Copyright 1983 by IheJournal of Personality and Social Psychology1983. Vol. 45. No.3. 587-597American Psychological Association. Inc.Marital Interaction: Physiological Linkage. and Affective ExchangeJohn M. GottmanRobertVV. LevensonUniversity of Illinois' .Indiana UniversityThirty married couples were studied during naturalistic interactions to determinethe extent to which variation in marital satisfaction could be accounted for byphysiological and affective patterns between and within spouses. The authorshyPothesized that (a) compared to nondistressed couples' interactions, distressedcouples' interactions would show greater physiological interrelatedness or "linkage," more negative affect, and more reciprocity of negative affect and (b) thesedifferences would be more pronounced when the interaction was high in conflict(discussing a marital problem) as opposed to low in conflict (discussing the eventsof the day). Heart rate, skin conductance, pulse transmission time, and somaticactivity from both spouses were analyzed using bivariate time-series techniquesto derive a measure of physiological linkage. Self-report affective data (obtainedusing a video-recall procedure) were analyzed using sequential analyses to derivea measure of affect reciprocity. The hypotheses were strongly supported; 60% ofthe variance in marital satisfaction was accounted for using measures of physiological linkage alone. Additional nonredundant variance was accounted for bythe other physiological and affective measures.Social interaction provides a rich, naturalistic, and theoretically advantageous context for studying the relations among physiological, affective, and behavioral phenomena. Unfortunately, the demands associatedwith laboratory experimentation extract significant compromises that may escalate untilthe experimental context bears little relationto natural dyadic interaction. For example,interaction between two strangers may besubstituted for interaction between intimates;a carefully "programmed" confederate mayreplace the second person; and finally, theconfederate may be replaced by a film, a photograph, an audio recording, or a situationcreated in the subject's own imagery. In thisexperiment, naturalistic interaction within ahusband-wife dyad was used as the contextfor studying physiological and affective responses. We expected that physiological andaffective patterns within these dyads wouldreflect both the level of marital satisfactionand the extent of conflict in the interaction.Affective Patterns in Distressed,Dissatisfied MarriagesInteractions between spouses in distressedand nondistressed marriages have been consistently found to differ in terms of affectivepatterns. Birchler, Weiss, and Vincent (1975)observed greater negative affect (during problem solving and conversation) and less positive affect (during problem solving) in distressed couples. Similarly, there have beenseveral reports of more humor, more laughter, more support, and less defensiveness innondistressed than in distressed familiesThe authors would like to thank Patricia S. Meek and (Mishler & Waxler, 1968; Riskin & Faunce,Carole Mannheim for their invaluable assistance with 1970). Gottman (1979) proposed a model inthis project. This research was supported by U.S. PublicHealth Service Grant NIMH299I 0 and by National In- which marital satisfaction is related to threestitute of Mental Health Research Scientist Development dimensions: (a) positive-negative affect, (b)Award K2-OO257 to the second author.negative-affect reciprocity (i.e., patterned exRequests for reprints should be sent either to Robert changes of negative affect between the husW. Levenson, Department of Psychology, Indiana Uniband and wife), and (c) asymmetry. He hyversity, Bloomington, Indiana 47405 or to John M. Gottman, Department of Psychology, University of Illinois, pothesized that interaction in dissatisfiedmarriages would be characterized by less posChampaign, Illinois 61820.587

588ROBERT W. LE'l'ENSON AND JOHN M. GOTTMANitive affect, more negative affect, more reciprocity of negative affect, and greater asymmetry of predictability of affect (Le., onespouse would dominate the exchange of affect) than in satisfied marriages. Researchsupporting this model has been reported inindependent laboratories in the United States(Gottman, 1979, 1980; Margolin & Wampold, 1981), Germany (Revenstorf, Vogel,Wegener, Halweg, & Schindler, 1980), and theNetherlands (Schaap, 1982).We hypothesized that the patterns of observable affect and affect exchange that typifydissatisfied marriages would be accompaniedby parallel patterns in physiological response.Because negative emotions are intimatelyassociated with physiological activation (e.g.,the sympathetic nervous system activity associated with fear and anger), we hypothesized that patterned exchanges of negativeaffect within the dyad would activate physiological systems and produce parallel patterning of physiological responses betweenthe spouses. We found no research on maritalinteraction that provided evidence directlyrelevant to this point but did find severalstudies of dyadic interaction that includedphysiological measurement and providedsome support for this hypothesized relationbetween negative affect and