Journal of Personality and Social Psychology1982, Vol. 43, No. 6, 1230-1242

Copyright 1982 by the American Psychological Association, Inc.0022-3514/82/4306-l230$00.75

Exit, Voice, Loyalty, and Neglect:Responses to Dissatisfaction in Romantic Involvements

Caryl E. Rusbult, Isabella M. Zembrodt, and Lawanna K. GunnUniversity of Kentucky

A typology of characteristic responses to dissatisfaction in romantic relationshipsis discussed, and hypotheses concerning the determinants of each category ofresponse are outlined. It is argued that the four primary reactions to relationshipdecline are exit, voice, loyalty, and neglect. Three investment model variables(Rusbult, 1980a) should predict the conditions under which each response ismost likely to be enacted: (a) the degree of satisfaction with the relationship priorto the emergence of problems, (b) the magnitude of the individual's investmentof resources in the relationship, and (c) the quality of the best available alternativeto the relationship. Four studies provided generally consistent support for thehypotheses. As predicted, to the extent that prior satisfaction was high, voice andloyalty were more probable, whereas exit and neglect were less probable. Simi-larly, increases in investment size encouraged voice and loyalty, whereas lowerlevels of investment appeared to inspire exit or neglect responses. More attractivealternatives promoted exit while hampering loyalist behavior. These results arein agreement with investment model predictions. However, there seemed to beno (or, at best, a weak) relation between alternative quality and voice or neglectreactions to dissatisfaction.

How do individuals respond when theybecome dissatisfied with their romantic in-volvements? Under what circumstances arethey likely to actively discuss problems, andunder what circumstances are they likelysimply to end their relationships? When arepeople likely to react with quiet loyalty, andwhen do they respond with benign neglect?Social scientists have proffered numeroustheories designed to describe the develop-ment of romantic relationships (Clore &Byrne, 1974; Levinger & Snoek, 1972; Mur-stein, 1976; Saegert, Swap, & Zajonc, 1973),but insufficient attention has been given tothe phenomenon of relationship decline.

Some researchers have explored the im-pact of a variety of concrete factors (e.g., in-come, education, age) on specific responses

This research was supported by a grant to the firstauthor from the University of Kentucky Research Foun-dation. The authors wish to express their sincere grati-tude to Anne Andrews, Susan Banta, Deborah Dempsey,Randall James, Ricky Jones, Marsha Webster, and LynnWest for their help in conducting the studies and toMichael Hubbard and David Lowery for useful com-ments on an earlier draft of this article.

Requests for reprints should be sent to Caryl Rusbult,Department of Psychology, University of Kentucky,Lexington, Kentucky 40506.

to dissatisfaction, such as divorce and break-ups (Hill, Rubin, & Peplau, 1976; Levinger& Moles, 1979), communication style andcontent (Fineberg & Lowman, 1975; Wills,Weiss, & Patterson, 1974), extrarelationshipsexual involvements (Glass & Wright, 1977;Jaffe & Kanter, 1976), and the expression ofnegative affect or hostility (Billings, 1979;Gottman, Notarius, Markman, Bank, &Yoppi, 1976). Researchers working withinthis tradition have identified a wide range ofpotential responses to dissatisfaction andhave explored the impact of numerous basicfactors on these behaviors. However, few ofthese authors have developed systematic tax-onomies of this domain of behaviors, norhave they constructed abstract models of theconditions that promote one response overalternative reactions.

Other psychologists have proposed theo-ries of the development and deterioration ofrelationships. For example, in their discus-sion of the "depenetration process," Alt-man and Taylor (1973) explore the processof conflict management and suggest that de-penetration involves "a cycling between ex-cessive interaction and withdrawal" (p. 174),which leads to more constricted exchange.

1230

RESPONSES TO DISSATISFACTION 1231

Levinger (1979) describes relationship growthand dissolution in terms of attractions (pos-itive minus negative) and barriers (commit-ments, obligations, termination costs) inher-ent in a given romantic involvement. Rus-bult (1980a) also proposes that theindividual's level of commitment to main-tain a relationship is in large part a functionof three simple factors: declining satisfaction,increases in alternative quality, or "divesti-ture," which lead to declining commitmentand relationship dissolution. Each of theseapproaches provides an interesting theoreti-cal account of growth and decline processes,but all focus primarily on individuals' stay/leave decisions to the exclusion of alternativereactions to dissatisfaction.

Thus, one body of literature provides arich description of the range of available re-sponses to decline, and a second traditionpresents theoretical models of the process ofdeterioration. The present article attempts tointegrate these two approaches. Its goals areto (a) outline a simple typology of reactionsto declining satisfaction and (b) propose atheory that delineates a set of abstract pre-dictors of these responses.

The present model draws on knowledgeaccumulated in research and theory, reportedabove, and emerges directly from several ad-ditional sources: (a) Hirschman's (1970,1974)theory of responses to decline in economic/political organizations, (b) Rusbult and Zem-brodt's (in press) typology of responses todeclining satisfaction in romantic associa-tions, and (c) Rusbult's (1980a) investmentmodel of satisfaction and commitment inexchange relationships.

In his discussion of reactions to decline infirms, organizations, and states, Hirschman(1970) attempted to outline a simple typol-ogy of responses to dissatisfaction. He pro-posed that three characteristic responses todeteriorating satisfaction exist: (a) exit—end-ing the relationship, (b) voice—actively andconstructively expressing one's dissatisfac-tion, with the intent of improving conditions,arid (c) loyalty—passively but optimisticallywaiting for conditions to improve.

Rusbult and Zembrodt (in press) per-formed a multidimensional-scaling study ofresponses to declining satisfaction in ongo-ing, adult romantic involvements and found

that these three categories characterize be-haviors in romantic relationships. They alsoidentified a fourth logical response to dissat-isfaction, neglect: passively allowing a rela-tionship to atrophy. These four categoriesappeared to provide a fairly comprehensive,yet simple, description of the domain of re-actions to deteriorating satisfaction. The fol-lowing are examples of behaviors represen-tative of each category of response:

Exit—formally separating, moving out of a joint resi-dence, deciding to "just be friends," getting a divorce.

Voice—discussing problems, compromising, seekinghelp from a therapist or clergyman, suggesting- solu-tions to problems, asking the partner what is botheringhim or her, trying to change oneself or change thepartner.

Loyalty—waiting and hoping that things will improve,"giving things some time," praying for improvement.

Neglect—ignoring the partner or spending less time to-gether, refusing to discuss problems, treating the part-ner badly emotionally or physically, criticizing thepartner for things unrelated to the real problem, "justletting things fall apart," (perhaps) developing extra-relationship sexual involvements.

