JACK W. BREHM2

University of Minnesota

THE importance of the choice situation isreflected in the considerable amount oftheory and research on conflict. Conflict

theory has generally dealt, however, with thephenomena that lead up to the choice. Whathappens after the choice has received littleattention. The present paper is concernedwith some of the consequences of making achoice.

Previous consideration of the consequencesof choice have been limited to relatively un-specified hypotheses (1, 3) or to qualitativeanalysis (4). However, a recent theory byFestinger (2) makes possible several explicitpredictions. According to this analysis of thechoice situation, all cognitive elements (itemsof information) that favor the chosen alterna-tive are "consonant," and all cognitive ele-ments that favor the unchosen alternative are"dissonant" with the choice behavior. Further-more, other things being equal, the greater thenumber of elements favoring the unchosen al-ternative (i.e., the greater the relative attrac-tiveness of the unchosen alternative) thegreater the resulting "dissonance." When "dis-sonance" exists, the person will attempt toeliminate or reduce it. Although space limita-tions preclude further discussion of the theory,it may be said that several derivations are pos-sible concerning the consequences of making achoice. The present study was designed to testthe following:

1. Choosing between two alternatives cre-ates dissonance and a consequent pressure toreduce it. The dissonance is reduced by makingthe chosen alternative more desirable and theunchosen alternative less desirable after thechoice than they were before it.

2. The magnitude of the dissonance and the1 This paper is based on a thesis offered in partial ful-

fillment of the requirements for the Ph.D. degree at theUniversity of Minnesota. The author wishes to thankhis advisor, Dr. Leon Festinger, for his invaluable aidin the formulation and execution of the study. He alsowishes to thank the Laboratory for Research in SocialRelations, which supported the study.

1 The author is now in the Department of Psy-chology, Yale University.

consequent pressure to reduce it are greater themore closely the alternatives approach equaldesirability.

3. Exposing a person to new relevant cogni-tive elements, at least some of which are con-sonant, facilitates the reduction of dissonance.

METHODIn order to test these hypotheses, a procedure was

required in which each subject would: (a) rate each ofa variety of objects on desirability, (6) choose betweentwo of tie objects rated, and (c) rate the desirability ofeach object again. Also, the extent to which the choicealternatives approached equal desirability had to besubject to control.

These requirements were met in the following way.The 5s were asked, in the context of consumer research,to rate the desirability of each of eight manufacturedarticles. As payment for taking part in the research,each 5 was given a choice between two of the ratedarticles. After the S had made his choice, and with theobjects then out of sight, he was asked to rate eachagain. Thus a measure of change in desirability wasavailable for the chosen and unchosen alternatives, andfor articles not involved hi the choice.

Design and Procedure

Subjects and rationale. The 5s consisted of 225 femalestudents, mostly sophomores, from elementary psychol-ogy classes at the University of Minnesota. The initialinstructions were designed to convince 5 that she de-served, and would receive, payment for participating.Thus, 5 was told that: (a) the task was contract workfor several manufacturers rather than a regular psycho-logical experiment; (6) the experimenter and professorhi charge were profiting substantially from the project;(c) the 5 herself was being asked to spend several morehours on the project; (d) for participation she wouldreceive a product of one of the manufacturers.

The objects. After 5 agreed to participate, it was ex-plained that there were eight manufactured articles,each of which was to be rated as to its desirability. Eachobject was then taken out of its box, briefly described,and shown to 5.

The objects were new-looking and made by differentmanufacturers. They ranged hi retail value from about$15 to about $30. They consisted of the following: anautomatic coffee-maker, an electric sandwich grill, asilk-screen reproduction, an automatic toaster, a fluores-cent desk lamp, a book of art reproductions, a stopwatch, and a portable radio.

The rating scale. Objects were rated by marking acontinuous line on which eight identifying points werespaced equally. These points were accompanied bywritten statements from "extremely desirable" to "defi-

384

POSTDECISION CHANGES IN DESIRABILITY OP ALTERNATIVES 385

nitely not at all desirable." All eight scales appearedside by side on the same sheet.

