diversification bias: explaining the discrepancy in …diversification bias 35 deed, large quantity...

Journal of Experimental Psychology: Applied1995, Vol. I, No. 1,34-49

Copyright 1995 by the American Psychological Association, Inc.I076-898X/95/S3.00

Diversification Bias: Explaining the Discrepancy in VarietySeeking Between Combined and Separated Choices

Daniel Read and George LoewensteinCarnegie Mellon University

Recent research has revealed a pattern of choice characterized as adiversification bias: If people make combined choices of quantities of goods forfuture consumption, they choose more variety than if they make separatechoices immediately preceding consumption. This phenomenon is explored ina series of experiments in which the researchers first eliminated severalhypotheses that held that the discrepancy between combined and separatechoice can be explained by traditional accounts of utility maximization. On thebasis of results of further experiments, it was concluded that the diversificationbias is largely attributable to 2 mechanisms: time contraction, which is thetendency to compress time intervals and treat long intervals as if they wereshort, and choice bracketing, which is the tendency to treat choices that areframed together differently from those that are framed apart. The researchersdescribe how the findings can be applied in the domains of marketing andconsumer education.

Most normative models of choice imply that it isbetter to make choices in combination rather thanseparately. For example, at dinner tonight weshould choose the wine with an eye to the entree,and before going out we should select a matchingtie and jacket. Indeed, we may want to combine allfour choices. The restaurant we select may influ-ence the costume we choose and vice versa. Theadvantages of combined choice stem from comple-mentarity and substitutability between goods. Onlywhen choices are taken together can such interac-tions be accommodated optimally.

Although the view that combined choices arebest is normatively appealing, there appears to be

Daniel Read, Department of Engineering and PublicPolicy, Carnegie Mellon University; George Loewen-stein, Department of Social and Decision Sciences,Carnegie Mellon University.

We thank Itamar Simonson, Shane Frederick, KarenJenni, Jerry Joseph, Steve Klepper, Elizabeth Mannix,Sally Sleeper, and Lowell Taylor for their assistance.

Correspondence concerning this article should beaddressed to Daniel Read, who is now at the Depart-ment of Advertising, University of Illinois at Urbana-Champaign, 103 Gregory Hall, 810 South Wright Street,Urbana, Illinois 61801. Electronic mail may be sent viaInternet to [email protected].

at least one situation in which combined choicesactually leave consumers worse off. This occurswhen consumers select many goods of the samekind to meet future consumption needs. Recentresearch by Simonson (1990) has shown that ifconsumers combine their purchases in this way,they will choose more diverse bundles (or showmore variety seeking) than if they choose themindividually when the goods are to be consumed.In one experiment, Simonson's subjects chosethree snacks for consumption on 3 specific days.Some scheduled the snacks in advance (simulta-neous choice), while others chose one on each day(sequential choice).1 Subjects showed much morevariety seeking when making simultaneous choices(64% chose three different items), than whenmaking sequential choices (9%). The same patternwas also observed in a subsequent study (Simonson& Winer, 1992) examining actual family purchasesof yogurt. Within the same family, as purchasequantity increased so did the variety chosen; in-

1 We used the terms simultaneous and sequentialchoice to refer to choices made using experimentalarrangements similar to Simonson's. When speaking inmore general terms or when referring to differentexperimental arrangements, we use the terms combinedAnd separated choice.

34

DIVERSIFICATION BIAS 35

deed, large quantity purchases often containedunusual flavors that were never bought separately.It appears that consumers plan much more diver-sity for themselves than they will subsequentlywant. We refer to this as the diversification bias.

Our main goal is to explain the underlying causeof the discrepancy in variety seeking betweensimultaneous and sequential choice. We are par-ticularly concerned with whether the discrepancycan be justified normatively or whether it is, as ourtitle correctly implies that we conclude, a type ofchoice bias.

Classical Model of Variety Seeking

To provide a context for our work, we begin witha discussion of the classical economic model ofvariety seeking, drawing attention to those ele-ments of the model that we feature in our subse-quent analysis. According to the classical model,variety seeking arises from object-specific satiationor the diminishing rate of marginal return to consump-tion. According to this view, the optimal bundle ofgoods contains variety because the benefit from anadditional unit of a specific good (i.e., its marginalreturn) decreases as a function of the number ofunits of that good one already possesses.

