behavior exchange theory and marital decision making

8/7/2019 Behavior Exchange Theory and Marital Decision Making

1/10

Journal of Personality and Social Psychology1976, Vol. 34, No. 1, 14-23

Behavior Exchange Theory and Marital Decision MakingJohn Gottman, Cliff Notarius, Howard Markman,Steve Bank, and Bruce YoppiIndiana UniversityMary Ellen RubinSouth C entral Indiana M ental Health Founda tion, Bloomington, Indiana

Distressed and nondistressed couples in two studies made decisions on high-and low-conflict tasks. They continuously coded both the intended impact oftheir own behavior and the impact of their spouse's behavior. In Study 1 dis-tressed couples did not differ from nondistressed couples on how they intendedtheir behavior to be received. However, the behavior of distressed spouses wasactually received more negatively by their partners than the behavior of theirnondistressed counterparts. The couples in Study 2 also behaved in a wayconsistent with a communication deficit explanation of distressed marriages;that is, distressed couples' behavior was likely to be coded as more negativethan they intended. Task effects and a reciprocity hypothesis were also tested.Data from Study 1 showed no conflict effect, but the results of Study 2 sug-gested that high-conflict tasks may be a better means for discriminating dis-tressed from nondistressed couples than low-conflict tasks. The data on recip-rocity indicate only minimal support for the view that distressed marriage ischaracterized by less positive or more negative reciprocity than nondistressedmarriage.

Recently, two different hypotheses havebeen proposed to describe conflict resolutionin distressed and nondistressed marriages. Onehypothesis, based on behavior exchange the-ory, is that it is more likely tha t nondistressedcouples will produce behaviors coded as posi-tive by observers than will distressed couples.This hypothesis has recently received somesupport (Birchler, Weiss, & Vincent, 19 75).A second hypothesis is that there is agreater reciprocity of positive exchange innondistressed than in distressed marriages.Reciprocity of positive exchange has been sug-gested as the central characteristic of success-ful marital interaction in the clinical literatureon marriage counseling (Azrin, Naster, &

This research was partially supported by GrantR01 MH 24594 SP from the National Institute ofMental Health. The authors wish to thank the Men-tal Health Center, the Center for Human Growth,and the Center for Counseling and PsychologicalServices for their cooperation, and Steve Asher, JackBates, Fran Cherry, and Alice Eagly for their com-ments on the manuscript.Requests for reprints should be sent to John Gott-man, who is now at the Department of Psychology,University of Illinois, Children's Research Center, SIE. Gerty Drive, Champaign, Illinois 61820.

Jones, 1973; Lederer & Jackson, 1968; Rap -pap ort & Harre ll, 1972; Stua rt, 1969 ; Weiss,Hops, & Patterson, 1973).The two hypotheses are independ ent. Highbase rates of positive behaviors do not implyreciproc ity. For example, Birchler et al.'s(1975) finding of higher rates of positive be-havior in nondistressed compared with dis-tressed couples does not imply that reciprocityis different for the two groups of couples. Al-though nondistressed couples may seem to bereciprocating positive behavior more fre-quently than distressed couples, that may onlybe an artifact of the higher probability ofpositive behaviors in nondistressed couples.By emitting more positive responses, nondis-tressed couples increase the probability thatone partner's positive response will be fol-lowed by the other partner's positive response.What needs to be demonstrated is that a par-ticular consequent code in a sequence can bepredicted from a particular antecedent code.The test for reciprocity is that the diagonalelements of the first-order Markov matrix ofconditional probabilities must be significantlygreater than the relative frequency of the con-sequent code (Patterson, 1974). For example,the conditional probability (p ) of a conse-

14


2/10


3/10

16 GOTTMAN, NOTARIUS, MARKMAN, BANK, YOPPI, AND RUBINcedure, it is the couple who codes the behav-iors exchanged, not independent observers. Itis possible that what is coded as a warm smileby two observers is perceived as a sarcasticsmirk to a spouse, and conversely, a "put-down" or "interruption" that is coded as nega-tive in Birchler et al. (1975, p. 352) is notalways seen as negative by the couple.The concept of the couple coding their ownbehavior raises the question of whether dis-tressed couples differ from nondistressed cou-ples in the intended positivity of their mes-sages. According to the widely acceptedcommunication deficit explanation of maritaldistress, distressed couples are presumed tointend their messages to be received as farmore positive than they are in fact received.However, it could be that distressed couplesintend their messages to be more negative, orless positive, than nondistressed couples. Ifthe messages were then received as more nega-tive, or less positive, distressed couples wouldbe communicating well; their spouses wouldreceive the messages as intended. A communi-cation deficit explanation of marital distresswould seem more appropriate if intended posi-tivity of messages differed markedly from thecoding of the messages received. In this casemessages received would not be coded as thesender intended them to be coded.