After the eight objects had been placed on a table hifront of 5, E explained the rating procedure. It wasstressed that desirability meant the net usefulness ofthe object after one had taken into consideration notonly its attractiveness and quality but also how muchthe 5 herself needed such an article. When it was clearto 5 how to rate, she was encouraged to inspect eacharticle carefully and to take as much time as she likedin rating. The 5s spent from 5 to 20 minutes inspectingand rating. Most 5s spent about 15 minutes.

The choice. When 5 had finished rating the objects,E explained the method of payment. He said that 5 wasto get one of the eight objects she had just rated, butbecause 5s would tend to choose the more attractiveobjects and there weren't enough to go around, thechoice would be limited. In order to be fair to everyonea list of pairs of objects had been made up and each 5was given a choice between the two objects of a pairpicked at random. The E then pretended to look at aschedule to see which two objects 5 could choose be-tween. In actuality, the objects offered for choice weredetermined by the degree of dissonance to be created,as is explained below. The E then told 5 which two ob-jects she could choose between. As soon as 5 indicatedher choice, the chosen object was put back hi its box,the box was closed, securely tied with string, and putwith whatever personal belongings 5 had with her. Thisroutine was designed to convince 5 that she was gettingthe article.

The manipulation of dissonance. The two objects of-fered for 5's choice were picked in the following manner.One was always an article that had been rated fairlyhigh in desirability, i.e., at about 5,6, or 7 on the 8-pointscale (where 8 represents "extremely desirable"). Tocreate high dissonance (High Diss condition), the otherobject was always nearly as desirable as the first, i.e.,only 3^ to lj<j scale-points lower. For medium disson-ance (Med Diss) the alternative was always about 2scale-points lower, and for low dissonance (Low Diss)the alternative was always about 3 scale-points lowerin desirability.3 If an 5's ratings were such that the pre-determined condition could not be created, an alterna-tive condition was used, or, if that was not possible, the5 was discarded.

One control condition (Gift condition) was includedto determine the effect on change hi desirability ratingsof being given one of the objects without having tochoose between two. In this condition E explained thatto keep from running out of any one article, 5 wouldget one that had been randomly determined previously.As with a chosen object, the gift object was put hi itsbox, tied with string, and put with 5's personal belong-ings. The gift item was always picked to correspond hidesirability rating with the first of the choice articles,i.e., with a rating of about 5, 6, or 7 on the scale.

Providing new information. After 5's chosen or giftitem had been put with her things, E explained that fourof the manufacturers were interested in finding out what

1 Since preliminary analysis revealed no reliable dif-ferences between the Med and Low Diss conditions,they were combined and called Low Diss.

strikes people as being good or bad about their products.To accomplish this they had given samples of their prod-ucts to an independent research organization to have anobjective appraisal made of each product. The E thensaid that he wanted 5 to read the research report foreach of these four products, and when she had finishedhe would ask her what struck her as being good or badabout each, and also which comments would be good foradvertising the product. He then took four fictitiousresearch reports from a folder and handed them to 5.While 5 read, E put away the remaining seven objects.

Each "research report" was on a separate sheet andconsisted of a short paragraph of supposedly factualmaterial stating two or three good and two or three badpoints about the object. For example, the report for thegrill read as follows: "This grill is versatile, grills toast,sandwiches, hot dogs, frozen waffles, etc. Waffle platesmay easily be attached (cord and optional waffle platesare not supplied, -these require additional purchases).The grill plates may be damaged if kept heated too long(7 or 8 mm.) hi closed position. The heat indicator dialfluctuates, usually underestimating amount of heat. Theother surface is durable, easy to clean, won't rust."

For about half the 5s in both the High and Low Dissconditions, the four research reports included the choicealternatives, and for all other 5s, they did not includethe choice alternatives. In the Gift condition, the fourresearch reports always included the gift item. Thoseconditions in which the research reports included thealternatives will be referred to as Info (Information) andthe remaining will be referred to as No Info.

