Rebound from Marital Conflict and Prediction *

JOHN MORDECHAI GOTTMAN, Ph.D.t ROBERT WAYNE LEVENSON, Ph.D.*

Marital interaction has primarily been laboratory reported that primarily nega­ examined in tl. . context of conflict resolu­ tive affect predicted divorce (Gottman, tion. This study investigated the predictive 1993, 1994; Gottman & Levenson, 1992), a ability of couples to rebound from marital subsequent 9-year longitudinal study we conflict in a subsequent positive conversa­ conducted with newlyweds (with a newer tion. Results showed that there was a great version of our coding system) deal of consistency in affect across both found that low levels of positive affect conversations. Also examined was the during the first few months of ability of affective interaction to predict also predicted later divorce (Gottman, divorce over a 4-year period, separately in Coan, Carrere, & Swanson, 1998). With each of the two conversations. It was this newer emotion coding system focus­ possible to predict divorce using affective ing just on affect, our laboratory could now variables from each conversation, with code not only the conflict-resolution conver­ 82.6% accuracy from the conflict conversa­ sation, but also the couple's discussion of tion and with 92.7% accuracy from the the "events of the day" conversation, and positive rebound conversation. we now had a sensitive instrument for measuring positive as well as negative Fam Proc 38:287-292, 1999 affect. This nonconflict aspect of marital inter­ action has received surprisingly little at­ BSERVATIONAL investigations of mari­ tention. Two exceptions are Gottman tal interaction tend to sample O (1979, 1980) and Birchler, Weiss, and Vm­ couples' behavior using conversations de­ cent (1975). Gottman (1979, 1980) re­ signed to elicit conflict and problem solv­ ported that it was possible to discriminate ing (see Gottman, 1979). Although our distressed from nondistressed couples whether they talked about a conflict issue * This research was supported by grant MH35997 in their marriage or worked on a "fun from the National Institute of Mental Health to the deck" task in which the admonition to the authors, and by Research Scientist Award K2MH00257 to the first author. couples was to look over the deck of items, t James Mifflin Professor of Psychology, Depart­ plan, reminisce, and have a good time. ment of Psychology, University of Washington, Box Birchler, Weiss, and Vincent (1975) used a 351525, Seattle WA 98195; e-mail:johng@u. self-report diary measure of "pleases" and washington.edu. "displeases," a precursor of the :j: Director, Institute for Personality Research, Department of Psychology, University of California, Observation Checklist. In the home envi­ Berkeley. ronment, distressed partners recorded sig- 287 Process, Vol. 38, No.3, 1999 © FP, Inc. 288 / FAMILY PROCESS nificantly fewer pleasing and significantly affect than the MICS, for both positive greater displeasing events than was the and negative affect (Gottman, 1996). This case for nondistressed partners. The ratio system also had the advantage that, un­ of pleases to displeases at home discrimi­ like the MICS, it could easily be used to nated the groups quite dramatically: the code both conflict and nonconflict interac­ ratio was 29.66 for nondistressed and 4.30 tion. for distressed couples. Also, during the Four years later, the couples in our conflict resolution interaction, these au­ study were again contacted and their mari­ thors computed a similar ratio using the tal status was assessed. This made it Marital Interaction Coding System possible for us to ask the questions of (MICS), an observational coding system: whether one could assess the most salient distressed couples produced an average of couples' affective behavior in terms of 1.49 positives per minute, while nondis­ divorce prediction equally well in situa­ tressed couples produced an average of tions other than conflict resolution, and 1.93 positives per minute, a significant whether the patterns of prediction varied difference. Birchler, Weiss, and Vincent as a function of the type of conversation. also had a period of 4 minutes of "free METHODS interaction" in which couples were in­ structed to "talk about anything while we Participants are setting up the equipment." There were Couples were originally recruited in significant differences in these conversa­ 1983 in Bloomington, Indiana, using news­ tions between distressed and nondis­ paper advertisements. Approximately 200 tressed couples only on negative, but not couples who responded to these advertise­ positive interaction (distressed negative ments completed a demographic question­ rate 1.07 "negatives" per minute; nondis­ naire and two measures of marital satisfac­ tressed negative rate .73 "negatives" per tion, for which they were paid $5.00. From minute). this sample, a smaller group of 85 couples We reasoned that if the conflict discus­ was invited to participate in the labora­ sion is followed by a positive conversation, tory assessments and to complete a num­ we can expect that there may be some ber of additional questionnaires. The