Theoretically, these responses differ fromone another along two dimensions. The firstdimension is constructiveness/destructive-ness. Whereas voice and loyalty are construc-tive responses that are generally intended tomaintain and/or revive the relationship, exitand neglect tend to be relatively destructive.The second dimension is activity/passivity.Exit and voice are active behaviors (i.e., theindividual is doing something about the re-lationship), whereas loyalty and neglect aremore passive responses.1 The current mean-ing of the term passive may differ somewhatfrom common usage. Here, passive refers tothe impact of the behavior on the problemat hand and may not necessarily be descrip-tive of the behavior itself. For example, aneglectful response'such as destructive criti-cism may be overtly active, but it is passiveand destructive in regard to the future of thecurrent relationship.

Rusbult's investment model (1980a) iden-tifies some basic variables that may predictthe conditions under which each of these fourresponses should occur. The model is basedon interdependence theory (Kelley & Thi-

1 In the most extreme cases, of course, loyalty andneglect are "nonresponses," or the complete absence ofactivity, and are behaviorally indistinguishable.

1232 C. RUSBULT, I, ZEMBRODT, AND L. GUNN

baut, 1978; Thibaut & Kelley, 1959) and em-ploys traditional exchange-theory concepts(c.f. Blau, 1964; Romans, 1961). That is, themodel assumes that humans are motivatedto maximize outcomes, points to the impor-tance of relationship structure as a determi-nant of behavior in dyads, and makes use ofexchange-theoretical constructs such as re-wards, costs, alternatives, and investments.Thus, it is similar in many respects not onlyto interdependence theory but also to thework of Levinger (1979), Altman (1974),Blau (1964), and Becker (1960). In the past,the investment model has been utilized todescribe the development of satisfaction andcommitment in exchange relationships suchas romantic associations (Rusbult, 1980a),friendships (Rusbult, 1980b), and jobs (Far-rell & Rusbult, 1981). The three variablesthat predict degree of commitment to rela-tionships should also determine the condi-tions under which exit, voice, loyalty, andneglect are likely to be enacted. These threevariables are (a) the degree to which the in-dividual was satisfied with the relationshipprior to its decline, (b) the magnitude of theindividual's investment of resources in therelationship, and (c) the quality of the indi-vidual's best alternative to the current rela-tionship.

In general, increases in prior satisfactionshould promote constructive rather than de-structive responses to dissatisfaction. Givenhigh prior satisfaction, the individual is likelyto believe that it is desirable to restore therelationship to its previous state and that con-structive responses are more likely than de-structive ones to produce "payoffs." Thus,greater prior satisfaction should induce voiceand loyalty while discouraging exit and ne-glect. The extant research provides indirectsupport for these assertions. It has been dem-onstrated that greater marital satisfaction,happiness, and adjustment (each of whichimplies greater prior satisfaction) are asso-ciated with voicelike behaviors such as su-perior communication (Beir & Sternberg,1977; Murphy & Mendelson, 1973), greaterself-disclosure (Critelli & Dupre, 1978), moreexpressions of love and affection (Fineberg& Lowman, 1975), more frequent exchangesof pleasurable behaviors (Wills et al., 1974),and greater perceptual accuracy between

partners (Sporakowski & Hughston, 1978).Decreases in satisfaction seem to be asso-ciated with neglect responses such as theexpression of negative affect, hostility, andbelligerent complaints (Billings, 1979; Gott-man, Markman, & Notarius, 1978; Hawkins,1968); negative attributions about partners'communications (Gottman et al., 1976); theenjoyment of fewer shared recreational ac-tivities (Birchler, Weiss, & Vincent, 1975);and the greater likelihood of extramarital coi-tus (Bell, Turner, & Rosen, 1975; Glass &Wright, 1077). It also appears that greaterhomogamy, higher socioeconomic status(SES), higher esteem for partner, and in-creased satisfaction tend to produce morestable, durable relationships (i.e., loyalty),whereas greater marital discord, unequal in-volvement, low homogamy, and husbands'low income and employment stability arepredictive of divorce and separation (i.e.,exit; Cherlin, 1979; Cutright, 1971; Hill etal., 1976; Todres, 1978). Collectively, theseresults provide support for the hypothesisthat greater prior satisfaction promotes voiceand loyalty and inhibits exit and neglect.

Increases in investment size should alsoencourage constructive behaviors while dis-couraging destructive responses. "Invest-ments" refers to the resources the individualhas put directly into the relationship that arethen intrinsic to that involvement (e.g., self-disclosing, spending time with the partner,investing emotion) or resources that are ex-trinsic but have become indirectly connectedto the association (e.g., shared material pos-sessions, shared recreational activities, mu-tual friends). Individuals who have investedhighly in their relationships have much tolose by abandoning them and should there-fore be more likely to enact constructive re-sponses intended to maintain their relation-ships. Unfortunately, few researchers haveexplored the phenomenon of investment.However, relationships of greater duration(i.e., greater investment size) have been shownto exhibit higher levels of communication(i.e., voice; Krain, 1975), and younger per-sons, who have perhaps invested less in theirrelationships, are more likely to separate anddivorce (i.e., exit; Bloom, Hodges, Caldwell,Systra, & Cedrone, 1977). Additionally, Rus-bult (1980a) found that individuals who had

RESPONSES TO DISSATISFACTION 1233

invested more heavily in their relationshipswere more likely to report commitment totheir partners and relationships (i.e., exhibitloyalty) and less likely to report intentionsto end their involvements (i.e., exit). Evi-dence from research on nonromantic rela-tionships also supports these hypotheses. Ex-periments on the entrapment phenomenonconsistently demonstrate that persons whohave numerous or sizable resources attachedto a line of action are very likely to persistin that behavior (i.e., remain loyal; Rubin& Brockner, 1975; Staw, 1976; Staw & Fox,1977; Teger, 1980). The literature on effortjustification also suggests that individualswho have exerted greater effort for an activ-ity/person are likely to report greater attrac-tion and commitment to that activity (i.e.,exhibit loyalty; Aronson & Mills, 1959; Ger-ard & Mathewson, 1966; Wicklund, Cooper,& Under, 1967)., Alternative quality should determinewhether the individual's response to dissat-isfaction will be active or passive. Good al-ternatives (an alternative relationship or sol-itude) both provide the individual with themotivation to do something (be active, voice,or exit; "shape up or ship out") and give himor her a source of power for effecting changesin the relationship—the threat posed by anattractive alternative can serve as a source ofpower. In the absence of a good alternative,the mc-re probable reaction is to passivelywait for the relationship to improve (i.e., re-main loyal) or passively allow it to die (i.e.,neglect it). Research evidence suggests thatgood alternatives—in the form of an extra-relationship sexual partner (Glass & Wright,1977; Hunt, 1969; Jaffe & Kanter, 1976), thewife's greater actual or potential earnings(Cherlin, 1979; Ross & Sawhill, 1975), orgreater physical attractiveness relative topartner (White, 1980)—are positively asso-ciated with breakups or divorce. Rusbult(1980a) also found that better alternativesencouraged separation (i.e., exit) and dis-couraged commitment to maintain relation-ships (i.e., loyalty).