In the present investigation, couples inter-acted using a "talk table," a device we con-structed in which only one person could speakat a time. After speaking, the speaker codedthe "intended impact" of his or her message.Before speaking, the listener coded the "actualimpact" of the message. The talk table neednot constrain the behavioral repertoire of theinteracting dyad as in Speer (1972); the talktable is apparently an inexpensive, simple wayto study behavior exchange in marriages, es-pecially compared with detailed ratings ofvideotapes (as in Birchler et al., 1975). How-ever, before this can be confidently asserted,the talk table must demonstrate its ability toelicit different patterns of exchange in differ-ent kinds of couples.One purpose of the present investigationwas to assess the effect of the experimentaltask on the coding of the interaction. The ef-ect of situational context on marital andfamily interaction has been either ignored

or minimized (Riskin & Faunc e, 1 972 ). Forexample, Haley (1964, 1967) placed no im-portance on the tasks he chose for families towork on. Jacob and Davis (1973) reportedconsiderable stability across experimentaltasks to the structure of talk and interruptionsin father-mother-child interactions. One po-tentially important dimension of contextualvariation that has been ignored is the degreeto which the decision-making task inducesconflict. For example, all of Jacob and Davis'(1973) tasks were low-conflict tasks. It wouldseem theoretically important to ascertainwhether the variables derived from behaviorexchange theory can discriminate nondis-tressed couples from distressed couples in low-conflict as well as high-conflict tasks. Thepresent investigation employed two sets oftasks that induced either high or low conflict.The tasks ranged in content from a consensualranking of a list of preferred dog breeds tothe discussion of an actual unresolved maritalproblem.

STUDY 1Method

Subjects and selection. Thirty couples responded toan advertisement asking for couples who eitherclassified their relationship as "mutually satisfying"or "experiencing marital difficulties." All coupleswere offered $10 for participating, and it was madeclear that no therapy would be offered. In addition,other distressed couples were recruited from localcampus and community mental health centers. Ofthe IS distressed couples who participated in thestudy, 11 were referred from clinical sources and 4responded to the advertisement.Two major definitions of marital distress havebeen used, self-report of satisfaction and whether ornot the couple comes to the attention of publicagencies such as marriage counselors or divorcecourts. Diverse measures of marital satisfaction havebeen shown to tap the same basic factor (Burgess,Locke, & Thomes, 1971). The Locke-Wallace MaritalRelationship Inventory (MRI) is the most widelyused of these inventories; it has excellent discrimina-tive validity in cross-sectional studies (Navran, 1967)and good predictive validity in longitudinal studies(Terman & Wallin, 1949). Using the cutoff scores onthe MRI recommended by Burgess et al. (1971, pp .330-331), only high-scoring nondistressed couplesand low-scoring distressed couples were included inthe analyses. Specifically, distressed couples in whichat least one spouse's MRI score was less than 85were included, and nondistressed couples in whichboth husband and wife MRI scores were greater than102 were included. A distressed couple was selectedfor analysis only if at least one spouse was dissatisfied