goal spillover from conflict to nonconflict inter­ of this two-stage sampling was to insure action. We thought that the couple's abil­ that we came close to obtaining a distribu­ ity to rebound from negative affect that is tion of marital satisfaction in which all generated by a conflict discussion might parts of the distribution would be equally turn out to be a predictor of the eventual represented. Complete sets of usable fate of the marriage. The investigation of physiological data were obtained from 79 this contention was the purpose of the of these 85 couples. These 79 couples could present article. be described as follows. were Many marital interaction observational about 32 years old (SD = 9.5 years); (b) coding systems (such as the MICS), were were about 29 years old (SD = 6.8 specifically designed for problem-solving years). They were married an average of 5 interaction, and they may not be appropri­ years (SD = 6.3 years). The average mari­ ate for coding the nonconflict interaction. tal satisfaction for husbands was (average For this report, we employed an observa­ of Locke-Wallace and Locke-Williamson tional system, the specific affect coding scales) = 96.80 (SD = 22.16); and for system (SPAFF) that obtained consider­ wives the average marital satisfaction ably more detail and specificity in coding was 98.56 (SD = 20.70). GOTTMAN AND LEVENSON / 289 Procedures be silent and not interact, and during which we obtained baseline physiological Interaction Session measures (not discussed in this report). Couples arrived in the laboratory after Details of the procedures for setting up having been apart for at least 8 hours. these conversations are available upon They had three 15-minute conversations: request. (1) events of the day; (2) conflict resolution (discussion of a problem area of continu­ Followup ing disagreement); and (3) a pleasant Four years after the initial assessment, topic. In this article we report on only the the original subjects were recontacted and latter two conversations. at least one spouse (70 husbands, 72 The conversations were always in the wives) from 73 of the original 79 couples order shown above for the following rea­ (92.4%) agreed to participate in the fol­ sons. We wanted couples to have the events lowup. completed a set of ques­ of the day conversation first because we tionnaires assessing marital satisfaction, wanted to sample this kind of everyday and items relevant to possible marital nonconflict interaction, and we wanted to dissolution. The two dichotomous vari­ begin our laboratory session with a re­ ables, serious considerations of divorce in union conversation that would seem natu­ the 4 years since Time-l and Time-2, and ral and help make subjects comfortable actual divorce will serve as the external with the laboratory situation. It was also criterion variables in our report. the most natural way to start the couples' conversation after they had been apart for Data Coding and Analysis 8 hours. In pilot work in which we began The videotapes of the interaction were with the conflict conversation, we found coded using the Specific Affect Coding that there was an undesirable spillover of System (SPAFF), which focused on spe­ negative affect into the events of the day cific . Coders were first trained discussion. using the Ekman and Friesen (1978) We followed this events of the day con­ FacialAction Scoring System, with a set of versation with the couples' conflict discus­ our own audiotapes for recognizing affect sion. After filling out a problem inventory, in the voice, and a set of videotapes for the spouses were interviewed about an detecting specific features in affect using area of continuing disagreement in their paralinguistic, contextual, linguistic, and marriage and asked to discuss this area kinesic channels. However, the training and try their best to resolve the issue in went beyond specific features and observ­ the next 15 minutes. The conflict discus­ ers were taught to use a Gestalt approach sion was followed by the couple filling out to recognizing specific emotions in all an inventory of positive topics and an channels combined. The initial training of interview in which they were asked to coders took over 200 hours. Coders classi­ identify a topic that they would both enjoy fied each speech act (usually a phrase) discussing. The plan for the post-conflict within a turn at speech as affectively conversation was twofold: to debrief sub­ neutral, as one of five negative affects jects so they could recover from the con­ (anger, contempt/disgust, sadness, fear, flict conversation, and to assess the whining), or as one of four positive affects amount of recovery. Each conversation (affection/caring, humor, interest/curios­ was preceded by a 5-minute pre-conversa­ ity, and joy/enthusiasm). Coding manuals, tion period in which couples were asked to training and test video and audiotapes are