In what way can these hypotheses betested? Because effecting experimental ma-nipulations of critical features in romanticinvolvements was judged from the outset tobe unethical and/or nearly impossible, alter-

native methodologies were by necessity ex-plored. Ideally, a combination of longitudi-nal studies and careful observation of mar-ried couples' problem-solving behaviorsshould be undertaken (such research is pres-ently being planned). The present article re-ports four studies designed as a preliminarytest of these hypotheses. Two studies employrole-playing methodologies and two exploreresponses to dissatisfaction in cross-sectionalsurveys of real, adult romantic involvements.Both role-playing experiments (Cooper, 1976;Darroch & Steiner, 1970; Freedman, 1972)and survey studies (Campbell & Stanley,1963; Kidder, 1981) have methodologicalweaknesses: Role-playing methods possesslow realism, and survey approaches are in-capable of producing sound informationconcerning cause-effect relations. But thestrengths of one method correspond to .theweaknesses of the other. The role-playingstudies are highly controlled and clearly dem-onstrate causal relations, whereas the surveyspossess greater real-world validity. Thus,through converging operations, these studiesmay provide a preliminary test of the currentmodel.

Study 1

Study 1 was designed to establish thecausal impact on exit, voice, loyalty, andneglect of changes in investment model fac-tors. Degree of prior satisfaction, investmentsize, and alternative quality were orthogo-nally manipulated in a role-playing essay thatdescribed a once-satisfying relationship onthe decline. Subjects were asked to placethemselves in the position of the essay pro-tagonist (the dissatisfied individual) andcomplete a questionnaire concerning theirreactions to and probable behavior in thatsituation.

Method

Subjects. Subjects were 128 undergraduates from in-troductory psychology classes at the University of Ken-tucky. Each of 16 subjects recruited for every experi-mental session was randomly assigned to one of eightexperimental conditions. There were equal numbers ofmales and females within each condition.

Procedure. The experimenter informed subjects thatthey were to read an essay describing a fictional situationand asked that they attempt to place themselves in theposition of the major character. Essays read by male and

1234 C. RUSBULT, I. ZEMBRODT, AND L. GUNN

female subjects were identical except for changes in thesex of the major character, current partner, and alter-native partner. In females' essays, the protagonist (Sarah)was described as a 21-year-old college student who en-joyed reasonable success socially and academically.Sarah had dated her current partner, Robert, for 3months and had disclosed some of her personal feelingsto him. However, this relationship had recently becomeless satisfying because Robert had started drinkingheavily at parties and had said rude and abusive thingsto Sarah and her friends. Additionally, Sarah had re-cently met a man named John who was interested indating her. The essay ended with the following state-ment: "What are you going to do about your relationshipwith Robert? Think about this situation for a few min-utes, reading the essay a second time if necessary. Thencomplete the attached questionnaire." At the end of thesession subjects were debriefed and thanked for theirparticipation.

Independent variables. Prior satisfaction was manip-ulated through variations in the description of the pro-tagonist's feelings for his or her partner before the part-ner developed a drinking problem. In the low-prior-sat-isfaction condition the current partner was described asmarginally attractive on a variety of dimensions, and theprotagonist was said to be only "moderately attractedto Robert . . . you think that you are not really in lovewith him, and aren't really as satisfied with the relation-ship as you have been with past ones." In contrast, inthe high-prior-satisfaction condition the current partnerwas said to possess numerous attractive qualities, andthe involvement was described as "relatively rewarding. . . you are fairly strongly attracted to Robert . . . youare coming to love him." Investment size was variedthrough changes in the description of the amount andtype of the protagonist's self-disclosures to the currentpartner. In the high-investment-size condition the pro-tagonist was said to have disclosed about "the most pri-vate, secret aspects" of her life, whereas in the low-in-vestment-size condition the protagonist had disclosedlittle. Alternative quality was manipulated throughchanges in the description of the alternative person(John/Lisa) who was interested in dating the protagonist.In the high-alternative-quality condition John was de-scribed as an extremely attractive person who engagedin many activities Sarah found to be enjoyable. In thelow-alternative-quality condition the alternative was de-scribed as only moderately appealing.

Dependent variables. All subjects wrote responses tothe following open-ended question: "Please describe theway in which you are likely to respond to your currentdissatisfaction with Robert." One blank, lined page wasprovided for subjects' responses. Two trained coders,naive to the experimental condition that elicited eachresponse, scored each essay for degree of exit, voice, loy-alty, and neglect (1 *= not at all, 5 = definitely; the re-spective interrater reliability coefficients were .82, .85,.75, and .68). Subjects then completed 17 Likert-typemanipulation checks and dependent measures. Endpoints on the seven-point scales were "not at all/ex-tremely," "nothing/a great deal," or "not at all/defi-nitely." The prior-satisfaction manipulation checks askedhow satisfying the relationship was, how attracted to therelationship the subject was, and the extent to which thesubject liked/loved the partner before the drinking prob-

lem emerged. The investment-size manipulation checksrequired that subjects report the degree to which theydisclosed their private feelings to the partner, the partnerwas aware of the subject's "private self," and the subjecthad invested emotionally in the relationship. For thealternative-quality variable, the manipulation checksasked how satisfying the subject thought a relationshipwith the alternative would be, how attracted to that re-lationship the subject was, and the extent to which thesubject liked/loved the alternative person. Two measuresof each response to dissatisfaction were obtained. Forexit, the measures were "I will end the relationship withRobert" and "I am motivated to end my current rela-tionship." The voice measures were "I will try to discussthe problem with Robert" and "I will openly discuss theproblems with our relationship." The two measures ofloyalty were "I will remain quietly loyal to my relation-ship with Robert" and "I don't plan to do anything butexpect that things will improve." For neglect, the twomeasures were "I will passively allow the relationshipwith Robert to deteriorate" and "I don't plan to do any-thing, and expect that things will become worse."

Results and Discussion

Reliability of measures. Measures withineach set of items were statistically related toone another: A series of reliability analysesrevealed significant alphas for the prior sat-isfaction (.89), investment size (.95), and al-ternative quality (.86) manipulation checksand for the Likert and coded-essay measuresof exit (.80), voice (.82), loyalty (.66), andneglect (.65). Because all reliability coeffi-cients exceeded lowest acceptable levels(Nunnally, 1967), a single averaged measurewas created for each set.

Manipulation checks. Three-factor anal-yses of variance (ANOVAS) were performedon each set of manipulation checks. Com-pared to subjects in the low-prior-satisfactioncondition, subjects in the high-prior-satisfac-tion condition reported greater satisfactionwith the partner prior to the development ofhis/her drinking problem (low M = 4.99,high M = 6.04), F(l, 120) = 57.69,p < .001.Subjects in the high-investment conditionstated that they had invested more heavilythan their low-investment counterparts (lowM=1.82, high M=6.13), F(l, 120) =615.15, p < .001. Subjects in the high-alter-native-quality condition reported greater at-traction to their alternatives than did thosein the low-alternative-quality condition (lowM= 4.54, high M = 5.42), F(l, 120) = 39.38,p < .001. No other main effects or interac-tions were significant. Thus, all three inde-

RESPONSES TO DISSATISFACTION 1235

pendent-variable manipulations appear tohave been successful.

