|Home | About | Journals | Submit | Contact Us | Français|
Using statistically imputed data to increase available power, this article reevaluated the long-term effects of divorce mediation on adults’ psychological adjustment and investigated the relations among coparenting custody conflict, nonacceptance of marital termination, and depression at 2 occasions over a decade apart following marital dissolution. Group comparisons revealed that fathers and parents who mediated their custody disputes reported significantly more nonacceptance at the 12-year follow-up assessment. Significant interactions were observed by gender in regression models predicting nonacceptance at the follow-up; mothers’ nonacceptance was positively associated with concurrent depression, whereas fathers’ nonacceptance was positively associated with early nonacceptance and negatively associated with concurrent conflict.
The emotional story of any given divorce likely includes chapters of anger, sadness, guilt, regret, relief, and—one hopes—happiness. As adults physically rearrange the fabric of their life together, the process of uncoupling and actions geared toward resolution and healing are carried on for years following the actual divorce event. After a period of both emotional and physical upheaval, most adults cope successfully with the dissolution of their marriage (Amato, 2000; Aseltine & Kessler, 1993; Booth & Amato, 1991; Hetherington & Kelly, 2002; Lorenz et al., 1997), and some individuals report opportunities for growth and increased independence (Marks, 1996). However, individual- and couple-level difficulties can become fixtures of postdivorce life, and a subset of individuals report ongoing conflict with and attachment to their former spouse up to several years after the formal end of their marriage (Emery, 1994; Kitson & Holmes, 1992; Kitzman & Emery, 1994; Maccoby & Mnookin, 1992; Wang & Amato, 2000).
One of the primary problems in developing a long-term empirical picture of how adults psychologically adjust to the divorce experience is participant attrition; longitudinal samples sizes typically dwindle as individuals move away or cite reasonable explanations for not participating in follow-up research (e.g., not wanting to “revisit” a painful emotional experience, not having time for detailed psychological assessments). Fortunately, relatively novel, sophisticated, and accessible means exist for statistically dealing with problems of incomplete data and enhancing the available power for follow-up studies (Rubin, 1987; Schafer, 1997; Schafer & Olsen, 1998). In this article, we use the methods of multiple imputation (MI) to statistically impute incomplete data to reconsider the long-term treatment effects of a divorce mediation intervention (see Emery, Laumann-Billings, Waldron, Sbarra, & Dillon, 2001) and to examine the continued relations among coparenting conflict, nonacceptance (of marital termination), and depression in the decade following marital separation. Although the statistical methods used in this study are not yet in wide use within the larger family psychology or psychology treatment literatures, such approaches have much potential for addressing the problems of long-term sample attrition and limited power in longitudinal designs, in that they allow researchers to make the most of limited or incomplete data.
Divorcing men and women commonly report elevated levels of depression, anxiety, substance abuse, and poor health-related habits (Amato, 2000; Aseltine & Kessler, 1993; Booth & Amato, 1991; Gotlib & McCabe, 1990; Lorenz, Simons, & Chao, 1996; Simons & Johnson, 1996). Although many of the consequences of divorce can be viewed as the same problems that cause declines in marital quality, longitudinal evidence indicates that a number of the psychological and physical correlates are specific to marital dissolution (Amato, 2000; Bruce, 1998; Menaghan & Lieberman, 1986). Increased rates of illness, morbidity, and suicide are especially notable for divorced men (Burman & Margolin, 1992; Hu & Goldman, 1990; Kiecolt-Glaser & Newton, 2001). In addition to these negative consequences, there is considerable evidence that disturbed parenting practices (resulting from the stress of marital dissolution) place children at risk for maladjustment (Hetherington, Bridges, & Insabella, 1998).
A central topic in the long-term study of divorce is whether parents can, over time, successfully cope with the personal stressors of marital dissolution while also renegotiating an acrimonious coparenting relationship with their former partner. Few areas of family relations are as robustly associated with children’s psychological well-being as the extent to which parents openly argue; scream and yell; and deride, undermine, and belittle each other both in front of their children and behind closed doors (Buehler et al., 1997). Determining the factors associated with continued conflict among divorced adults is thus a timely endeavor for more completely understanding both adult and child adjustment over time.
Differing from general marital acrimony and co-parenting problems in intact families (cf. Margolin, Gordis, & John, 2001), this form of conflict occurs between divorced or separated parents entrenched in child custody disputes and/or ongoing battles regarding appropriate parenting practices. Conflict between divorced parents focused on child custody, caretaking, and parenting issues represents a unique domain of interparental discord, and these disputes play a major role in children’s postdivorce adjustment (Grych & Fincham, 1993; Hetherington, 1999; Johnston, Kline, & Tschann, 1989; Johnston & Roseby, 1997; Kitzman & Emery, 1994; Kline, Johnston, & Tschann, 1991). Conflict after divorce appears to focus more on children and may occur more often in their presence after separation (Emery, 1994).