physiologicallinkage.Interpersonal PsychophysiologyThere is limited, but consistent, evidencethat physiological responses between members of an interacting dyad can show considerable relatedness or linkage. In the 1950sseveral studies were reported of the physiological responses of psychotherapists andtheir clients during interviews. DiMascio,Boyd, Greenblatt, and Solomon (1955) reported that heart rates of the therapist anda neurotic patient undergoing long-term psychotherapy "often varied together and atother times varied inversely from each otherother" (p. 9). DiMascio, Boyd, and Greenblatt (1957), using the first 12 interviews.witha patient in therapy, found that the therapist'sand patient's heart rates' moved in similardirections as the levels of "tension" in theinterview varied but moved in opposite directions when the patieritexpressed "antag-onism" toward the therapist. In a study ofthe effects of an examiner's praise or criticismof female neurotic patients' responses to adiagnostic test, Malmo, Boag, and Smith(1957) found that the amplitude of the electromyogram (EMG) obtained from the chinof the examiner and the patient both fell following praise and remained constant following criticism. In Kaplan and Bloom's (1960)review of this literature, they interpretedthese results as indicating physiologicalcomponent of "empathy." This theme wasrestated by Ax (1964) when he speculatedthat empathy might be thought of "as an autonomic nervous system state which tends tosimulate that of another person" (p. 12).More recently, Kaplan, Burch, and Bloom( 1964) reported two studies of correlationsin skin conductance among dyads constituted on the basis of sociometrically determined "liking" and "disliking." In the firststudy, using male subjects, one four-persongroup in which all subjects disliked eachother, a second four-person group in whichall subjects liked each other, and a third fourperson group comprising a mixture of likingand disliking were studied during five 45minute discussion sessions. The data wereanalyzed in terms of two-person dyads. Results indicated that dyads from the group thatdisliked each other were more likely to showsignificant correlations in skin conductancecompared to the other two groups. In the second study, female subjects interacting indyads (10 liking, 10 disliking, and 10 neutraldyads) in two 20-minute discussion sessionswere used. Again, subjects paired on the basisof mutual dislike were most likely to showsignificant correlations in skin conductance.Methodological ProblemsWe felt that the results from the studies ofinterpersonal psychophysiology were encouraging but not without problems. First, inmany of these studies an attempt was madeto characterize the physiological relation ina dyad on the basis of a single autonomicnervous system (ANS) measure; a morebroadly based set of physiological measureswould provide a better test of this relation.Second, it is important to determine whetherthe physiological relation in a dyad provides

PHYSIOLOGY AND AFFECT IN DYADS589information beyond that which is provided among spouses in dissatisfied marriages wouldby other measures, such as simple self-report be more likely to manifest negative affect andmeasures. Although it would certainly be in- negative-affect reciprocity. Finally, we hyteresting to find that physiological measures pothesized that the information obtainedprovide essentially the same information as from studying patterns of physiological acself-report measures, the additional expense tivity would provide information that wasand obtrusiveness of physiological measures nonredundant with that which could be obwould hardly be justified unless they pro- tained by simply studying affective patterns.vided additional, nonredundant information.Third, the use of running correlations to asMethodsess physiological relatedness between mem- Subjectsbers of a dyad has serious problems. BecauseMarried couples were recruited using two brief adthese kinds of observations are not indepenvertisements placed in the Bloomington, Indiana, newsdent, the significance tests of the correlation paper. To obtain a representative sampling of maritalare incorrect; errors of nearly 500% have satisfaction levels, the first advertisement asked for "coubeen found in such instances by Hibbs ples needed for research project studying marriage,"(1977). Further, within each member's data whereas the second asked for "couples having difficultymarital problems." The approximately 100 rethere are autocorrelations (i.e., cyclicity) that solvingspondents to the advertisements were further screenedmust be controlled for before inferences re- on the telephone to ensure that English was their nativegarding correlational relations across sub- language and that they could meet our scheduling rejects can be drawn. This is particularlY'im- quirements. In this initial telephone contact, the subjectsportant in studies in which the manipulation were told that the experiment would involve coming tothe laboratory to discuss neutral topics and problem(or dyadic grouping) could be expected to areas in their marriage and that videotape, self-report,increase the autocorrelation within each sub- and physiological data would be obtained. The first 30ject's data (see