4/10

BEHAVIOR EXCHANGE 17with the marriage (low MRI), and a nondistressedcouple was selected for analysis only if both spouseswere satisfied with the marriage (high MRI). Usingthese cutoff scores, in Study 1 10 distressed couplesand 6 nondistressed couples were selected for analysisfrom the sample of 30 couples, all of whom hadcompleted the experimental procedures.1 Coefficientalpha for the sample of Study 1 on the MRI was .88.The two groups of couples did not significantlydiffer in age, educational level, or the number ofyears married. Couples were, on the average, 24.95years old and had been married an average of 3.22years.Procedure. Interview teams were composed of onemale graduate student in clinical psychology and onefemale undergraduate. There were four such teams,and each couple was interviewed by one of theseteams. Spouses individually filled out a problem in-ventory, a demographic information sheet, and theMRI. The problem inventory asked each spouse torate the perceived severity of a list of problemareas. The couple was then asked to agree jointlyon the three most salient current problem areas orareas of disagreement in their marriages. Next, thecouple was asked to describe each of the problemareas separately and to provide a specific descriptionlike a play-by-play account of an incident, providingsituational context and a typical conversation illus-trating each problem. This was done in order to es-tablish a rapport with the interviewing team and toprovide specificity to the problem to be discussed onthe talk table. The talk table is a double slopingtable. A toggle switch on the side of the table oper-ated by the couple lit a button on the side of thespouse who had the floor to speak. There were tworows of five buttons. The five buttons on the leftwere used by the speaker to code the "intended im-pact" of his or her message; the five buttons on theright were used by the receiver to code the "impact"of the message received. The buttons were labeled"super negative," "negative," "neutral," "positive,"and "super positive." Although partners could see oneanother, a metal shield blocked the buttons from theview of the other person so that neither spouse couldsee the codes assigned by their partners throughoutthe experiment.

In order to familiarize themselves with its use,each couple was then asked to converse for a fewminutes using the talk table. When the couple indi-cated they were comfortable with the talk table, theexperimental tasks began.Each couple then completed three low-conflicttasks: (a) The choice questionnaire (Haley, 1964)is a consensus decision-making task that requires anagreed-upon ranking of personal preferences usinglists such as new cars or breeds of dogs, (b) ThreeThematic Apperception Test cards (Locke, 1961)required the couple to create jointly one story foreach set of cards, (c) One of two tasks was ad-ministered first individually and then again forconsensual rankin g; one task (called NASA) involvedrank ordering 15 items for their survival value fora life-and-death trip to the moon. Correct answers

had been supplied by the National Aeronautics andSpace Administration (Hall, 1971). The other taskdesigned for this study (called the food task) in-volved rank ordering 10 items according to theirnutritional value. Correct answers had been sup-plied by a nutrition specialist at Indiana University.A pilot study with 147 undergraduates, conducted todetermine whether there was a sex bias to the tasks,revealed no significant sex differences in knowledgeon either of the tasks. The couple also completed twohigh-conflict tasks: (a) The Inventory of MaritalConflict (IMC) (Olson & Ryder, 1970) has beenfrequently used in marital research. The IMC is ahigh-conflict consensus decision-making task, in whichthe couple is presented with three short vignettes ofmarital conflict and required to agree on which spousein the vignette is primarily responsible for the problem,(b ) The couple was also asked to discuss a currentproblem that they had agreed in the interview wasmost salient and asked to try to come to a mutuallysatisfactory resolution of this problem. Within eachsession the five tasks were randomly ordered. Dis-tressed couples were told tha t a report would be writ-ten that they could read and discuss with the inter-viewing team and that could be sent to the couple'stherapist with their written permission.Results and Discussion

Multivariate analysis of variance of spouses'coding of their par tners' behavior (impac t),combined over all five tasks, with distressed-nondistressed as a between-subjects factor andhusband-wife as a within-subjects factor re-sulted in a significant effect for the distressfactor, Wilks-Lambda F(4, 25) =4.54, p


5/10

18 GOTTMAN, NOTARIUS, MARKMAN, BANK, YOPPI, AND RUBINTABLE 1

UNIVARIATE F RATIOS AND M E A N S FOR RELATIVEFREQUENCY OF SPOUSE'S CODING OF PARTNER'SBEHAVIOR, STUDY 1

Variable**Super negativeNegativePositiveSuper positive

a Neut ra l is not included

2.0610.66*:7.47*7.36*in th e

Distressedaverage.06

* .24.34.02

Nondis-tressedaverage.01.11.50.17

analysis of multivariate F\all five variables form a linearly dependent set, since they sumto 1.0.b degrees of freedom = 1*p < .05.** p < .01 ., 28.

