Romantic relationships engender positive and negative emotions, which can be powerful drivers of both adaptive and maladaptive behavior. Negative interactions and emotions afford prime contexts where intimate partner violence (IPV) may occur (McNulty & Helmuth, Reference McNulty and Hellmuth2008). Thus, the ability of both partners to regulate their emotions—involving not overreacting to their partner's negative affect yet supportively reacting to their positive affect, especially during times of disagreement—could protect individuals from engaging in physical IPV. Both psychological aggression and physical IPV have been directly linked to poorer emotion regulation abilities and higher levels of trait anger and hostility among couples (Norlander & Eckhardt, Reference Norlander and Eckhardt2005; Shorey, Brasfield, Febres, & Stuart, Reference Shorey, Brasfield, Febres and Stuart2011; Stith, Smith, Penn, Ward, & Tritt, Reference Stith, Smith, Penn, Ward and Tritt2004). Likewise, couples’ physical aggression has been shown to be positively associated with and predicted by couples’ self-reports of negative communication behaviors and feeling emotionally and physiologically overwhelmed or flooded, buttressing the association between violence and an inability to regulate emotions (Cornelius, Shorey, & Beebe, Reference Cornelius, Shorey and Beebe2010).

Couples’ communication patterns and observed affect, and their associations to the overall quality and stability of couples’ relationships, have long been a focus of research, given their importance for couples’ well-being and family stability. Measures of affect include both global or static measures (i.e., the overall level or amount of positive and negative affect expressed during conversation; Ackerman et al., Reference Ackerman, Kashy, Donnellan, Neppl, Lorenz and Conger2013; Gordis, Margolin, & Vickerman, Reference Gordis, Margolin and Vickerman2005; Johnson et al., Reference Johnson, Cohan, Davila, Lawrence, Rogge, Karney and Bradbury2005; Shortt, Capaldi, Kim, & Laurent, Reference Shortt, Capaldi, Kim and Laurent2010) and time-dynamic measures (i.e., the second-to-second unfolding of or dynamic changes in positive and negative affect states across time; Burman, Margolin, & John, Reference Burman, Margolin and John1993; Gottman, Swanson, & Murray, Reference Gottman, Swanson and Murray1999). Each of these measurement approaches has been used primarily to understand how interactions between partners impact, or are impacted by, macro-level relationship factors (e,g,, relationship satisfaction, psychological aggression, physical IPV, and relationship cessation; Johnson et al., Reference Johnson, Cohan, Davila, Lawrence, Rogge, Karney and Bradbury2005; Shortt et al., Reference Shortt, Capaldi, Kim and Laurent2010). Studies of observed affective interactions during couples’ conflict have usually involved samples of convenience (e.g., college student, flyer recruited, and distressed couples) and have often been limited to married couples (vs. including cohabiting and dating couples) or to samples selected by some other characteristic (e.g., samples with a violent husband). It is not clear how characteristics of dynamic affect reactivity during couples problem-solving discussions found in such studies would generalize to community samples, or be related to trait hostility and IPV perpetration in community samples. The purpose of the current study was to examine these questions for an at-risk (due to higher levels of delinquency in neighborhoods where they resided in childhood) community sample of young men and their women partners.