Gottman, 1981; Gottman & acceptable couples formed the subject sample used inthe experiment. Each couple was paid 30 for 4 hoursRingland, 1981).The present study addressed these prob- of participation.lems by including mUltiple physiological Apparatusmeasures, including a procedure that obPhysiological data. The selection of physiologicaltained self-report affective measures continvariables represented a compromise between two opuously without interrupting the flow of be- posing considerations: our desire to constrain and enhavior, employed more appropriate time-se- cumber subjects as little as possible and our desire tories analytic techniques for determining the obtain as comprehensive a physiological assessment asdegree of physiological relatedness across possible within the technical limits imposed by simultaneously recording data from two subjects. These consubjects, and determined the extent of non- siderationsled us to select four physiological dependentredundancy between physiological and affec- measures obtained from three kinds of recording devicestive measures.placed on the surface of the subject's skin and from aFour primary hypotheses were tested. First, fourth device attached to the subject's chair. On-lineas indicated earlier, we hypothesized that analysis of the physiological data was accomplished usingsystem consisting of an eight-channel Grass Model 7physiological linkage would occur in situa- apolygraphand a PDP 11/10 minicomputer, which wastions in which negative affect was prevalent equipped with analog-to-digital capabilities and was proand in which this affect was exchanged or grammed to process the following measures from eachreciprocated. between members of the inter- subject continuously and to average the measures intoacting dyad. Second, we hypothesized that IO-sec measurement periods:Heart rate measured by the interbeat interval (IBI).physiological linkage would be greater whenBeckman miniature electrodes with Redux paste weredyads were engaged in a high-conflict interplaced in a bipolar configuration on opposite sides ofaction that would elicit more negative affect,the subject's chest. The time interval between successuch as discussing marital problems, comsive R waves on the electrocardiogram was measuredat a resolution of I msec. IBI was included becausepared to a relatively low-conflict interaction,it is an ANS function that can be altered by bothsuch as discussing the events of the day.sympathetic and parasympathetic influences and beThird, we hypothesized that physiologicalcause changes in rate of contraction are one of thelinkage would be negatively correlated withbasic ways by which the heart regulates its output ofmarital satisfaction, because interactionsblood to the body.

590ROBERT W. LEVENSON AND JOHN M. GOTTMANPulse transmission time to the finger (PTT). PTT wasdetermined by measuring the time interval betweenthe R wave and the arrival ofthe pulse pressure waveat the middle finger ofthe nondominant hand. A GrassPTT photoplethysmograph was used to detect thepulse wave, and the resolution was 1 msec. PTT wasincluded because it reflects different cardiovascularactivities than IBI (i.e., changes in the force of theheart's contraction and changes in the distensibility ofthe arteries between the heart and the finger, bothmediated primarily by the sympathetic branch of theANS). Changes in contractile force are the other basicmethod (besides heart rate) by which the heart regulates its output.Skin conductance level (SCL). A constant-voltage device was used to pass a small voltage between Beckmanregular electrodes attached to the middle phalanges ofthe first and third fingers of the nondominant hand.The electrolyte was Beckman electrode paste. SCL recorded from this site was included because it measuresthe activity of sweat glands thought to be related toemotionality. It is under the control of the sympatheticbranch of the ANS but has a different stimulationchemistry than does the sympathetic cardiovascularfunctions involved in IBI and PTT.General somatic activity (ACT). An electromechanicaltransducer attached to the platform under each subject's chair detected movement in any plane. This totally unobtrusive measured provided a global index ofsomatic muscle activity.These four measures provide a reasonable breadth ofmeasurement because they reflect the activity of fourphysiological systems (heart, vasculature, sweat glands,muscles). In addition, IBI, PTT, and SCL are all sensitiveto sympathetic nervous system activity, which has anevolutionary connection with negative emotions, suchas fear and anger, by virtue of its role in preparing theorganism for the adaptations of "fight" and "flight."Self-rating of affect. A simple positive/negative affective measure was used in keeping with our experiencethat individuals who are, inexperienced in rating theirown affect find more complex affective measures toodifficult. Empirical evidence (e.g., Osgood, Suci, & Tannenbaum, 1957) indicates that the positive/negative dimension accounts for most of the variance in emotionaljudgments. Furthermore, we wanted to obtain a continuous report of affect (as opposed to discrete pretest andposttest measures) that reflected the moment-to-momentaffective changes during