Consider the case of a consumer choosing be-tween oranges and candy bars. Figure 1 shows a

Cuxtybro 4

family of indifference curves for such a consumer,depicting preferences for bundles of the two goods.Indifference curves are isoquants, depicting combi-nations of goods having identical utility to theconsumer, with higher curves depicting greaterutility. In most cases, indifference curves are con-vex toward the origin, indicating that as the rela-tive amount of one good in a bundle increases, theamount of the other good needed to compensatefor some loss of that good declines. If oranges areinitially preferred to candy bars, as is the case forthe consumer whose preferences are depicted inFigure 1, an orange will be the first item chosen.However, there will then be less benefit to herfrom a second orange than from the first candy bar;if she is at the point (1,0), with one orange and nocandy bars, then choosing a candy bar will put heron a higher indifference curve, point (1, 1), thanselecting a second orange (2,0).

Applied to the problem of variety seeking, theindifference curve analysis becomes somewhatmore complicated because consumption typicallyoccurs over time, and satiation for a particulargood is likely to dimmish as the interconsumptioninterval (the time interval between acts of consump-tion) increases. In our example, if you have justeaten an orange it might well make an immediatesecond orange less appealing, but it will probablyhave little effect on the pleasure of an orangeeaten tomorrow. Indeed, if the interconsumptioninterval is sufficiently large, the same benefit willbe obtained from consuming a second unit of anitem as from the first. Such a case, in whichconsumption of one unit has no effect on themarginal benefit of a second, is represented bystraight line indifference curves. Thus, as in Figure2, the effect of the interconsumption interval canbe described in terms of changes in the convexityof the indifference curve, with separation in timeleading to reduced convexity.2

A common functional form for indifferencecurves is given by the constant elasticity of substitu-tion (CES) formula:

k = (1)

Oranges

Figure 1. Indifference curves for two goods, orangesand candy bars, with oranges preferred to candy barsand choices to be made simultaneously.

2 We assume that tastes do not change over time andthat temporal discounting is constant for both goods.Although taste change can complicate the issue, in thiscontext its primary effect would be to add noise.

36 DANIEL READ AND GEORGE LOEWENSTEIN

Cudybon 1 -

Orates

Figure 2. Indifference curves for oranges and candybars with varying convexity. Note that convexity de-creases as the interconsumption interval increases.

where k is the overall utility level of a particularindifference curve, r (0 < r < 1) describes theconvexity of the indifference curve (a measure ofthe degree of object-specific satiation), a\ and a2

are the initial relative preferences for the twoitems, and q\ and q2 are the quantities consumed ofeach good.

If convexity decreases exponentially with theinterconsumption interval, as seems plausible, thenwe can express r as a function of time (/) as follows:

(2)

where r0 is the parameter applicable to simulta-neous consumption, and 6 is an exponential decayparameter (0 > 0). Increasing the interconsump-tion interval will increase r, thereby flattening theindifference curve. When the interconsumptioninterval is long r approaches 1 and k, the utility of abundle of two goods, approaches (a^q\ + a^2)-The utility of the bundle is then maximized bychoosing a bundle consisting exclusively of thegood having the highest initial preference.

If the discrepancy in variety seeking betweensequential and simultaneous choice reflects amisprediction of tastes, it might stem from atendency to overpredict satiation: although eatingthree candy bars of the same kind may seem quiteunappealing in prospect, it may not be bad at all ifthey are consumed on different days. In terms of

the indifference curve analysis, this suggests thatsimultaneous choices are based on a more convexindifference curve (i.e., one having a lower value ofr) than are sequential choices.

Explanations for the Diversification Bias

Explanations for the discrepancy between choicemodes in amount of variety seeking can be dividedaccording to whether the underlying decision pro-cesses are thought to be unbiased or biased.Unbiased decision processes, enabling an indi-vidual to maximize net utility from consumption bychoosing more variety during simultaneous thansequential choice, are compatible with traditionalmodels of rationality or utility maximization. Wediscuss three hypotheses maintaining that theobserved discrepancy in variety seeking betweensequential and simultaneous choice are due tosuch unbiased processes. Alternatively, biases arisewhen the decision processes adopted lead to sys-tematic deviations from optimality. We presenttwo hypotheses holding that response format ef-fects on diversification reflect biased decision pro-cesses.

