A Theory of Marital Dissolution and Stability

Research is presented on the prospective longitudinal prediction of marital dissolution. First, a cascade toward marital dissolution is described. Second, the cascade is predicted with variables from a balance theory of marriage. Third, there are process and perception (the distance and isolation cascades) cascades related to the cascade toward dissolution. The importance of "flooding" is discussed, as well as a mechanism through which negative perceptions (which are 2 dimensional) become global and stable and through which the entire history of the marriage is recast negatively. The role of physiology is outlined. A theory is presented in which a "core triad of balance" is formulated in terms of 3 weakly related thermostats (connected by catastrophe theory) and related to the distance and isolation cascade. Implications for a minimal marital therapy are discussed. Correspondence may be addressed to John Mordechai Gottman, Department of Psychology, University of Washington, NI—25, Seattle, Washington, 98195. Received: November 4, 1992 Revised: December 14, 1992 Accepted: December 18, 1992 Today separation and divorce are common phenomena. Separation appears to be a trustworthy road to divorce rather than to reconciliation. When couples separate, about 75% of these separations will end in divorce ( Bloom, Hodges, Caldwell, Systra, & Cedrone, 1977 ). Current estimates put the divorce rate in the United States somewhere between 50% (as of 1970; Cherlin, 1981 ) and a startling 67% (as of 1989; Martin & Bumpass, 1989 ). The divorce rate for second marriages tends to be about 10% higher than the rate for first marriages ( Glick, 1984 ). Researchers now know that separation and divorce have strong negative consequences for the mental and physical health of both spouses. These negative effects include an increased risk for psychopathology, an increased number of automobile accidents–some resulting in fatalities–and an increased incidence of physical illness, suicide, violence, homicide, and mortality from diseases (for a review, see Bloom, Asher, & White, 1978 ). Marital disruption may not merely be related to these negative life events, it may actually be among the most powerful predictors of them ( Holmes & Rahe, 1967 ). There is even evidence from one large sample 9-year epidemiological prospective study on the predictors of dying or staying alive that the stability of a marriage is the best predictor of these two phenomena, even when controlling for factors such as initial health and health habits ( Berkman & Syme, 1979 ; Berkman & Breslow, 1983 ). The Berkman and Syme study suggests that the health-buffering effects of marriage are all granted to men, consistent with suggestions made by Bernard (1982) . However, recent research ( Gottman & Levenson, 1992 ; Kiecolt-Glaser et al., 1987 ; Kiecolt-Glaser et al., 1988 ) suggests that the opposite is true, namely, that an ailing marriage or marital dissolution most strongly affects the physiological functioning of women. Studies are not consistent in this regard (e.g., cf. Levenson & Gottman, 1985 ), but, in general, the evidence does not support Bernard's contention. To resolve this inconsistency of costs and buffers offered by marriage as a Journal of Family Psychology © 1993 by the American Psychological Association June 1993 Vol. 7, No. 1, 57-75 For personal use only--not for distribution. Page 1 of 19 11/27/2000 http://spider.apa.org/ftdocs/fam/1993/june/fam7157.html function of gender, Gottman (in press) found that the illness of women is directly affected by marital distress, whereas for men it is mediated through loneliness. He speculated that this result is probably due to the relatively meager social support systems of men as compared with those of women; these systems for most men are restricted to their wives. It is probably the case that an ailing marriage negatively affects the physiology of both partners. Also, there is now convincing evidence to suggest that marital distress, conflict, and disruption are associated with a wide range of deleterious effects on children, including depression, withdrawal, poor social competence, health problems, poor academic performance, and a variety of conduct-related difficulties ( P. A. Cowan & Cowan, 1987 , 1990 , 1992 ; Easterbrooks, 1987 ; Emery, 1982 , 1988 ; Emery & O'Leary, 1982 ; Forehand, Brody, Long, Slotkin, & Fauber, 1986 ; Gottman & Katz, 1989 ; Hetherington, 1988 ; Hetherington & Clingempeel, 1992 ; Hetherington, Cox, & Cox, 1978 , 1982 ; Howes & Markman, 1989 ; Katz & Gottman, 1991a , 1991b ; Peterson & Zill, 1986 ; Porter & O'Leary, 1980 ; Rutter, 1971 ; Shaw & Emery, 1987 ; Whitehead, 1979 ). There is evidence from two U.S. national probability samples that adults who experienced a divorce as children are under considerably more stress than those who did not ( Kulka & Weingarten, 1979 ; Glenn & Kramer, 1985 ). These adults report less satisfaction with family and friends, greater anxiety, that bad things more frequently happento them, and that they find it more difficult to cope with life's stresses in general. There is also evidencefor a reasonably reliable phenomenon of the intergenerational transmission of divorce, but the effect isnot large (e.g., see Pope & Mueller, 1979 , p. 109), and this relationship is not found in