interactions. To accomplish this,the husband and wife returned to the lab in separatesessions to view the videotape of their interaction andto provide affect ratings (see procedures below). The ratings were obtained by having the subject manipulate adial whose pointer moved on a 180 scale divided intonine divisions, with the legends "very negative" at 0 ,"neutral" at 90 (this was at the center of the fifth division), and "very positive" at 180 . The dial was attached to a potentiometer in a voltage-dividing circuitthat was monitored by the computer to enable determination of the dial position. The subject was asked toadjust the dial position as often as necessary while viewing the videOtape so that it always reflected the way heor she felt during the initial interaction session. The computer sampled the dial position every 5 msec andaveraged these readings into IO-second measurementperiods.'Video data. During the interaction session (see below), the spouses were seated facing each other in twocomfortable chairs in the center of a small room. 1\vomonochrome video cameras and two electret microphones were mounted on shelves on the wall oppositeeach chair and slightly above head level. The video signals from the two cameras were combined into a splitscreen image using a video mixer and then routed to atime-code generator that superimposed the elasped timeat a resolution of 10 msec. The audio signals from thetwo microphones were combined using an audio mixer.The resultant video and audio signals were recorded ona reel-to-reel video recorder. All mixing and recordingequipment were located in the room adjacent to the subject room, which also contained the physiological-datarecording equipment.Synchronization of physiological, q/fective, and videodata, To enable synchronization between these threedata sources,. it was necessary to establish a common"time zero." This was accomplished in the interactionsession by having the computer remotely start the videotime-code generator at the same moment that it startedtiming the physiological and affective data. In the videorecall session (see below), subjects viewed this videotapewhile a new set of video, 'affective, and physiological datawas obtained. To synchronize these new data with thedata from the interaction session, a strip of reflective foilwas placed on the back side of the videotape of the interaction session at the point at which the video timecode was started. A locally constructed optoelectric device was used to detect this foil strip and send a signalto the computer, which then started the time-code generator for the new video recording and started the newphysiological and affective data recording. In this mannerthe same time zero was established for all data sets.ProcedureInteraction session. To ensure that sufficient interaction would occur, couples were scheduled for the initiallaboratory session at a time when they would not haveinteracted with each otlwr for 8 hours. For most couplesthis meant scheduling the session in the early eveninghours .after work. When the subjects arrived the femaleexperimenter (an advanced doctoral student in clinicalpsychology) provided them with an overview of the procedure, asking them to interact as naturally as possibleand to avoid trying to effect any particular kind of behavior. In partIcular, they were asked to speak to eachother rather than to the cameras and not to worry aboutexplaining events or identifying names for our benefit.A female technician entered the room and attached thephysiological recording devices to the wife, explainingthe function of each device. Then the recording deviceswere attached to the husband. After verifying the qualityof all physiological signals, the experimenter returned topresent the instructions for the first interaction segment.The first segment began with a 5-minute period ofsilence that served as a preinteraction baseline. After 5minutes, a small amber light on a table beside them was

PHYSIOLOGY AND AFFECT IN DYADSilluminated, indicating that the conversation should begin. The couple was asked to engage in their typical conversation about the events of the day as if they were homealone at day's end and to continue the conversation untilthe amber light was extinguished (after ,IS minutes ofconversation). The experimenter left the room, the videorecording was started, and the computer started the timecode generator, beginning the 20 minutes of physiologicaldata collection.The experimenter returned to the room after the firstsegment to administer a series of four questionnaires thateach spouse completed separately over a 20-minute Pc,riod. The questionnaires included a demographic measure, a problem inventory in which they rated the severity of a number of common marital problems, andtwo inventories of marital satisfaction (Burgess, Locke,& Thomes, 1971; Locke & Wallace, 1959).'The experimenter examined the couple's answers tothe problem inventory and selected a problem area thatboth spouses rated as being important. This problem wasdiscussed with them to attempt to focus the conflict ona single issue, to elicit their feelings about the conflict,and to detetmine whether the problem was a source ofcontinuing conflict for them. If the issue had alreadybeen resolved, if