Hypotheses That Do Not Imply Bias

Diversification. Simonson (1990) and Kahn andLehmann (1991) suggested that people seek vari-ety because they are risk averse and uncertainabout their preferences.3 Choosing variety reducesthe likelihood of repeatedly consuming somethingundesirable. With reference to the classical model,this implies that consumers are uncertain about

3 Simonson (1990) mentioned an additional hypoth-esis to account for the discrepancy in variety seekingbetween simultaneous and sequential choice. Accordingto the time-and-effort hypothesis, people choose morevariety during simultaneous choice because it is bothquicker and less effortful to choose several differentitems than to choose several items of the same kind. Intwo experiments we found that, contrary to this hypoth-esis, participants took less time to select a certainnumber of items when they were told that all had to beidentical (four of the same type of candy bars) thanwhen they were not so constrained (they could choosewhichever four candy bars they liked). Given the lack offace plausibility or empirical support for the time-and-effort hypothesis, we conserve space by not presentingthe details of these experiments.


the utility (k) of each bundle of goods. Therefore,they choose a bundle that is unlikely to have thelowest utility, although at the cost of virtuallyeliminating the bundle that has the highest utility.Diversification is an unbiased decision rule be-cause the expected utility of such low-risk assort-ments can be greater than that of riskier bundlescontaining only one kind of item (even if it is theone they expect to like best). On this view, diversi-fication is a response to uncertainty about prefer-ences for goods. Consequently, the diversificationhypothesis predicts that variety seeking shoulddecline if such uncertainty is reduced.

Information seeking. People may also seek vari-ety because it provides information that can helpto inform future choices. If we always consume thesame kind of item, we may miss out on betteralternatives. Information seeking may have asmaller influence on sequential choices becausemany people may find they like the first good theychoose sufficiently to decide that it is not worthexchanging the certain pleasure from that good fora potentially less enjoyable good along with someextra information. The information-seeking ac-count implies that consumers making simulta-neous choices will choose less variety if they havealready obtained, or believe that they will obtain,all the information that variety seeking could yield.

Cognitive capacity limitations. Both the diversi-fication and information-seeking hypotheses sug-gest that people seek out variety in simultaneouschoice for some normatively justifiable reason, butchange their minds when they discover how muchthey like their first selection. An alternative ac-count that does not imply bias is that peoplechoose less variety in sequential choice becausethey are unable, due to cognitive capacity limita-tions, to implement their desire for variety. It iswell known that our ability to cope with complexproblems is sharply limited by our capacity forprocessing information (Baddeley, 1976; Kahne-man, 1973; Miller, 1957). Indeed, capacity limita-tions may be one of the primary reasons whypeople adopt suboptimal decision rules or heuris-tics (Simon, 1957); these heuristics do the best jobpossible given people's capacity constraints. Thelimitations that are most relevant for variety seek-ing are those on memory: To cany out a plannedsequence of consumption, one must be able toremember the plan.

Unlike the next two hypotheses, the cognitivecapacity limitation hypothesis views the sequentialchoice condition as the source of the observeddiscrepancy between simultaneous and sequentialchoice. It posits that variety seeking in the sequen-tial choice condition is attenuated by cognitivelimitations and predicts that if these limitations arereduced, variety seeking during sequential choicewill increase. Support for this hypothesis would notchallenge the normative view that decisions arebest made simultaneously. Instead it would implythat such simultaneous choices, although they maybe optimal, are difficult to implement.

Hypotheses That Imply Bias

Time contraction. This hypothesis proposes thatpeople subjectively shrink the interconsumptioninterval when making simultaneous choices, thusexaggerating the impact of satiation on their pref-erences.

We know well that we can become satiated oneven the most desirable foods, so that if we are toconsume several dishes within a short interval weshould choose a variety. Yet satiation is fleeting,and our preferences typically return to their precon-sumption level within a short time. The timecontraction hypothesis holds that we do not givesufficient weight to the interconsumption interval,treating even lengthy intervals as if they were verybrief. An experiment reported by Kahneman andSnell (1992) may have demonstrated such a bias.At a specified time on each of 7 consecutive days,their subjects ate a bowl of yogurt while listening tothe same piece of music. On the 1st day, theystated how much they liked both the yogurt andthe music and then predicted how much theywould like them on the 7th day. Although theypredicted that they would like both the music andthe yogurt much less, there was no consistentpattern of decline. In fact, they liked the yogurtmore on the 7th day than on the 1st, and their dropin liking for the music was smaller than they hadexpected. Perhaps these subjects imagined a muchsmaller interconsumption interval than was speci-fied and therefore overpredicted their degree ofsatiation. Other work by Kahneman and coauthorson retrospective evaluations of sequences of plea-sure and pain has uncovered a systematic patternof duration neglect, in which retrospective ratings of


the overall utility of sequences of pleasure andpain are insensitive to the durations of thosesequences (Kahneman, Frederickson, Schreiber,& Redelmeier, 1993; Frederickson & Kahneman,1993). Time contraction, if demonstrated, could beviewed as an analogous phenomenon associatedwith prospective judgments.