Responses to dissatisfaction. A three-fac-tor multivariate analysis of variance (MAN-OVA) was performed on the four response-to-dissatisfaction measures.2 The results of thisanalysis are displayed in Table 1. As pre-dicted, high prior satisfaction produced agreater tendency toward voice and loyaltyand a lesser tendency toward exit and neglectthan did low prior satisfaction, F(4, 117) =11.19, p < .001. Compared to subjects in thelow-investment-size condition, those in thehigh-investment-size condition were morelikely to enact voice and loyalty than exit andneglect, F(4, 117) = 3.01, p < .02, althoughthe impact on neglect was not statisticallysignificant. The multivariate effect of alter-native quality on the four responses was sig-nificant, F(4, 117) = 3.59, p < .009, but theonly significant univariate effects were for theexit and neglect measures. Additional anal-yses revealed that sex of subject produced nosignificant main effects or interactions, andthere were no other statistically significanteffects. Thus, hypotheses concerning the im-pact of prior satisfaction and investment sizewere generally supported, although the im-pact of investment size on neglect was notstatistically significant. Alternative qualitysignificantly influenced exit and neglect butnot voice or loyalty.

Study 2

The second study was designed to deter-mine whether the results of Study 1 hold inreal, ongoing romantic involvements. A cross-sectional survey of college-age subjects' ro-mantic involvements was performed to ob-tain measures of each model parameter.Multiple regression and related techniqueswere employed to test predictions concerningthe determinants of each response to dissat-isfaction.

Method

Subjects. Subjects were 85 University of Kentuckyundergraduates who completed the questionnaire forcourse credit. Fifteen to 20 subjects attended each ex-perimental session.

Procedure. The experimenter explained that subjectswere to complete a questionnaire designed to explorethe ways in which people react when they become dis-

Table 1Mean Exit, Voice, Loyalty, and Neglect for EachExperimental Manipulation: Study 1

Level ofindependent

variable

Variable High Low

Prior satisfactionExitVoiceLoyaltyNeglect

Investment sizeExitVoiceLoyaltyNeglect

Alternative qualityExitVoiceLoyaltyNeglect

3.154.732.642.34

3.454.722.S42.18

3.854.532.282.51

4.26**4.04**2.03**3.08**

3.98*4.05**1.94*2.49

3.54*4.242.452.71*

Note. Higher numbers indicate a greater propensity toenact that response to dissatisfaction. The possible rangeof each measure (after appropriate transformations) was1 to 6 (two scales ranged from 1 to 7, two from 1 to5). Each mean is the average of four measures of thatresponse.* p< .05. **p< .001.

satisfied with their romantic involvements. She de-scribed the range of relationships subjects were allowedto describe (any duration or degree of seriousness) andstated that the dissatisfaction described in the question-naire could concern any problem from the most trivialannoyance to the most severe incompatibility. After fin-ishing the questionnaire, which required approximately30 minutes to complete, subjects were debriefed.

Questionnaire. Subjects wrote one-page essays in re-sponse to the following: "Please think of a time whenyou became dissatisfied with your romantic relationship.Describe that situation and your feelings, and especiallyyour response to the situation." Two trained codersscored each essay using the same categories employedin Study 1 (extent of each response on five-point scales;the interrater reliability coefficients were .90 for exit, .79for voice, .54 for loyalty, and .74 for neglect). In addition,a set of seven-point Likert-type scales was constructedto measure each parameter of the investment model.For each parameter, the variable was briefly defined, aset of items representing concrete examples of the vari-

2 Separate analyses of the Likert-type measures andthe coded-essay measures were initially performed, andthese separate analyses revealed nearly identical patternsof results. Therefore, a single averaged measure of eachresponse to dissatisfaction was used in the analyses.

1236 C. RUSBULT, I. ZEMBRODT, AND L. GUNN

Table 2Correlations of Investment Model VariablesWith Exit, Voice, Loyalty, and Neglect: Study 2

Variable Exit Voice Loyalty Neglect

Prior satis-faction -.39** .30* .37** -.39**

Investmentsize -.29* .39** .40** -.31*

Alternativequality .35** -.06 -.25* .14

*p<.05. **;><.001.

able was provided (e.g., for investment size, "How muchtime do you spend together"), and several abstract mea-sures of the variable were obtained. Unless otherwiseindicated, end anchors on the scales were "not at all/extremely" or "nothing/a great deal." The abstract mea-sures of prior satisfaction were "Before you became un-happy with your relationship, how satisfied with it wereyou," "How attracted to your partner were you," "Howpositive were your feelings about it," and "How muchdid you care about your partner." Alternative qualitywas measured by the following abstract measures: "Howsatisfying did you. think your alternatives (another re-lationship or solitude) would be," "How did your bestalternative compare to your ongoing relationship" (1 =alternative was much worse, 7 = alternative was muchbetter), and "How did your best alternative compare tothe lifestyle you ideally desire" (1 = very far from ideal,7 = very close to ideal). Three items served as abstractmeasures of investment size: "All things considered, towhat extent were there important objects, persons,events, or activities associated with the relationship thatyou would have lost (in some "sense) by ending it" (1 =none, 7 = a great many), "What was the size of yourinvestment in your relationship" (1 = invested nothing,7 = invested a great deal), and "What was the size ofyour investment in your relationship compared to whatmost people invest in their relationship" (1 = much lessthan others, 7 = much more than others). The measuresof exit, voice, loyalty, and neglect were identical to thoseused in Study 1 (two measures of each response), exceptthat they were phrased in the past tense in this experi-ment (i.e., "What did you do?" rather than "What willyou do?").

Results and Discussion

Reliability of measures. Reliability anal-yses revealed significant alphas for the priorsatisfaction (.89), alternative quality (.69),and investment size measures (.90) and forthe Likert-type and coded-essay measures ofexit (.86), voice (.80), loyalty (.65), and ne-glect (.74). Therefore, a single measure ofeach variable was formed by averaging eachsubject's scores on the measures that com-posed every set.

Responses to dissatisfaction. Multiplecorrelations between the four response-to-dissatisfaction measures and each investmentmodel predictor were computed.3 Exit, voice,loyalty, and neglect were significantly cor-related with prior satisfaction (R = .50), in-vestment size (R = .51), and alternative qual-ity R = .38). Relevant zero-order correla-tions are presented in Table 2. Consistentwith predictions, both prior satisfaction andinvestment size were positively correlatedwith voice and loyalty and negatively cor-related with exit and neglect. However, al-though alternative quality encouraged exitand discouraged loyalty (as predicted), thecorrelations with voice and neglect were notstatistically significant.

Study 3

Study 3 was designed to extend the gen-eralizability of Study 1 by replicating thatstudy while operationaljzing investmentmodel variables in different ways. Also, newitems concerning exit, voice, loyalty, andneglect were created in an attempt to im-prove measurement. Finally, we intended toprovide yet another opportunity to explorethe impact of alternative quality and to de-termine if this factor continued to produceonly weak effects on responses to dissatisfac-tion.

