Marital Conflict Behaviors and Implications for Divorce over 16 Years

Abstract

This study examined self-reported marital conflict behaviors and their implications for divorce. Husbands and wives (N = 373 couples; 47% White American, 53% Black American) reported conflict behaviors in years 1, 3, 7, and 16 of their marriages. Individual behaviors (e.g., destructive behaviors) and patterns of behaviors between partners (e.g., withdrawal-constructive) in Year 1 predicted higher divorce rates. Wives’ destructive and withdrawal behaviors decreased over time, whereas husbands’ conflict behaviors remained stable. Husbands reported more constructive and less destructive behaviors than wives and Black American couples reported more withdrawal than White American couples. Findings support behavioral theories of marriage demonstrating that conflict behaviors predict divorce and accommodation theories indicating that conflict behaviors become less negative over time.

A large percentage of marriages end in divorce with estimates ranging from 40 to 50% (Bramlett & Mosher, 2002; Rogers, 2004). Behavioral theories of marriage have suggested that conflict behaviors have important implications for couples’ evaluations of their marriages and divorce (Karney & Bradbury, 1995; Kelly, Fincham, & Beach, 2003). Indeed, destructive conflict behaviors (e.g., criticism, yelling) predicted increased divorce (Gottman, Coan, Carrere, & Swanson 1998; Orbuch, Veroff, Hassan, & Horrocks, 2002). Much less is understood about the implications of constructive (e.g., calm discussion, listening) and withdrawal behaviors (e.g., keeping quiet, leaving to cool down) for divorce. In addition, little is known about whether these conflict behaviors remain stable or change over time among couples who stay married.

This study specifically examined self-reported destructive, constructive, and withdrawal behaviors among newlywed White and Black American couples in their first year of marriage (1986) and again in years 3, 7, and 16. The study had three goals. First, we considered whether self-reported conflict behaviors used in the first year of marriage predict how long couples remain married over 16 years. This will provide vital information on whether conflict behaviors, other than the often studied destructive behaviors, are influential for divorce over time. Second, we assessed whether conflict behaviors present in the early years of marriage are consistent or change over time among couples who stay together. This will allow for an unprecedented longitudinal investigation regarding whether couples in long-term marriages become better able to manage conflict over time. Lastly, we examined whether the implications of conflict behaviors for divorce and conflict behavior trajectories vary by gender and race. It is critical to examine marriages among diverse groups of people, as the meaning of conflict behaviors and their implications may vary by gender and race (Orbuch et al., 2002).

Implications of Conflict Behaviors for Divorce

Researchers often categorize conflict behaviors as destructive, constructive, or withdrawal (Crohan, 1996; Kurdek, 1995; Oggins, Veroff, & Leber, 1993; Pasch & Bradbury, 1998). Destructive behaviors include overtly negative reactions to marital problems such as yelling, insults, criticism, belligerence, and contempt. Constructive behaviors involve overtly positive reactions such as saying nice things, calmly discussing the problem, and actively listening. Withdrawal behaviors entail disengaging from the conflict or person and may include leaving the situation or keeping quiet. These three categories are certainly not the only way of grouping conflict behaviors but they do subsume other categorizations. For example, researchers have defined conflict behaviors as negative or positive affect expression (Gottman et al., 1998), hostile or warm (Matthews, Wickrama, & Conger, 1996), and negative, positive, or disengaged (Smith, Vivian, & O’Leary, 1990). Regardless of the terms, conflict behaviors may influence couples’ marriages.

According to behavioral theories of marriage, destructive behaviors result in negative evaluations of marriage and declines in marital satisfaction and stability, whereas constructive behaviors lead to improvements in evaluations of marriage and increases in marital satisfaction and stability (Karney & Bradbury, 1995; Kelly et al., 2003). Gottman and colleagues (1994, 1998) found that destructive behaviors (e.g., criticism, defensiveness, and contempt) used in observed interactions predicted divorce among newlyweds up to 7 years later and among longer married couples (married an average of 5 years) up to 14 years later (Gottman & Levenson, 1992, 2000, 2002). Although there are several critiques of Gottman’s work due to nonrandom samples and a lack of control variables (Stanley, Bradbury, & Markman, 2000), other studies have revealed similar results. For example, studies show that destructive behaviors used premaritally and in the first year of marriage predicted divorce from 4 to 14 years later (Clements, Stanley, & Markman, 2004; Orbuch et al., 2002; Rogge & Bradbury, 1999) and greater observed destructive behaviors (i.e., hostility) in long-term marriages (married an average of 18 years) predicted divorce up to five years later (Matthews et al., 1996).

Less work has examined links between constructive conflict behaviors and marital longevity, and the limited research findings are contradictory. Gottman and colleagues (1998) concluded that constructive conflict behavior (i.e., active listening) was not effective due to low frequencies in observed interactions and no association with divorce. On the contrary, Matthews et al. (1996) found that low levels of observed constructive behaviors predicted divorce in long-term marriages.

Similarly, the findings are inconsistent with regards to withdrawal conflict behaviors and divorce. Observations of withdrawal predicted divorce up to 7 years later among newlyweds (Gottman, 1994; Gottman et al., 1998) and up to 14 years later among longer married couples (Gottman & Levenson, 2002). Gottman and Krokoff (1989) suggested that withdrawal may lead to the deterioration of marriages over the long term perhaps because problems are left unresolved and that there is increased distance and alienation. In contrast, others have found no association between withdrawal and later marital outcomes (Bradbury, Campbell, & Fincham, 1995; Pasch & Bradbury, 1998).

In addition to specific behaviors, certain related interactional patterns between the husband and wife, such as demand-withdraw and negative reciprocity, may be predictive of divorce. For example, the demand-withdraw pattern, in which one partner makes demands (e.g., criticizes, makes complaints) and the other partner attempts to withdraw from the interaction, predicted decreased marital satisfaction over 2.5 years (Heavey, Christensen, & Malamuth, 1995). Negative reciprocity, in which both partners use destructive behaviors portended early divorce (within the first 7 years of marriage; Gottman & Levenson, 2002).