Since the time contraction hypothesis is basedon the assumption that people do not consider howthe size of the interconsumption interval willinfluence satiation, it predicts that the level ofvariety seeking will be reduced if a large intercon-sumption interval is made salient. This predictionis tested in several experiments reported below.

Choice bracketing. This hypothesis holds thatthe discrepancy in variety seeking occurs becausesimultaneous choices are presented together andare thus framed as a type of portfolio choice,whereas sequential choices are considered inisolation.

There is considerable evidence that subjectstend to frame problems according to how theproblems are presented to them (Tversky & Kahne-man, 1981). Simultaneous choices are presented tosubjects in the form of a package, and perhaps themost straightforward choice heuristic applicable tosuch packages is diversification. In the sequentialchoice condition, in contrast, subjects are pre-sented with the choices one at a time, and thenatural choice heuristic applicable to a singlechoice is to choose the single most preferredalternative. A similar mechanism was alluded to bySimonson, who noted that "in addition to simplify-ing decisions, making multiple choices simulta-neously enables consumers to implement, rathereasily, global strategies for the different selec-tions" (Simonson, 1990, p. 160).

Experiments

Experiment 1 was designed primarily to deter-mine whether the discrepancy in variety seekingbetween simultaneous and sequential choice couldbe attributed to unbiased decision processes. Aswill be shown, none of these hypotheses wassupported. In subsequent studies, we investigatedthe validity of the time contraction and choice-bracketing hypotheses, concluding that these pro-cesses are major determinants of the diversifica-tion bias.

Experiment 1

Overview

Participants made sequential and simultaneouschoices of three snacks, just as in Simonson's(1990) original study, with the addition of severalmanipulations. The experiment constitutes a repli-cation of Simonson's original study, tests of severalof the hypotheses described above, and an investi-gation of whether people are aware of the discrep-ancy between choice formats.

Tests of the causes of variety seeking. The diver-sification hypothesis suggests that variety seekingis analogous to holding a diverse investment portfo-lio and that it will be reduced if uncertainty aboutthe goods is reduced. To reduce uncertainty abouttheir preferences, we gave some simultaneouschoice subjects a pretaste of all the snacks beforethey made their choices. If the diversificationhypothesis is correct, less variety seeking shouldoccur in this pretaste condition.

The information-seeking hypothesis holds thatpeople seek variety to discover what the world hasto offer and thereby to update their preferences. Itimplies that if information is made availablethrough some extrachoice mechanism, there willbe less motivation to seek variety during choice. Totest this prediction, a group of participants (post-taste group) was given samples of all the snacksoffered immediately after making simultaneouschoices. Since subjects in this condition obtain atthe outset all the information possible from varietyseeking, the information-seeking hypothesis pre-dicts that variety seeking will be reduced in thiscondition, as well as in the pretaste condition inwhich they receive samples before choosing.

The cognitive capacity limitation hypothesis holdsthat the choice conditions differ primarily in theease of intertransactional thinking. That is, al-though sequential choice subjects would like tochoose the same amount of variety as simultaneouschoice subjects, they are unable to act on thisdesire because they can neither remember whatthey have already chosen nor anticipate what theywill choose. To test this hypothesis, we included asequential choice condition in which participantswere told at the outset that they would be makingthree weekly choices and were also reminded ateach session of their previous choices. If varietyseeking is hindered by capacity limitations, then


this aware sequential choice group should showmore variety seeking than an unaware group.