Method

Subjects. One hundred twelve University of Ken-tucky undergraduates participated in partial fulfillmentof the requirements for an upper division psychologycourse. Each of 30 to 50 subjects recruited for everyexperimental session was randomly assigned to one ofeight experimental conditions. There were equal num-bers of males and females in each condition.

Procedure. The procedure was the same as in Study1. Subjects read an essay describing a fictional situation,placed themselves in the position of the major character,and completed a questionnaire as they would if theywere the essay protagonist. The protagonist was de-scribed as a reasonably successful individual involvedin a deteriorating romantic relationship. The source ofthe dissatisfaction was the same as in Study 1—the part-ner's drinking problem.

Independent variables. Again, prior satisfaction wasmanipulated through variations in the description of the

3 Preliminary analyses revealed similar patterns of re-sults for the Likert-type and the coded-essay measures,so an averaged measure of each response was used.

RESPONSES TO DISSATISFACTION 1237

protagonist's relationship before the partner developeda drinking problem. However, the specifics of the de-scription differed from those previously employed, Inthe low-prior-satisfaction condition, the current partnerwas described as "okay," and the protagonist was saidto be "attracted to Robert, but you aren't as satisfiedwith the relationship as you would like to be. All thingsconsidered, you would call it an 'okay' relationship." Incontrast, in the high-prior-satisfaction condition, thecurrent partner was described as the kind of partner thatSarah had always dreamed about: All things considered,Sarah felt "that Robert was the ideal match," and shewas "very strongly attracted to him." In Study 1, in-vestment size was varied through changes in the amountand type of self-disclosures the protagonist had made toher partner; the invested resource was put directly intothe relationship, according to the essay protagonist's ownchoice. In the present study, the invested resource wasextrinsic; the resource was initially extraneous but laterbecame connected to the relationship and would havebeen lost if the relationship had ended. In the high-in-vestment-size condition, the two were tennis doublespartners who had won numerous matches together. Be-cause their coach was convinced that they could "go far"as doubles partners, they had signed up for several tour-naments in the spring and summer. Thus, the protag-onist's future in tennis, which was said to be importantto her, became connected to her future with Robert. Inthe low-investment-size condition, the two were said toplay occasional doubles matches with friends, but bothstill preferred to play singles tennis in tournaments.Thus, although she enjoyed playing tennis with Robert,Sarah's future in tennis was in no way connected to herfuture with Robert. In Study 1, alternative quality wasmanipulated through changes in the attractiveness of aspecific alternative date. In the present experiment, al-ternatives were varied through changes in the protago-nist's evaluation of the general quality of the field ofalternatives. In the high-alternative-quality condition,the protagonist felt that "there were a number of at-tractive, available men at the University . . . the Uni-versity is a wonderful place to meet all sorts of inter-esting, fun men . . . 'there are a lot of fish in the sea.' "In the low-alternative-quality condition, the protagonistbelieved that her "standards were quite high—the avail-able men at the University were apparently attractive toothers, but not to [her] . . . 'a good man is hard tofind.'"

Dependent variables. Two types of measures wereobtained in the experiment. First, all subjects completeda questionnaire consisting of 17 nine-point Likert-typescales. This questionnaire was similar to that employedin Study 1 and contained manipulation checks on thethree independent variables and measures of exit, voice,loyalty, and neglect. Unless otherwise noted,' end pointson the scales were "not at all/extremely" or "nothing/a great deal." The prior-satisfaction manipulation checksasked subjects to report degree of attraction to the part-ner, satisfaction with the relationship, and liking/lovingfor the partner before problems emerged in the rela-tionship. The manipulation checks on the investmentsize variable concerned the extent to which the protag-onist's career was "connected" to his or her relationship,his or her enjoyment of tennis would be affected by achange in the relationship, and it was important to con-

tinue playing tennis with the partner. The alternative-quality manipulation checks assessed degree of attrac-tion to other persons at the University, anticipatedsatisfaction derived from dating these persons, and theextent to which available others compared favorably tothe ideal. The exit, voice, loyalty, and neglect measuresin this portion of the questionnaire were identical tothose employed in Study 1 (two measures of each re-sponse). The second set of measures was an additionalquestionnaire designed specifically to obtain better mea-sures of exit, voice, loyalty, and neglect. This question-naire contained 20 five-point Likert-type scales: fiveitems for each response to dissatisfaction. The endpointsfor each item were 1 = I would definitely not do this and5 = 1 would definitely do this. The exit measures were"I would end the relationship," "I would tell my partnerto leave," "I would try to figure out ways to get out ofit," "I would drop him/her like a hot potato," and "Wewould go our separate ways." The voice measures were"I would talk to my partner about what was botheringme," "I would suggest things that I thought would helpus," "I would ask my partner what was bothering him/her," "I would ask my partner what I was doing wrong,"and "I would try to fix things up." The loyalty measureswere "When the problems emerged, I would wait, hop-ing things would get better," "I would hope that if I justhung in there things would get better," "I would waitpatiently," "I would never consider ending the relation-ship—I would wait for things to improve some," and"I would wait to see what would happen." The neglectmeasures were "I guess I would just sort of let things fallapart," "I would get angry and wouldn't talk at all,""I would watch t.v., and we probably wouldn't talkmuch," "I would start treating my partner badly," and"I would allow the relationship to 'die a slow death.' "These items were randomly ordered, and a single ran-dom order was employed across all subjects.4

Results and Discussion

Reliability of measures. Reliability anal-yses revealed significant alphas for the ma-nipulation checks designed to measure priorsatisfaction (.95), investment size (.92), andalternative quality (.95) and for both sets ofmeasures of exit (.79), voice (.80), loyalty(.76), and neglect (.66). Therefore, a singleaveraged measure was computed for each setof items.

Manipulation checks. The manipulationchecks were subjected to a three-factor AN-OVA. Compared to low-prior-satisfaction-condition subjects, high-prior-satisfactionsubjects reported greater attraction, love, andsatisfaction prior to the relationship's decline

4 Because results for open-ended essay responses andstructured questionnaire items were equivalent in Stud-ies 1 and 2, no open-ended essay responses were ob-tained in this study.

1238 C. RUSBULT, I. ZEMBRODT, AND L. GUNN

(low M = 5.25, high M = 7.97), F( 1, 104) =383.54, p < .001. High-investment-size sub-jects felt more "connected" to their partnersthan did their low-investment-size counter-parts (low M = 2.21, high M= 7.01), F(l,.104) = 415.64, p < .001. Subjects in the high-alternative-quality condition perceived thatthe field of alternatives was generally moreattractive and appealing than did those in thelow-alternative-quality condition (low M =2.68, high M = 6.94, F(l, 104) = 416.39, p <.001. No other main effects or interactionswere significant. Thus, the experimental ma-nipulations appear to have been successful.