Overall, contradictory results regarding constructive and withdrawal behaviors may be due to the use of predominately observational methods of small and nonrandom samples. Furthermore, studies frequently have not controlled for important sociodemographic factors that differentiate couples. The present study sought to examine behavioral theories of marriage by investigating three types of conflict behaviors and their implications for divorce using a larger, more diverse, sample.

Conflict Behaviors over Time

Another major question in the marital literature is whether the behaviors present in the early years of marriage are consistent or change over time. Behavioral theories of marriage, such as social learning theory or the enduring dynamics model, have proposed that couples enter marriage with individual differences that remain consistent over time and eventually predict divorce (Caughlin, Huston, & Houts, 2000; Huston, Caughlin, Houts, Smith, & George, 2001). Thus, according to these theories, couples either show consistent conflict behaviors or increases in conflict behaviors over time. In support of these theories, research examining couples over relatively short periods of time (2 to 4 years) revealed destructive and constructive conflict behaviors were stable over time (Gottman & Levenson, 1999; Kelly, Huston, & Cate, 1985).

In contrast, developmental and accommodation theories have suggested that conflict behaviors may become less negative over time as couples become more tolerant of one another (Carstensen, Gottman, & Levenson, 1995; Lindahl, Clements, & Markman, 1998). In one study of long-term marriages, couples retrospectively reported that their marriages became more enjoyable, tolerant, comfortable, and that communication improved over time (Robinson & Blanton, 1993). Carstensen and colleagues (1995) found that older married couples were more affectionate and less negative with one another compared to younger couples. Similarly, Lindahl et al. (1998) found that self-reported use of withdrawal and destructive behaviors decreased over 10 years of marriage among a small sample of 36 couples. The researchers hypothesized that, over time, couples become more tolerant and better able to use constructive rather than destructive or withdrawal behaviors to resolve conflict.

Conflict Behaviors: The Contexts of Gender and Race

Longitudinal studies of marital conflict and divorce often do not compare husbands and wives in the same models and often lack racial and ethnic diversity (Karney & Bradbury, 1995). Yet, the consequences of conflict behaviors and the trajectories of conflict behaviors over time may vary by both gender and race. The implications of conflict behaviors for divorce may vary depending on whether the husband or wife uses the behavior but the research is inconsistent. Orbuch et al. (2002) found that both husbands and wives who reported using destructive behaviors had higher rates of divorce. Others studies reveal gender differences in these links. For example, Gottman et al. (1998) found that wife’s negative start up (i.e., destructive) and lack of de-escalation of negative affect by the husband was associated with divorce.

Studies also have shown that the use of conflict behaviors varies by gender. In particular, wives used more destructive behaviors and husbands used more withdrawal and constructive behaviors (Carstensen et al., 1995; Levenson, Carstensen, & Gottman, 1994). These findings are mostly from young White American couples in the earlier years of their marriages. There is less information regarding whether husbands and wives have different trajectories of conflict behaviors over time in a marriage.

In addition, the implications and use of conflict behaviors may vary between White and Black American couples. Compared to White Americans, Black Americans begin their marriages with greater risk factors, stressors, and challenges that may spill over into their marriage (LaVeist, 2005; Orbuch, Veroff, & Hunter, 1999). For example, Black American couples reported lower income, were more likely to have premarital children, and more likely to have cohabited before marriage than White American couples (Orbuch et al., 2002). These stressors may influence the implications of conflict and the types of behaviors that couples use.

Researchers have found conflicting results regarding whether there are race differences in the consequences of conflict behaviors for marriage. Orbuch, Veroff, and Holmberg (1993) discovered that the behavioral display of conflict was a predictor of marital well-being in the early years of marriage among White American couples, but not Black American couples. They argued that Black Americans interpreted behaviors typically defined as destructive (e.g., interrupting, negative tone of voice) as acceptable and less problematic. Conversely, when examining the implications of conflict for divorce, Orbuch and colleagues (2002) found that destructive conflict behaviors increased the risk of divorce over 14 years among both Black and White American couples.

A small number of studies have revealed race differences in conflict behaviors (Broman, 2005; Oggins et al., 1993). Black American couples reported more destructive and withdrawal behaviors than White American couples (Broman, 2005; Oggins et al., 1993). It is interesting to note, however, that Adelmann, Chadwick, and Baerger (1996) discovered that older Black American couples reported using fewer destructive behaviors than younger Black American couples, suggesting that destructive behaviors may decrease over time. It is not clear, however, whether there are race differences in conflict behaviors that persist over time in marriages.

Other Sociodemographic and Life Course Factors

We also considered sociodemographic and life course variables that may influence divorce and conflict behaviors as covariates. Sociodemographics including lower education, lower income, and more equal ratios of income between the husband and wife have predicted higher divorce rates, although the effects varied in some studies (Orbuch et al., 2002; Rogers, 2004).

In addition, several life course factors predict divorce. Couples who had children prior to marriage, who cohabited before marriage, who grew up with divorced parents, and who married younger had higher rates of divorce (Amato, 1996; Bumpass, Martin, & Sweet, 1991; Clements et al., 2004; Orbuch et al., 2002; Rodrigues, Hall, & Fincham, 2006). These variables also have been associated with conflict behaviors. For example, husbands and wives from divorced families of origin had greater marital conflict (Sanders, Halford, & Beherens, 1999; Whiton, Rhodes, Stanely, & Markman, 2009).

Present Study

Although studies have shown that destructive conflict increased the risk of divorce, many unanswered questions remain. The findings regarding constructive and withdrawal behaviors have been inconsistent and most studies have examined observed conflict behaviors within small samples of White American couples. In addition, prior studies have not assessed conflict behaviors over long periods of time nor controlled for the sociodemographic and life course factors that differentiate couples. The current study addresses the gaps in the literature by investigating self-reported conflict behaviors among a diverse sample of Black American and White American newlywed husbands and wives over 16 years of marriage.