The time contraction hypothesis holds thatpeople psychologically underestimate the length ofinterconsumption intervals. Concerns about satia-tion, which are a reasonable basis for varietyseeking when interconsumption intervals are short,then justify high variety seeking. We hypothesizedthat the interconsumption interval is not verysalient to consumers making simultaneous choices,but that if this salience were increased, varietyseeking would decrease. To increase this salience,we included an additional sequential choice condi-tion in which long-interval choices were contrastedwith short-interval choices. Based on the timecontraction hypothesis, we predicted that peoplewould contrast the two interconsumption intervals,thereby becoming more aware of the low level ofsatiation that would remain after 1 week.

The time contraction hypothesis also makes aprediction about the order in which goods arechosen. McAlister (1982) and Pessemier and Han-delsman (1984) argued that variety seeking duringsequential choice is partly due to increasing satia-tion for familiar goods. Satiation is reflected in therate at which people change from one good toanother in a series. Even if one chooses only twodifferent goods in a long series, satiation for thosegoods could be reflected in a pattern of alternatingconsumption (i.e., anABABAB pattern), but not intwo short series of uniform consumption (e.g., thepattern AAABBB). Although Pessemier and Han-delsman were concerned with actual satiety, thetime contraction hypothesis is based on imaginedsatiety. When people simultaneously schedule aseries of consumption occasions, they imagine howthey would feel at each time and, if they believethey will be satiated for their favorite good, choosean alternative. Imagine that someone's favoritegood is A and second favorite is B. At Time 1 theywill choose A. However, if they imagine that theywill still be satiated for ,4 at Time 2 they will makeB their second choice. At Time 3, they will imagineless satiation for A, and so will be more likely tochoose it again. Consequently, the time contrac-tion hypothesis predicts that simultaneous choicesubjects who choose two items of one kind and oneof another will likely schedule them in the orderABA. In the sequential choice condition, this willnot happen. Since there will be no actual satiationat Time 2 for goods eaten at Time 1, those who

choose two of one kind will have no preference forthe order ABA over any other order. In short,amongst those subjects who choose two differentitems, the time contraction hypothesis predictsthat simultaneous choice subjects will be morelikely to schedule them in the order ABA than willsequential choice subjects.

Do simultaneous choice subjects change theirminds? A central prediction of the view thatconsumers are better off making sequential choicesis that those making simultaneous choices will laterwant to change their choices so as to have lessvariety than they had planned. We tested thisprediction in the following way: During the secondand third session, we asked the simultaneouschoice subjects if they would want to change thesnack planned for that day and, if so, what theypreferred. The participants were not allowed toactually change what they received.

Awareness. Are simultaneous choice subjectsaware that they would probably choose less varietyif they chose sequentially? Simonson (1990) com-pared one group of subjects who predicted thechoices that they would make during sequentialchoice (e.g., "What will you choose next week?")with a standard simultaneous choice group. Hefound that subjects predicted less variety seekingduring sequential choice than actually occurredduring simultaneous choice. This suggests thatsubjects were at least partially aware of the amountof variety they would want at the moment ofconsumption. Simonson did not compare his pre-diction group to a standard sequential choicecondition, leaving some doubt about the degree ofthis awareness. To test for awareness, we includeda sequential choice condition in which subjectsduring the first session predicted what they weregoing to choose in subsequent sessions.

Method

Participants. Three hundred seventy-eight stu-dents in 16 undergraduate classes in economicsand history participated in the experiment. Toapproximately equate the number of participantsin each condition, two classes each were assignedto five experimental conditions and three each tothe remaining conditions. Because the classes metweekly, the length of the interconsumption inter-val was 1 week.


Materials. To reproduce the conditions of Si-monson's (1990) original study as closely as pos-sible, we offered our participants similar choices:Snickers bars, Oreo cookies, milk chocolate withalmonds, tortilla chips, peanuts, and cheese-peanut butter crackers. For the conditions requir-ing samples, we wrapped small portions of eachitem in a strip of foil.

Conditions. Depending on the group to whichthey were assigned, each participant selected threesnacks in one of the following experimental condi-tions:

1. Standard sequential choice. Participants choseone item on each of the 3 weeks. They were nottold that there would be more than one choice.

2. Sequential choice with awareness. Partici-pants knew that they would be making threechoices on 3 different weeks, and all the choiceswere made on the same form, so that they hadreliable knowledge about previous choices.

3. Prediction. Participants made their first choiceand then predicted what they would choose onsubsequent weeks. They were informed that theirpredictions were nonbinding. The second andthird sessions were conducted in the same way asthe standard sequential choice condition. Specifi-cally, they were not reminded of their predictions.