The second rating. After 5 had finished answeringquestions about the research reports, E said that themanufacturers were interested in finding out how evalu-ations of their products changed after a person hadlooked them over and then left the store. To do this, itwas necessary to rate each object again now that 5 hadlooked them over and they were all out of sight. The 5was asked to reconsider each item carefully and thenrate each in the same manner as the first time. To mini-mize the effects of memory for the first ratings, thesecond rating scales were given one at a time instead ofall on one sheet.

Upon completion of the second rating, the experi-ment was fully explained. Only two or three of all theparticipants showed resentment at not getting the ob-ject. With these E went into more detail about thereasons for designing such an experiment.

Assignment of Si to experimental conditions. Withinthe limits of availability, 5s were scheduled and assignedto the different conditions at random. Order of assign-ment was varied so as to cancel out differences due tochange hi the effectiveness of E with time and practice.There were 27 5s hi each of the High Diss conditions,33 in Low Diss—No Info, 30 in Low Diss—Info, and30 in the Gift condition. In addition, there were 48 whochose the object initially rated lower, and were thereforeeliminated from consideration.4

••Analysis of these data showed that their deletioncould not account for the main experimental results.Indeed, the changes in desirability were, if anything,in the direction of reducing dissonance.

386 JACK W. BREHM

RESULTS AND DISCUSSION

Scoring the Desirability Ratings

It will be remembered that each object wasrated, both before and after the choice, on ascale of desirability. These ratings were as-signed numerical values corresponding to theirlinear position on the scale, with 1.0 represent-ing "not at all desirable," and 8.0 representing"extremely desirable." The values were foundto the closest tenth of an interval.

Any change in desirability of an object couldthen be found by comparing values of the firstand second ratings. A difference between thesevalues was marked positive if it indicated anincrease in desirability, and negative if it indi-cated a decrease. However, since reduction ofdissonance may be accomplished either byraising the desirability of the chosen, or lower-ing the desirability of the unchosen object,change in dissonance was measured by the al-gebraic difference in change of ratings betweenchosen and unchosen objects.

Since ratings are less than perfectly reliable,some of the change from first to second ratingmust be attributed to regression effects, whichshould be greater in the case of initial ratingsnear the ends of the scale than for those nearthe middle. So as to correct for regression, firstand second ratings of all objects not involvedin a choice (or as a gift) were correlated sepa-rately for objects for which information wasgiven and for objects for which informationwas not given. Prediction equations were thenused to determine the expected amount ofregression for any given initial rating.5 To esti-mate the true change in desirability, the ex-pected regression was algebraically subtractedfrom the actual change in rating. If, for exam-ple, a person's rating of the chosen article didnot change, but the expected regression was adecrease of .20, then the net change in desira-

* The obtained correlation coefficients were, for ob-jects without information, .95, and for objects with, .89.The mean rating changes were .05 and — .07, respec-tively. This method of estimating regression assumesthat it is linear throughout the scale. Comparison ofactual mean regression calculated from individual scaleintervals with those obtained from the prediction equa-tions indicates a slight flattening of regression in thelower part of the scale. However, estimates of regressionfor the chosen and unchosen alternatives computedfrom individual scale intervals yield essentially the sameresults as estimates obtained from the product-momentcorrelation.

TABLE 1MEANS or INITIAL RATINGS, RATING CHANGES,

EXPECTED REGRESSION AND COHKECTEDRATING CHANGES

N

In Choice

Ini-tialrat-ing

Ratingchange

Not inChoice

Re-gres-sionCAT-557)

Cor-rectedRatingChange

No Information

Low dissonanceChosenUnchosenChange in dis-

sonancefHigh dissonance

ChosenUnchosenChange in dis-

sonance

3333

2727

5.983.54

6.195.23

.33*-.14-.47

.20-.66**- .86**

-.05.10

+ .15

-.06.00

+ .06

.38*-.24-.62*

.26-.66**-.92**

Information

Low dissonanceChosenUnchosenChange in dis-

sonanceHigh dissonance

ChosenUnchosenChange in dis-

sonanceGift condition

3030

2727

30

6.003.47

6.055.07

5.91

-.30.07

+ .37

-.04-.64**-.60*

- .40**

534)