Although we recognize the limitations of self-reports of behavior (e.g., memory biases, Robinson & Clore, 2002), they provide a useful and important tool for measuring marital processes that may not be observable. For example, it is difficult to examine withdrawal in the laboratory as many of the behaviors, such as leaving the room, are not possible in this setting (Roberts, 2000). In addition, self-reported perceptions such as attributions and emotions are highly associated with objective marital outcomes including destructive behaviors in the laboratory and divorce (Bradbury, Beach, Fincham, & Nelson, 1996; Karney & Bradbury, 2000; Orbuch et al., 2002).

In summary, the current study had three overall aims. First, we examined whether the conflict behaviors reported in the first year of marriage predict divorce over 16 years. We hypothesized that consistent with behavioral theories of marriage, couples who reported greater constructive and less destructive and withdrawal behaviors would have lower divorce rates than couples who reported less constructive behaviors and greater destructive and withdrawal behaviors (Gottman & Levenson, 2002; Matthews et al., 1996). We also predicted that husbands and wives who engaged in destructive reciprocity or demand-withdrawal patterns (destructive-withdrawal, constructive-withdrawal) would have higher divorce rates (Gottman, 1994; Heavey et al., 1995).

Secondly, we considered whether conflict behaviors remain stable or change over time among couples who stay married. Due to the extended length of this study, we postulated that spouses would accommodate to one another and report less destructive and withdrawal behaviors and greater constructive behaviors over time (Lindahl et al., 1998).

Thirdly, we assessed gender and race differences in the implications of conflict behaviors for divorce and in the use of conflict behaviors over the course of marriage. Based on previous work, we predicted that conflict behaviors would have similar effects on divorce for husbands and wives and Black American and White American couples (Orbuch et al., 2002). We also hypothesized that wives would report more destructive, less constructive, and less withdrawal behaviors than husbands (Carstensen et al., 1995; Oggins et al., 1993), and that Black American couples would report more destructive and withdrawal behaviors than White American couples (Broman, 1995; Oggins et al., 1993).

Method

Participants

Participants were from the Early Years of Marriage Project (EYM), a longitudinal study of married couples. Couples were selected from those who applied for a marriage license in Wayne County, Michigan from April through June 1986. Eligible couples included those of the same race, in their first marriage, with a wife younger than 35. All eligible Black American couples and a random sample of White American couples were asked to participate. A total of 66% of those contacted participated. The sample included 373 (174 White American and 199 Black American) newlywed couples in Year 1 (1986). On average, husbands were age 27 and wives were age 25 and they were between the first four to nine months of the marriage. Table 1 includes marital status of participants over time and Table 2 includes a description of study variables by Year.

Table 1.

Marital Status of Couples in each Year of the Study (n = 373)

	Year 1	Year 3	Year 7	Year 16
Marital Status	n (%)	n (%)	n (%)	n (%)
Married	373 (100)	304 (81.5)	242 (64.9)	183 (49.1)
White American	174 (100)	155 (89.1)	135 (77.6)	108 (62.1)
Black American	199 (100)	149 (74.9)	107 (53.8)	75 (37.7)
Divorced/Separated		52 (13.9)	108 (29.0)	172 (46.1)
White American		16 (9.2)	35 (20.1)	63 (36.2)
Black American		36 (18.1)	73 (36.7)	109 (54.8)
Ineligible				14 (3.8)
White American				3 (1.7)
Black American				11 (5.5)
Unknown		17 (4.6)	23 (6.2)	4 (1.1)
White American		3 (1.7)	4 (2.3)	0 (0.0)
Black American		14 (7.0)	19 (9.5)	4 (2.0)

Note: Ineligible respondents were deceased/widowed or severely ill. There were 4 couples who were separated or divorced in Year 14, but remarried their Year 1 spouse by Year 16. Percentages were calculated by taking the number of participants in the particular category over the number of participants in the first year (e.g., Total, White American, or Black American).

n = 373 represents the total number of couples.

Table 2.

Descriptive Statistics on Sociodemographics, Life Course Variables, and Conflict Behaviors (n = 746)

		Year 1				Year 3				Year 7				Year 16
		White		Black		White		Black		White		Black		White		Black
Variables		Hus.	Wife	Hus.	Wife	Hus.	Wife	Hus	Wife	Hus.	Wife	Hus.	Wife	Hus.	Wife	Hus.	Wife
Sociodemographics and life course variables
Education in years	M	13.3	13.0	12.9	13.1
	SD	1.9	1.9	1.8	1.8
Premarital children	M	.2	.2	.5	.5
	SD	.4	.4	.5	.5
Cohabited in months	M	7.1	7.1	15.7	15.8
	SD	14.1	14.5	24.0	24.4
Not raised w/ 2 parents	M	.2	.3	.5	.6
	SD	.4	.5	.5	.5
Age at marriage	M	25.5	23.6	26.9	24.6
	SD	3.6	3.6	4.3	3.9
Income	M	36705	35239	26656	25504	41754	40592	36851	36174	55505	52820	52527	50277	68553	68923	62705	60545
	SD	18328	16940	16522	16687	18257	18103	18390	18541	18448	18217	19683	23189	17738	17381	21287	23230
Income ratio wife/hus.	M	.6	.6	.8	.8	1.0	1.0	.7	.7	1.1	1.1	.9	.9	.4	.4	1.1	1.1
	SD	.4	.4	2.0	1.9	3.5	3.5	.7	.7	5.5	5.5	1.2	1.2	.5	.5	1.6	1.6
Conflict behaviors
Destructive	M	1.8	2.0	2.0	2.2	1.9	2.0	2.0	2.2	1.8	1.9	1.8	2.1	1.8	1.7	1.8	2.0
	SD	.7	.8	.8	.8	.7	.8	.8	.7	.6	.7	.5	.7	.7	.6	.7	.7
Constructive	M	2.9	2.6	2.9	2.6	2.6	2.5	2.9	2.6	2.6	2.5	2.7	2.6	2.7	2.6	2.8	2.6
	SD	.6	.6	.7	.7	.6	.6	.7	.6	.6	.6	.6	.7	.5	.6	.6	.7
Leaving withdrawal	M	1.7	1.6	2.2	2.2	1.9	1.9	2.3	1.9	2.2	2.2	2.1	2.0	1.9	1.9	2.1	1.8
	SD	1.0	1.0	1.2	1.3	1.0	1.0	1.2	1.1	1.0	1.1	1.0	1.1	1.1	1.1	.9	1.0
Quiet withdrawal	M	2.0	2.3	2.3	2.4	2.1	2.1	2.2	2.2	2.1	2.3	2.3	2.4	2.0	2.0	2.4	2.1
	SD	1.1	1.1	1.2	1.2	.9	1.1	1.1	1.0	1.0	1.1	1.0	1.1	1.0	1.1	1.2	1.1

Note: n = 746 is the total number of participants in Year 1.