4. Standard simultaneous choice. In the first ofthe three classes, participants chose a snack forthat day and for the two subsequent classes.

5. Simultaneous choice with pretaste. This condi-tion was identical to standard simultaneous choiceexcept that participants tasted a small sample ofeach snack before making their three choices.

6. Simultaneous choice with posttaste. Partici-pants were informed that immediately after mak-ing their choices, they would receive a sample of allthe snacks.

7. Simultaneous choice preceded by hypotheti-cal immediate consumption. Participants were firstinstructed to imagine that they were receiving allthree choices immediately. They made hypotheti-cal choices under these conditions and were theninformed that, in fact, their selections would begiven to them during three classes spaced at least 1week apart. They then made their actual choices inthe same manner as the standard simultaneouschoice conditions.

During the second and third classes in all thesimultaneous choice conditions, we asked partici-pants whether, if given the opportunity, they would

prefer a different snack than the one they hadpreviously chosen and, if so, which one theywanted. We emphasized that they could not actu-ally change from their initial choice and that theywould receive the snack they had originally se-lected for consumption on that date.

Procedure. The experiment was conducted inthe closing minutes of each class session. Duringthe first session, the experimenter first placed oneof each snack on the instructor's desk, allowing thestudents to see what they could get. Responsesheets containing instructions were distributed, aswell as foil-wrapped samples to those participantswho were to receive a pretaste or a posttaste. Allforms contained a list of available snacks, as well asinstructions specific to each condition. Once partici-pants had completed the forms, they returnedthem to the experimenter, who gave them thesnack they had chosen for that day.

During the second and third classes, the experi-menter distributed the forms and the participantsresponded and were given the snacks of theirchoice. The six snacks were not displayed again.

Results

Table 1 reports the mean level of variety seekingin all conditions, along with the number of peoplechoosing different numbers of items. The level ofvariety seeking was coded as 1 if the participantchose the same snack three times, 2 if the partici-pant chose one snack twice and another a singletime, and 3 if the participant chose three differentsnacks. All of the analyses we report include only

Table 1Variety Seeking in All Conditions of Experiment 1

Number of differentkinds chosen

One Two Three

Group % n % M

Sequential choiceSequential-awarePredictionSimultaneous choice 18Simultaneous-pretaste 17Simultaneous-post-

taste 20Simultaneous after

immediate 17

46 19 46 19 8 3 1.6150 25 34 17 16 8 1.6638 18 31 15 31 15 1.94

10 38 21 45 25 2.278 38 18 46 22 2.29

9 45 20 34 15 2.14

9 31 16 52 27 2.35


participants for whom we were able to calculatethe level of variety seeking—the simultaneouschoice participants and the sequential choice par-ticipants who attended all three classes. An overallanalysis of variance (ANOVA) comparing thelevel of variety seeking in all experimental condi-tions confirmed significant differences betweensome of them, F(6, 332) = 7.98, p < .0001,justifying the use of planned comparisons. Unlessotherwise stated, analyses compare the mean levelof variety seeking between conditions.

We strongly replicated Simonson's (1990) majorfinding: there was more variety seeking in thestandard sequential choice condition than in thestandard simultaneous choice condition, f(87) =4.6, p < .001. We next turn to an examination ofthe specific hypotheses.

Unbiased hypotheses. As is evident in Table 1,the simultaneous choice conditions did not differamong themselves in level of variety seeking, F(3,196) < 1. Two of the hypotheses proposing thatthe discrepancy between sequential and simulta-neous choice was attributable to unbiased pro-cesses were dependent on such differences; there-fore these hypotheses were unsupported. Accordingto the diversification hypothesis, variety seekingshould be lower in the pretaste condition, whereasthe information seeking hypothesis predicts that itshould be lower in both the pretaste and posttasteconditions. Note that although the mean varietyseeking in the posttaste condition is nonsignifi-cantly lower than the other three simultaneouschoice conditions, this cannot be construed assupport for the information seeking hypothesis,which, as already noted, also predicts less varietyseeking following a pretaste.

The cognitive capacity limitation hypothesis pre-dicted that informing people that they wouldreceive one choice per week and reminding themof their previous choices would increase theirvariety seeking. This did not occur, as can be seenby comparing the standard and aware sequentialchoice conditions, f(89) < 1.