sents means and univariate F ratios for dis-tressed and nondistressed couples.These results are consistent with the dataobtained by Birchler et al. (1975) for ob-servers' coding of the couple's behavior. Birch-ler et al. found that distressed couples weremore likely to emit negative and less likely toemit positive codes than nondistressed couples.Furthermore, the results support a communi-cation deficit model of marital distress. Bylooking at Table 1 and Table 2 in combina-tion, it can be seen that the distributions ofintent and impact are similar for nondis-tressed couples. This was not the case for dis-tressed couples. Both groups of couples werenearly identical in the way they intended theirbehavior to be coded. Therefore, for dis-tressed couples, their behavior was coded asfar less positive than was their intention.Conflict effect. To check the low-high con-flict manipulation, the videotapes of fourcouples, two distressed and two nondistressed,were randomly selected and coded for agree-ment and disagreement. Th e first 2 minutes ofeach task were coded, with coders tallyingeach occurrence of an agreement or disagree-ment. There were four pairs of coders; eachpair coded only one couple. Coders were blindto the task hypothesis and as to whether thecouple whose tape they were coding was dis-tressed or nondistressed. The total number ofagreements and disagreements were calculatedseparately for each coder, summing over high-and low-conflict tasks. Coders were firsttrained using an interaction category systemdeveloped in our laboratory (Gottman,Notarius, & Markman, Note 1) and reached

criterion on an achievement test, which testedtheir knowledge of agreement and disagree-ment codes. One coder was designated thecoder and the other the reliability checker.Correlations between coder and reliabilitychecker over the four couples were .93 foragreement and .93 for disagreement. In thelow-conflict tasks, there were avera ges of 10.33agreements and 6.92 disagreements; in thehigh-conflict tasks, there were averages of 5.38agreements and 8.00 disagreements, x2(*) 4.23, p < .05. In research on family interac-tion, agreement and disagreement have beenshown to be the best consistent discriminatorsbetween distressed and nondistressed families(Riskin & Faunce, 1972). Agreement-dis-agreement ratios less than one are character-istic of distressed families, and ratios greaterthan one are characteristic of nondistressedfamilies. For the low-conflict tasks, the agree-ment-disagreem ent ratio was 1.49, whereas forthe high-conflict tasks it was .67.To assess the effect of high- or low-conflicttasks, the five received impact codes were con-sidered on a 5-point Likert scale ranging from1 (super negative) to 5 (super positive). Thesame Likert scale was constructed for the in-tent codes. Scale scores were separately aver-aged for the three low-conflict tasks and thetwo high-conflict tasks. An analysis of vari-ance was performed on the impact data andon the intent data, with two levels of distress(distressed-nondistressed), two levels of sex(husband-wife), and two levels of conflict(high-low). As expected, there was a signifi-cant F ratio for the impact data on thedistress factor, F(l,56) = 15.37, p < .001.Distressed couples averaged 3.04 and nondis-tressed couples averaged 3.52 over all tasks.

TABLE 2UNIVARIATE F RATIOS AND M E A N S FORFREQUENCY OF SPOUSES ' INTENT,STUDY 1

VariableSuper negativeNegativePositiveSuper positive

.181.72

.53

.12

Distressedaverage.00.09.49.05

RELATIVE

Nondis-tressedaverage.00.05.55.03

a Degrees of freedom = 1, 28. All Fs are nonsignificant. Neu-t ral is not included in the analysis of multivariate F; all fivevariables form a linearly dependent set, since they sum to 1.0.


6/10

B E H A V I O R E X C H A N G E 19There was not a significant conflict effect,F ( l , 56) = 3.86, p > .05, although the meanswere in the predicted direction (high-conflictM = 3.17; low-conflict M - 3.39). There wasno significant sex main effect or any signifi-cant interactions. On the inten t variable, therewere no significant effects for any factor or forthe interactions.It was thus possible to discriminate dis-tressed from nondistressed husbands and wiveson how positively they coded their spouse'sbehavior regardless of the level of conflict thetask induced. These results support the Jacoband Davis (1973) findings and extend thesefindings to high-conflict tasks.