Among research that has examined coparenting conflict prospectively, Maccoby and Mnookin’s (1992) investigations of 1,100 California families who filed for divorce found that over a third of the sample reported continued coparenting conflict 1.5 years after the marital separation. A quarter of this representative sample of divorced parents had a conflicted relationship almost 4 years after filing for divorce, and it also appeared that, for some families, intense anger was replaced with disengagement, characterized by both low cooperation and conflict. In an effort to deal with their anger toward each other, a subset of parents went their separate ways or developed parallel parenting strategies that involved as little contact as possible (see also Ahrons, 1994; Ahrons & Rodgers, 1987; Hetherington, 1999). With the exception of Macobby and Mnookin’s (1992) and Ahron’s (1994) work, little empirical research has investigated whether overt coparenting conflict persists or desists in the years following divorce or separation, especially among high-conflict parents (Amato, 2000). In their 10-year follow-up study, Wallerstein and Blakeslee (1989) reported that conflict among divorced parents remained high, with roughly one half of the women and one third of the men in their sample expressing intense anger toward their former partner. Given the clinical perspective of this work (see Wallerstein & Blakeslee, 1989, p. 309), it is unclear how these estimates were derived and how they relate to well-established measures of coparenting conflict.
When considering how parents navigate the many hurdles of divorce, it is fitting to note that problems of coparenting do not occur in a vacuum; such difficulties often go hand in hand with other divorce-related problems. Emery and colleagues (Emery, 1994; Emery et al., 2001; Emery, Matthews, & Wyer, 1991; Emery & Sbarra, 2002) have suggested that divorcing adults often become tangled in child custody disputes as a means of contesting the end of a relationship. Contesting child custody—especially for noninitiators of divorce, who are usually men (Braver, Whitley, & Ng, 1993; J. D. Gray & Silver, 1990)—provides a legally sanctioned means of maintaining contact with a former spouse and battling a partner’s decision to leave the marriage. Of course, many parents have legitimate reasons for contesting custody (e.g., alimony problems, visitation concerns). However, several investigations have revealed that attachment to a former spouse, typically conceptualized as nonacceptance of the end of the relationship, is among the central dynamics of adult adjustment following divorce (Amato, 2000; Berman, 1988; Kitson, 1982; Kitson & Holmes, 1992; Masheter, 1991; Wang & Amato, 2000; Weiss, 1975). Kitson and Holmes (1992), for instance, reported that whereas 70% of adults reported feeling a longing for their partner and a failure to accept the permanence of the divorce at the time of filing for separation, almost 30% of adults reported continued attachment to their former spouse up to 4 years after the separation (with 13% feeling more attached and less accepting of the divorce over time). Wang and Amato (2000) found that, up to 5 years after divorce, women reported less attachment to their ex-spouse, and initiators and individuals with a new partner also reported less preoccupation with their former partner and better general divorce adjustment (see also Black, Eastwood, Sprenkle, & Smith, 1991; Kiecolt-Glaser et al., 1988; Thompson & Spanier, 1983). To date, no investigations have documented the extent to which adults remain preoccupied with their former partner or nonaccepting of divorce over periods longer than 5 years. Furthermore, no systematic studies have examined whether continued preoccupation with a former spouse is associated with increased conflict or depression over time.
In the present study, we explore these issues by examining the relations among coparenting conflict, nonacceptance, and depression over two periods of time separated by 12 years. In addition, we reevaluate the long-term associations between these outcomes and whether child custody disputes were either mediated or litigated 12 years earlier. Previously published studies with this high-conflict sample focused on the role of divorce mediation in adults’ psychological adjustment, postdivorce family life (e.g., nonresidential parent contact and involvement with the child), and satisfaction with the administration of justice (Emery, 1994; Emery et al., 2001; Emery, Matthews, & Kitzman, 1994; Emery et al., 1991; Emery & Wyer, 1987). Divorce mediation is a time-limited form of dispute resolution that, in the past 2 decades, has been increasingly recognized as a means of promoting efficiency in the dispute resolution process (compared with adversarial court litigation), improving party satisfaction, and raising compliance with court agreements (C. J. A. Beck & Sales, 2001; Irving & Benjamin, 1995; Kelly, 1996). In comparison with adversarial settlement procedures such as court litigation, mediation also is hypothesized to promote individual well-being and preserve family relationships that continue despite the end of marriage (Emery, 1994). Among the main problems in conducting long-term evaluations of the potential lasting effects of mediation are small sample sizes and high participant attrition (C, J. A. Beck & Sales, 2001; Emery, 1994). However, advances in quantitative methodology now make possible the imputation of incomplete data, thus allowing mediation researchers to address the long-standing problem of limited power in follow-up studies (see Schafer & Olsen, 1998). Using these methods, we conduct the same group comparisons (mediation vs. litigation; mothers vs. fathers) reported in our first 12-year follow-up article (Emery et al., 2001). In reanalyzing with the imputed sample, our intention is to be able to make more refined and sophisticated statements about the long-term effects of divorce mediation for adult psychological adjustment using the best available statistical tools. Beyond the mediation/litigation and mother/father comparisons, all other analyses are unique to this article. Overall, this study is concerned with understanding the percentage of adults reporting continued divorce-related difficulties at the long-term follow-up, reevaluating the long-term effects of divorce mediation and differences in the psychological adjustment of mothers and fathers, and, finally, understanding the long-term associations of coparenting conflict, nonacceptance, and depression. The work is guided by the following hypotheses.