every study. Forexample, in the Kelly and Conley (1987) 35-year prospective study these relationships were notstatistically significant (although the trend was there). Despite these negative consequences of marital dissolution, researchers still do not understand howmarriages may be patterned in some way that spells their eventual destruction. In decade review papers,both White (1990) and Price-Bonham and Balswick (1980) noted that little is known about the maritalprocesses that may be predictive of divorce. The research agenda is relatively clear. What needs to beknown is whether there are specific trajectories toward marital dissolution or marital stability that aresystematically related to the qualities of a marriage. Furthermore, this knowledge must come fromprospective longitudinal studies rather than from retrospective accounts of failed marriages (e.g.,Ponzetti & Cate, 1988 ; Vaughn, 1990 ), as reconstructions of the past are notoriously unreliable.Critical in this research agenda are two goals: good prediction of which couples will be on whichtrajectory and a highly specific empirically based theory of the marital processes associated withdissolution. Previous Studies Predicting Dissolution Empirical research has not been very successful at predicting which married couples will separate ordivorce and which married couples will stay together. The epidemiological attempts at understanding thechanges in divorce in this century have not tried to predict which couples in a cohort might divorce.Instead, demographic correlates of stability have been studied, and epidemiologists have attempted todiscover variables that would show the same kinds of patterns over time as the divorce rate time-seriesand that could reasonably account for variation in divorce rates over time (see Cherlin, 1981 ).Lamentably, studies attempting to specify interactional behaviors and processes that are antecedents ofmarital dissolution have been quite rare. The current lack of knowledge concerning which patterns ofmarital interaction lead to marital dissolution stems in part from the fact that, in most studies, divorceand separation have been viewed as independent rather than dependent variables. This means that thesestudies have been primarily concerned with the effects of marital dissolution on other variables and onthe adjustment of spouses and children to marital dissolution.Page 2 of 19 11/27/2000http://spider.apa.org/ftdocs/fam/1993/june/fam7157.html Of the many published studies to date with the terms marital separation or divorce in their titles, onlysix are prospective longitudinal studies that have attempted to predict future separation and divorce( Block, Block, & Morrison, 1981 ; Bentler & Newcomb, 1978 ; Constantine & Bahr, 1980 ; Fowers &Olson, 1986 ; Kelly & Conley, 1987 ; Sears, 1977 ). What did these studies find? Fowers and Olson(1986) were completely unable to predict divorce or separation with their instrument called PREPARE,which assesses disagreements between prospective spouses in 11 different areas. Unfortunately, theseauthors also combined separated and divorced couples into one group. This is a mistake because thecorrelates of separation may not be identical to the correlates of divorce, and it is important to keep themapart for the time being until more is known. The PREPARE instrument does not discriminate betweencouples who were less happily married but stayed together and couples who separated or divorced.Fowers and Olson reported only a discriminant analysis between the happily married and those whoseparated or divorced, thus confounding two factors, marital status and marital happiness. ForPREPARE, they claimed as validity evidence its high correlations with marital satisfaction, so it is likelythat what they were in fact measuring with PREPARE was one aspect of marital satisfaction (probablyconsensus, as the instrument ostensibly assesses consensual agreement on 11 areas of marriage).Markman, Floyd, Stanley, and Storaasli (1988) critiqued the commonly held belief that compatibilitybetween prospective spouses ensures marital stability. They quoted G. Levinger (1966) as suggestingthat what may count is not compatibility, but how the couple handles the inevitable incompatibilitiesthey will encounter. Then there is a prospective study by Sears (1977) . There is an interesting history to the Searsprospective study. Terman and Oden (1947) had reported that a self-report measure of emotionalstability administered to gifted subjects when they were 7 to 14 years old was related to maritalhappiness 18 years later (the relationship, although statistically significant, was only 0.25). In fact, in1940 Terman actually developed a "marriage aptitude test" to give to his young geniuses. Theconcurrent correlation of Terman's marital aptitude score with marital satisfaction was 0.62, and therelationship was mainly due to what he called "neuroticism" items, not childhood family background. Itturned out, in fact, that the members of the couples who divorced between 1940 and 1946 had maritalaptitude scores that were a standard deviation below those who stayed married. Sears later related these1940 marital aptitude scores to marital outcome in 1972 (When the average age of the original subjectswas 62). The correlations were 0.28 for women and 0.12 for men. Although for women this correlationwas