The conceptual model regarding the posited direct associations of each partner's trait hostility with each partner's IPV perpetration and indirect associations via affect reactivity is shown in Figure 1. The hypothesized associations are predicated on stress and emotion regulation theory in couples (Laurent & Powers, Reference Laurent and Powers2007), prior evidence of the negative effects of dysregulation or poor emotional control for relationships (Gottman & Levenson, Reference Gottman and Levenson1992), and in dynamic developmental systems (DDS) theory (Capaldi, Kim, & Shortt, Reference Capaldi, Kim, Shortt, Putallaz and Bierman2004; Capaldi, Shortt, & Kim, Reference Capaldi, Shortt, Kim, Pinsof and Lebow2005). The DDS framework emphasizes the simultaneous examination of individual characteristics (e.g., psychological traits) of both partners, and relationship-specific factors (e.g., relationship processes) to furthering our understanding of the development and occurrence of IPV in couples. Specific to the current study, the DDS theory is utilized within an actor–partner interdependence model (APIM; Kenny, Kashy, & Cook, Reference Kenny, Kashy and Cook2006) framework to examine whether the associations between each partner's trait hostility (individual characteristics) and physical IPV perpetration may operate through couples’ negative and positive affect dynamics (relationship processes). A dyadic approach is utilized allowing for both actor effects, examining the extent to which one's own hostility and affect reactivity predict one's own IPV perpetration, and partner effects, examining the extent to which IPV perpetration by a man or woman can be predicted by his or her partner's hostility and affect reactivity. This thus allows each partner to be simultaneously both a perpetrator and victim of IPV.

Figure 1.

Conceptual model: Men's and women's positive and negative affect reactivity, trait hostility, and physical intimate partner violence (IPV) perpetration. (+) denotes positive and (–) denotes externalizing negative affect reactivity. Solid and dashed lines denote actor and partner effects, respectively. Correlations between men's and women's hostility and all affect reactivity variables are not shown for parsimony.

IPV and negative emotionality (e.g., anxiety, anger, and hostility) are strongly associated for both men and women (Moffitt, Krueger, Caspi, & Fagan, Reference Moffitt, Krueger, Caspi and Fagan2000; White & Widom, Reference White and Widom2003). In a meta-analytic review of male perpetrators of physical IPV, perpetrators versus nonperpetrators and moderate-to-high perpetrators versus low-to-moderate perpetrators of physical IPV were found to have elevated levels of anger and hostility (Norlander & Eckhardt, Reference Norlander and Eckhardt2005). A review by Corvo and Johnson (Reference Corvo and Johnson2013) found that psychological risk factors, including anger and hostility, tend to be the strongest predictors of IPV. They are relatively proximal predictors, compared for example to family-of-origin risk factors or adolescent behaviors (Capaldi, Knoble, Shortt, & Kim, Reference Capaldi, Knoble, Shortt and Kim2012). In the current study, men's and women's trait hostility were included in the prediction models for two reasons. First, to examine whether negative and positive affect reactivity would add to prediction of IPV over and above this proximal trait, as poor emotional control and hostility are associated (Fosco, Caruthers, & Dishion, Reference Fosco, Caruthers and Dishion2012); and second, to test whether negative and positive affect reactivity mediate the association of trait hostility and IPV perpetration. Thus, it was hypothesized that negative affective reactivity would be positively associated with trait hostility and physical IPV and positive affective reactivity would be negatively associated with trait hostility.

Regarding effects of trait hostility versus affect reactivity, Madhyastha, Hamaker, and Gottman (Reference Madhyastha, Hamaker and Gottman2011) examined moment-to-moment spousal influence on affect during problem solving for two samples each recruited via news advertising. The first sample (N = 124) was selected to provide a uniform distribution of marital satisfaction and participants were on average aged in their 40s; the second was of newlyweds (N = 130) married in the last 6 months, both in first marriage and childless. Although they found some evidence of influence, they also found evidence that for many people their own initial state and emotional inertia dictated the outcome of the conflict discussion more than the moment-to-moment affect of the spouse. Thus, the current study represents an important further test of the hypothesis that, within couples, individuals show emotional reactivity related to their partner's affective expression and that such reactivity is related to both their overall level of hostility and their perpetration of physical IPV.