Reciprocity. Four analyses of covariancewere performed to compare p ( W + / H + )with p ( W + ) , p ( H + / W + ) w i t h p ( H + ) ,p ( W - / H - ) with p ( W - ) , and p ( H - /W) with p (H) for distressed and nondis-tressed couples. Probabilities were separatelycomputed for each couple. In each case theunconditional probability was the covariate.Because of the relatively low frequencies ofthe super-positive and super-negative codes,these were combined with positive and nega-tive codes, respectively, for the sequentialanalysis. The test of the reciprocity hypothe-sis of distressed versus nondistressed marriageis the significance of the group F ratio on thatportion of the conditional probability thatcannot be predicted by the unconditionalprobability. For example, the residual in re-gressing p ( W + / H + ) o n p (W+) must begreater in nondistressed couples than in dis-tressed couples. This means th at knowing tha tthe husband has just coded the wife's behavioras positive increases the prediction over thebase rat e of W + that th e wife will subse-quently code the husband's behavior as posi-tive. The direction of the prediction is ob-viously reversed for negative reciprocity; thatis, there should be more negative reciprocityin distressed than in nondistressed couples.

Table 3 presents the results of the analysesof covariance. There is some evidence thatthere is more positive reciprocity in nondis-tressed couples than in distressed couples.Though the degree of gain in prediction fromconditional to unconditional probabilities issmall for nondistressed couples, the analysisof covariance controls for the possibility that

TAB LE 3RECIPROCITY ANALYSES, STUDY 1

Variable DistressedPositive reciprocity

P (W+)p ( W+/ H +)p ( H +/ W+)Negative reciprocityp (W)p ( W - / H - )p ( H - )p ( H - / W - )

.340.430.387.350

.298.450.291.398

Nondis-tressed.541.561.635.687.161.167.063.075

ReciprocityX Groups Fh

4.68*.192

.263.04

a P (W + ) = th e probability of a positive wife code, p (W )= the probability of a negative wife code; p (H + ) = the prob-ability of a positive husband code, p (H ) = the pro babi lity ofa negative husband code.b Calculated by covariance analysis, with unconditional baserate as the covariate .*p < .05.

when unconditioned base rates are initiallylow, greater gain in prediction may be morelikely.STUDY 2

MethodSubjects and selection. An independent sample of14 clinic and 16 nonclinic couples was recruited for

participation in the research project. All clinic coupleswere seeking therapeutic assistance for marital prob-lems at either of two community mental healthcenters; nonclinic couples responded to a press re-lease that briefly described the project. Both clinicand nonclinic couples were paid $10 for participation.Precisely the same cutoff scores on the Locke-WallaceM R I as in Study 1 were used to select 12 distressedand 8 nondistressed couples for analysis; all 30couples completed the experimental procedures. Hus-bands and wives in the two groups did not signifi-cantly differ in age, educational level, number ofyears married, or salary. Couples in the present studywere on the average 32.50 years old, had been mar-ried an average of 9.44 years, had completed an aver-age of 13.95 years of schooling, and earned an aver-age salary of 11.44 thousand dollars. Thus, thecouples in Study 2 had been married appro xim ately6 years longer than the couples in Stud y 1 and ha dapproximately 2 years less of schooling. The majordifference between the samples in the studies seemsto be the length of time the couples were married.

Procedure. Based upon the results of Study 1, theprocedure was modified as follows: First, two taskswere selected: the Inventory of Marital Conflict(IMC) as a high-conflict task and the food task asa low-conflict task. The IMC was selected becauseit is a standardized task, it appeared to induce ahigh degree of conflict in Study 1 in both group s(nearly all couples fought considerably on this task),and it could be expanded to include additionalvignettes to ensure a sufficient sample of interactions


7/10

20 GOTTMAN, NOTARIUS, MARKMAN, BANK, YOPPI, AND RUBINTAB LE 4

UNIVARIATE F RATIOS AN D M E A N S FO R RELATIVEFREQUENCY OF SPOUSE'S CODING OF PARTNER'SBEHAVIOR, STUDY 2

Variablea p\>Nondis-Distressed tressedaverage average

Super negativeNegativePositiveSuper positive2.983.8219.87*.21

.05.18.30.12.00.11.55.15

a Neutral is not included in the analysis of multivariate F;all five variables form a linearly dependent set, since they sumto 1.0.b degrees of freedom = 1, 36.*p < .001.

for stable estimates of conditional probabilities. Thefood task was selected because it induced little con-flict in either group in Stud y 1. The I M C w as ex-panded to include three additional vignettes.All sessions were conducted at one of two com-munity mental health centers nearest the couple'sresidence. It may be the case th at the lack of differ-ences on inten t variables in Study 1 may have re-sulted from the couples not understanding that theywere to code how they intended their messages to betaken by their spouses. More explanation and aspecific example was provided for the intendedpositivity codes of the talk table than had beenprovided in Study 1.Results and Discussion