A small but reliable subsample of participants will report continued ongoing coparenting conflict and nonacceptance at the 12-year follow-up. Specifically, we expect that 15% of the (nonimputed) sample will endorse the most extreme response choices on the acrimony and nonacceptance scales. This estimate reflects the notion that most adults ultimately fare well after divorce but that some individuals become embroiled in continued coparenting difficulty and stuck in patterns of yearning for a former partner (Amato, 2000; Emery, 1994; Hetherington & Kelly, 2002).
Using imputed data to make group comparisons on the three constructs of psychological adjustment (i.e., coparenting conflict, nonacceptance, and depression) at the 12-year follow-up, we expect that fathers will report less coparenting conflict and more nonacceptance, whereas mothers will report greater depression. These predictions derive primarily from the earlier findings from this study, in which men who mediated reported less coparenting conflict at the 18-month follow-up but also tended to be more non-accepting of relationship termination (Emery et al., 1994; Emery et al., 1991). For depression, we expect that the present results will follow generally reported gender differences in mood disturbance (A. T. Beck, Steer, & Garbin, 1988), but there is no a priori reason to believe that either mothers or fathers will report clinically elevated depression at the follow-up.
Imputation of the incomplete data also allows for the reanalysis of treatment group differences between the mediation and litigation samples. With additional power to detect group differences, we predict that individuals who initially mediated their child custody disputes will report significantly less coparenting conflict and nonacceptance at the 12-year follow-up than parents who litigated but that no demonstrable differences will be observed for mean levels of depression. As noted in the first follow-up report (Emery et al., 2001), mediation interventions are not a psychological panacea designed to reduce all problems of psychological adjustment. Instead, the services received by this sample were pointed, with the main aims of reducing immediate coparenting difficulties and assisting parents in making the transition out of marriage. In turn, we expect that mediation will help parents step onto a trajectory of positive adaptation with respect to these interpersonal difficulties but not necessarily result in long-term group differences in general psychological adjustment (e.g., depression).
From a clinical perspective, it has been argued that parents often contest child custody as means of contesting the end of their marriage (Emery, 1994; Emery & Sbarra, 2002). To date, however, no empirical research has tackled this question. Using the imputed sample, we expect nonacceptance (at both the initial assessment and follow-up) to be a significant predictor of continued coparenting conflict in regression models that control for initial conflict. Similarly, we predict that coparenting conflict will be the strongest predictor of continued nonacceptance. In addition, given research demonstrating a strong association between nonacceptance and depression in the years immediately following divorce (Kiecolt-Glaser et al., 1987; Wang & Amato, 2000), we also predict that these relations will hold 12 years later and that depression (at the follow-up) will be a significant predictor of concurrent nonacceptance. Finally, because fathers in this sample were more nonaccepting of the end of their marriage (Emery et al., 1991), we also expect an interaction by gender for both coparenting conflict and nonacceptance at the 12-year follow-up. Specifically, we anticipate that fathers, but not mothers, will exhibit a significant positive association between conflict and nonacceptance at the follow-up.
Participants were a subsample of 71 families who had requested a child custody hearing from a juvenile and domestic relations district court in Central Virginia between 1983 and 1986. Of these 71 families, 35 were randomly assigned to divorce mediation and 36 to traditional litigation to resolve their custody dispute in the initial study (see Emery & Wyer, 1987). By virtue of seeking custody settlements, the sample represents a relatively conflicted group of divorced parents; approximately 10% of divorced parents seek litigation settlements to solve child custody disputes (Emery, 1994). The initial assessments took place an average of 5 weeks after dispute resolution. The long-term follow-up assessment took place an average of 12 years and 2 months after the initial custody decision (range 10.6–13.3). (We also conducted a 1-year follow-up; however, these data are not included in the present analyses; see Emery et al., 1994). The original sample consisted of 61 mothers and 57 fathers; the 12-year, nonimputed follow-up sample included 34 mothers and 40 fathers. Table 1 displays the available sample sizes in previously published reports and the current sample. The discrepancy between the sample sizes in the first two columns is explained by the fact that the data were incomplete even within the first study (Emery et al., 1991); for the conflict and nonacceptance outcomes, a does not apply option was available to participants in each assessment. As a result, in the initial study, 26 participants reported on their satisfaction with the administration of justice but indicated that the conflict and nonacceptance questions, for whatever reasons, did not apply to them. Because no data were available for these participants in the initial study, we therefore omitted them from the imputation analyses. In other words, we only imputed data at the follow-up for participants for whom we had complete data at the initial assessment. The third column of Table 1 represents the sample sizes reported in the nonimputed 12-year follow-up (Emery et al., 2001). The fourth column, identical to the second, reflects the imputed sample size in the current study.
At the 12-year follow-up, the nonimputed sample had the following demographic characteristics. The average age was 39 years for mothers (range = 30 to 54 years) and 43 years for fathers (range = 30 to 62 years). Seventy-nine percent of the sample was White, with the remainder being Black. The sample was largely of lower socioeconomic status. Average reported annual family income at the 12-year follow-up was $27,687 for the men in the sample (range = $3,000 to $60,000); women’s family income averaged $22,532 (range = $12,000 to $80,000). (Data were collected in 1997–1998.) The average number of years of education for the men was 11.7 years; for women it was 11.9 years. Approximately three quarters of both the men and the women had a high school education or less. Most families were from a Protestant religious background. About half of the men and women were either remarried or living with a partner at the time of the long-term follow-up interview. Both the men and the women had an average of 2.5 biological children (range = 1 to 7). The average age of the target child was 17 years. Two thirds of the target children were male. No significant differences were found between the mediation and litigation groups on any of these background characteristics. At the Time 3 assessment, fathers were slightly older than mothers, t(l, 37) = 4.07, p < .001, but no other significant differences existed between parents.