statistically significant, the prediction was quite weak. Kelly and Conley (1987) were stronglyinfluenced by the weak Sears result and added the commendable innovation that the personalities oftheir subjects should not be assessed by self-report but rather by other people in the subjects' friendshipnetworks. Next is a prospective study of divorce ( Block et al., 1981 ) that emerged from the landmark longitudinalstudy of child development by the Blocks. The Blocks used these data to search their data to determinewhether they had any precursor predictors of parental divorce. In the Block et al. article, one variableemerged as a predictor. Parental disagreement about child rearing practices from 57 families when thechild was 3.5 years old significantly discriminated between the intact and divorced groups 10 years later. Constantine and Bahr (1980) , in a 6-year longitudinal study, interviewed over 3,500 men between theages of 14 to 24 years by telephone. The study used an 11-item subscale of the 29-item scale of Rotter(1966) ; an example of the items was "Many of the unhappy things in people's lives are partly due to badluck." Men who divorced or separated in the 6-year period were compared with men who remainedmarried. The total score on the locus-of-control measure was not significantly related to marital stability.Although there was a similar lack of statistical significance on the Fate scale (belief that events are dueto luck or fate) and the Personal scale (belief that a person's own behavior influences hisreinforcements), the Leadership scale discriminated the two groups: The stable group had a moreinternal orientation (meaning that the men who remained married thought that becoming a leader wasPage 3 of 19 11/27/2000http://spider.apa.org/ftdocs/fam/1993/june/fam7157.html due to skill and not luck). It is hard to know what to make of this result. Bentler and Newcomb (1978) used the Bentler self-report personality test with 77 newlywed coupleswho were followed for 4 years. Separated and divorced couples were combined into one category. Theyfound that couples who remained married were more similar in age, interest in art, and attractivenessthan couples who separated or divorced, although husband—wife differences were not statisticallysignificantly different across the groups for any dimension. Men who separated or divorced describedthemselves as more extraverted, more invulnerable, and more orderly than men who stayed married.Women who separated or divorced described themselves as less clothes conscious and less congenialthan women who stayed married. These personality dimensions do not provide a clear theoreticalpicture, nor do they tell us about the dynamics of marriage that might be associated with a longitudinalcourse toward dissolution. Kelly and Conley (1987) had acquaintances of the couple rate the partners' personalities. Theirs was aprospective 35-year longitudinal study of marital stability. They reported that the men who remained married were more conventional, less neurotic, and had greater impulse control than those whodivorced. A similar pattern was found for women, with the additional finding that women who stayedmarried were judged as higher in emotional closeness and lower tension in their families of origin.Unfortunately, Kelly and Conley reported no statistical tests in their article. However, in aprepublication version of the article that they sent to me ( Kelly & Conley, 1985 ), they reported that thestably married and divorced groups differed by about two thirds of a standard deviation on thediscriminant function. It is frustrating not to have statistical tests for this important study. Nonetheless,this latter result can be translated into an equivalent point-biserial correlation coefficient of 0.25 (seeGlass & Stanley, 1970 , p. 163, Equation 9.5). These results suggest that adjusted people have morestable marriages. A rival hypothesis is that the friends of the couples, who presumably have intimateknowledge of their friends and the marriages, used the personality items to tap a dimension of distressedmarital interaction patterns or negative perceptions of the marriage that they observed in socializing withtheir friends. That is, the validity of the friends' ratings of personality variables was not first clearlyestablished. Terman, Buttenweiser, Ferguson, Johnson, and Wilson (1938) found that the judgments of acouple's friends of their marital satisfaction correlated about 0.60 with the couple's actual maritalsatisfaction. The findings from this handful of studies are not conclusive. In fact, they are somewhat hard tointegrate. Although effect sizes in these studies were not particularly large, the fact that they had anyability to predict dissolution at all is encouraging for additional efforts at longitudinal prediction usingthe same and other methods. An important methodological improvement would be the addition of directobservation of marital behavior, which could provide greater descriptive clarity in prospectivelongitudinal research and might account for greater amounts of variance in marital dissolution. Recent Findings From Laboratories of Levenson and Gottman In this article I summarize some of the major findings on the longitudinal study of couples to date andpresent a theoretical formulation that organizes these results. The basic paradigm in our