To examine how similar our sample was to the U.S. population in general, we used the General Social Survey (GSS) data from 1980 – 1994. We divided the GSS sample into Black and White Americans and selected only those who were first married and ranged in age from 25 to 37. We found no differences between the EYM sample and the GSS sample in income, education, parental status, likelihood of cohabitation, and employment status (See Orbuch et al., 2002). The representative makeup of our sample enhances the significance of our findings.

Participants completed face-to-face interviews in their homes with race-matched interviewers in Years 1, 3, 7, and 16 of their marriages. Spouses were interviewed separately and then together as a couple. The response rate varied across the waves with an average of 80% of the original sample participating (range 70% to 93%). We calculated response rate by dividing the total number of husbands or wives interviewed by the number who were married. Husbands and wives who completed at least one wave and reported a conflict were included in the present analyses (343 of the husbands and 351 of the wives).

Procedure

The measurement procedure for assessing perceptions of conflict behaviors varied in Year 1 compared to Years 3, 7, and 16. In Year 1 each spouse independently described a recent disagreement and then indicated what conflict behaviors they used during the disagreement. In Years 3, 7, and 16 the couples participated in an interview together and reported on the same disagreement but completed questionnaires separately regarding the types of behaviors used. In particular, couples recalled the disagreement they had, talked it over, and agreed on what the conflict was about. The interviewer then read a list of different behaviors and the husband and wife separately completed questionnaires.

Because of the change in methodology, husbands and wives in the same dyad could report on a different disagreement in Year 1, whereas husband and wife dyads in Years 3, 7, and 16 were required to report on the same disagreement. We argue that although the conflict behavior methodology may have changed, a previous study (Acitelli, Douvan, & Veroff, 1993) found no significant differences in marital well-being or conflict behaviors between couples who reported congruent conflicts in Year 1 and those who did not. Similar to previous studies of Year 1 and 3 conflict behaviors, we were interested in individual variations in conflict behaviors used and not the specific dynamics of particular conflicts (Acitelli et al., 1993; Crohan, 1996).

Measures

Conflict behaviors

The questionnaire included a list of 11 different behaviors. Participants indicated how true each statement was of the disagreement from 1 (not at all true) to 4 (very true). The items were meant to measure destructive, constructive, and withdrawal behaviors. Destructive behaviors included four items: I yelled and shouted at my spouse, I insulted my spouse or called him or her names, I brought up things that happened long ago, and I had to have the last word. Constructive behaviors included five items: I calmly discussed the situation, I listened to my spouse’s point of view, I tried hard to find out what my spouse was feeling, I tried to say nice things, and I tried to make my spouse laugh. Withdrawal behaviors included two items: I went away for a while to cool down before we talked it out and I suddenly became very quiet and pulled away.

We created separate mean scores for destructive and constructive behaviors. The scales were internally consistent across the Years (destructive α range = .60 to .74; constructive α range = .69 to .81).

We considered whether to combine the withdrawal items to make a withdrawal behavior score, but the items were not highly correlated and had low internal consistency, indicating that they most likely represent disparate approaches to conflict. Examining the items separately is in line with past research, which used these items separately and successfully (Crohan, 1996; Oggins et al., 1993). Further, these items were differentially associated with marital quality, which provided evidence of their validity and distinctiveness (Crohan, 1996). Henceforth, we refer to these two constructs as leaving withdrawal and quiet withdrawal. Although single item scales are a potential weakness, studies have shown that single items can be as reliable and valid as multiple item scales, especially when the item assesses something concrete rather than abstract (Bergkvist & Rossiter, 2007). We consider these items to be concrete because participants were asked about their own behavior in response to a specific situation. Yet it is important to note that these measures are perceptions of conflict behaviors rather than measures of actual behavior.

Gender, race, and time

We coded gender as 0 (husband) or 1 (wife). We coded race as 0 (White American) or 1 (Black American). Time represented the year of marriage coded as 1, 3, 7, and 16. We centered time with Year 1 so that 0 represented the first year of marriage.

Sociodemographic and life course covariates

Education included the highest grade of school or year of college that husbands and wives had each completed in 1986. We included two income measures that were recorded each year. Participants reported how much all of the family members in the household made before taxes from 1 (none or less than $2,999) to 22 ($75,000 and over). We recoded each category as the midpoint to approximate a continuous variable ranging from $1,500 to $80,000 with an average income of $30,933. We also considered the ratio of the wife and husband’s income by dividing the wife’s personal income by the husband’s personal income in each Year.

We included premarital parental status of the husband and the wife (0 = no child before marriage or 1 = had a child before marriage), whether the husband or the wife lived with both biological parents up to the age of 16 (0 = raised with both parents or 1 = not raised with both parents), and the number of months the couple cohabited before marriage. We also included the age of the husband and wife in 1986 as a continuous variable. Table 2 provides descriptive information regarding the couples by race and gender.

Marital status

In Year 16 we tracked the respondents using several different methods to find out their marital status (mail, phone, field, and marital records). We obtained marital status information regarding all but 1.1% of the original sample irrespective of whether the participants completed the interviews. By Year 16, 46% of the couples had divorced or separated from their original spouse whereas 49% had remained married. See Table 1 for marital status information.

