analyses reported in Humbad et al. (2010) in the last two columns. The constrained model was an excellent fit to the data (chi-square value = 5.4, 8 df, CFI = 1, and RMSEA = 0).

Table 4.

Multivariate Results: Predicting Relationship Adjustment from Externalizing Psychopathology and Relevant Personality Variables

Predictors	Path Estimates for Constrained Model		Humbad et al. (2010) Path Estimates for Constrained Model
	a (aw= am)	p (pw= pm)	a (aw= am)	p (pw= pm)
Externalizing Psychopathology	−.15**	−.12**	−.04**	−.03**
Communal Positive Emotionality	.21**	.05*	.23**	.11**
Negative Emotionality	−.15**	−.03	−.13**	−.04*
Constraint	.05*	.01	.07**	.02

Note. N = 794 couples. All predictor variables were placed into the same model. Estimates for each predictor variable therefore quantifies its unique association with relationship adjustment while controlling for the other predictor variables. This model was fit constraining actor (a) and partner (p) effects to be equal across women (w) and men (m; i.e., a_w= a_m; p_w= p_m) but different for each predictor variable. This model provided an excellent fit to the data, with χ2 (8) = 5.4, confirmatory fit index (CFI) = 1, and root-mean-square error of approximation (RMSEA) = 0. Path estimates of the constrained multivariate model from Humbad et al. (2010) are provided in the last two columns for comparison purposes. Standardized variables were used, thus path coefficients can be interpreted as standardized estimates.

^*p < .05

^**p < .01

Consistent with our hypothesis, greater levels of externalizing psychopathology remained significantly associated with lower relationship adjustment for individuals and their spouses even while controlling for relevant personality traits. Moreover, actor effects remained statistically significant for all personality variables, and communal positive emotionality also had a significant partner effect.

A multivariate APIM examining externalizing psychopathology, observer-reported positive affect and negative affect, and relationship adjustment was also estimated. As seen in Table 5, a model in which actor and partner effects were constrained across gender was first fit (“Constrained Model”) and provided a reasonable fit to the data (chi-square value = 21.9, 6 df, CFI = .973, RMSEA = .058), but given the significant chi-square value and the relatively smaller CFI, a fully saturated model in which actor and partner effects varied for men and women was also estimated. Results of this model suggest that for men and women, after controlling for the effects of observer-reported positive and negative affect, greater externalizing psychopathology remains significantly associated with lower levels of relationship adjustment for individuals and their partners. Interestingly, only women’s levels of observer-reported positive affect remained significantly associated with their own relationship adjustment (women’s partner effects and men’s actor and partner effects were nonsignificant). Greater observer-reported negative affect in women remained significantly associated with lower self-reported relationship adjustment for themselves and their partners, whereas only the actor effect for men was significant (i.e., greater observer-reported negative affect in men was associated with lower self-reported relationship adjustment for men). These results collectively suggest externalizing psychopathology has unique associations with relationship adjustment beyond both self-reported measures of personality traits and observer coded measures of individual characteristics.

Table 5.

Multivariate Results: Predicting Relationship Adjustment from Externalizing Psychopathology and Observer-reported Positive and Negative Affect

Predictors	Path Estimates for Constrained Model		Path Estimates for Unconstrained Model
	a (aw= am)	p (pw= pm)	aw	pw	am	pm
Externalizing Psychopathology	−.24**	−.14**	−.20**	−.16**	−.25**	−.11**
Positive Affect	.13**	.03	.24**	.08	.05	−.03
Negative Affect	−.19**	−.08**	−.28**	−.14**	−.11*	−.01

Note. N = 794 couples. All predictor variables were placed into the same model. Estimates for each predictor variable therefore quantifies its unique association with relationship adjustment while controlling for the other predictor variables. This model was fit constraining actor (a) and partner (p) effects to be equal across women (w) and women (m; i.e., a_w= a_m; p_w= p_m) but different for each predictor variable. This model was a poor fit to the data, with χ2 (6) = 21.9, confirmatory fit index (CFI) = .973, and root-mean-square error of approximation (RMSEA) = .06. Therefore, path estimates for the unconstrained or fully saturated model (i.e., a and p paths varied by gender) are presented and should be interpreted. Standardized variables were used, thus path coefficients can be interpreted as standardized estimates.

^*p < .05

^**p < .01

Externalizing Psychopathology, Personality, and Observer-Reported Satisfaction

A final analysis evaluated associations between externalizing psychopathology, personality traits, and observer-reported overall satisfaction. Given each couple had only one satisfaction code, the model was a multiple regression where we were able to test the plausibility of constraints. We first freely estimated coefficients for men and women for each predictor and then compared those coefficients to a second model with constrained estimates across gender. The change in R-square between these models was small (change in r-square = .004) and the constrained model fit the data well (chi-square value = 2.5, 4 df, CFI = 1, RMSEA = 0). Thus, the more parsimonious model where predictors were constrained across gender was used.