Fam. Proc., Vol. 38, Fall, 1999 290 / FAMIlY l'IW( 'EBB available from the first author. The num­ Divorce Prediction ber of onsets of each code (that is, the Using the SPAFF data, two discriminant number of episodes) for each code, collaps­ function analyses were conducted, one ing across speech acts within a turn at separately for each conversation. Codes speech; hence, for example, two consecu­ were entered into the discriminant func­ tive speech acts by a that tion only if their univariate F-ratios were received the same code would be collapsed statistically significant. For the conflict into one. The kappa coefficient of reliabil­ discussion, four SPAFF codes were en­ ity, controlling for chance agreements, tered: husband interest and contemptJ was equal to 0.75 for the entire SPAFF disgust, and interest and sadness, coding. with canonical correlation 0.43, and X2 (4) = 13.23,p = .0102. The percent correct RESULTS prediction in the 4 years of stability and divorce was 82.6%. For the subsequent Comparisons of Conversations positive rebound conversation, four SPAFF To compare whether the conversations codes were entered, husband anger, hus­ succeeded in inducing different amounts band contemptJdisgust, wife affection, and of positive and negative emotionality, the wife anger, with canonical correlation total amounts of negative and positive 0.59, with X2 (4) = 27.56,p < .00001. The affect were computed for husband and percent correct prediction in the 4 years of wife separately. Positive affect was de­ stability and divorce was 92.7%. fined as the sum of humor, affection, interest, and joy. Negative· affect was Profile Analysis defined as the sum of anger, contempt! Profile analysis was conducted by exam­ disgust, whining, sadness, and fear. Neu­ ining group means for each variable as tral affect was excluded from these compu­ well as correlations of each variable with tations. A series of paired t-tests were the discriminant function (see Table 1). In conducted. the conflict discussion, the results of the In comparing the conflict conversation profiles were that, at Time-I, couples who with the positive conversation, for the stayed married had more husband inter­ conflict conversation there was: less hus­ est, less husband contemptldisgust, more band positive affect, t(71) 8.38,p < .001, = wife interest, and less wife sadness than with means of 49.61 for the positive conver­ couples who eventually divorced. In the sation and 21.15 for the conflict conversa­ positive rebound conversation, the results tion; there was less wife positive affect, of the profiles were that at Time-I, couples t(71) = 8.76,p < .001, withmeansof50.83 who stayed married had less husband for the positive conversation and 20.99 for anger, less husband contemptldisgust, the conflict conversation; there was more less wife anger, but also less wife affection husband negative affect, t(71) = 8.54, p < than couples who divorced . .001, with means of 12.27 for the positive conversation and 42.11 for the conflict conversation; there was more wife nega­ Consistency of Affective Behavior tive affect, t(71) = 10.71, p < .001, with Table 2 is a summary of the correlations means of 13.87 for the positive conversa­ across the same affects across the two tion and 49.42 for the conflict conversa­ conversations. The highest correlation tion. Hence, the conflict discussion gener­ was obtained for overall amount of affec­ ated more negative affect and less positive tive behavior, with lower correlations for affect than the positive conversation. specific affects. In predicting from the (iOTTlIIA:\ AND LEVENSO~ / 291

TABLE 1 The divorce prediction results also sug­ Profile Analyses: Means [or Va ria hie.' in the Dimrce gest that patterns of emotional interac­ Prediction Discriminant Function Analyses tion can predict divorce or marital stabil­ Still ity with fairly high accuracy in either of Variable Together Divorced the conversations. In the conflict discus­ Conflict Conversation sion, the predictors were husband inter­ Husband Interest 10.88 4.63 est, husband contempt/disgust, wife inter­ Husband ContempU est, and wife sadness. Disgust 4.62 10.13 The divorce prediction results were even Wife Interest 12.14 3.75 better in the positive rebound conversa­ Wife Sad 2.18 5.88 Positive Conversation tion, 92.7% compared to 82.6% correct Husband Anger 1.06 8.29 classification. In the positive rebound con­ Husband ContempU versation, the results of the profiles were Disgust 2.05 11.14 that, at Time-I, couples who stayed mar­ Wife Anger 2.13 8.71 ried had less wife anger, less husband Wife Affection 2.52 5.86 contemptJdisgust, but also less wife affec­ Note: In this computation, the old SPAFF interest tion than couples who divorced. The only code was estimated as interest plus validation. Data are the number of onsets (episodes). finding that is difficult to explain is the finding that, for couples who divorced, wives were more affectionate in the posi­ conflict discussion to the positive conversa­ tive conversation than wives in tion, the highest correlations were for the that stayed intact. However, we can pro­ amounts of nonneutral affect for both pose the hypothesis that since among husband and wife. The specific affects that demonstrated consistency were husband humor, interest, joy, anger, contemptJ TABLE 2 disgust, whining, fear, but not