-.41.07

+ .48

-.42-.23+ .19

-.40

.11

.00-.11

.38*-.41*-.79**

.00

* Significantly different from zero at the .05 level.•• Significantly different from zero at the .01 level,t A minus sign indicates decrease in dissonance.

bility was considered an increase of .20. Ratingchanges altered in this manner will be referredto as corrected rating changes. The measure ofprimary interest is the Corrected Change inDissonance. Both corrected and uncorrectedmean rating changes for the chosen and un-chosen objects, as well as the corrected and un-corrected Change hi Dissonance, are presentedhi Table 1. Rating changes which were foundby t test to be significantly different from zeroare starred.

Changes in Desirability Ratings

The effect of the amount of dissonance. Ac-cording to Hypothesis 1, making a choice cre-ates dissonance and a consequent pressure to

POSTDECISION CHANGES IN DESIRABILITY OF ALTERNATIVES 387

re-evaluate the alternatives in order to reducethe dissonance. Examination of CorrectedChange hi Dissonance scores in Table 1 revealsa significant reduction of dissonance in all butthe Low Diss-Info condition. It may also benoted that reduction of dissonance is accom-plished both by raising the desirability of thechosen alternative and by lowering the desira-bility of the unchosen alternative.

According to Hypothesis 2, the magnitudeof the dissonance and consequent pressure toreduce it is greater the more closely the alter-natives approach equal desirability. Thus thereduction of dissonance should be greater inthe High than in the Low Diss conditions.Data hi Table 1 further indicate that in the NoInfo condition the Corrected Change in Dis-sonance for High Diss (—.92) shows greaterreduction than that for Low Diss (—.62),though the difference is not statistically sig-nificant. In the Info condition the CorrectedChange score for High Diss (—.79) showsgreater reduction than that for Low (—.11)and is significant at the 5 per cent level of con-fidence. Thus, the data clearly support Hy-pothesis 2.

The effect of new cognitive elements. Accordingto Hypothesis 3, exposing a person to newrelevant cognitive elements, at least some ofwhich are consonant, facilitates the reductionof dissonance. The magnitude of the reductionshould therefore be greater in the Info than inthe No Info condition.

It will be noted from Table 1 that the Cor-rected Change hi Dissonance scores show nogreater reduction of dissonance hi the Info thanhi the No Info conditions. In fact, the amounthi High Diss-No Info is greater than that hiHigh Diss-Info, and similarly, that hi LowDiss-No Info is greater than that in LowDiss-Info, though neither of these differencesis significant. The expectation that there wouldbe greater reduction of dissonance hi the Infothan in the No Info condition is, therefore, notsupported.

The second expectation concerning the effectof new cognitive elements was that the differ-ence hi reduction of dissonance between Highand Low Diss conditions would be greater hithe Info than hi the No Info condition. It willbe observed hi Table 1 that the difference inCorrected Change hi Dissonance between Highand Low Diss hi the No Info condition (.30) is

not significant, while the same difference hi theInfo condition (.68) is significant at the 5 percent level. However, the second-order differ-ence for the Info and No Info conditions is notstatistically significant. Thus, the expectationthat there would be a greater difference hi re-duction of dissonance between High and LowDiss conditions in the Info condition is sup-ported only by trends hi the data.

The effect of ownership. Previous studies havefound that when a person is given an object hetends subsequently to see it as more desirable.This may be called the effect of ownership. Thepresent results may perhaps be regarded assimply a reflection of this phenomenon. TheCorrected Rating Changes for the unchosenalternative hi each experimental condition pro-vide pertinent data. It is clear that changes hidesirability of the unchosen alternative are dueto choice rather than ownership. In Table 1 itmay be seen that the Corrected RatingChanges of the unchosen object hi the LowDiss conditions are for No Info, —.24 and forInfo, .00, neither of which is a significantchange. For the High Diss conditions thesefigures are —.66 and —.41, respectively, andare statistically significant. It is therefore evi-dent that desirability ratings of the unchosenalternative not only show a decrease, but alsoreflect the difference between High and LowDiss conditions hi the pressure to reduce dis-sonance. At least this part of the present re-sults is thus not attributable to ownership.