Responses to dissatisfaction. A three-fac-tor MANOVA was performed on the exit,voice, loyalty, and neglect measures.5 A sum-mary of the results of this analysis is pre-sented in Table 3. The multivariate effect onresponses to dissatisfaction was significantfor the prior satisfaction, F(4, 101) = 10.64,p< .001, and investment size factors,

Table 3Mean Exit, Voice, Loyalty, and Neglect forEach Experimental Manipulation: Study 3

Level ofindependent

variable

Variable High Low

Prior satisfactionExitVoiceLoyaltyNeglect

Investment sizeExitVoiceLoyaltyNeglect

Alternative qualityExitVoiceLoyaltyNeglect

3.215.813.892.54

3.515.413.702.65

3.695.443.572.84

4.09**4.98**3.23*3.07**

3.90*5.383.32*2.96*

3.615.353.452.78

Note. Higher numbers indicate a greater propensity toenact that response. The possible range of each measure(after appropriate transformations) was approximately1 to 7 (two scales ranged from 1 to 9, five from 1 to 5).Each mean is the average of seven measures of that re-sponse.*p<.05. **p<.001.

101) = 3.10, p < .02, but not for the alter-native quality variable, F(4, 101) = .20, p <.94, ns. As predicted, prior satisfaction sig-nificantly promoted voice and loyalty anddiscouraged exit and neglect. Greater invest-ment in a relationship encouraged loyaltyand inhibited exit and neglect but did notsignificantly affect voice behaviors. These re-sults are in basic agreement with the hy-potheses and with the results of Studies 1 and2. However, although the alternative qualitymanipulation appeared to have created theintended conditions, this variable did not sig-nificantly affect any of the four responses todissatisfaction. Additional analyses revealedno significant main effects or interactions in-volving sex of subject, and there were no sig-nificant interactions in the three-factor anal-ysis above.

Study 4

Study 4, a cross-sectional questionnaire,was identical to Study 2 except for two vari-ations. First, questionnaire items were re-written in an attempt to obtain better mea-sures of each investment model parameter.Particular care was taken on the alternativequality measures, which yielded a lower re-liability coefficient in Study 2 than did otherclusters of measures. Second, the new re-sponse-to-dissatisfaction scale (from Study 3)was added to the existing questionnaire inorder to measure exit, voice, loyalty, andneglect better.

Method

Subjects. Subjects were 77 undergraduates from anintroductory sociology course. The questionnaire wascompleted during a regular lecture session, so all subjectsperformed the task simultaneously.

Procedure. The experimenter explained the purposeand nature of the questionnaire, as in Study 2. Subjectswere asked to describe a relatively dissatisfying experi-ence (of trivial or major importance) in a romantic re-lationship of any duration or degree of involvement. Thequestionnaire required approximately 40 minutes tocomplete. After all subjects had finished, students weredebriefed and thanked for their cooperation.

Questionnaire. As in Study 3, the questionnaire con-tained two types of measures: regular seven-point Likert-

5 Preliminary analyses revealed identical patterns offindings for the two types of measures of response todissatisfaction, so a single averaged measure of each re-sponse was used in the analyses.

RESPONSES TO DISSATISFACTION 1239

type items, designed to measure investment model vari-ables (both concrete and abstract) and responses to dis-satisfaction (these were modified from those used inStudy 2), and the response-to-dissatisfaction scale de-veloped for Study 3. Unless otherwise indicated, endanchors were "not at all/extremely," "none/many," or"nothing/a great deal." Both concrete and abstract mea-sures of prior satisfaction and investment size were iden-tical to those employed in Study 2. Several new concretemeasures of alternative quality were added, and previousmeasures were revised in an attempt to make their mean-ing clearer (e.g., the second measure was changed to"Which was better, your relationship or your alterna-tives?" 1 = alternatives were much worse, 7 = alterna-tives were much better). The exit, voice, loyalty, andneglect -measures were the same as those employed inthe previous studies. The seven-point questionnaireitems were identical to those in Studies 1, 2, and 3 (twomeasures of each response). The response-to-dissatisfac-tion scale, a second means of measuring each reaction,was identical to that used in Study 3 except that itemswere phrased in the past tense.6

Results and Discussion

Reliability of measures. Reliability anal-yses revealed significant alphas for the setsof items designed to measure prior satisfac-tion (.89), investment size (.86), alternativequality (.79), exit (.82), voice (.88), loyalty(.85), and neglect (.72).7 Therefore, the mea-sures employed in the remaining analyseswere averaged scores for items within eachcluster.

Responses to dissatisfaction. Each invest-ment model variable was regressed onto theexit, voice, loyalty, and neglect measures.The resulting multiple correlations were sig-nificant for prior satisfaction (R = .68), in-vestment size (R - .61), and alternative qual-ity (R = .57).8 Zero-order correlations be-tween investment model variables andresponses to dissatisfaction are presented inTable 4. Consistent with the experimentalhypotheses, both prior satisfaction and in-vestment size were positively correlated withvoice and loyalty and negatively correlatedwith exit and neglect. Also as predicted, al-ternative quality was positively related to exitand negatively related to loyalty. However,alternative quality was not significantly cor-related with voice or neglect.

General Discussion

The, results of the four studies presentedin this article provide fairly consistent evi-dence concerning the impact of investment

Table 4Correlations of Investment Model VariablesWith Exit, Voice, Loyalty, and Neglect: Study 4

Variable Exit Voice Loyalty Neglect

Prior satis-faction

Investmentsize

Alternativequality

-.48**

-.27*

.54**

.56**

.59**

-.14

.49**

.38**

-.48**

-.45**

-.38**

.19

*p<.05. **p<.001.

model variables on exit, voice, loyalty, andneglect. Across all four studies, variations inprior satisfaction produced the hypothesizedeffects: Subjects who had been more satisfiedwith their relationships before problemsemerged were more likely to respond to theirdissatisfaction with voice and loyalty andwere less likely to respond with exit or ne-glect. The investment size variable produceda similar pattern of results: Among highlyinvested persons, voice (three of four studies)and loyalty (all four studies) were more prob-able responses, whereas exit (all four studies)and neglect (three of four studies) were lesslikely. Thus, individuals who were initiallyrelatively happy with or highly invested intheir relationships were more likely to enactfairly constructive responses when they be-came dissatisfied. Constructive reactions in-cluded behaviors such as discussing prob-lems, compromising, adopting an activeproblem-solving orientation, or simply wait-ing patiently for conditions to improve.Lower levels of satisfaction or investmentwere promotive of relatively destructive be-haviors such as ignoring the partner, quietly

6 In Studies 1 and 2, open-ended essay responses andstructured questionnaire items yielded equivalent find-ings, so open-ended responses were not obtained in thisstudy.

7 For all four studies, the results of statistical analyseson the investment model and response-to-dissatisfactionmeasures were consistent with standard requirements fordiscriminant and convergent validity. Further infor-mation concerning scale characteristics is available fromthe first author.