The details of the original field study involving random assignment to mediation/litigation conditions are described at length elsewhere (Emery, 1994; Emery et al., 1991, 2001). Parents were approached at random about either attempting mediation or participating in an evaluation of the court’s services (litigation control group). Following random assignment, families either proceeded through the usual court settlement processes or entered the mediation service. Mediation took place inside a courthouse, was conducted by one of four pairs of male and female comediators, and was limited to no more than six 2-hr sessions (average = 2.4 sessions). Only 4 of 35 mediation cases had a contested court hearing, whereas, in contrast, 26 of 36 cases in the adversary group proceeded into litigation (Emery et al., 1991). For the 12-year assessment, attempts were made to locate all participants from the original sample through original telephone numbers, phone numbers of friends and family members, local phone books, and Internet search directories. In addition, members of the research team canvassed participants’ neighborhoods and places of work listed in the original files. If participants were located and gave consent for their participation, they received $50 compensation for a 2-hr interview session. The assessment consisted of a 20–30 min interview, followed by 60–90 min of written questionnaires and then a final 10-min interview. Interviews were conducted in the participants’ home unless they requested to come into the laboratory or complete the questionnaire through the mail. For parents, the interviews and questionnaire materials included questions about relitigation or other changes in child custody, narrative accounts of the events of the initial separation and the custody dispute, information about remarriage, and descriptions of the coparenting relationship (Emery et al., 2001).
The Acrimony Scale (AS) is a 25-item measure of coparenting conflict between separated or divorced parents that yields a single acrimony score, the mean of all items, with higher scores indicating greater conflict and more coparenting difficulties (Shaw & Emery, 1987). Examples items include, “I have angry disagreements with my former spouse,” and, “Visitation is a problem between myself and my children’s mother/father.” Responses are made on a 4-point Likert scale rating the degree to which each statement characterizes the relationship (1 = almost never to 4 = almost always). Items are worded in a counterbalanced format to control for response bias. This scale has high internal consistency (α = .86) and test–retest reliability (r = .88; Emery, 1982). At the initial assessment, the alpha for the AS was .87 for fathers and .83 for mothers, and mothers’ and fathers’ AS scores were correlated .29 (n = 60, p < .05). At the 12-year follow-up, the alpha for the AS was .88 for fathers and .91 for mothers, and mothers’ and fathers’ AS scores were correlated .41 (n = 40, p < .05). Correlations between initial AS and follow-up AS were .41 (n = 33, p < .05) for fathers and .32 for mothers (n = 41, p < .05).
The Nonacceptance of Marital Termination Scale (AMT) is an 11-item scale that taps a range of feelings about accepting the end of the marriage (see Kitson, 1982). It yields a single score, the mean of the items. Higher scores indicate greater nonacceptance. Example items include “I find myself thinking a lot about my former spouse” and “I feel that I will never get over this separation/break-up.” Responses are made on a 4-point Likert scale rating the degree to which each statement characterizes current feelings, ranging from 1 = not at all my feelings to 4 = very much my feelings. At the follow-up, the alpha for the AMT was .80 for mothers and .75 for fathers, and mothers’ and fathers’ AMT scores were correlated .05 (n = 37, ns). Correlations between the AMT at the initial assessment and the follow-up were .40 for mothers (n = 46, p < .01) and .46 (n = 38, p < .01) for fathers.
The Beck Depression Inventory (BDI) is a 21-item inventory that assesses components of depressive states (A. T. Beck et al., 1988). Its single scale is the mean of the items, with higher scores indicating more depression. At the follow-up, the alpha for the BDI was .92 for mothers and .76 for fathers, and the correlation between mothers and fathers was .40 (n = 39, p < .05). Correlations between the BDI at the initial and follow-up assessments were .22 (n = 47, ns) for mothers and .19 (n = 38, ns) for fathers.
Although a majority of the participants in the original sample participated in the 12-year follow-up, a substantial number did not. We examined selection effects by comparing follow-up participants with nonparticipants on the initial coparenting conflict, nonacceptance, and depression measures. For mothers, significant attrition group differences were found for fathers’ reports of early coparenting conflict and nonacceptance. Mothers who remained in the study had been married to men who reported significantly less coparenting conflict at the initial assessment (M = 47.11, SD = 11.51) than the partners of mothers who did not participate at the follow-up (M = 53.67, SD = 11.76), F(1, 57) = 4.04, p < .05. In other words, mothers who did not complete the follow-up assessment were married to men who reported more conflict at the initial assessment than the partners of women who did complete the follow-up assessment. The opposite was true for fathers’ early nonacceptance. Women who remained in the study were married to men who reported more nonacceptance at the initial assessment (M = 28.20, SD = 9.21) than the partners of women who left the study (M = 23.00, SD = 6.03), F(1, 58) = 4.37, p < .05. Finally, at the initial assessment, coparenting conflict was significantly lower for men who remained in the study (M = 46.93, SD = 12.13) than for men who did not participate in the follow-up (M = 53.23, SD = 10.31), F(1, 58) = 3.75, p < .05.