laboratories isone that collects simultaneous synchronized videotapes and psychophysiological measures. Couples arevideotaped talking about the events of the day, a major area of continuing disagreement in theirmarriage, or a pleasant topic or they spend 24 hr in an apartment laboratory as they normally would athome. The tapes are later coded for specific emotions, facial expressions of emotion, problem-solvingbehaviors, and visual gaze patterns. After the interaction, a video recall procedure is used to obtain acontinuous self-report of affect. Couples are also interviewed about their perceptions of specificmoments in the interaction and information is obtained during these moments about their experience,Page 4 of 19 11/27/2000http://spider.apa.org/ftdocs/fam/1993/june/fam7157.html thoughts, expectations, thoughts about the partner, hopes, and attributional processes. Several years later couples are recontacted, and researchers perform a variety of follow-up assessments about maritalquality and stability. What are the basic results? There Is a Trajectory or Cascade Toward Divorce The problem with doing a 35-year longitudinal study such as Kelly and Conley's (1987) is that in 35years people may not be very interested in your variables because the field may have changed so much.However, there is an important methodological problem that needs to be solved in conducting short-termlongitudinal research on divorce. This is a problem whenever one attempts to predict a rare event. Forexample, in the research on myocardial infarction, it would be helpful to have a set of precursorpredictor variables such as angina, shortness of breath, specific pain in the throat or left arm, and so onthat are easier to predict than the relatively rare event of a heart attack. What needs to happen is that aGuttman scale of these precursors needs to be identified. One would then expect that the predictors ofthe trajectory toward the rare event would predict the precursors better than the rare event. This wouldbe the expected pattern if the trajectory were being predicted. Such a scale of precursor events has been developed. In two longitudinal studies ( Gottman & Levenson,1992 ; Gottman, in press ), Gottman and Levenson have discovered that there is a specific trajectorytoward marital dissolution that forms a Guttman scale of precursor predictor variables. This scalesuggests that couples who divorce remain unhappily married for some time, seriously considerdissolution, then actually separate and then divorce (see Figure 1 ). This analysis revealed that in Study 1, the model in Figure 1a fit these data well, with a nonsignificant χ2 4, N = 73 = 7.09, p = .13 , and a normed Bentler-Bonnett goodness-of-fit statistic of 0.994 (which issufficiently close to 1.0 to indicate a good fit); the model for Study 2 in Figure 1b also fit the data well,with χ 2 3, N = 52 = 4.63, p = .20 and a Bentler-Bonnet norm of .998. Although this cascade may seemobvious, it actually has been the subject of some debate as to whether there is a continuum betweenresearch on marital satisfaction and marital dissolution. The Cascade Toward Marital Dissolution Can Be Predicted With Just Two Variables, Which AreBased on a Balance Theory of Marriage Using an observational system (called the Rapid Couples Interaction Scoring System, or RCISS) forTime 1 data the accumulated positivity minus negativity in the two speakers' interaction over time isgraphed and couples are divided into two groups, those whose curves have positive speaker slopes(balance favors positivity), called "regulated," and all other couples (balance does not favor positivity),called "nonregulated." These two groups of couples can be distinguished with other observationalsystems as well ( Gottman & Levenson, 1992 ), and they are significantly different in their trajectoriestoward dissolution or stability over time. To give the reader an idea of the results, we found that in aregression using the husband and wife's speaker RCISS slopes, nonregulated couples were significantlymore likely to have been unhappily married at both Times 1 and 2, F (2, 70) = 10.01, R = 0.46, p < .001;F (2, 70) = 6.39, R = 0.38, p < .01, respectively; husbands and wives were more likely to have seriouslyconsidered dissolution, F (2, 70) = 3.60, R = 0.29, p < .05; F (2, 70) = 4.68, R = 0.33, p < .05,respectively; and husbands and wives were more likely to have separated and divorced, F (2, 70) = 5.03,R = 0.34, p < .01; F (2, 70) = 7.09, R = 0.40, p < .01, respectively–(see Gottman, 1993 ). Discriminantanalyses suggested that both positive and negative RCISS speaker behavior contributed to the predictionand that the ratio between negative and positive RCISS codes was the best discriminator. Hence, there issome evidence that a balance model provides a suggestion that what is being regulated in regulatedmarriages is a balance between positivity and negativity.Page 5 of 19 11/27/2000http://spider.apa.org/ftdocs/fam/1993/june/fam7157.html Not All Negativity Is Equally Corrosive: The Four Horsemen of the Apocalypse There were some negative acts that were more predictive of dissolution than others. For example, angerwas not predictive of separation or divorce, but the husband's defensiveness, contempt, and stonewalling(the listener's withdrawal from interaction) were predictive of divorce ( r = 0.42, p < .001; r = 0.25, p< .05; r = 0.29, p < .01; respectively), whereas the wife's criticism was predictive of separation ( r =0.36, p < .001), and her criticism, defensiveness, and contempt were predictive of divorce ( r = 0.26, p< .05; r = 0.38, p < .001; r = 0.34, p < .01; respectively). By means of a global specific emotions codingsystem, the wife's contempt and disgust were found to be particularly predictive of marital separation ( r= 0.24, p < .05). On the basis of only facial expression coding with Ekman and Friesen's (1978) EmotionFacial Action Coding System (EMFACS) system, the frequency of the wife's facial expressions ofdisgust correlated 0.51 ( p < .001) with the number of months the couple was to separate in the next 4years. A structural model supported a process cascade in which criticism leads to contempt, which leadsto defensiveness, which leads to stonewalling. The results suggest that these four processes areparticularly corrosive to marital stability. We also found that stonewalling is characteristically male and that a man's stonewalling is associated with the physiological arousal of both spouses ( Gottman, inpress ). By means of both the RCISS observational coding and the Marital Interaction Coding System(MICS) coding, it was possible to construct structural equations models of this process cascade (seeFigure 2 ). Figure 2 shows that these four variables for both the RCISS and the MICS form a Guttman-like scale. For the RCISS, χ 2 2, N = 73 = 0.00, p = 1.00 , whereas for the MICS χ 2 2, N = 73 = 0.82, p =0.66 . These figures show that there is considerable consistency in this Guttman-like scaling ofprocesses. These analyses show that such a Guttman-like scaling model is consistent with these data. There Is a Process Cascade That Relates to the Couple's Perception of the Relationship and ThatAlso Predicts Dissolution A set of five questionnaires formed a cascade, using structural equations modeling. This cascade iscalled the distance and isolation cascade. The theoretical speculation is that this cascade begins withflooding. Flooding is measured with a questionnaire in which the subject endorses items that claim thatthe partner's negative emotions are unexpected ("seem to come out of nowhere"), unprovoked, intense,overwhelming, and disorganizing and that the partner will do anything to terminate the interaction (e.g.,run away). The other variables in the distance and isolation cascade are perceiving the marital problemsas severe, thinking that it is better to work out problems alone, having arranged their lives as more inparallel (they don't eat together as often as they used to, they are more likely to have separate friends,etc.), and loneliness in the marriage. The flooding variable is correlated significantly with behavior (e.g.,RCISS speaker slopes). There are gender differences in flooding that suggest that men are flooded byless intense negative affects and behaviors than women (merely criticism is necessary for men to feelflooded, but contempt is necessary for women to feel flooded). This cascade is illustrated by thestructural model of Figure 2b . The model in Figure 2b fits the data, with χ 2 1, N = 52 = 1.69, p = .19 ,Bentler-Bonnett normed index = .998. During Marital Conflict, the Perception of Positive Affect Is One Dimensional, Whereas thePerception of Negative Affect Is Two Dimensional In a study of marital violence conducted with 158 couples (in collaboration with N. Jacobson), weexamined the couples' self-report of their emotional experience during their most positive and mostnegative moments in the interaction (determined by the video recall rating dial). We found ( Gottman, inpress ) that their self-ratings of their emotional experience (on the Rushe affect checklist) factored intoone dimension for positive moments, with all the positive affects loading on this factor. However, fornegative moments, there were two factors. One factor could be summarized as hurt and righteousPage 6 of 19 11/27/2000http://spider.apa.org/ftdocs/fam/1993/june/fam7157.html indignation (sadness, anger, and contempt); it contained thoughts of retaliation. The second factor couldbe described as hurt and perceived attack (internal whining, innocent victimhood, fear, and worry). Bothtypes of responses are likely to represent distress-maintaining cognitions. The theoretical challenge is to test whether these momentary perceptions of the interaction are related tomore global and stable cognitions about the partner and the marriage. This would provide an importantlink with attribution theory research on marriage. The well-known attribution effects in marriage (e.g.,see Fincham, Bradbury, & Scott, 1990 ) suggest that in unhappy marriages people develop hypothesesabout their partner's behavior that are very hard to disconfirm. Positive behavior is attributed to fleeting,situational causes, whereas negative behavior is attributed to the stable and global negative traits of thepartner. Holtzworth-Munroe and Jacobson (1985) used indirect probes to investigate when couplesmight "naturally" search for causes of events and what they conclude when they do search for causes.They found evidence for the hypothesis that distressed couples engage in more attributional activity thannondistressed couples and that attributional thoughts primarily surround negative impact events. Theyconcluded