.8 Preliminary analyses revealed similar patterns of re-sults for the two types of measures of response to dis-satisfaction, so an averaged measure of each responsewas formed for the analyses.

1240 C. RUSBULT, I. ZEMBRODT, AND L. GUNN

allowing the relationship to decay, or endingthe relationship.

Results concerning the impact of varia-tions in the quality of the individual's alter-natives were inconsistent. In three of the fourstudies, higher quality alternatives encour-aged exit, and in two studies, good alterna-tives discouraged loyalty. (In Study 3, wherealternatives were operationalized as gener-alized expectations, no significant effects onresponses to dissatisfaction were obtained.)Thus, there is some evidence that better al-ternatives (at least, better concrete alterna-tives) promote exit and inhibit loyalty. How-ever, alternative quality significantly influ-enced neglect behaviors in only one studyand had no significant impact on voice re-sponses. In future research it might be fruitfulto examine the effects of a desirable concretealternative as opposed to no alternative ora definitely undesirable alternative. The pres-ent research contrasted desirable and neutralconcrete alternatives (Study 1) and desirableand undesirable generalized alternatives(Study 3). It may be that in the present stud-ies, alternative quality was not manipulated(or measured, in Studies 2 and 4) as well asit could have been or that only certain sortsof alternatives strongly encourage voice (ordiscourage neglect). Alternatively, these find-ings may suggest that alternative quality ismost strongly predictive of simple stay/leavedecisions. This conclusion seems warrantedin light of past research concerning the im-pact of alternatives on commitment. Rusbult(1980a) carried out two studies in which bet-ter alternatives significantly decreased com-mitment to maintain romantic relationships.Because exit and loyalty should be promotedby the same factors that encourage the ex-tremes of low versus high commitment, thesetwo sets of findings are consistent with oneanother: Good alternatives encourage exitand low commitment to maintain relation-ships, whereas poor alternatives promote loy-alty and high commitment. However, thisline of reasoning is clearly post hoc, and theseissues should be explored further.

Although the present studies provide gen-erally good support for the experimental pre-dictions, these findings should be viewed asa preliminary test of the proposed model.Problems with role playing and survey data

have been previously discussed, and becauseof these problems, the generalizability ofthese results may be limited. However, sig-nificant effects were obtained using a rela-tively inexperienced subject population (col-lege students) who were asked to employ in-trospection and/or retrospection. This(somewhat remarkable) fact suggests that themodel may be fruitfully applied to the studyof more long-standing involvements utilizingmore sophisticated and sensitive methodol-ogies.

The directions for future research shouldbe clear. First, the relation between ques-tionnaire reports of response to dissatisfac-tion and actual couple behaviors should beexplored. For example, do voice responseson a questionnaire "translate" into superiorproblem-solving skills among real couples?Do neglect responses translate into spouseabuse/neglect or extrarelationship affairs? Arelated issue concerns the generalizability ofthe current findings to more long-standinginvolvements (married couples or more ex-tended adult relationships). Does severity ofdissatisfaction influence individuals' respon-ses? The source of dissatisfaction in both role-played studies was a relatively serious one—the partner's drinking problem. The impactin the survey studies of severity of dissatis-faction was also not explored. Future re-search should examine the impact of degreeof dissatisfaction and should explore inter-actions of this factor with other variables,particularly alternative quality. Do responsesto dissatisfaction differ as a function of ageor relationship duration? Also, the presentresearch was designed to explore responsesto dissatisfaction from the point of view ofone member of the relationship. It would beinteresting to explore this phenomenon fromthe point of view of the dyad: What exchangeprocesses are involved, what is the impact ofone individual's response on the behavior ofhis or her partner, and so on. Finally, it seemscritical that research be designed to exploretemporal sequencing of responses to dissat-isfaction. Do persons characteristically re-spond initially with one response (perhapsloyalty) and later cycle through other alter-native behaviors (perhaps voice, followed byneglect and exit)? In this regard, Altman andTaylor's (1973) comments on the cycling of

RESPONSES TO DISSATISFACTION 1241

conflict may suggest some interesting predic-tions.

Collectively, the four studies reported inthis article provide good evidence concerningthe relation between investment model fac-tors and responses to dissatisfaction in ro-mantic relationships. The present researchdeals not only with stay/leave decisions butdescribes a more complex, comprehensiveset of possible reactions to relationship de-cline. The current model is capable of inte-grating previously unrelated research on suchdiverse behaviors as marital communicationand conflict resolution, separation and di-vorce, romantic satisfaction and commit-ment, spouse abuse and neglect, and extra-relationship sexual involvements. This ap-proach not only presents an abstract typologyof the range of possible reactions to relation-ship decline but also offers a theory concern-ing the conditions under which each responseought to be more or less probable—the in-vestment model. Because the investmentmodel emerged from the exchange traditionwithin social psychology (Kelley & Thibaut,1978; Thibaut & Kelley, 1959), it should bepossible to relate the present approach toother exchange models of developing rela-tionships (Altaian & Taylor, 1973; Levinger,1976; Levinger & Snoek, 1972). The invest-ment model of the development (cf. Rusbult,1980a) and decline (cf. Rusbult & Zembrodt,in press) of romantic involvements may thusprove to make a useful contribution to theliterature on interpersonal processes.

References

Altman, I. The communication of interpersonal atti-tudes: An ecological approach. In T. L. Huston (Ed.),Foundations of interpersonal attraction. New York:Academic Press, 1974.

Altman, I., & Taylor, D. A. Social penetration: The de-velopment of interpersonal relationships. New York:Holt, Rinehart & Winston, 1973.

Aronson, E., & Mills, J. The effect of severity of initi-ation on liking for a group. Journal of Abnormal andSocial Psychology, 1959,59, 177-181.

Becker, H. S. Notes on the concept of commitment.American Journal of Sociology, 1960, 66, 32-40.

Beir, E. G., & Sternberg, D. P. Marital communication.Journal of Communication, 1977,27, 92-97.

Bell, R. R., Turner, S., & Rosen, L. A multivariate anal-ysis of female extramarital coitus. Journal of Marriageand the Family, 1975, 37, 375-384.

Billings, A. Conflict resolution in distressed and non-

distressed married couples. Journal of Consulting andClinical Psychology, 1979, 47, 368-376.

Birchler, G. R., Weiss, R. L., & Vincent, J. P. Multi-method analysis of social reinforcement exchange be-tween maritally distressed and nondistressed spouseand stranger dyads. Journal of Personality and SocialPsychology, 1975, 31, 349-360.

Blau, P. M. Exchange and power in social life. NewYork: Wiley, 1964.

Bloom, B. L., Hodges, W. J., Caldwell, R. A., Systra, L.,& Cedrone, A. R. Marital separation: A communitysurvey. Journal of Divorce, 1977, /, 7-19.

Campbell, D. T., & Stanley, J. C. Experimental andquasi-experimental designs for research. Chicago:Rand McNally, 1963.

Cherlin, A. Work life and marital dissolution. In G.Levinger & O. C. Moles (Eds.), Divorce and separa-tion. New York: Basic Books, 1979.