The propensity of one partner to reciprocate the other's negative affect quickly and consistently during conflict discussions may be a key aspect of emotional reactivity and risk for IPV. Burman et al. (Reference Burman, Margolin and John1993) studied a sample of convenience recruited via advertising of 65 married couples (in their middle to late 30s on average), with a child aged 3–18 years, who were recruited for a study of marital conflict and selected to fit particular conflict types (physically aggressive, verbally aggressive, withdrawing, and nondistressed/low conflict). They found that although most of the couples entered into angry affective states to some degree during conflictual interactions, for couples who reported physical IPV (vs. couples characterized as verbally aggressive, withdrawn, or low conflict), men and women (but particularly women) were more likely to reciprocate their partner's angry affect.

In addition to the study by Burman et al. (Reference Burman, Margolin and John1993), there is further evidence that there may be some gender differences in negative affect reactivity during discussions of conflicts, from work regarding demand and withdrawal behavior during marital interactions. Christensen and Heavey (Reference Christensen and Heavey1990) found for a sample of convenience (recruited through clinics and advertising) of 31 married couples (with a child aged 7–12 years, 18 of whom had a diagnosis of attention-deficit/hyperactivity disorder) that the wife–demand/husband–withdraw pattern was more likely than the husband–demand/wife–withdraw pattern. Patterns varied somewhat according to whether the issue under discussion involved the wife demanding a change of the husband's behavior or vice versa. However, the men were overall more withdrawn than were the women, suggesting that men may be more conflict avoidant than women, and thus possibly less likely to show negative affect reactivity to their partner's negative affect. Christensen, Eldridge, Catta-Preta, Lim, and Santagata (Reference Christensen, Eldridge, Catta-Preta, Lim and Santagata2006) examined demand–withdraw patterns for a cross-cultural sample of young university student couples in four countries recruited via class announcements and social contacts. Self-reports of communication patterns indicated that woman–demand/man–withdraw was significantly more likely than man–demand/woman–withdraw.

Eldridge, Sevier, Jones, Atkins, and Christensen (Reference Eldridge, Sevier, Jones, Atkins and Christensen2007) examined demand–withdraw communication among 182 married couples recruited via news advertising and seeking marital therapy (N = 68 severely distressed, N = 66 moderately distressed) or nondistressed couples (N = 48). The men and women were, on average, aged in their 40s and had been married for an average of 10 years. Using self-reports and rating scales, rather than moment-by-moment coding, Eldridge et al. found that greater demand–withdraw during relationship problem-solving discussions was associated with greater distress and that overall, wife–demand/husband–withdraw was greater than husband–demand/wife–withdraw. They found that behaviors varied by which partner was seeking change and which carried the burden of change; wife–demand/husband–withdraw was greater than husband–demand/wife–withdraw for wife-chosen topics, but not for husband-chosen topics, where there was no significant difference between the two interaction patterns.

This current study was also influenced by and builds on the work of Dr. Tom Dishion in a number of ways. Much of his work tested aspects of the coercion model (Dishion, Patterson, & Kavanagh, Reference Dishion, Patterson, Kavanagh, McCord and Tremblay1992), including work with his mentor Dr. Gerald Patterson. The coercion model involves interactions, initially within the family of origin, which reinforce problem or antisocial behaviors. These mechanisms were posited to involve negative reinforcement, whereby a young child would use aversive responses (termed coercive behaviors) to terminate aversive behaviors of parents or siblings (Patterson, Reid, & Dishion, Reference Patterson, Reid and Dishion1992). For example, a parent or sibling makes a request, the child ignores it, which on further repeated or escalated requests may result in yelled refusals to cooperate. If the parent or sibling then backs off from the demand, this reinforces the child's strategy. Positive reinforcement involves gaining someone's cooperation through aversive behavior (e.g., nagging and shouting at the mother until she lets the child have candy). Coercion theory posits that these practiced behaviors are a basis of antisocial and aggressive behaviors, as well as of related problem behaviors (e.g., substance use). Much of Dr. Dishion's work involved observation of influences in dyadic interactions and development of real-time codes to capture such moment-by-moment exchanges and social influences, including the Peer Process Code (Dishion et al., Reference Dishion, Crosby, Rusby, Shane, Patterson and Baker1989). A particularly innovative aspect of his work was to observe how exchanges based around positive reinforcement resulted in increases in problem behaviors for male deviant peer dyads (Dishion, Spracklen, Andrews, & Patterson, Reference Dishion, Spracklen, Andrews and Patterson1996).