that nondistressed couples engaged in relationship-enhancing attributions, whereas distressed couples engaged in distress-maintaining attributions. Distress-maintaining attributions maximize the impact of negativity and minimize the impact of positivity of the partner's behavior. Moreover,Holtzworth-Munroe and Jacobson pinpointed an important gender difference: Distressed husbandsgenerated more attributions than nondistressed husbands, but distressed and nondistressed wives did notdiffer in this regard. They suggested that normally men may not engage in much attributional activitybut that they outstrip women once relationship conflict develops. Relationship-enhancing attributions were responses to positive partner behavior in both groups ofcouples. Relationship-enhancing attributions minimize the impact of the negative behaviors andmaximize the impact of the positive behaviors of the partner. In an experimental study by Jacobson, McDonald, Follette, and Berley (1985) , distressed and nondistressed couples were randomly assignedinstructions to act positive or to act negative. Jacobson et al. found that distressed couples were likely toattribute their partners' negative behavior to internal factors, whereas nondistressed couples were likelyto attribute their partners' positive behavior to internal factors. Thus, these attributions, once established,make change less likely to occur. Behaviors that should disconfirm the attributional sets tend to getignored, whereas behaviors that confirm the attributional set receive attention. Attributional processes may tap the way couples think in general about the marital interaction as itunfolds with time. Attributions and general thought patterns about negative behaviors may thus betheoretically useful in providing a link between the immediate patterns of activity seen in behavioralinteraction and physiological response and more long-lasting and more global patterns that span longertime periods. It might be the case that these more stable aspects of the marriage are better predictors oflong-term outcomes, such as divorce, than those that can be obtained from behavioral observation. Thecontent dimensions of negative attributions that have been studied include locus (partner, self,relationship, or outside events), stability (e.g., due to partner's trait or to a state that is situationallydetermined), globality (how many areas of the marriage are affected), intentionality (negative intent–selfish vs. unselfish motivation), controllability, volition, and responsibility (e.g., blameworthiness).Fincham, Bradbury and Scott (1990) reviewed experimental evidence for this phenomenon andconcluded that, by and large, these patterns had been pretty well established by research. For attributionsabout negative events, 100% of the studies reviewed supported differences between happily andunhappily married couples on the two dimensions of globality and selfish versus unselfish motivation. Negative Momentary Perceptions of the Interaction Were Related to Attributions and to Flooding These Rushe self-ratings of emotional experience during couples' most negatively rated moments wererelated to flooding and to the Fincham-Bradbury attribution items assessing globality, stability, andPage 7 of 19 11/27/2000http://spider.apa.org/ftdocs/fam/1993/june/fam7157.html selfishness of the partner's negative behavior. Hence, these momentary negative perceptions are related to more lasting cognitions about the marriage. Flooding and Hypervigilance Through Escape Conditioning I suggest that flooding leads to emotional (escape) conditioning, which causes a state of hypervigilanceto the cues conditioned to the flooding. As has been proposed in the attributional literature withoutexplanation, I suggest that the flooded person then begins to distort ambiguous cues in the biaseddirection of seeing them as threatening or frustrating (producing either thoughts of innocent victimhoodor righteous indignation). This causes the well-known negative attributional state in which it is verydifficult for the partner to disconfirm the negative attribution by positive acts because they are attributedto fleeting, situational causes, whereas the negative acts of the partner are attributed to lasting, stabletraits of the partner. A nondisconfirmable hypothesis thus emerges. Once flooding occurs, the distanceand isolation cascade begins, leading to an emotional divorce within the marriage. Flooding is related toviewing one's marital problems as severe, believing that it is better to work out problems alone, arranging the marriage so that the partners' lives are more in parallel rather than interdependent, andloneliness. There is one further generalization of negative cognitions that we have detected. Furtherstabilization of negative cognitions may lead to the nondisconfirmable hypothesis, which generalizes toa global negative view of the entire relationship, its history, meaning, and philosophy. Buehlman,Gottman, and Katz (1992) presented evidence that the couple's view of their history (coded from theOral History Interview; Krokoff & Gottman, 1984 ) significantly predicted divorce and correlated withbehaviors and other negative affects that predict marital dissolution. Physiological Reactivity and Chronically High Levels of Physiological Arousal Are Related toDivorce Prediction In the Buehlman et