the conflict was too diffuse to focus ona single issue, or if the couple felt extremely uncomfortable discussing the topic, then this procedure wasrepeated for another high-priority topic until an acceptable topic had been identified.After the experimenter left the room, the second interaction segment began with a 5-minute silent baseline.The couple was signaled by the amber light to begindiscussing the problem area in an attempt to reach somesolution or compromise. Video and physiological datawere collected in the same manner as in the events-ofthe-day segment, and this discussion continued until theamber light was extinguished (after IS minutes).The experimenter reentered the room and helped thecouple resolve any remaining tension resulting from theprevious interaction. The technician then entered' andremoved the recording devices. Each spouse was thenscheduled for a separate video-recall session.Some mention should be made concerning our deci sion not to counterbalance the orders of the two interactional segments. Pilot testing revealed that if the problem-area segment came first, then negative affect persisted and contaminated the events-of-the-day segment.Video-recall sessions. Each spouse returned to thelaboratory, for a separate session at a convenient timewithin 3 to 5 days2 after the interaction session. For this,session, a video monitor was placed in the position previously occupied by the spouse's chair. The experimenterprovided an overview of the session's procedure, carefully explaining the use of the affect rating dial. Then thetechnician attached the physiological recording devicesto obtain the same set of measures obtained in the original session. A video recording of the spouse viewing thetape of the interaction session was also obtained.The session consisted of the spouse's viewing the tapeof the two 20-minute, interaction segments (separated bya lO-minute rest periOd) from the first session and providing the affect ratings. After the video-recall session forthe second spouse, the payment for the entire experimentwas made.'591Data AnalysisPhYSiological data. For the purposes of the presentreport, we worked with the four physiological measures(IBI, PTT, SCL, ACT) obtained in the interaction session. 3 Each dependent variable was averaged into lO-secperiods for the husband and wife during the events-ofthe-day and problem-area interactions. Each of these in, These measures of marital satisfaction have a venerable 40-year history, beginning with the work of Terman,Buttenwieser, Ferguson, Johnson, and Wilson (1938).They have been employed in longitudinal and cross-sectional studies, have been refined over the years, and havebeen shown to have reasonable levels of con'Struct andpredictive validity. Historically, most measures of marital satisfaction have been highly correlated. Burgess andWallin (1953) suggested that the various constructs usedby researchers to assess people's attitudes toward theirmarriages were tapping one basic dimension; this conclusion has held up. Every few years a new measure ofmarital satisfaction is proposed (e.g., Spanier, 1976), butit invariably correlates very highly with the older measures (i.e., Burgess, Locke, & Thomes, 1971; Locke &Wallace, 1959) that we used.2 The decision to have subjects return to the laboratoryfor a separate recall session meant introducing an unwelcome time lapse between obtaining the physiologicaland affective data sets; this could introduce error thatwould work against our hypotheses. Several considerations went into this decision. We did not want to impedethe natural course of the interaction by having subjectsadjust the affect rating dials during the interactions. Weconsidered having subjects view the tapes immediatelyfollowing the interaction, but this would have made foran extremely long and arduous session, and the resultantirritation would undoubtedly distort the affect ratings.Despite the fact that subjects reported no difficulty recapturing their affective experience when watching thevideo tape, we cannot definitively assert that the ratingsobtained after this delay are as accurate as ratings thatmight have been obtained sooner. Pertinent to this question are the results of some recently completed preliminary analyses of the relation jJetween subjects' physiological responses during the interaction session and during the video-recall session. Thus far we have found thesephysiological data to be highly related; for example, astatistical measure of the coherence between the husband's skin conductance level data in the two sessionswas significant for 56 of the 60 interaction segments(events of the day and problem area for each of the 30husbands). This indicates that subjects may "relive" the 'original experience physiologically when viewing the videotape of the interaction. An even more relevant test willbe made when our coding of the affective content of theinteraction section is completed; we expect these "objective" ratings to confirm the subjects' self-ratings reported here.3 Currently in progress are analyses of the physiological data collected during the video-recall sessions anddetailed coding of the behavioral and affective data inthe video records. These results will be reported at a laterdate.