More recently, Dr. Dishion extended his work in examining coercive processes within key relationships across the earlier life span to examination of processes in couples’ relationships, via observations of interactions. This work involved, first, a study of influences of observed positive and negative emotions on depressive symptoms using the Specific Affect Coding System (Ha, Dishion, Overbeek, Burk, & Engels, Reference Ha, Dishion, Overbeek, Burk and Engels2014), and second, a study examining prediction from disrupted parenting in early adolescence to use of coercive strategies—assessed as negative-hostile interactions, coercion (i.e., the way partners dismiss, invalidate, criticize, put down, or show contempt for each other), and commitment—at age 28–30 years (Ha, Otten, McGill, & Dishion, Reference Ha, Otten, McGill and Dishion2019). Ha et al. (Reference Ha, Otten, McGill and Dishion2019) found that disruptive parenting in early adolescence predicted romantic partner coercive strategies 15 years later, and that deviancy training with peers also predicted such coercive strategies. Thus, the study identified processes at key developmental stages related to risk for aggressive and destructive interactions in romantic relationships.

Method

Participants and procedures

Men were originally recruited at ages 9 to 10 years for the longitudinal Oregon Youth Study (OYS; 74% recruitment, N = 206). The boys were at increased risk for problem behaviors because of elevated neighborhood risk of delinquent behavior associated with the location of their public school. The men were assessed nearly annually to age 37 years with a participation rate of at least 88%. At age 18 years and approximately biannually thereafter, eligible men were invited to participate in the OYS–Couples Study of couples’ adjustment with their romantic partners. Data for the current study is from OYS–Couples Study assessment Wave 4 (91% participation rate) and includes N = 156 heterosexual couples (one same-sex couple was excluded to clarify the focus on men's and women's behaviors within male–female couples) when the men were ages 25 to 27 years, and Specific Affect Coding (SPAFF; Coan & Gottman, Reference Coan and Gottman2007; Gottman, McCoy, Coan, & Collier, Reference Gottman, McCoy, Coan and Collier1995) was conducted. Average ages were 26.14 (SD = 0.62) and 24.94 (SD = 4.07) for the men and women, respectively. Regarding ethnic origin, 90.4% of the men and 82.7% of the women were European American, with the next most common group being Hispanic. The average relationship length was 3.49 (SD = 2.85) years. Regarding relationship status, 42.3% (N = 66) of the couples were married, 37.2% (N = 58) were cohabiting, and 20.5% (N = 32) were dating. Average annual income for the couples was $38,160 (SD = 2,280).

Following internal review board approval and written informed consent, men and women completed interviews and questionnaires separately and videotaped discussions together (total duration of 34 min). Discussion topics included a warm-up (5 min), party planning (5 min), interpersonal problem solving (14 min total, 7 min for each partner's topic), and personal goals (10 min total, 5 min for each partner's topic). Each partner selected his or her relationship problem issue by rating a list of issues that tend to cause disagreements for couples (e.g., partner not doing share of chores) using the Partner Interaction Checklist (Capaldi, Reference Capaldi1991). Problem-solving discussions of the men's topics preceded the women's topics.

Measures

Observed affect

Men's and women's affect during the problem-solving discussions was coded using the SPAFF (Coan & Gottman, Reference Coan, Gottman, Coan and Allen2009; Gottman et al., Reference Gottman, McCoy, Coan and Collier1995). Affect codes were assigned on the basis of a gestalt of verbal content, verbal tone, context, facial expressions, and body movements. Each partner's affect was coded separately during multiple passes through the video for each discussion topic, as each individual was assigned a code during both speaking and listening turns. In each pass, coders focused on a single person of the dyad assigning SPAFF codes with the associated session time stamps to indicate changes in affect. An individual remained in a particular affective state until a change in affect was observed and coded (e.g., from neutral to positive affect), thus yielding continuous-time measures of each person's affect (i.e., time series) during the discussions. Partner's files were merged using session time.