al. (1992) report, some of the Oral History variables predictive of divorce correlatedwith physiological reactivity (elevation over and above a baseline). In this study, we also found (but didnot report in Buehlman et al.) that chronic levels of physiological arousal in the husband (in addition toreactivity) were strongly predictive of divorce. In particular, in this study we included a situation inwhich couples relaxed with their eyes closed before the interview that set up the conflict discussion. Wethought that arousal differences on this baseline would be a more reliable reflection of chronic activationthan the eyes-open baseline obtained after the interview and before the conflict discussion. We found asignificant result for the husband's heart rate in correlation with divorce. This was also true in the eyes-closed baseline that preceded the interview about the marital conflict, when, presumably couples wouldbe less likely to be autonomically aroused. No other physiological variable was significantly correlatedwith divorce. However, the husband's baseline eyes-closed heart rate was also a significant predictor of separation ( r = − .28, p < .05). The wife's baseline activity level was a significant predictor of thehusband's later considerations of dissolution ( r = .41, p < .01); this was also the case for her activitylevel during the interaction ( r = .44, p < .01), and with her considerations of dissolution for the eyes-closed baseline ( r = .29, p < .05). A set of t -tests conducted on the husband's interbeat interval showedthat these correlations were associated with large Time-1 differences in heart rate between couples whoeventually divorced and couples who stayed together. For the eyes-closed baseline, t (51) = 2.62, p< .05; for the eyes open baseline, t (51) = 2.69, p < .01. The means for the eyes-closed baseline for thehusbands of the couples who eventually divorced compared to those who remained stable differed byover 11 beats per minute (BPM). This represents quite a large effect in the psychophysiologicalliterature, in which heart rate effects are typically of the order of 4 to 5 beats a minute. The divorcedmean was 84.49 BPM and the stable couples' husband's mean was 73.43 BPM; for the conflictinteraction, t (51) = 2.57, p < .05, divorced mean = 85.50 BPM, and stable mean = 75.20 BPM.Page 8 of 19 11/27/2000http://spider.apa.org/ftdocs/fam/1993/june/fam7157.html At the time of this writing, the precise role of physiology remains open, as there is not consistency from study to study as to whether husbands or wives in ailing marriages are more aroused or morephysiologically reactive (cf. Gottman & Levenson, 1992 ; Gottman, in press ). A speculative view of therole of physiology in relation to information processing and access to creative and non-overlearnedbehavior patterns and cognitions can be found in Gottman (1990) . A more complete discussion of thisquestion can be found in Gottman (in press-b, Chapter 14) . There Are Three Types of Stable Marriages Gottman (1993) presented a typology of marriages, using the RCISS speaker and listener cumulativepoint graph slopes. There was evidence for three types of stable couples: The volatile couple was thehighest in emotional expressivity, the validating couple intermediate, and the conflict-avoiding couplewas the lowest. They could be distinguished from one another in the amount and timing of persuasionattempts: Volatile couples began their persuasion in the first third of the interaction when feelings werefirst being expressed and remained high throughout the interaction; validating couples peaked in themiddle third (the usual arguing phase, see Gottman, 1979 ); conflict-avoiding couples seemed never toengage in persuasion attempts. All three groups of stable couples had a positive-to-negative ratio of 5:1(no significant differences between the three groups), whereas for unstable marriages the ratio was 0.8:1(i.e., there was more negativity than positivity). Gottman (1993) analyzed these results using a predatorprey model of behavior, suggesting that in a marriage there is a balanced behavioral ecology; hesuggested that some set points in this ecology favor stability, whereas others favor instability. Gottman (in press) proposed the hypotheses that these are the entire range of necessary adaptations to ensure astable marriage and that all unstable marriages represent failures at being able to negotiate one of thesestable styles (i.e., all "mixed" styles will be unstable longitudinally). These three styles of marriage aredifferent on many other dimensions, and the derived typology strongly resembles Fitzpatrick's (1988) typology of happily married couples. Fitzpatrick used a questionnaire to divide couples into three puretypes, and, in a series of studies, she correlated the typology with behaviors, attitudes, and perceptions.Fitzpatrick's samples did not include unhappy couples, nor were couples followed longitudinally. The existence of the typology suggests that variables that Raush, Barry, Hertel, and Swain (1974)originally proposed as potentially dysfunctional in a marriage are not actually dysfunctional. Inparticular, Raush et al. proposed the idea that both escalating quarrels about seemingly trivial issues andthe avoidance of conflict are potentially harmful in a marriage. However, volatile couples do in factengage in