592ROBERT W. LEVENSON AND JOHN M. GOTTMANteractions consisted of a 5-minute "baseline" (thirty 10sec periods) and 15 minutes of interaction (ninety 10sec periods). Thus, the total physiological data setconsisted of 57,600 data points (30 couples X 2 spouses X2 interaction segments X 4 dependent variables X 120periods). We computed the average of the husband's andthe wife's scores on the two marital-satisfaction inventories administered in the interaction session to derivean overall measure of marital satisfaction (M 111.7,SD 21.8).For the traditional and time-series analyses we planned,several preliminary data reductions were needed. First,simple summary statistics (means and standard deviations) were calculated for each dependent measure foreach spouse for the following segments: (a) baseline priorto events of the day, (b) events of the day, (c) baselineprior to problem area, and (d) problem area. Then thefour dependent measures from each subject for each 15minute interaction segment were normalized by computing z scores for the ninety 10-sec periods using themean and standard deviation from the 5-minute baselinethat preceded that segment.The normalized data were used in bivariate time-seriesanalyses to determine the extent of each couple's physiological linkage ilsing the four physiological measuresobtained from a husband and wife during a given interaction segment. The interested reader is referred to Gottman (1981, chapters 23-25) for a complete descriptionof these techniques and their mathematical basis. Because some readers may not be familiar with time-seriesanalysis, a brief example will be provided using a husband's and a wife's IBI. The time-series analyses providedus with two chi-square values. One represents the extentto which the husband's pattern of IBI accounts for variance in the wife's pattern of IBI beyond the varianceaccounted for by her own pattern ofIBI (thus controllingfor the autocorrelation problem mentioned earlier). Thesecond represents the extent to which the wife's patternof IBI accounts for variance in the husband's pattern ofIBI beyond the variance accounted for by his own patternofIBI. These chi-square values were converted to z scoresto enable comparison across measures (these z scorescan also be used to test significance). In a similar manner,a pair of z scores was computed for each of the remainingthree physiological variables (PTT, SCL, and ACT). Theresulting set of eight z scores were used to determine theextent of the "multivariate" physiological linkage between a couple during that interaction segment.Self-report of affect. Simple summary statistics andnormalized scores were computed in the same way thatwas used for the 'physiological data. We also needed toderive an affective measure that would reflect "reciprocity of affect." On an a priori basis, criteria were established for determining whether each spouse's self-reported affect in each of the ninety lO-sec periods waspositive, negative, or neutral. To be coded positive, theraw score average had to be greater than or equal to 6.0(referenced to the original 9-point affect-rating dialscales), and the z score had to be greater than or equalto .5. Thus, a positive code meant that the pointer wasactually on the positive portion (the raw score criterion)of the dial and was positive relative to the subject's rangeof ratings (the z score criterion). To be coded negative,the raw score average had to be less than or equal to 4.0and the z score had to be less than or equal to -.5. Thenumber of IO-sec periods during the interaction t

the variance in marital satisfaction was accounted for using measures of physi ological linkage alone. Additional nonredundant variance was accounted for by the other physiological and affective measures. Social interaction provides a rich, natu ralistic, and theoretically advan