Multivariate analysis of variance of impactdata, with distressed-nondistressed as a be-tween-subjects factor and husband-wife as awithin-subjects factor resulted in a significanteffect for the distress factor, Wilks-LambdaF(4, 33) = 7.01, p < .001 , and no significantsex or Sex X Distress interaction. Table 4 pre-sents the univariate F ratios that contributedto the distress multivariate F ratio. Nondis-tressed spouses were more likely to code theirpartner's behavior as positive than their dis-tressed counterparts. These results replicatethose of Study 1. A comparison of the multi-variate F ratios suggests that the results areeven stronger here. This may be a function ofthe differences in the length of marriage in thetwo studies, since marital happiness is knownto decrease with length of time married withinthe age ranges of the current two studies(Burgess et al., 1971).Multivariate analysis of variance of spouses7intent resulted in no significant distress, sex,or Sex X Distress F ratios. There were nosignificant differences in intention for either

husbands or wives on any category. This pat-tern of no differences is similar to that ob-tained in Study 1. For comparison with Study1, Table 5 presents the means and univariateF ratios for distressed and nondistressedcouples.These results lend strong support to a com-munication deficit explanation of marital dis-tress: In the two studies, although distressedand nondistressed couples do not significantlydiffer in the way they intend their messages tobe received by their spouses, they do signifi-cantly differ in how the messages are actuallyreceived.Conflict effect. Conflict effects were assessedin the same manner as in Study 1. An analysisof variance on the impact data found no sig-nificant main effect for conflict, 7^(1,72) =2.29, p > .05, although, again, results were inthe predicted direction (high-conflict M =3.48, low-conflict M = 3.57). This result isconsistent with that of Study 1. As expected,there was a significant F ratio for the impactdata on the distress factor, F(l,72) = 14.57,p < .001. Distressed couples averaged 3.33,and nondistressed couples averaged 3.72 overall tasks.There was a significant Distress X Conflictinteraction for the impact data, F(l,72) =8.44, p < .01 . It w as easier to d iscriminatedistressed from nondistressed couples on thehigh-conflict task than on the low-conflicttask. Distressed couples averaged .70 loweron impact scores than nondistressed coupleson the high-conflict task; distressed coupleswere .13 higher on impact scores than nondis-tressed couples on the low-conflict task. Therewere no significant main effects or interactioneffects on the intent variable.

TABLE 5UNIVARIATE F RATIOS AND M E A N S FO R RELATIVEFREQUENCY OF SPOUSES' INTENT, STUDY 2

Variable 8 FbNondis-Distressed tressedaverage average

Super negativeNegativePositiveSuper positive.66.805.43*.23

.00.04.41.14.00.02.60.10

a Neutral is not included in the analysis of multivariate F;all five variables form a linearly dependent set, since they sumto 1.0.b Degre es of freedom = 1, 36.*p < .05.


8/10

BEHAVIOR EXCHANGE 21The presence of G roup X Conflict inte rac-tion effects in the second study and their ab-sence in the first study is interesting. It maybe that the food task is a different kind oftask for the couples of Study 2, since the hus-bands in this study are working and henceare less likely to have a role in food prepara-tion than husbands in Study 1, who are morelikely to be in school and share in the house-work. The pilot study with both tasks, men-tioned in the Procedure section of Study 1,found no significant sex differences in knowl-edge on either of the food or NASA tasks;however, a subsequent pilot study with collegestudents found significant sex differences in

self-ratings of confidence (on a 9-point scale)in the solution for the NASA task, /(68) =2.16, p < .04, with females less confident thanmales. It may be that working husbands inStudy 2 have as little confidence in theirknowledge of food as the female undergradu-ates had of their knowledge of space. Bothgroups of husb ands m ay have yielded m ore onthis task and consequently been perceived asmore positive by their wives. It may also bethe case that the length of marriage factorwas related to the obtained interaction effects.Perhaps couples who have been married longer(Study 2) have learned to exchange morepositive messages in low-conflict situationsand to avoid situations of high conflict. Thisis consistent with the clinical picture of the"stable-unsatisfactory' 7 marriage (Lederer &Jackson, 1968) or the "united-front" couple(Kramer, 1968), who can present a satisfac-tory image of themselves to the public in so-cial situations but are unable to hide behindthis image when discussing real marital issues.