Regarding the coding of different affect states, SPAFF has three overarching mutually exclusive categories of positive, neutral, and negative affect composed of 18 individual affect codes. To ensure adequate interrater reliability of the assignment of affect codes, 15% of couples’ topic discussions were independently coded by two different coders (κ = .83). Positive affect encompassed joy, humor, affection, validation, and interest. Internalizing negative affect serves a different function from externalizing negative affect (Shortt et al., Reference Shortt, Capaldi, Kim and Laurent2010). Thus, men's and women's negative affect was subdivided into externalizing negative affect (which included anger, domineering, belligerence, defense, disgust, contempt, criticism, and threats) and internalizing negative affect (which included fear or tension, sadness, whining, and stonewalling). However, transitions into and out of internalizing negative affect accounted for less than 2% of the total transitions and were therefore too sparse to be modeled accurately. Hence, men's and women's affect was collapsed into one of three categories: positive, neutral, and externalizing negative affect (the latter is hereafter referred to as negative affect). The time-varying transitions among these affective states were analyzed. Only affect during the two interpersonal problem-solving discussions was included, and men's and women's affective reactivity was analyzed within their own chosen topic discussion only. Thus, the men's affective reactivity, given their partners’ affect, was assessed from coding during the man's chosen relationship problem; the women's affective reactivity, given their partners’ affect, was assessed from coding during the woman's chosen relationship problem.

Trait hostility

Men's and women's trait hostility was measured by self-reports using the five-item hostility subscale (e.g., easily irritated) of the Brief Symptoms Inventory (Derogatis & Spencer, Reference Derogatis and Spencer1982). Likert response scales included five categories that ranged from 1 = not at all to 5 = very much. Men's and women's reliability equaled α = .75 and α = .55, respectively.

Physical IPV

Men's and women's physical IPV was measured using the perpetration and victimization physical assault subscales of the Conflict Tactics Scale (Straus, Reference Straus1979). Six items were answered twice by both partners, first in relation to their own perpetration (e.g., I pushed my partner) and second in relation to their victimization (e.g., My partner pushed me). Response scales included eight categories and were scored to reflect the approximate number of times in the last year physical IPV occurred (i.e., the annual frequency scores as described by Straus, Hamby, Boney-McCoy, & Sugarman, Reference Straus, Hamby, Boney-McCoy and Sugarman1996), where the midpoints of the response categories were used as follows: 0 = never or less than once a year and not in past year but before that; 1.5 = once or twice a month; 9 = several times a year but less than monthly; 18 = once or twice a month; 25 = several times a month, one or more times a week but less than daily, and daily). Reliability of men's and women's self-reported perpetration equaled α = .80 and α = .78, respectively, and self-reported victimization equaled α = .82 and α = .84, respectively. Composite IPV scores were created by summing over the six items.

Self-reported perpetration and partner-reported victimization composite scores were significantly associated (r = .95, p < .001 for men's perpetration and r = .75, p < .001 for women's perpetration), indicating congruence between self-reports and partner reports of each partner's behaviors. The men's perpetration report and the women's victimization report scores were averaged to denote the men's perpetration with a corresponding calculation to denote the women's perpetration. Scores were then rounded to the highest integer value to denote the number of times in the last year that the men and women, respectively, had perpetrated physical IPV. Finally, given that only three women and two men reported perpetration of physical IPV at a rate of greater than seven times in the last year, physical IPV scores were capped at seven. These scores were then used in the subsequent analyses. Frequencies of men's and women's rates of physical IPV perpetration in the past year are in Table 1.

Table 1.