escalating quarrels about almost any issue; they seem to have an ethic of honest exploration ofall feelings on any issue at any time. However, they balance this direct confrontation and persuasionwith high levels of positivity (affection and humor). Conflict-avoiding couples also are quite stablelongitudinally; what is central to their marriage is the absence of persuasion attempts, the affirmation ofshared beliefs, and the minimization and isolation of conflict (e.g., "agreeing to disagree"). Each type ofmarriage is likely to represent a necessary adaptation that ensures a stable marriage, each with its ownrewards and costs and each with its own comfort level of emotional expression. Cultures are likely todiffer in the distribution (and preference for) these three types. Theoretical Formulation I now present the outline of a theory that integrates these results using a theoretical framework of"systems of systems" recently devised by the Swiss theoretical physicist, Roland Fivaz ( Fivaz, 1991 , inpress ). Some of my theoretical formulation is speculative. Fivaz's thinking is based on a generalizationof thermodynamics, which concerns the physics of heat. In Fivaz's thinking, a simple set of conjugatevariables is required to describe the behavior of a wide class of systems. I do not define the termconjugate in general, but in Newtonian mechanics, position and momentum are conjugate variables. InPage 9 of 19 11/27/2000http://spider.apa.org/ftdocs/fam/1993/june/fam7157.html Fivaz's general notation, these conjugate sets of variables are represented by P and Q, and I speak in thistheory of P-space and Q-space. He calls the P variables the flow variables, and the Q variables the ordervariables. Because these ideas are rather esoteric–except, perhaps for those readers who are familiarwith some physics–I think that placing them in the realm of marriage and making them more concreteshould make them more accessible. In the theory I wish to formulate, the P variables are simply the cumulative (or integrated) sum over timeof positive minus negative behaviors (such as in the RCISS point graphs or emotion codes). In general,one can think of each partner in a marriage as having a built-in "meter" that measures the totality ofaccumulated negativity in this interaction. As in the case of the point graphs, negativity is reduced orbalanced by the positivity expressed or received. To summarize, the P variables measure the total flowand accumulation of overall negativity over time, as the interaction proceeds. I then suggest that there isa threshold slope of these cumulated variables such that if the threshold is exceeded, this will affect the"perception" of the interaction. First, a bit more about the general character of the Q variables. In physics the P variables reflect kineticenergy (or flow), and the Q variables reflect potential energy (or order). Analogously, I suggest that theQ variables, the equivalent of "potential energy" in my theory, must represent some form ofdisplacement from an equilibrium state of perceived well being. These order or perception variables arethe Q variables. In my work, they are operationalized by means of the rating dial and the video recallprocedure (see Gottman & Levenson, 1985 ), as well as by means of techniques such as the talk-table( Gottman, Notarius, Gonso, & Markman, 1976 ), the communication box (see Floyd & Markman,1983 ; the Markman & Baccus, 1982 , work with Black couples; and the Ferraro & Markman, 1981 ,work with deaf couples), thought listing ( Ickes, Robertson, Tooke, & Teng, 1986 ; Ickes, Stinson,Bissonnette, & Garcia, 1990 ), and the Rushe self-ratings of affective experience when couples areinterviewed about specific moments in the interaction ( Gaelick, Bodenhausen, & Wyer, 1985 ). In my formulation, the Q variables were initially thought of as dichotomous, as either the perception ofwell-being and safety in the relationship (feeling loved and respected–in which case one can think of Q= 1) or the opposite perception. However, as I mentioned earlier in this article, on the basis of work wehave done using a video recall interview and the Rushe ratings, this perception of "non-well-being" is made up of either moments of feeling hurt and under attack, that is, perceived threat (Q = − 1) or moments of hurt and anger (also Q = − 1). These two responses in Q space are called righteousindignation and innocent victimhood. In the former moment, anger and contempt are central and in thelatter fear is central; both kinds of moments are blended with sadness and disappointment. Both kinds ofresponses in Q space represent distress-maintaining cognitions, and the work of Zillmann (1979)suggested that, once provoked, men are less likely to physiologically soothe themselves than women andmore likely to engage in these distress-maintaining cognitions, unless they can retaliate. If theseZillmann gender differences can be generalized to marriages, they pinpoint some important problems. To explain how to operationalize this theory, I suggest that the Q space represents the "subtext" in a person's perception as the interaction unfolds. Generally, this is a quiescent subtext, as long as Q = 1; theperception of neutrality and well-being are subtext that are not very prominent in awareness. When Q = − 1, perceptions shift and a new cognitive—emotional process is engaged.