Overall, there seems to be evidence that theconflict level of the task influences howspouses code each other's behavior. In thefirst study it was equally easy to discriminatedistressed from nondistressed couples whetherthey were discussing an unresolved personalissue or rank ordering a list of dog breeds.In the second study couples could be dis-criminated only on the high-conflict tasks.These results may account, in part, for thedifficulty of obtaining consistent differencesacross studies between distressed and nondis-tressed families (Jacob, 1975). It would bewise to sample systematically from low- and

TAB LE 6RECIPROCITY ANALYSES, STUDY 2

Variable* DistressedPositive reciprocityp (W+)p (W+/H+)

p (H+/W+)Negative reciprocityp(W-)p ( W - / H - )p ( H - / W - )

.448.536.405.512.223.354.235.331

Nondis-tressed

.700.787.701.756.141.195.079.121

ReciprocityX GroupsFb

2.22.58

.61

.15a P (W + ) = probability of a po sitive wife code, p (W )

= probability of a negative wife code, p (H + ) = probability ofa positive husband code, p (H ) = probability of a negativehusband code.b Calculated by covariance analysis, with unconditional baserate as the covariate.

high-conflict tasks in future investigations ofcouples and family interaction. The results ofthe present investigation would suggest thathigh-conflict tasks are more reliable for dis-crimination between distressed and nondis-tressed couples.Reciprocity. Table 6 presents the results ofthe reciprocity analyses. Reciprocity variablesdid not discriminate between distressed andnondistressed couples. In addition, the onesignificant positive reciprocity effect of Study1 was not replicated.The results of these two studies must bringinto question the current marriage counselingimage of positive reciprocity as characteristicof nondistressed marriages and not character-istic of distressed marriages. The high baserate of behavior that is positively coded byobservers or by spouses may seem to beequivalent to reciprocity, but it is not. Thedata from the present investigation supporta "bank account" model of nondistressed mar-riage rather than a reciprocity model. In abank account model, a nondistressed marriagediffers from a distressed marriage in that thereare more positive "deposits" than negative"withdrawals." In a nondistressed marriagethe consequent positive impact codes are notcontingent upon the spouse's antecedent cod-ing. Perhaps it is precisely this lack of re-ciprocity in a context of high positive ex-change that characterizes stable positive in-teraction in nondistressed couples.


9/10

22 GOTTMAN, NOTARIUS, MARKMAN, BANK, YOPPI, AND RUBINTh e results of the present investigation havedemonstrated the usefulness of the talk tableas a means for operationalizing behavior ex-change theory. Results obtained are generallyconsistent with more expensive coding studies.Furthermore, the power of variables obtainedby the couple coding their own behavior todiscriminate between distressed and nondis-tressed couples is somewhat greater than thatobtained by observer-coding studies. This re-sult may have some theoretical import. Per-haps the differences between observer andspouse coding could account for some ofthe unreliability inherent in present codingsystems. If one observer focuses on the

facial expressions of the sender of a message,a positive code could result from a read-ing of intention; if the other coder focuseson the message's impact on the receiver, anegative code could result from a reading ofthe impact of the message. It would be fruit-ful to control the cues that observers use andto study those instances when observers andspouses are discrepant. This would be one wayto study the development of a private messagesystem in close relationships in which manymessages do not mean to the stranger whatthey do to the person for whom they were in-tended. Work along these lines is in progressin our laboratory.

One limitation of the present investigationdeserves mention. Because of the arrange-ments with referring agencies and the natureof the advertisement, both couples and in-terviewers were aware of which couples wereseeking or starting marriage counseling. TheMRI scores were, however, not available toeither the interviewers or the eouples. Thisaspect of the design of the present investiga-tion introduces an experimenter expectancycriticism and the possibility that couples maybe presenting themselves in order to be con-sistent with their classification. Subsequentinvestigations could improve on the currentprocedures by advertising for couples withoutrecruiting couples from therapeutic agencies.Couples could be split on MRI scores, and theinterviewers could remain blind to a couple'sclassification. Such a procedure, however, maydraw different samples of couples than thepresent investigation, which used a conver-gence of two definitions of marital distress.