affection, Correlations ofAffective Behaviors Across Conversations and sadness; for the wife the specific affects that demonstrated consistency were Variable Conflict & Positive humor, interest, anger, contemptJdisgust, Husband whining, and fear, but not affection, joy, Total Emotionality .71*** and sadness. Humor .28** Affection .18a DISCUSSION Interest .50*** ExcitemenUJoy .24* Comparisons of positive and negative Anger .52*** affect across the conversations show that, ContemptlDisgust .40*** despite the conversations having occurred Whining .65*** in a fixed order, the conversations were Sadness .18a significantly different in the manner ex­ Fear .72*** Wife pected by the experimental inductions. Thtal Emotionality .73*** The conflict conversation induced signifi­ Humor .26* cantly more negative and less positive Affection .00 affect. Also, despite the positive conversa­ Interest .46*** tion having followed the conflict conversa­ ExcitemenUJoy .14 Anger .38*** tion, the affect it induced was still far ContemptlDisgust .38*** more positive and less negative than the Whining .48*** affect the conflict conversation induced. Sadness .08 Most couples were able to rebound from Fear .60*** the negativity of marital conflict. a p < .10; * P < .05; ** P < .01; *** P < .001 Fam. Proc., Vol. 38, Fall, 1999 292 / FAMILY PROCESS couples who divorced the higher level of well in inducing less positive and more wife affection is contextualized by higher negative affect, and the positive induction levels of husband contempt/disgust, it may was also successful. However, at this time, represent their compliance to being dero­ until there is further research, we cannot gated by their husbands, and so their draw conclusions about the nature of the affection may be an index of a process that couple's affective behavior conversations is not very functional; their affection in independent of their order. Nonetheless, the face of their husband's disgust and we are quite encouraged by the divorce contempt may represent a time bomb for prediction results from each ofthe conver­ this ailing marriage. We have observed a sations. similar pattern with physically abused wives. In these marriages the wife's affec­ REFERENCES tion in the face of her husband's contempt Birchler, G., Weiss, R., & Vincent, J. (1975). and disgust toward her may represent Multimethod analysis of social reinforce­ compliance, fear, and holding on to a ment exchange between maritally distressed dream that the contempt is ephemeral. and nondistressed spouse and stranger dy­ These results suggest that marital inves­ ads. Journal of Personality and Social tigators and therapists might benefit from Psychology 31: 349-360. considering the construct of a couple's Ekman, P., & Friesen, w.v. (1978). Facial resilience in being able to rebound from Action Coding System. Palo Alto CA: Consult­ the negative affect during marital conflict. ing Psychologists Press. The results of this investigation also pro­ Gottman, J.M. (1979). Marital interaction: vide support for the consistency ofemotion­ Experimental investigations. New York: Aca­ demic Press. ally based marital interaction patterns Gottman, J.M. (1980). The consistency of across consecutive conversations that var­ nonverbal affect and affect reciprocity in ied a great deal in the amount of positive marital interaction. Journal of Consulting and negative affect that they induced. and Clinical Psychology 48: 711-717. Despite these overall differences in emo­ Gottman, J.M. (1993). The roles of conflict tionality, couples were quite consistent in , escalation or avoidance in most emotions across these conversations. marital interaction: A longitudinal view of An intentional design of the present five types of couples. Journal of Consulting investigation was that the order of conver­ and Clinical Psychology 61: 6-15. sations was not randomized or counterbal­ Gottman, J.M. (1994). What predicts divorce? anced. Our intention was to assess spill­ Hillsdale NJ: Lawrence ErlbaumAssociates. Gottman, J.M. (1996). What predicts divorce: over of negative affect from the conflict to The measures. Hillsdale NJ: Lawrence the positive conversation. We cannot say ErlbaumAssociates. with confidence whether the results would Gottman, J.M., Coan, J., Carrere, S., & be the same if the behavior observed dur­ Swanson, C. (1998). Predicting marital hap­ ing the positive conversation was not a piness and stability from newlywed interac­ function of the conversation that preceded tions. Journal of Marriage and the Family it. For example, would the same or differ­ 60: 5-27. ent effects be obtained in divorce predic­ Gottman, J.M., & Levenson, R.w. (1992). tion on the positive conversation if it had Marital processes predictive oflater dissolu­ been the only conversation in the study? tion: Behavior, physiology, and health, Jour­ At this time, we cannot say. The results of nal of Personality and Social Psychology 63: 221-233. the t-tests do suggest that order effects, if they exist at all, were not large. The Manuscript received March 11, 1999; accepted conflict induction seemed to work quite April 6, 1999. j