To impute complete data for the follow-up sample, we used NORM (Schafer, 1999), a software program designed to assist researchers in following the MI guidelines outlined by Rubin (1987; Schafer, 1997), and followed the procedures outlined by Schafer and Olsen (1998). It is important to note that NORM (Schafer, 1999) was not designed with the intention of imputing dependent data (i.e., repeated measures on the same individuals). Although this is an obvious concern for the present study and may yield “too significant” results (see Schafer & Olsen, 1998), we used this tool for imputing data with the mindset that even less than ideal MIs are better than largely incomplete data (this point was described in detail by Rubin, 1987), and we revisit these issues in the Discussion section when considering the observed results.
The basic strategy of MI is to progress through a series of data estimation steps resulting in multiple complete data sets. The analyses of interest are conducted in each data set and then pooled according to Rubin’s (1987) rules for combining estimates and standard errors from the multiple data sets. For the present study, the data were first submitted to the estimation algorithm within NORM, which is a general method for obtaining maximum-likelihood estimates of parameters from incomplete data (Schafer & Olsen, 1998). To generate the complete data sets, we used all available data from the initial assessment, including the AS, AMT, and BDI measures. We used the parameter estimates from the EM run as data values in the data augmentation (DA) step. Once we diagnosed successful convergence of the DA (by examining the series autocorrelation plots of the parameters), we then specified that NORM draw random parameters and create a complete data set once every 60 iterations until six complete data sets were generated. We then analyzed the data using standard statistical software, and we concatenated the estimates and their standard errors from each data set and submitted them to NORM to assess statistical significance, taking into account the overall rate of missingness. This approach for analyzing the multiple results of imputed data is outlined by Rubin (1987). Finally, it is important to note that NORM assumes that the mechanism of missingness is ignorable or missing at random (MAR). The MAR situation means, according to Schafer and Olsen (1998), that “the probabilities of missingness may depend on data values that are observed but not ones that are missing” (p. 552). Although this is an assumption of the MI procedure, a MAR situation is tenable until there is reason to believe that data are missing completely at random, which is a situation that cannot be easily dealt with using MI procedures. Given the selective attrition effects, it is likely that a MAR situation is operating in the follow-up data.
To evaluate the first hypothesis, that 15% of parents in the follow-up sample would report ongoing coparenting difficulties and problems accepting marital termination, we conducted basic descriptive analyses on the AS and AMT items at the 12-year follow-up. For these analyses, the percentages of mothers and fathers responding in the almost always or almost never categories (depending on the wording of the item) at the follow-up assessment were of the most interest. These response choices indicate the most severe forms of conflict or the least cooperation and successful coparenting. At the follow-up, an average of 18.10% of mothers (SD = 14.69%, Mdn = 15.00%, range = 0.00%–50.00%) responded in the most conflicted category. Notable among the responses was that, although 0.00% of mothers reported not being adjusted to the divorce, 41.00% responded that they can never talk to their former spouse about problems with the children, and 19.00% reported ongoing angry disagreements with their former partner. As indicated by a paired-samples t-test for the 25 AS items, fathers endorsed significantly fewer extreme scores than mothers, t(24) = 2.14, p = .041, indicating that they reported less coparenting conflict than mothers at the follow-up. An average of 13.10% of fathers (SD = 11.74%, Mdn = 10.00%, range = 0.00%–39.00%) responded in the most conflicted category across all AS items. As with the mothers, 0.00% of fathers reported not being adjusted to the divorce; however, 14.50% reported feeling hostile toward their former partner almost 12 years after their initial child custody dispute.
Similar descriptive analyses were conducted for nonacceptance. The AMT contains one item that assesses sentiments about divorce in general rather than being specific to an individual’s current circumstances. Given our desire to understand current rates of adjustment and not sentiments about divorce in general, we omitted this item from the nonacceptance analyses. At the follow-up, an average of 7.31% of mothers endorsed the very much my feelings rating for the nonacceptance items (SD = 10.55%, Mdn = 4.55%, range = 0.00–36.00%). Zero percent of mothers found themselves spending a lot of time thinking about their former partner, wondering what their ex-partner was doing, or feeling like they would never get over the divorce. However, 6.50% of mothers continued to feel like the divorce was a horrible mistake, and 36.00% reported never being glad they made the break. An average of 13.00% of fathers endorsed the most extreme continued nonacceptance (SD = 12.20%, Mdn = 11.00%, range = 0.00–40.00%), and this rate was significantly larger than the pattern reported by mothers, t(29) = 3.94, p = .003. Forty-four percent of fathers reported that they were not at all glad to have made the break over a decade ago.
The second and third hypotheses concern gender and treatment group differences on the three outcome variables of interest with the imputed data samples. Table 2 displays the average group means and effect sizes to emerge from the analyses with each of the six imputed data sets. The probability estimates reported in the text reflect the significance testing probability according to Rubin’s (1987) rules for combining imputed data and correcting for the studywise rate of missingness. These analyses were conducted as follows: Six data sets were generated from the MI run, mean group comparisons were conducted in each data set, the resulting estimates and standard errors were submitted to NORM, and the resulting probabilities generated via Rubin’s (1987) rule for assessing and combining parameter estimates are reported in the text.