Results of the regression analysis are presented in Table 6. Surprisingly, only one predictor was significantly associated with observed relationship adjustment: negative emotionality. Greater negative emotionality was associated with lower levels of observer-reported satisfaction after controlling for all other variables (b = −.16, p < .01). Coefficients for all other predictors were small in size (i.e., ranging from .01-.05) and nonsignificant. Only 7% of the variance in observer-reported satisfaction was explained by this model.

Table 6.

Regression Results: Predicting Observer-reported Satisfaction from Externalizing Psychopathology and Personality

Predictors	Estimate (b)	Standard Error
Externalizing Psychopathology	.01	.03
Communal Positive Emotionality	.05	.03
Negative Emotionality	−.16**	.03
Constraint	−.01	.03

Note. N = 794 couples. All predictor variables were placed into the same regression and were constrained across gender. The R² value was .07. This model was a good fit to the data, with χ2 (4) = 2.5, confirmatory fit index (CFI) = 1, and root-mean-square error of approximation (RMSEA) = 0. Standardized variables were used, thus betas can be interpreted as standardized estimates.

^*p < .05

^**p < .01

Discussion

The primary goal of the current study was to extend analyses conducted in prior work (e.g., Donnellan et al., 2007; Dyrenforth et al., 2010; Humbad et al., 2010; Marshal, 2003; Savard et al., 2006) using another large sample of couples to evaluate associations between externalizing psychopathology, personality, and relationship adjustment. These analyses were guided by the framework of the vulnerability-stress-adaptation model (VSA; Karney & Bradbury, 1995), in which personality traits and psychopathology can both serve as enduring vulnerabilities (and strengths, in some cases) that can distally affect relationship quality. The current sample also used observational data to address concerns with shared method variance. Specifically, in addition to self-reported measures of personality and psychopathology, we also had observer-reported measures of each individual in the relationship, and we were able to use these observer-reports to test whether they also served as enduring vulnerabilities (or strengths) for relationship adjustment. We also examined whether personality and externalizing psychopathology were associated with observer-reported measures of relationship quality.

Our findings were largely consistent with those reported in Humbad et al. (2010), such that externalizing psychopathology and personality traits were associated with relationship adjustment, with estimates close in magnitude to those reported in Humbad et al. (2010). Specifically, we found that higher levels of both externalizing psychopathology and negative emotionality (i.e., the dispositional tendency to experience negative emotions) were associated with poorer relationship adjustment in both partners. Higher levels of both communal positive emotionality (i.e., the dispositional tendency to experience positive emotions) and constraint (i.e., the dispositional tendency to be behaviorally restrained) were associated with greater relationship adjustment in both partners. Moreover, given the considerable work linking personality traits to externalizing psychopathology (e.g., Krueger & Eaton, 2010), we estimated a model with both personality traits and externalizing psychopathology. Consistent with Humbad et al. (2010), there were unique effects of externalizing psychopathology on relationship adjustment in these analyses.

There were a few key differences between results in the current study and those found in Humbad et al. (2010). First, externalizing psychopathology was more strongly associated with relationship adjustment in the current study than in Humbad et al. (2010). One reason might be due to differences in externalizing measures across the two studies. Whereas Humbad et al. (2010) used a summation of DSM lifetime symptom counts based on clinical interview across three clinical disorders (adult antisocial personality disorder, conduct disorder retrospectively reported, and alcohol dependence), the current study used a broader (and more normally-distributed) measure of externalizing psychopathology (i.e., Adult Self Report; Achenbach & Rescorla, 2003) that covered items measuring a variety of aggressive, rule-breaking, and intrusive behaviors that are not all DSM symptoms of a specific disorder. Moreover, in the current study, participants reported only on their behaviors in the last six months, whereas in Humbad et al. (2010) symptoms were assessed across the lifespan. These substantive methodologic differences (a more normally-distributed measure of psychopathology assessed closer in time) could thus account for the stronger associations observed here. Yet another contributing factor may be shared-method variance, given that externalizing psychopathology and relationship adjustment were both based on self-report measures (whereas in Humbad et al., 2010, externalizing psychopathology was based on clinical interviews and relationship adjustment was based on self-report). Shared-method variance is always an issue with mono-informant designs, and thus future work should continue to use multiple informants to reduce such concerns. A final, more minor difference between the current and prior work is that the current study found evidence for partner effects with constraint (i.e., husbands’ and wives’ constraint was negatively associated with their partner’s report of relationship adjustment), an association that was not significant in Humbad et al. (2010). Nevertheless, we argue that the current study should be considered a relatively strong replication of Humbad et al. (2010), given that the overall patterns generally held across the studies.