Descriptive statistics of study variables

Data analytic plan

Model 1: Couples’ time-varying affect dynamics

The second-to-second changes in affect for men and women during their respective problem-solving discussions were modeled using advanced survival analysis techniques (Gardner & Griffin, Reference Gardner and Griffin1989). Specifically, time-varying influence models were used to predict the rate with which men and women transitioned from neutral into either positive or negative affect states during their own topic discussions as a function of their partners’ time-varying affect (Stoolmiller & Snyder, Reference Stoolmiller and Snyder2006). Note that these are only two of the six possible transitions among affect states. Transitions from positive to either neutral or negative and from negative to either neutral or positive were not examined. Neutral affect was chosen as the reference category for both statistical and substantive reasons. Statistically, the low base rates of positive-to-negative transitions (1.6% of men's and 1.0% of women's transitions) and negative-to-positive transitions (1.2% of men's and 0.8% of women's transitions) were too sparse to model accurately. Substantively and more important, choosing neutral affect as the reference category allowed for the examination of how the time-varying changes in partners’ supportive, positive affect and unsupportive, negative affect (vs. neutral affect) may provoke the other partner to transition out of neutral affect into either positive or negative affect.

Survival analyses are time-to-event models, where an estimate of the likelihood of an event occurring over a particular time interval is obtained, provided that the event has not yet occurred during that interval. Formally, this quantity is known as the hazard rate and is denoted as h(t), which is a function of time, t. The higher the estimated value of the hazard rate, the sooner the event occurs. Events in the current analyses were men's and women's transitions from neutral into either positive or negative affect during the problem-solving discussions of their own respective topics. The survival analysis models used have three advanced components, namely, multiple events, competing risks, and multilevel analyses. First, multiple events indicate that multiple transitions from neutral into either positive or negative affect occurred across the course of the 7-min discussions. This differs from traditional survival analyses where events, such as death, can only occur once (cf. Allison, Reference Allison1984). Second, the models also allow for competing risks. Specifically, from neutral affect, it is possible to transition into positive affect or negative affect, but not both; thus, the positive and negative affect states have competing risks of occurring. Third, the models are multilevel; intraindividual variability in the second-to-second changes in observed affect states (Level 1) were nested within persons (Level 2).

Estimates of men's and women's positive and negative affect reactivity during their respective topic discussions were allowed to correlate and were simultaneously estimated in one model using Mplus version 7 (Muthén & Muthén, Reference Muthén and Muthén1998–2012). The baseline hazard rates represent the average rates with which men and women transitioned from neutral into positive or negative affect states, respectively (i.e., grand means). Tests of congruent affect dynamics captured the rate with which men and women transitioned into negative (or positive) affect states, given their partners were also negative (or positive), whereas tests of incongruent affect dynamics captured the rate with which men and women transitioned into negative and positive affect states, given that their partners were in opposing affect states (e.g., transition into positive, given that their partner was negative). Finally, two random intercept effects (more formally referred to as strata effects, see Appendix A) were included for both men and women, indicating the deviation above or below the overall baseline hazard rates (i.e., the grand means) of positive and negative affect transitions, respectively. These random effects were correlated within person for both men and women and were expected to be negatively related; that is, the person who transitions the fastest into positive affect is expected to be the slowest to transition into negative affect. Correlations across partners and across positive and negative affect reactivity estimates were also allowed in the model. It was expected that those men and women who transition into positive affect states the fastest would have partners who also were quick to transition into positive affect states but slow to transition into negative affect states (and similarly for negative transitions). See Appendix A for further details regarding the event history model.

The person-specific estimates (i.e., random effects) of the rates with which men and women transitioned from neutral into positive and negative affect states (conditioned on the average partners’ affect) then served as mediating variables in the subsequent APIM (Figure 1) analysis examined in Model 2. This approach has the limitation of less consistent estimates due to unaccounted-for measurement error in the estimates (Buonaccorsi, Reference Buonaccorsi2010). However, attempting the simultaneous estimation of both the survival analysis (which modeled men's and women's affect reactivity) and the associations among men's and women's affect reactivity and their trait hostility and perpetration of physical IPV resulted in an unwieldy model (i.e., convergence issues due to the number of estimated parameters).