REFERENCE NOTE1. Gottman, J., No tarius, C , & Markm an, H . Th ecouples interaction scoring system (CISS). Un-published manuscript, 1976. (Available from JohnGottman, Department of Psychology, Universityof Illinois, Children's Research Center, 51 E.Gerty Drive, Champaign, Illinois 61820.)

REFERENCESAlexander, J. F. Defensive and supportive communi-cations in family systems. Journal of Marriage an dthe Family, 1973, 35 , 613-617.Azrin, N . H., Naster, B . J., & Jones , R. Reciprocitycounseling: A rapid learning-based procedure formarital counseling. Behavior Research and Therapy,1973, 11, 365-382.Birchler, G. R., Weiss, R. L., & Vincent, J. P. Multi-method analysis of social reinforcement exchangebetween maritally distressed and nondistressedspouse and stranger dyads. Journal of Personalityand Social Psychology, 1975, 31 , 349-360.Burgess, E. W., Locke, J. J., & Thomas, M. M. Th efamily. New York: Van Nostrand Reinhold, 1971.Gergen, K. The p sychology of behavior exchange .Reading, Mass.: Addison-Wesley, 1969.Haley, J. Research on family patterns: An instru-ment measurement. Family Process, 1964, 3, 41-65.Haley, J. Speech sequences of normal and abnormalfamilies with two children present. Family Process,

1961,6, 81-97.Hall, J. Decisions, decisions, decisions. PsychologyToday, November, 1971, 51-88.Jacob, T. Family interaction in disturbed and normalfamilies: A methodological and substantive review.Psychological Bulletin, 1975, 82 , 33-65.Jacob, T., & Davis, J. Family interaction as a func-tion of experimental task. Family Process, 1973, 12 ,415-427.Kramer, C. H. The theoretical position: diagnosticand therapeutic implications. In Charles Kramer,Bernard Liebowitz, Robert Phillips, Sylvia Schmidt,& James Gibson (Eds .), Beginning phase of fam ilytreatment. Chicago: Family Institute of Chicago,1968.Lederer, W. J., & Jackson, D. D. The mirages ofmarriage. New York: Norton, 1968.Locke, H. J. Prediction adjustment in marriage: Acompa rison of a divorced and happily marriedgroup. New York: Holt, 1961.Navran, L. Communication and adjustment in mar-riage. Family Process, 1967, 6, 173-180.Olson, D. H., & Ryder, R. G. Inventory of maritalconflicts (IMC): An experimental interaction pro-cedure. Journal of Marriage and the Family, 1970,32 t 443-448.Patte rson, G. R. A basis for identifying stimuli whichcontrol behaviors in natural settings. Child De-velopment, 1974, 45 , 900-911.Rappaport, A. F., & Harrell, J. A behavioral-exchangemodel for marital counseling. Family Coordinator,1972,2.7,203-212.


10/10

BEHAVIOR EXCHANGE 23Raush, H . L., Barry, W . A., Hertel, R. K., & Swain,M. A. Communication, conflict and marriage. SanFrancisco: Josey-Bass, 1974.Riskin, J., & Faunce, E. E . An evaluative review of

family interaction research. Family Process, 1972,11, 365-455.Speer, D. C. Marital dysfunctionality and two-personnon-zero-sum game behavior. Journal of Person-ality and Social Psychology, 1972, 21 , 18-24.Stuart, R. B. Operant interpersonal treatment formarital discord. Journal of Consulting and Clini-cal Psychology, 1969, 33, 675-682.Terman, L. M., & Wallin, P. The validity of marriage

prediction and marital adjustment test. AmericanSociological Review, 1949, 14 , 497-504.Thibaut, J. W., & Kelley, H. H. The social psychologyof groups. New York: Wiley, 1959.Weiss, R. L., Hops, H., & Patterson, G. R. A frame-work for conceptualizing marital conflict: A tech-nology for altering it, some data for evaluating it.In Leo A. Hamerlynck, Lee C. Handy, & Eric J.Mash (Eds.), Behavior Change: The Fourth BanffConference on Behavior Modification. Champaign,111.: Research Press, 1973.

(Received April 27, 1975)

behavior exchange theory and marital decision making

Documents