For gender (collapsed across treatment groups), a large effect was observed for the mean comparison of nonacceptance: Fathers reported greater mean nonacceptance at the follow-up. This effect was statistically significant according to Rubin’s (1987) rules for pooling imputed data, t = 2.08, p < .05. According to the effect size analysis, mothers endorsed more depression symptoms than fathers (between a small and a medium effect), and this difference also was statistically significant according to Rubin’s (1987) estimation procedures, t = −1.90, p < .05. The difference between mothers’ and fathers’ reports of conflict at the follow-up was small; accordingly, analysis of the aggregate sample was not significant, t = 0.17, p = .86.
For the treatment groups (collapsed across gender), small to medium effects, according to Cohen’s (1988) conventions, were observed for coparenting conflict and nonacceptance. At the 12-year follow-up, mothers and fathers who mediated reported less coparenting conflict and more nonacceptance than parents who litigated their custody dispute. No mean differences were observed for depression scores. Submitting the estimates and standard errors back into NORM revealed that only the nonacceptance parameter was statistically significant, t = −2.12, p < .05. Corrected for average missingness, no significant differences were observed for conflict, t = 1.60, p = .11 or depression, t = −0.04, p = .97.
The final hypothesis predicted that early and later nonacceptance would be a significant predictor of continued coparenting conflict at the 12-year follow-up and, similarly, that concurrent conflict and depression would predict nonacceptance at the follow-up. Because the potential interaction between gender and nonacceptance in the prediction of conflict was also of interest, we ran each of the two regressions models separately for men and women.1 For the prediction of coparenting conflict at the follow-up, conflict at the initial assessment was entered as the first step in a hierarchical model. Treatment group (i.e., mediation or litigation), depression at the initial assessment and follow-up, and nonacceptance at the initial assessment and follow-up were entered in the second step of each model. When the outcome was nonacceptance at the follow-up, nonacceptance at the initial assessment was entered in the first step, followed by the other predictor variables in the second step. Once the models were run in each of the 12 data sets (6 each for mother and fathers), the estimates were submitted back into NORM (Schafer, 1999) to pool the data for significance testing, according to Rubin’s (1987) rules. For each significant parameter, the pooled estimates and associated significance values are reported.
For mothers, none of the variables were significant predictors of conflict at the follow-up once the standard errors were corrected for within- and between-imputations variance. For regression models predicting mothers’ nonacceptance, depression at the follow-up was a significant predictor (average β = .22), t(55) = 2.24, p < .05. With controls for earlier nonacceptance, women reporting more depression at the follow-up also reported more nonacceptance of marital termination. For fathers, none of the variables was significantly associated with conflict at the follow-up. However, both nonacceptance at the initial assessment (average β = .36), t(51) = 2.19, p < .05, and conflict at the follow-up (average β = −.22), t(51) = −2.88, p < .05, were associated with current reports of nonacceptance. The greater fathers’ nonacceptance was immediately after the initial custody resolution, the greater was their continued report of nonacceptance 12 years later. In contrast to the expectation that nonacceptance would be positively associated with conflict, there was a significant negative relation between nonacceptance and coparenting conflict at the 12-year follow-up. The more fathers reported fighting with their former partner over issues of coparenting, the more accepting of marital termination they were at the follow-up.
The chief aims of this study were to assess the rate of continued divorce-related difficulties in a sample of highly conflicted adults, to reexamine potential long-term differences between parents who mediated and litigated their custody disputes using statistical techniques for imputing missing data, and to apply these methods to increase power for regression models evaluating the long-term relations among coparenting conflict, nonacceptance, and depression. To our knowledge, this research is among only a handful of long-term empirical studies of coparenting conflict and nonacceptance following divorce (Amato, 2000; Ahrons & Rodgers, 1987; Hetherington & Kelly, 2002; Maccoby, Buchanan, Mnookin, & Dornbusch, 1993) and is the first to apply novel methods of statistical imputation to evaluate the long-term effects of divorce mediation.
The first hypothesis was offered to determine, with the nonimputed data, what percentage of parents remain stuck or canalized in their divorce-related problems more than a decade after the official end of their marriage. Consistent with the notion that psychological resilience is more of a rule than an exception following divorce (Amato, 2000; Hetherington & Kelly, 2002; Wang & Amato, 2000), both mothers and fathers who remained in this study adjusted well to the dissolution experience over time. For example, in terms of raw response rates, 0% of mothers and fathers reported never adjusting to the divorce at the 12-year follow-up, and 0% of mothers and 4% of fathers reported feeling like they would never get over the divorce. Despite the risks associated with divorce, most of the high-conflict parents who remained in this study fared well over time. These empirical results stand in contrast to the qualitative findings of Wallerstein and Blakeslee (1989), who reported that in their sample of 52 families, one half of the women and one third of the men were still intensely angry at their former spouses at the 10-year follow-up assessment. One methodological explanation for this discrepancy is that the current study uses psychometrically reliable and valid instruments for assessing coparenting conflict rather than relying on detailed clinical interviews (Wallerstein & Blakeslee, 1989).