Analysis Strategy

We used a repeated measure ANOVA to examine which behaviors couples used most often across Years. To investigate race and gender differences in sociodemographics, life course variables, and conflict behaviors in Year 1, we used 2 × 2 ANOVAs for the continuous variables and logistic regressions for the categorical variables. We calculated correlations to describe how the sociodemographic and life course variables were associated with divorce and conflict behaviors in Year 1.

Next, to assess whether conflict behaviors used in Year 1 of marriage predicted divorce, we estimated Cox regression models. Cox regression models allow for the examination of the risk of divorce as a function of multiple predictors. Risk is assessed in terms of whether the event occurs as well as when it occurred in continuous time. Cox regression models produce hazard ratios, which are interpreted as the hazard function that corresponds to each unit change in the predictor. A hazard ratio greater than one represents an increase in the rate of the event, whereas a ratio less than one refers to a decrease. The outcome was the number of years until divorce. All models included the following covariates: mean couple income in Year 1, mean couple education in Year 1, income ratio in Year 1, whether the couple had premarital children, how many months the couple cohabited, mean couple age, and separate variables for whether the husband and wife were raised in two parent households.

We estimated three sets of Cox regression models to examine: (a) main effects of conflict behaviors, (b) interactions between husband and wife conflict behaviors, and (c) interactions between conflict behaviors and race. The first model included two blocks of predictors: (a) the covariates, and (b) husbands’ and wives’ use of all four strategies as separate predictors. Next, we estimated a series of models testing interactions between the husband and wife conflict behaviors. This involved separate models testing all possible combinations of behaviors, two behaviors at a time, to reduce overestimating the models. For example, we tested whether there was an interaction between destructive and quiet withdrawal by including the main effects husband and wife reports of quiet withdrawal and destructive behaviors as well as four interactions (husband destructive X wife quiet withdrawal, wife destructive X husband quiet withdrawal, husband destructive X wife destructive, husband quiet withdrawal X wife quiet withdrawal). Finally, to test whether there were race differences in the behaviors that predict divorce, we estimated a model that included all possible race X conflict behavior interactions.

Before estimating the Cox regressions, we tested the proportional hazard assumption for each predictor variable by calculating correlations between the Schoenfeld residual for each covariate and rank ordered survival time (Kleinbaum & Klein, 2005). Significant correlations indicate that the variable is time-dependent and violates the proportionality assumption. We found only one violation (husband’s destructive behaviors), which we corrected for by entering the interaction between husbands’ self-reported destructive behaviors and time (the number of years to divorce; Singer & Willett, 2001).

We used multilevel models to examine whether conflict behaviors varied over the 16 years of marriage. Multilevel models are ideally suited for dyadic and longitudinal data. These models account for correlated errors due to interdependency between dyad members and repeated measures of individuals over time (Kenny, Kashy, & Cook, 2006). The data also can be unbalanced with data from only one dyad member or data for fewer than the total number of time points. We estimated several multilevel models to determine the best fit. The models included the recommended 2 levels for longitudinal dyadic data, in which level 1 refers to time and level 2 refers to the couple (Kenny et al., 2006). First we estimated models in which the slopes and intercepts were allowed to vary between couples (random intercepts and slopes). There was zero or nonsignificant variance between the slopes of the couples for all behavior types. Thus, in the final model we allowed the intercepts, but not the slopes, to vary between couples. The model also allowed for correlated errors between husband and wives.

To examine whether conflict behaviors varied by gender, race, and time, we estimated four multilevel models, one predicting each of the four behavior types. The predictors included gender, race, time, and all possible two-way interactions. Because the three-way interactions were not significant, they were removed from the models. We included all covariates in the models including Year 1 education, income in each Year, income ratio in each Year, premarital children, months cohabited, not raised with both parents, and age at marriage.

Because of the change in method and the relatively large number of husbands and wives who did not report a conflict in Year 1 (29% of husbands and 21% of wives), we estimated a series of models to determine whether and how to include participants who did not recall a conflict in Year 1. The patterns of results differed by gender, age, and time between the participants who recalled a conflict in Year 1 and those who did not. Thus, we included all participants irrespective of reporting conflict in Year 1 and included the variable as a covariate in the models (0 = did not recall a conflict in Year 1, 1 = recalled a conflict in Year 1). In preparation for the multivariate analyses, we explored the correlations across all study variables and found multicollinearity (r > .50) would not be a concern for subsequent analyses.

Results

Description of the Data

Similar to national statistics, 46% of the couples had divorced by Year 16 of the study. More Black Americans divorced than White Americans (Table 1).

As shown in Table 2, across Years husbands and wives used constructive behaviors most often followed by withdrawal and destructive behaviors (F (1, 1625) = 240.30, p < .01). A preliminary examination of the conflict behaviors in Year 1 using 2 × 2 ANOVAs revealed that wives used more destructive (F (1, 555) = 6.82, p < .01) and less constructive behaviors (F(1, 555) = 28.31, p < .01) than did husbands. Black Americans used more destructive (F (1, 555) = 6.45, p < .01), quiet withdrawal (F (1, 555) = 4.70, p < .01), and leaving withdrawal (F (1, 555) = 34.52, p < .01) in Year 1 than did White Americans.

Analyses examining gender and race differences in sociodemographic and life course variables revealed that wives were younger (F (1,555) = 41.10, p < .01) and more likely to grow up without both parents (B = .56, SE = 18, odds ratio (559) = 1.76, p < .01) than were husbands. Black American couples were older at marriage (F (1,555) = 14.38, p < .01), were more likely to have premarital children (B = 1.26, SE = .19, odds ratio (555) = 3.53, p < .01), cohabited a greater number of months (F (1, 555) = 26.02, p < .01), were more likely to have been raised without both parents (B = 1.34, SE = .19, odds ratio(559) = 3.80, p < .01), and reported lower household income across Years (F (1, 1570) = 63.02, p < .01) than did White Americans across waves. There were no gender or race differences in education or race difference in income ratio.