It still might be true, however, that part orall of the gain hi desirability of the chosen alter-native is due to ownership rather than choice.To check this possibility, some 5s were simplygiven an object that corresponded hi initialrating to the chosen alternative of the choiceconditions. If the effect of ownership were toaccount for any part of the gain hi desirabilityof the chosen alternative, then this gift objectshould also increase hi desirability. The rowlabeled "Gift" hi Table 1 presents the meanchanges hi desirability ratings for this condi-tion. It is clear from the Corrected RatingChange of .00 that none of the gain hi desira-bility of the chosen alternative may be attrib-uted to the effect of ownership.

Changes in desirability and avoidance of con-flict. A plausible alternative interpretation ofthe various desirability changes so far reportedshould be considered. According to this view

388 JACK W. BSEHM

a person tends to avoid conflict as unpleasant.Thus, if equally desirable alternatives are of-fered, conflict is created, leading to attempts toreduce it. It is clear that the conflict could bereduced by changing the desirability of thealternatives in order to make them less equalin desirability. One would expect the chosenalternative to increase and the unchosen alter-native to decrease in desirability. Furthermore,these changes would be proportional to theamount of conflict, which in turn would varywith the equality of desirability of the alter-natives. Consequently, one would expectgreater changes in the High Diss conditionsthan in the Low Diss conditions. The predic-tions are so far consistent with the obtainedresults.

However, there is an interesting distinctionbetween "the consequences of avoidance ofconflict" and "reduction of dissonance." Theamount of dissonance is a direct function of theproportion of relevant elements which are dis-sonant. It follows that the amount of disso-nance is limited by the proportion of relevantelements which are common to the alterna-tives. If all relevant elements are contained byboth alternatives, no dissonance is created bya choice between them. For example, choosingbetween identical automobiles would createlittle or no dissonance. Thus it may be saidthat, other things being equal, the greater theamount of overlap of cognitive elements, theless is the resulting dissonance. But conflictarises from an inability to determine whichalternative to choose, i.e., from approximatelyequal tendencies to choose both alternatives.Increasing the similarity of the choice objectswill not necessarily increase a person's abilityto choose between them. While conflict remainshigh, dissonance created by making the choicewould be relatively low.

In the course of the experiment 30 5s wereinadvertently given a choice between relativelysimilar objects, and the data for this type ofchoice were analyzed separately.6 The choicesdesignated as having large overlap of cognitiveelements were the following: a choice betweenany two of the coffee-maker, the toaster, andthe grill; or a choice between the art book andthe silk-screen reproduction. The changes indesirability from this type of choice may then

8 The author is indebted to Dr. Festinger, who firstnoted this difference.

TABLE 2MEAN RATING CHANGES (IGNORING REGRESSION) BOB

ALTERNATIVES WITH COGNITIVE OVERLAP

No Info Info

Low DissNChosenUnchosenChange in dissonancef

High DissNChosenUnchosenChange in dissonance

8.37

1.26+ .89

7.59.00

-.59

11-.14

.79+ .93

4-.70-.25+ .45

f A minus sign indicates decrease in dissonance.

be compared with those from the regular ex-perimental choices, which had less cognitiveoverlap. But first it will be pertinent to ex-amine data relevant to the amount of conflictexperienced by persons having these two typesof choice.