We also extended prior work by Humbad et al. (2010) by examining observer-coded measures of the relationship. Women coded as displaying higher levels of positive affect were more likely to report higher levels of relationship adjustment. Their partners were also more likely to report higher levels of relationship adjustment. In contrast, observational levels of positive affect in men didn’t seem to matter for their own reports of relationship adjustment or their partner’s reports. Greater observed negative affect was associated with lower levels of relationship adjustment for individuals and their partners. Finally, in a multivariate analysis, greater externalizing psychopathology remained significantly associated with lower relationship adjustment for individuals and their partners even after controlling for observer-reported positive and negative affect, a finding which follows the same pattern as the multivariate model examining self-reported personality, psychopathology, and relationship adjustment. This result strengthens the role of externalizing psychopathology as an enduring vulnerability for couples.

To further address concerns about shared-method variance, we also examined the relationship between personality, externalizing psychopathology, and an observer-reported measure of overall satisfaction of the romantic relationship. Although we predicted associations between the predictors and relationship quality would be consistent across the self-report and observer-report measure of the relationship, we instead found that only greater negative emotionality was significantly associated with lower levels of observer reported satisfaction. Observer-reported satisfaction was moderately to strongly correlated with scores on the dyadic adjustment scale (i.e., .43 and .39 for women and men, respectively, ps < .01) attesting to the validity of the observational measure. However, self-reported externalizing psychopathology and other personality traits besides negative emotionality were unrelated to observer-reported measures of the romantic relationship. This result may suggest that different ways of assessing relationship variables may yield different patterns of results. We hypothesize that perhaps negative experiences may have been most salient in helping coders to determine couples’ levels of satisfaction during the interactions, and negative dispositions may have driven negative experiences. For example, if a couple is highly conflictual (driven by their dispositions), it may be easier to notice that they are not satisfied in their relationship versus a couple who is getting along.² In addition, only 10% of the sample was rated at “Extremely Low” or “Low” for perceptions of satisfaction, whereas 63% of the coded sample was rated “High” or “Extremely High” for perceptions of satisfaction, suggesting that for the most part, couples were rated by coders as generally happy and satisfied in the current sample.

The current study has limitations that should be noted. First, as with Humbad et al. (2010), only cross-sectional associations are reported so there are limited constraints on causal inferences. However, prior work examining longitudinal relationships similar to those reported in the current study find that variation in personality and psychopathology precede variation in relationship satisfaction (e.g., Robins, Caspi, & Moffitt, 2002). Second, there is debate over the distinction between psychopathology and personality, with many arguing that personality traits underlie expressions of psychopathology (e.g., Krueger & Eaton, 2010; Krueger, Caspi, Moffitt, Silva, & McGee, 1996). Indeed, there was a substantial amount of overlap between externalizing psychopathology and certain personality traits in the current study. Longitudinal designs may prove valuable for addressing this issue as well.

In sum, the findings from the current study are consistent with existing work and results reported in Humbad et al. (2010) and ultimately highlight a role for externalizing psychopathology in relationship distress and dissatisfaction. Individuals high in externalizing psychopathology are more likely to engage in substance use, antisocial behaviors that may cause harm to others, and they are more likely to have traits that can cause significant interpersonal problems like deceitfulness, manipulativeness, and impulsivity. These kinds of behaviors are often distressing to partners and undermine the ability of couples to cope with conflict and constructively solve their problems. Although the outlook for couples in which one or both members are higher on externalizing symptoms may not sound promising (e.g., Capaldi & Owen, 2001; Jones & Weiser, 2014; Marshall, Jones, & Feinberg, 2011), there have been some treatments that could help couples who struggle with this range of symptoms. Behavioral Couples Therapy, for example, has been shown to be effective in treating substance use disorders in couples (Powers, Vedel, & Emmelkamp, 2008). In addition, Integrative Behavioral Couples Therapy (Jacobson, Christensen, Prince, Cordova, & Eldridge, 2000) may be beneficial for individuals to foster greater acceptance for their partners who may bring enduring vulnerabilities to their relationship that can lead to greater stress. Unfortunately, for antisocial personality disorder and the range of behaviors and traits associated with this disorder, there has been limited research on the effectiveness of different treatments (see e.g., Gibbon et al., 2010, for a review). Given the results of the current study, efforts to identify such treatments would likely prove beneficial given the negative associations between externalizing problems and relationship challenges.