Finally, we examined whether the covariates were associated with divorce (0 = married, 1 = divorced) and the conflict behaviors in Year 1 with correlations. Couples with less income (r (351) = −.16, p < .01), who were younger at marriage (r (355) = −.13, p < .01), who had premarital children (r (355) = .21, p < .01), with wives who were from households without both parents (r (355) = .18, p < .01), and who had less education (r (355) = −.29, p < .01) were more likely to divorce. Income ratio, cohabitation, and whether husbands grew up without both parents were not associated with divorce.

Covariates were also associated with conflict behaviors. Couples who had less income (r (535) = −.17, p < .01), who had premarital children (r (555) = .19, p < .01), who were from households without both parents (r (559) = .15, p < .01), and who were less educated (r (556) = −.18, p < .01) reported more destructive behaviors. Constructive and withdrawal behaviors were not associated with covariates.

Divorce as a Function of Year 1 Conflict Behaviors

Consistent with our hypothesis, husbands and wives who reported destructive conflict behaviors in Year 1 had higher divorce rates (Table 3). Greater leaving withdrawal also predicted higher rates of divorce but only when utilized by husbands. Inconsistent with our hypothesis, greater constructive behaviors predicted higher rates of divorce when utilized by wives.

Table 3.

Cox Regression Models Examining Divorce as a Function of Conflict Behaviors in Year 1 (n = 207)

	Model 1		Model 2	Model 3	Model 4
	Block 1	Block 2
Year 1 Predictor	HR	HR	HR	HR	HR
Conflict behaviors
Destructive
Husband		4.28**	6.37**
Wife		1.51*	2.02
Constructive
Husband		.86	2.46		.28*
Wife		1.62**	5.36**		.40
Leaving withdrawal
Husband		1.23*	1.20	.65	.45*
Wife		1.03	.98	1.05	.46*
Quiet withdrawal
Husband		.89	.82	.93
Wife		.97	.93	.59
Race (Black American)	2.81**	1.60	1.60	2.22**	2.58**
Couple education	.65**	.78*	.75**	.67**	.63**
Couple household income	1.00**	1.00	1.00	1.10*	1.00**
Income ratio (wife/hus.)	1.02	.68	.74	1.10	.95
Premarital par. status	.96	.67	.58*	1.01	1.10
Cohabitation in mos.	.99	1.01	1.01	1.00	1.00
Husband not raised with 2 parents	.95	.59*	.58*	.91	.84
Wife not raised with 2 parents	1.21	1.48	1.56	1.32	1.18
Couple age at marriage	.97	.98	1.00	.95	.96
	Model 1		Model 2	Model 3	Model 4
	Block 1	Block 2
Interactions	HR	HR	HR	HR	HR
Husband and wife same conflict behavior
Hus. constructive X wife constructive			.65*
Hus. destructive X wife destructive			.86
Hus. quiet X wife quiet			.03
Hus. leave X wife leave			1.02
Husband and wife leave X quiet
Husband leave X wife leave				1.02
Husband quiet X wife quiet				1.06
Husband leave X wife quiet				1.21*
Husband quiet X wife leave				1.06
Husband and wife constructive X leave
Hus. constructive X wife constructive					1.17
Husband leave X wife leave					1.09
Hus. constructive X wife leave					1.30*
Wife constructive X hus. leave					1.33*
Correction for proportional assumption
Hus. Destructive X Years to divorce		.14**	.13**
−2 Log likelihood	985.30	710.60	705.23	992.20	990.26
χ2	52.26**	418.59**	422.79**	67.09**	70.58**
χ2change from model with only covariates		274.70**	280.07**	15.70*	17.64*

Note: HR refers to hazard ratio. Models 2 and 3 χ²change are from χ^{2 =} 1007.90.

^*p < .05.

^**p < .01.

The main effects of leaving withdrawal and constructive behaviors should be interpreted with caution however, in light of the interaction findings. We tested models that included two conflict behavior types at a time and included interactions between the husband’s and the wife’s behaviors to test if patterns of behaviors differentially predicted divorce. We present the models that revealed significant interactions in Table 3 (Models 2 – 4). As we hypothesized and similar to the demand-withdraw pattern, there were significant interactions between constructive behaviors and leaving withdrawal. If either husband or wife used greater constructive behaviors and their partner used greater leaving withdrawal, they had higher divorce rates. There were also interactions between husband and wife constructive behaviors and husband and wife withdrawal behaviors. When both husband and wife reported greater constructive behaviors they had lower divorce rates. Finally, when wives used greater quiet withdrawal and husbands used greater leaving withdrawal, they had higher divorce rates.

To examine whether behaviors had similar effects on Black and White American couples, we entered interactions between conflict behaviors and race. As hypothesized, conflict behaviors appeared to have similar effects on divorce for Black and White American couples with one exception. One significant interaction between husbands’ use of quiet withdrawal and race (B = −.60, SE = .22, p < .01) indicated that whereas there was no association between quiet withdrawal and divorce among White American husbands (B = .16, SE = .21, HZ = 1.17, p = .45), Black American husbands had lower divorce rates when they used this strategy (B = −.35, SE = .02, HZ = .71, p < .05).

Interestingly, before the conflict behaviors were added to the model (Model 1; Block 1), couples had higher divorce rates if they were Black American, had lower income, and had less education. After the conflict behaviors were added to the model (Model 1; Block 2), only education and whether the husband was raised with both parents were significant predictors of divorce. The model significantly improved in fit between the covariate block and the block including conflict behaviors, indicating that conflict behaviors were significant predictors of divorce over and above race, sociodemographics, and life course covariates.

Conflict Behaviors over Time

Lastly, we assessed whether conflict behaviors changed over time and whether those changes varied by gender and race. Table 2 includes the means and Table 4 includes the multilevel models.

Table 4.