All 5s were asked, after the experiment wascompleted, if there was any conflict in makingthe choice, and if so, how much. Their re-sponses were categorized by the experimenteras "none," "little," "moderate," or "high."7

To obtain category frequencies large enoughfor a chi-square test, the categories were re-duced to two: "none" and "some" reportedconflict. It was then found that of all subjectswho had a high dissonance choice, 63 per centof those choosing between alternatives withoutcognitive overlap and 88 per cent of thosechoosing between alternatives with cognitiveoverlap, reported "some" conflict. This differ-ence is significant at the 6 per cent level by chisquare. Of those who had a low dissonancechoice, 47 per cent of those choosing betweenalternatives without overlap, and 25 per centof those choosing between alternatives withoverlap, report "some" conflict. The latterrelationship is in the opposite direction to thefirst but is not statistically significant. It maybe concluded that those who choose betweennearly equally desirable objects with largeoverlap of cognitive elements experience moreconflict than those choosing between dissimilar

7 The classification by E, who knew whether or notthe alternatives were nearly equal in desirability, maywell be biased in respect to the High versus Low Dissmanipulation. However, E did not expect to separateout choices involving cognitive .overlap so there is noreason to suspect a bias in regard to this variable.

POSTDECISION CHANGES IN DESIRABILITY OF ALTERNATIVES 389

objects. If the changes in desirability in thisexperiment are due to avoidance of conflict,then, one would expect greater changes wherethe alternatives are similar. On the other hand,if the changes are due to a pressure to reducedissonance, one would expect them to be lesswhere the alternatives are similar.

In Table 2 may be found the uncorrectedmean changes in desirability ratings for thechosen and unchosen alternatives which hadcognitive overlap. These may be comparedwith the uncorrected rating changes in Table 1.Corrections for regression are not necessarysince the important comparisons are betweenthe two High Diss conditions and between thetwo Low Diss conditions. It will be seen thatthe changes in the direction of reducing disso-nance (or avoiding conflict) are small or non-existent. For all those with a high dissonancechoice, 27 per cent of those choosing betweensimilar and 59 per cent of those choosingbetween dissimilar alternatives show ratingchanges in the direction of reducing dissonanceor avoiding conflict. This difference is signifi-cant, by an exact test, at the 11 per cent level.There is thus some evidence that choices be-tween alternatives with overlapping cognitiveelements create less tendency to change thedesirability of the alternatives in the expecteddirection. Since it has already been seen thatsuch choices are accompanied by more, ratherthan less, conflict, it appears that these changesin desirability reflect reduction of dissonancerather than avoidance of conflict.

SUMMARY AND CONCLUSIONS

The present experiment was designed toexamine some of the consequences of making adecision. Specific predictions about the conse-quences were based on a theory by Festinger.According to this theory, when a personchooses one of two alternatives, all of the itemsof information which favor the unchosen alter-

native will be dissonant. Thus a state of dis-sonance and pressure to reduce it are created.Examination of the possible ways in which dis-sonance may be reduced was limited in thepresent study to tendencies to re-evaluate thechoice alternatives after the decision.

Female 5s were asked to rate each of eightarticles on desirability, choose between two ofthem, and then rate each of the articles again.In addition, some 5s were exposed to a mixtureof good and bad information about the choicealternatives after the choice was made.

The results supported the prediction that-choosing between alternatives would createdissonance and attempts to reduce it by mak-ing the chosen alternative more desirable andthe unchosen alternative less desirable. A sec-ond prediction, that dissonance and conse-quent attempts to reduce it would be greaterthe more nearly the choice alternatives ap-proached equality, also received support. Thethird prediction, that exposure to new informa-tion containing at least some consonant ele-ments would facilitate reducing dissonance,did not receive clear support. A control condi-tion ruled out the possibility that the obtainedincrease in desirability of the chosen alterna-tive was due to ownership. Finally, some of thedata consistent with "dissonance theory" werefound not to be consistent with traditional"conflict theory."

REFERENCES

1. ADAMS, D. K. Conflict and integration. /. Pers.,1954, 22, 548-556.

2. FESTINGER, L. The relation between cognition andaction. Paper read at Symposium on Cognidon,Boulder, Colo., May, 1955.

3. LEWIN, K. Field theory in social science. New York:Harper, 1951.

4. MARTIN, A. H. An experimental study of the factorsand types of voluntary choice. Arch. Psychol.,1922, 22, No. 51.

Received July 15,1955.