Clore, G. L., & Byrne, D. A reinforcement-affect modelof attraction. In T. L. Huston (Ed.), Foundations ofinterpersonal attraction. New York: Academic Press,1974.

Cooper, J. Deception and role-playing: On telling thegood guys from the bad guys. American Psychologist,1976, 31, 605-610.

Critelli, J. W., & Dupre, K. M. Self-disclosure and ro-mantic attraction. Journal of Social Psychology, 19.78,106, 127-128.

Cutright, P. Income and family events: Marital stability.Journal of Marriage and the Family, 1971, 33, 291-306.

Darroch, R. K., & Steiner, I. D. Role-playing: An al-ternative to laboratory research? Journal of Person-ality, 1970,38, 302-311.

Farrell, D., & Rusbult, C. E. Exchange variables as pre-dictors of job satisfaction, job commitment, and turn-over: The impact of rewards, costs, alternatives, andinvestments. Organizational Behavior and HumanPerformance, 1981, 27, 78-95.

Fineberg, B. L., & Lowman, J. Affect and status dimen-sions of marital adjustment. Journal of Marriage andthe Family, 1975, 37, 155-160.

Freedman, J. L. Role-playing: Psychology by consensus.In A. G. Miller (Ed.), The social psychology of psy-chological research. New York: Free Press, 1972.

Gerard, H. B., & Mathewson, G. The effects of severityof initiation on liking for a group: A replication. Jour-nal of Experimental Social Psychology, 1966,2, 278-287.

Glass, S. P., & Wright, J. L. The relationship of extra-marital sex, length of marriage, and sex differences onmarital satisfaction and romanticism: Athanasiou'sdata reanalyzed. Journal of Marriage and the Family,1977, 39, 691-703.

Gottman, J. M., Markman, H., & Notarius, C. The ty-pography of marital conflict. Journal of Marriage andthe Family, 1978, 39, 461-477.

Gottman, J., Notarius, C., Markman, H., Bank, S., &Yoppi, B. Behavior exchange theory and marital de-cision making. Journal of Personality and Social Psy-chology, 1976, 34, 14-23.

Hawkins, J. L. Association between companionship,hostility, and marital satisfaction. Journal of Marriageand the Family, 1968, 30, 647-650.

1242 C. RUSBULT, I. ZEMBRODT, AND L. GUNN

Hill, C. T., Rubin, Z., & Peplau, L. A. Breakups beforemarriage: The end of 103 affairs. Journal of SocialIssues, 1976, 32, 147-168.

Hirschman, A. O. Exit, voice, and loyalty: Responses todecline in firms, organizations, and states. Cambridge,Mass.: Harvard University Press, 1970.

Hirschman, A. O. "Exit, voice, and loyalty": Furtherreflections and a survey of recent contributions. SocialScience Information, 1974, 13, 7-26.

Homans, G. C. Social behavior: Its elementary forms.New York: Harcourt, Brace & World, 1961.

Hunt, M. M. The affair. New York: World, 1969.Jaffe, D. T., & Kanter, R. M. Couple strains in com-

munal households: A four-factor model of the sepa-ration process. Journal of Social Issues, 1976, 32,169-191.

Kelley, H. H., & Thibaut, J. W. Interpersonal relations:A theory of interdependence. New York: Wiley, 1978.

Kidder, L. H, Selltiz, Wrightman, and Cook's researchmethods in social psychology (4th Ed.). New York:Holt, Rinehart & Winston, 1981.

Krain, M. Communication among premarital couplesat three stages of dating. Journal of Marriage and theFamily, 1975, 37, 609-618.

Levinger, G. A social psychological perspective on mar-ital dissolution. Journal of Social Issues, 1976,32,21-47.

Levinger, G. A social exchange view on the dissolutionof pair relationships. In R. L. Burgess & T. L. Huston(Eds.), Social exchange in developing relationships.New York: Academic Press, 1979.

Levinger, G., & Moles, O. C. (Eds.). Divorce and sepa-ration. New York: Basic Books, 1979.

Levinger, G., & Snoek, J. D. Attraction in relationships:A new look at interpersonal attraction. Morristown,N.J.: General Learning Press, 1972.

Murphy, D. C., & Mendelson, L. A. Communicationand adjustment in marriage: Investigating the rela-tionships. Family Process, 1973, 12, 317-326.

Murstein, B. I. Who will marry whom? New York:Springer, 1976.

Nunnally, J. C. Psychometric theory. New York:McGraw-Hill, 1967.

Ross, H. L., & Sawhill, I. V. Time of transition: Thegrowth of families headed by women. Washington,D.C.: Urban Institute, 1975.

Rubin, J. Z., & Brockner, J. Factors affecting entrap-ment in waiting situations: The Rosencrantz and

Guildenstern effect. Journal of Personality and SocialPsychology, 1975, 31, 1054-1063.

Rusbult, C. E. Commitment and satisfaction in roman-tic associations: A test of the investment model. Jour-nal of Experimental Social Psychology, 1980,76, 172-186. (a)

Rusbult, C. E. Satisfaction and commitment in friend-ships. Representative Research in Social Psychology,1980, 11, 78-95. (b)

Rusbult, C. E., & Zembrodt, I. M. Responses to dissat-isfaction in romantic involvements: A multidimen-sional scaling analysis. Journal of Experimental SocialPsychology, in press.

Saegert, S. C., Swap, W., & Zajonc, R. B. Exposure,context, and interpersonal attraction. Journal of Per-sonality and Social Psychology, 1973, 25, 234-242.

Sporakowski, M. J., & Hughston, G. A. Prescriptionsfor happy marriage: Adjustments and satisfactions ofcouples married for 50 or more years. Family Coor-dinator, 1978, 27, 321-327.

Staw, B. M. Knee-deep in the big muddy: A study ofescalating commitment to a chosen course of action.Organizational Behavior and Human Performance,1976, 16, 27-44.

Staw, B. M., & Fox, F. V. Escalation: The determinantsof commitment to a chosen course of action. HumanRelations, 1977, 30, 431-450.

Teger, A. I. Too much invested to quit. New York: Per-gamon, 1980.

Thibaut, J. W., & Kelley, H. H. The social psychologyof groups. New York: Wiley, 1959. ,

Todres, R. Runaway wives: An increasing North Amer-ican phenomenon. Family Coordinator, 1978,27, 17-21.

White, G. L. Physical attractiveness and courtship prog-ress. Journal of Personality and Social Psychology,1980, 39, 660-668.

Wicklund, R. A., Cooper, J., & Linder, D. E. Effects ofexpected effort on attitude change prior to exposure.Journal of Experimental Social Psychology, 1967, 3,416-428.

Wills, T. A., Weiss, R. J., & Patterson, G. R. A behav-ioral analysis of the determinants of marital satisfac-tion. Journal of Consulting Psychology, 1914,42, 802-811.

Received January 14, 1982Revision received May 3, 1982 •