Multilevel Model Results Predicting Self-Reports of Conflict Behaviors as a Function of Time, Gender, and Race (n = 694)

			Withdrawal
	Destructive	Constructive	Leaving	Quiet
Predictor	B (SE)	B (SE)	B (SE)	B (SE)
Intercept	2.249 (.224) **	2.851 (.198) *	1.804 (.322) **	1.773 (.309) **
Time	−.007 (.006)	.009 (.006)	−.012 (.010)	−.009 (.010)
Gender (Wife )	.245 (.051) **	−.260 (.051) **	−.089 (.091)	.103 (.088)
Race (Black American)	.085 (.071)	.116 (.061)	.308 (.101) **	−.064 (.098)
Time X Gender	−.011 (.005) *	.008 (.005)	−.009 (.010)	−.019 (.009) *
Time X Race	.002 (.007)	−.003 (.006)	−.028 (.011) **	.002 (.011)
Gender X Race	.111 (.059)	.079 (.059)	.097 (.106)	.134 (.102)
Education	−.011 (.013)	−.026 (.012) *	−.013 (.019)	−.010 (.019)
Income	−.001 (.001)	−.001 (.001) **	−.001 (.001)	−.001 (.001)
Income ratio (wife/hus.)	.004 (.009)	−.014 (.007)	.004 (.012)	.014 (.012)
Premarital par. status	.189 (.065) **	−.126 (.052) *	−.006 (.081)	.039 (.077)
Cohabitation in mos.	.001 (.001)	−.002 (.001)	.001 (.002)	−.001 (.002)
Not raised with 2 parents	.136 (.040) **	−.093 (.038) **	−.018 (.065)	.066 (.063)
Age at marriage	−.002 (.006)	.013 (.006) *	.012 (.009)	.021 (.009) *
Conflict Year 1	.039 (.053)	.035 (.049)	.242 (.082) **	.194 (.079) *
Variance between couples	.149 (.020) **	.066 (.013) *	.091 (.031) **	.047 (.029) *
Repeated effect	.042 (.016) **	.022 (.015) **	.063 (.048)	.138 (.048) **
Residual Variance	.305 (.016) **	.314 (.017) **	1.020 (.054) **	.942 (.050) **
−2 log likelihood	3105.1	2937.0	4584.1	4536.7

Note: n = 694 is the total number of participants who reported as least one conflict across years.

^*p < .05.

^**p < .01.

Destructive

Inconsistent with our hypothesis, there was no overall change in destructive behaviors over time but there was a time X gender interaction indicating that the change over time in destructive strategies varied by gender. As we hypothesized, wives used more destructive conflict behaviors than husbands. Partially consistent with our hypothesis, wives decreased their use of destructive behaviors over time (B = −.014, SE = .004, p < .01), whereas the husbands did not (B = −.003, SE = .004, p = .36). In addition, significant covariates indicated that respondents who grew up without both parents and respondents with premarital children used more destructive behaviors than respondents who grew up with both parents and respondents who did not have children prior to marriage.

Constructive

Contrary to our hypothesis, constructive behavior use was consistent over time. As we expected, husbands reported more constructive behaviors than wives but couples did not increase their use of these behaviors over time. Significant covariates indicated that respondents who were older at marriage, who grew up with both parents, who did not have children before marrying, who had more income, and more education used more constructive behaviors.

Withdrawal

Inconsistent with our hypothesis, withdrawal behaviors did not decrease over time among all couples but the trajectories varied by race and gender. As expected, Black American couples used leaving withdrawal more than White American couples, but the use of leaving withdrawal changed over time (time X race interaction). White American couples significantly increased in their use of this behavior over time (B = .02, SE = .007, p < .01), whereas Black American couples’ use of this strategy decreased over time albeit marginally (B = −.014, SE = .008, p = .09). Significant covariates indicated that couples who reported a conflict in Year 1 reported leaving withdrawal more than couples who did not report a conflict.

Quiet withdrawal changed over time but the changes varied by gender. Husbands showed no change over time (B = .002, SE = .007, p = .73), but partially consistent with our hypothesis, wives’ use of quiet withdrawal decreased over time albeit marginally (B = −.01, SE = .007, p =.06). In addition, couples who were older at marriage or who reported a conflict in Year 1 also reported using more quiet withdrawal.

Discussion

Behavioral theorists of marriage have proposed that destructive behaviors result in declines in marital satisfaction and stability, whereas constructive behaviors lead to improvements in marital satisfaction and stability (Karney & Bradbury, 1995; Kelly et al., 2003). Consistent with these theories, we found that over and above the sociodemographic and life course variables that vary within and between couples, conflict behaviors have important implications for divorce. As expected, husbands and wives who reported using more destructive behaviors had higher divorce rates (Gottman, 1994; Orbuch et al., 2002). Withdrawal behaviors predicted greater divorce rates but only as reported by husbands (Crohan, 1996). Constructive behaviors were associated with divorce but not in the expected direction; greater constructive behaviors among wives predicted greater divorce rates.

The unusual main effect associations made more sense when examined in light of the patterns of conflict behaviors between spouses and their effects on divorce. Unlike the consistently harmful effects of destructive behaviors for divorce, the implications of constructive and withdrawal behaviors for divorce vary depending on the context of the other partner’s behavior. Constructive behaviors were beneficial if used with a constructive spouse but ineffective in preventing divorce if used with a withdrawing husband or wife. Withdrawal was also unsuccessful among husbands and wives when both spouses withdrew.

The finding regarding the combined use of constructive and withdrawal behaviors is consistent with our hypothesis and previous research regarding the demand-withdraw pattern of marital interaction and its negative implications for marital satisfaction (Heavey et al., 1995). The present findings extend research by indicating that this pattern also has consequences for divorce up to 16 years later. Thus, if one spouse attempts to solve relationship problems with constructive behaviors, such as finding solutions, but the other spouse prefers to solve problems by leaving the situation, the pattern appears to have damaging effects on the longevity of marriage. We speculate that spouses who use constructive behaviors may perceive their partner’s leaving to cool down as a lack of investment in the relationship.