Despite evidence for resilience in the current sample, caution should be taken in concluding that positive adaptation is a given or that well-being is necessarily the absence of all problems. The response rates from the present study also are consistent with Macobby et al.’s (1993; Macobby & Mnookin, 1992) and Ahrons and Rodger’s (1987) descriptions of parents who become “disengaged” or evolve into “dissolved duos” over time. Maccoby and Mnookin (1992) reported that more than 40% of conflicted divorced couples became disengaged parents with little or no contact with each other (and, subsequently, their nonresidential children) 5 years after divorce. More than a decade later, the present findings suggest that this pattern may be reified once set in motion. For instance, 41 % of mothers and 28% of fathers in our follow-up sample reported that they can never talk with their former spouse about problems with the children. In contrast to avoidance or disengagement, a small percentage of parents remained actively hostile. Twenty-five percent of mothers and 10% of fathers reported “almost always” engaging in angry disagreements with their former partner. For a subgroup of the parents in this sample, the results suggest that continued coparenting difficulties and hostilities can become a reality of postdivorce life.
Using imputed data, we conducted group comparisons between mothers and fathers and the mediation and litigation samples. As noted, the imputation approach greatly enhanced the available statistical power to detect group differences in the psychological outcomes over time. For example, the follow-up sample size reported by Emery et al. (2001) included 82 total participants. With our use of imputed data, the sample size in the current study rose by 36, to 118, which is equal to that reported in first intervention assessment (Emery et al., 1991). Additionally, because selective attrition was observed on the outcomes of interest, imputing data on the basis of early scores allows for the empirical assessment of the likely scores for these individuals had they remained in the sample.
Although the effect size analysis revealed that litigation participants reported more conflict at the follow-up, this result did not hold once the average rate of missingness was accounted for according to Rubin’s (1987) rules for assessing the statistical significance of imputed results. The nonacceptance group effect was significant according to these standards. Compared with parents who litigated their custody disputes, participants who mediated their settlements reported significantly more feelings of longing for their ex-partner and feeling like the separation was a mistake at the 12-year follow-up. This effect was not observed in the nonimputed follow-up data (see Emery et al., 2001); however, the average nonacceptance scores were almost exactly the same in the two studies. Thus, had the entire sample participated in the follow-up, the experimental evidence from this study indicates that mediation causes continued nonacceptance. One possible explanation is that mediation works to enhance overall relationship quality among former partners, and the result of a good relationship is continued longing for one’s former partner. With respect to nonacceptance, parents who feel like they can work cooperatively with their former spouse may begin, over time, to wonder why they chose to make the split. An alternative explanation is equally tenable; namely, that adversarial litigation fosters such considerable ill will between ex-partners that individuals who litigated express little remorse, continued longing, or nonacceptance of the divorce. Given that nonacceptance is highly associated with depression (Kiecolt-Glaser et al., 1987; Wang & Amato, 2000), future research efforts are needed to investigate the factors keeping individuals stuck in a cycle of regret and yearning for a former partner.
As evidenced by the gender comparisons, nonacceptance at the follow-up was significantly greater for fathers than for mothers. In this sample, many more fathers than mothers did not want their marriage to end (Emery et al., 1991). Again, the mean nonacceptance scores derived from the MI techniques were very similar to the rates reported in the first 12-year follow-up without imputed data (Emery et al., 2001). Thus, once gains in power were added through imputation, the results indicate that fathers continue to report more nonacceptance than mothers. A number of studies suggest that men fare worse following the end of marriage, and it is commonly believed that men’s primary social support revolves around their wife, whereas women develop more sophisticated and supportive networks of friends and family (Burman & Margolin, 1992; Kiecolt-Glaser & Newton, 2001). As a result, when relationships dissolve, men struggle much more. Indeed, the absolute rate of fathers’ nonacceptance was substantial even at the 12-year follow-up. Although fathers reported large decreases in rates of nonacceptance over time (Emery et al., 2001), their average reports at the 12-year follow-up were greater than mothers’ reports at the initial assessment.
The regression analyses with the imputed sample reveal that, for mothers, depression at the follow-up was highly associated with concurrent nonacceptance, and, for fathers, early nonacceptance and co-parenting conflict were associated with concurrent nonacceptance. When MI is used, separating mothers and fathers for these analyses is equivalent to assessing an interaction between gender and the other independent variables. Thus, the three significant associations can also be described as significant interactions by gender, which suggests that the correlates of better or worse adaptation appear different for men and women. In contrast with the hypothesis that conflict would be positively associated with nonacceptance, the relation between these constructs was negative for fathers. Fathers who reported less ongoing coparenting conflict also reported more nonacceptance at the follow-up. We expected that conflict would be positively associated with nonacceptance largely because we believed that contesting custody disputes is a sanctioned means of contesting the end of marriage. For fathers, however, the findings suggest that a more systematic relation involves less conflict/better coparenting and more longing for a former partner. These findings raise the question of whether there are emotional costs of a good coparenting relationship among divorced parents, at least for the men in the current study. Emery and Coiro (1997) suggested that resilience among children of divorce is often associated with subtle negative consequences, including noted distress about the struggle and painful feelings about childhood. For fathers who feel able to successfully communicate with their former partner and have good coparenting relationships, one of the consequences may be a continued longing for their ex-partner and regret over past events.