Model 2: Couples’ affect dynamics, trait hostility, and physical IPV

In the model relating men's and women's trait hostility, affect dynamics, and physical IPV perpetration (Figure 1), actor effects are depicted as solid lines—denoting associations between one's own traits (e.g., hostility) and behaviors (e.g., IPV), whereas partner effects are depicted as dashed lines—denoting associations between partners traits and one's behaviors (e.g., IPV). The model was estimated as a path analysis in Mplus version 7 (Muthén & Muthén, Reference Muthén and Muthén1998–2012). The anteceding variables (men's and women's trait hostility) were correlated, as were the mediating variables (estimates of men's and women's positive and negative affect reactivity previously derived in Model 1). The outcome variables (men's and women's rates of physical IPV perpetration in the last year) were modeled using negative binomial distributions to account for the low prevalence of men's and women's physical IPV perpetration (19% and 20%, respectively) and the overdispersion in the data involving instances where the variability in the rate of physical IPV perpetration exceeds the average level of perpetration (Agresti, Reference Agresti2013). Overdispersion indicates that physical IPV perpetration was rare, but when it did occur, there was substantial variability in the rate with which it happened. Note that, given the use of count variables, there were no residual variances for men's and women's rates of physical IPV perpetration (Agresti, Reference Agresti2013); as such, the outcome variables could not be correlated.

Results

Sample descriptives

Sample descriptives for trait hostility, rate of IPV perpetration, and observed affect are presented in Table 1. Men's and women's average trait hostility and prevalence of physical IPV perpetration (19% and 20%, respectively) were similar. Of those men and women who perpetrated physical IPV, approximately one half of them committed one such act in the last year. Regarding observed affect, couples spent most of their discussion time in neutral affect and were most likely to transition into neutral affect (which accounted for approximately 80% of their total time and nearly 50% of their total transitions), followed by positive affect (which accounted for approximately 10% of their time and 38% of their total transitions) and negative affect (which accounted for approximately 9% of their time and 13% of their total transitions).

Model 1: Couples’ time-varying affect dynamics (Aim 1)

Model results for the survival analysis are presented in Table 2. During couples’ problem-solving discussions, men's and women's transitions into positive affect (from neutral affect) occurred at a faster rate if the partner was positive in affect and at a slower rate if the partner was negative in affect, thus exhibiting support for positive affect dynamics between partners. Alternatively, women (but not men) were significantly more likely to transition into negative affect if their partners were negative in affect, whereas men (but not women) were significantly less likely to transition into negative affect if their partners were positive. Thus, partial support was found for negative affect dynamics between partners, and support was found for the hypothesis that negative affect reactivity was more likely for women than for men.

Table 2.

Couples’ affect dynamics results (Model 1): Fixed effects and random effects

Note: (+) and (–), respectively, denote that men and women were more and less likely to transition.

^Mp < .10. *p < .05. **p < .01. ***p < .001.

Regarding the estimates of the random effects, both men and women exhibited significant variation in the rates with which they transitioned into negative and positive affect during their problem-solving discussions. Findings for covariances among transition rates (Table 2) indicated that those men and women who were the fastest to transition into negative affect were the slowest to transition into positive affect. In addition, rates of negative and positive transitions were positively associated for men and women within couples. Thus, those men and women who were quicker to transition negatively (or positively) during their own topic discussions had partners who were also quicker to transition negatively (or positively) during the partner's chosen topic discussion. The one exception was that men who transitioned into negative affect more slowly did not have partners who were faster to transition into positive affect during the women's topic discussions. Thus, support was found for the expected correspondence in affective reactivity across partners.