Our finding regarding the benefits of dual constructive behaviors between the husband and wife is inconsistent with Gottman’s conclusion that constructive behaviors, such as active listening, are ineffective (Gottman et al., 1998). We argue that contradictory results regarding the effects of withdrawal and constructive strategies on marital outcomes occur because the effects of specific conflict behaviors vary by gender and partner behaviors. We note that behaviors in our study were not necessarily used in response to the same conflict, but represent combinations of conflict styles. We think that this is particularly compelling and may indicate that not only are patterns of conflict behaviors influential for divorce when used in the same interaction, but also as combinations of general styles of conflict.

Finally, as we hypothesized, the majority of conflict behaviors had similar effects on divorce for Black and White Americans (Orbuch et al., 2002) and the conflict behaviors were better predictors of divorce than race. This finding is consistent with prior research, which revealed that the greater use of destructive conflict behaviors among Black American couples than White American couples accounted for race differences in marital satisfaction (Broman, 2005). Thus, the greater use of destructive and withdrawal behaviors in the early years of marriage among Black American couples may lead to higher divorce rates among these couples.

We found only one interaction between race and conflict behavior indicating that husband’s use of quiet withdrawal predicted lower divorce among Black Americans but not among White Americans. The effects of some conflict behaviors may vary by race given differential meanings of those behaviors (Orbuch et al., 1993). The use of withdrawal may be more acceptable among Black American couples due to the greater use of this behavior in the early years of marriage. The varied findings regarding quiet and leaving withdrawal also emphasize the importance of considering the differential implications of diverse types of withdrawal (Roberts, 2000).

Next, we assessed whether couples’ conflict behaviors remain stable or change over time. The results partially supported our hypothesis. Wives decreased the use of destructive behaviors and quiet withdrawal, whereas husbands’ use of these strategies remained stable over time. Constructive behaviors remained stable over time among both husbands and wives.

The decreases in withdrawal and destructive behavior support accommodation and developmental theories, indicating that behaviors become less negative over time. This is in contrast to social learning theories and research, which postulate that couples who are negative remain negative or increase in negativity over time (Caughlin et al., 2000; Huston et al., 2001). Studies that found stability in conflict behaviors over time examined marriages over 2 to 4 years (Gottman & Levenson, 1999; Kelly et al., 1985). Our study is more consistent with findings for long-term marriages, which show that relationships become more enjoyable and tolerant and have improved communication (Robinson & Blanton, 1993). Similarly, Lindahl and colleagues (1998) found that self-reported withdrawal and destructive behaviors decreased over 10 years.

The finding that constructive behaviors remained consistent rather than increasing over time was somewhat surprising but may be due to the high usage of this behavior among all couples over time. Because couples remained constructive and used fewer destructive and withdrawal behaviors over time, we believe the results are more supportive of accommodation theories than social learning theories.

There are several possible explanations for the greater change in conflict behaviors over time among wives. The problems that cause wives to use more destructive and quiet withdrawal behaviors early in marriage may be resolved over time. Alternatively, relationships and the quality of relationships may be more central to the lives of women compared to men (Almeida & Kessler, 1998). Wives over the course of marriage may recognize that the use of destructive and withdrawal conflict behaviors is not effective, nor beneficial to the overall well-being and stability of their marriages. Wives may also gain more effective conflict skills and become better able to express their negative feelings. Husbands may not show this same decrease in withdrawal and destructive behaviors over time because they use more constructive and less destructive behaviors than wives and thus have fewer negative behaviors to change.

Our findings also showed increased use of leaving withdrawal among White American couples, whereas Black Americans showed a decrease over time. Similarly, other cross-sectional studies have found that older Black American couples used fewer negative behaviors than young Black American couples (Adelmann et al., 1996). Further exploration is needed, but perhaps Black American couples become better able to express their negative feelings over time.

Consistent with previous literature and our hypotheses, we also found overall gender and race differences in behaviors. Wives reported more destructive behaviors and husbands reported more constructive behaviors (Carstensen et al., 1995). Black Americans reported more leaving withdrawal behaviors than White Americans (Oggins et al., 1993). However, with the exception of constructive strategies, these gender and race differences interacted with time indicating that the patterns in the first year of marriage do not remain consistent over time.

There are several possibilities for future research. Although this sample is similar to national data, newlywed couples today may differ from newlyweds in 1986. More recent cohorts are waiting longer to get married and are more egalitarian in their decision-making than are older cohorts (Amato, Johnson, Booth, & Rogers, 2003). Although the ratio of the wife’s and the husband’s incomes was not a predictor of divorce, couples who are more egalitarian may use different conflict behaviors compared to those couples for whom issues of inequality arise. Future work should consider additional variables that may predict conflict behaviors or divorce such as the topic of conflict, marital satisfaction, and commitment. Couples who fight about money and are satisfied with their marriages, for instance, may use different behaviors than couples who argue about infidelity and are dissatisfied with their marriages. In our more recent work we are examining associations among conflict behaviors and marital satisfaction. The dynamic interchange between conflict and satisfaction may have implications for divorce.

Future work should attempt to replicate these findings with different methods and measures. Researchers have found differences between self-report and observational measures of conflict behaviors (Lindahl et al., 1998). Self-reports and retrospective accounts of behavior are biased by memory and may reflect perceptions based on identity or beliefs about what should or ought to have happened (Robinson & Clore, 2002). Unfortunately, we did not collect data on when the conflict occurred and future work might include this information. Future work also should examine a greater number of withdrawal behaviors (e.g., cognitive reappraisal, drinking).

Overall, this study contributes to the literature in several important ways. Destructive as well as constructive and withdrawal behaviors have important implications for the longevity of marriage over 16 years. Wives appear to accommodate over the course of marriage using less destructive and withdrawal behaviors over time, whereas husbands’ conflict behaviors remain stable over time. In addition, we found race and gender differences in conflict behaviors in the early years of marriage were not consistent over time. We hope this study will lead to additional research on the complex dynamics of conflict between husbands and wives and the potential explanations for changes versus stability in conflict behaviors over time.