The association between depression and nonacceptance was anticipated; however, it was not expected that this relation would hold for mothers but not fathers. One explanation for these findings is that nonacceptance was rare among the mothers in this sample at the initial assessment (Emery et al., 1991). Hence, the emergence of longing for a former partner and regret over the divorce may represent a more extreme reaction for mothers, either stemming from or resulting in elevated levels of depression. For fathers, nonacceptance was an immediate and more commonplace reality of postdivorce life and, as such, was not associated with an atypical grief reaction, as it may be for mothers. Although this hypothesis requires further research attention, the evidence from this study is clear in suggesting differential patterns of functioning for men and women over time. Because we also found evidence that mediation causes nonacceptance over time, careful attention should be paid to the possible unanticipated consequences of coparenting mediation interventions. One resulting implication is that mediators should consider the psychological processes operating for both individuals and couples (and men vs. women) within mediation. Various approaches to family mediation place different emphasis on whether to attend to the psychological needs of clients (C. J. A. Beck & Sales, 2001). If an unanticipated effect of therapeutic mediation is nonacceptance, and if this construct is reliably associated with depression for women, mediators should be aware of the possibility that successful settlements can portend negative psychological consequences, including continued longing for a former partner and sadness.
The results of this study also raise the larger question of psychological change processes following divorce. A number of investigators have suggested that inquiry into how adults grieve the loss of their marriage can provide a mechanistic understanding of postdivorce psychological adjustment (Crosby, Lybarger, & Mason, 1986; Dillon & Emery, 1996; Emery, 1994; C. A. Gray & Shields, 1992; Hetherington & Kelly, 2002; Kitson & Holmes, 1992; Thompson & Spanier, 1983; Weiss, 1975). Given the interrelated nature of sadness, positive coparenting, and ongoing conflict in the present study, we suggest that future research concentrate on unveiling how adults move from states of relative distress toward resolution and the factors underpinning better or worse adaptation over time. One strategy for doing so is to use experience-sampling methodologies to capture the daily emotion regulation strategies individuals invoke to cope with relationship dissolution (cf. Bolger, Davis, & Rafaeli, 2003).
Despite the fact that this work is among the first long-term empirical investigations of custody conflict and nonacceptance among high-conflict divorced parents and the first to use statistically imputed data, some limitations should be noted. First, with respect to the imputation analyses, NORM (Schafer, 1999) was not designed to impute dependent data. Given the fact that we used earlier scores on the same individuals to impute later scores, results from the imputed analyses may be “overly” significant (see Schafer & Olsen, 1998). One piece of evidence suggesting that the imputed effects are not inflated is the consistency with the nonimputed outcomes. For example, the mean scores for the measures of psychological adjustment are very similar between the non-imputed follow-up and this current study, which imputed across six data sets. It appears that the imputation did not change the overall patterning of the results but instead served to raise the overall statistical power. Second, this study is based entirely on the self-report of adults who were highly conflicted at the time they initially petitioned the courts for a custody settlement. Because this degree of acrimony represents approximately 10% of divorcing couples (Emery, 1994), obvious caution is needed in extending the findings to all divorced adults. Moreover, as noted in other reports (Emery et al., 1991, 2001), the external validity of the mediation intervention may be called into question for a variety of reasons, and the results may not generalize to mediation conducted in other jurisdictions, with other samples, or with another underlying dispute resolution model.
This study used advances in statistical imputation of incomplete data to evaluate adults’ long-term psychological adjustment to divorce in a sample of families who were randomly assigned to either mediate or litigate child custody disputes in the middle 1980s. Consistent with the notion that most adults fare well following divorce, the results indicate that the majority of parents who remained in the study reported having adjusted well to the end of their marriage. At the same time, a subset of adults reported continued hostilities 12 years after initial child custody settlements. Using imputed data, we found that parents who mediated their custody disputes reported significantly more nonacceptance of marital termination than parents who litigated. Additionally, we found a significant Gender × Conflict interaction for predicting continued nonacceptance. Specifically, fathers reporting less conflict at the follow-up also reported greater nonacceptance; this relation did not hold for mothers. Given the significant relation between continued nonacceptance and less coparenting conflict among men and the causal effect of mediation on nonacceptance, we suggest that an unanticipated consequence of successful divorce mediation may be that it keeps parents attached in a way that is not always beneficial for their psychological health. Understanding the processes set in motion by custody mediation and the mechanisms associated with better or worse adjustment over time should continue to be a major thrust of research on how adults cope with marital transitions.
This work was supported in part by grants from the W.T. Grant Foundation to Robert E. Emery, PhD, and by National Research Service Award MH 12783-02 and a fellowship from the Center for Children, Families, and the Law at the University of Virginia to David A. Sbarra, PhD.
1We adopted this approach because potential interactions among imputation variables are not considered when NORM derives the MIs. Thus, to evaluate interactions with the imputed data, one must conduct entirely separate analyses between groups.
David A. Sbarra, Department of Psychology, University of Arizona.
Robert E. Emery, Department of Psychology, University of Virginia.