Model 2: Couples’ affect dynamics, trait hostility, and physical IPV (Aims 2 and 3)

Bivariate associations

Shown in Table 3 are the bivariate correlations among all of the study variables, including men's and women's affect reactivity scores (with higher values denoting faster affect transitions), which were estimated as random effects in the survival analysis (Model 1). Correlations indicated that more hostile men and women were quicker to react with negative affect and slower to react with positive affect, both within and across partners; the only exception was that women's positive affect reactivity was not related to men's trait hostility. Furthermore, men's and women's perpetration of physical IPV was positively related to both partners’ trait hostility levels, whereas only women's (but not men's) physical IPV perpetration was associated with women's quicker transitions into negative and slower transitions into positive affect.

Table 3.

Bivariate correlation matrix of the study variables

Note: Men's and women's externalizing negative and positive affect reactivity denote estimates of the random effects from the survival analysis (Model 1). Higher values denote quicker transitions into externalizing negative or positive affect states from neutral states. Physical intimate partner violence (IPV) denotes the number of times in the last year physical violence was perpetrated by men or women.

^Mp < .10. *p < .05. **p < .01. ***p < .001.

Path analysis results

Model results for the path analysis are presented in Table 4, including path coefficients and incidence rate ratios, which denote the change in the rate of IPV perpetration for a one-unit increase in the predictor variable. Note that effects are for the full path model and thus account for other associations in the model. Regarding prediction of rate of affect transitions from trait hostility, as hypothesized, more hostile men and women were quicker to transition into negative affect compared to less hostile men and women. In comparison, only women's (but not men's) higher trait hostility predicted slower positive affect transitions for both men and women, suggesting that only women's levels of trait hostility impaired the positive affect dynamics for couples during their problem-solving discussions.

Table 4.

Path analysis results (Model 2): Couples’ affect reactivity, trait hostility, and physical intimate partner violence (IPV) perpetration

Note: Incident rate ratios denote the change in the rate of IPV perpetration for a 1-unit increase in the corresponding predictor variable. Incident rate ratios above, equal to, and below 1, respectively, denote an increase, no change, and decrease in the rate of physical IPV perpetration.

^Mp < .10. *p < .05. **p < .01. ***p < .001.

Regarding prediction of physical IPV perpetration from trait hostility, both more hostile men and more hostile women perpetrated physical IPV at higher rates than those men and women who were less hostile. However, regarding partner effects, only women's hostility significantly predicted men's physical IPV perpetration. Men's trait hostility was not significantly predictive of women's IPV perpetration.

Regarding prediction of physical IPV perpetration from affect reactivity, women's externalizing negative affect reactivity predicted the women's IPV perpetration, whereas the corresponding association for men was not significant, as were effects of either men's or women's negative affect reactivity on their partners’ physical IPV perpetration. Associations of men's and women's positive affect reactivity with their IPV perpetration were not significant.

Findings regarding indirect effects of hostility on IPV mediated by affect reactivity (lower part of Table 4) were consistent with the findings for direct effects. The only significant indirect effect indicated that more hostile women were quicker to transitions into negative affect states and to perpetrate physical IPV at over twice the rate of women who were less hostile and showed lower levels of negative affect reactivity. The estimated covariance matrix for Model 2 denoting associations between each partner's hostility and negative and positive affect reactivity is shown in Appendix A.

In sum, more hostile men and women committed more acts of physical aggression toward their partners, but only women's hostility increased their risk for greater victimization. In general, there was support for the expectation that both women's hostility and their negative affect reactivity play a role in explaining IPV involvement, whereas positive affect reactivity was not predictive of IPV for either partner; thus, evidence did not support it as a protective factor. Furthermore, the rate with which women reacted to their partners’ negative affect by also becoming negative was found to partially explain the association between women's trait hostility and their IPV perpetration, thus identifying women's negative affect reactivity as a mediating, relationship process variable.