Time contraction. We made two predictionsbased on the time contraction hypothesis. Thefirst, that variety seeking would be lower if simulta-neous choice for delayed consumption followedchoice for immediate consumption, was not sup-ported. The second prediction, which was sup-ported, was that subjects in simultaneous choiceconditions who chose two items of one kind would

Table 2Proportion of ABA Judgments Made by ThoseChoosing Two of the Same Kind in All Conditionsof Experiment 1

Percentage Numberchoosing choosing

Group ABA two

Sequential choiceSequential-awarePredictionSimultaneous choiceSimultaneous-pretasteSimultaneous-posttasteSimultaneous after

immediate

213547676755

75

191715211820

16

commonly choose them in the order ABA rathertii&nAAB or BAA. Table 2 gives the proportion ofABA choices in all conditions. As predicted by timecontraction, the proportion of such choices ismuch higher in the combined simultaneous choiceconditions (65.3%) than in the sequential choicecondition (27.8%), x2(l) = 13.8,p < .001.

In hindsight, we suspect that the design of thesimultaneous choice preceded by hypothetical im-mediate choice condition was flawed because itintroduced another well-established and robusteffect that operated opposite to the predicteddirection: anchoring and insufficient adjustment.Participants' responses to the hypothetical immedi-ate consumption question revealed an extremelyhigh rate of variety seeking (M = 2.6).4 If partici-pants first anchored on their own initially highlevel of variety seeking and then failed to adjustsufficiently for the introduction of an interconsump-tion interval, the postadjustment level of varietyseeking might be higher than that of the conven-tional simultaneous choice conditions. This mighthave occurred even if the immediate/week delaymanipulation increased the salience of the inter-

4 The fact that the hypothetical immediate consump-tion question elicited a significantly greater level ofvariety seeking than the other simultaneous choiceconditions, f(51) = 4.6, p < .001, indicates that therewas some limit to time contraction (i.e., that subjects inthe simultaneous choice conditions did not completelyignore the interconsumption interval). If they had, thelevel of variety seeking in simultaneous choice wouldhave been indistinguishable from the level of varietyseeking associated with immediate consumption.


consumption interval and time contraction wasoperative. On the basis of the support implied byABA patterns and our belated recognition of thisconfound, we conducted further tests of timecontraction, which are reported below (see Experi-ments 2 and 3).

Awareness. Apparently participants were atleast partially aware of their preference for morevariety during simultaneous choice, because whenasked to predict what they would do, they antici-pated choosing significantly less variety than wasactually chosen by simultaneous choice subjects,/(246) = 2.67,p < .01, and they also predicted thatthey would want more variety than was chosen bysubjects in the sequential choice conditions,r(137) = 2.26, p < .05. Interestingly, for theprediction subjects, there was no difference be-tween the degree of predicted and actual varietyseeking (1.94 for actual choice, 1.90 for predictedchoice), t(28) < 1, suggesting that their predic-tions served as a plan which they then carried out.

Do simultaneous choice subjects change theirminds? Simultaneous choice subjects statedwhether they wanted something different thanthey had originally chosen and, if so, what. For the84 subjects providing us with complete data in thiscondition, we compared the amount of varietyoriginally chosen with their preferences at the timethey received the snacks. The mean variety of theirrevised choices (1.8) was significantly lower thanthat of their original choices (2.09), /(83) = 4.55,p < .001. Of the 37 people who changed theirminds at least once, fully 31 (84%) of themchanged in the direction of wanting less variety.Note that with six different types of candies,random changes in the snacks selected would befar more likely to move subjects in the direction ofwanting more variety than toward wanting less.Thus, this is a conservative measure of their desireto change to less variety. It is also conservativebecause participants were not actually given theoption of changing their minds, so they had littlemotive to acknowledge that their original choicewas "wrong."

Discussion

Simonson's finding that more variety is soughtduring simultaneous than sequential choice wasreplicated and proved quite robust. None of themany manipulations designed to remove factors

that might normatively justify this discrepancy hadany effect. Moreover, two results suggested thatsimultaneous choices are actually biased. First, thesimultaneous choices were made in the order ABAmuch more frequently than sequential choices.This suggests that subjects believed that theywould experience greater satiation than actuallyoccurred. Second, simultaneous choice subjectswho indicated that they would want to change theiroriginal choices were likely to change in thedirection of less variety rather than more. Again,this suggests that people believed that they wouldgrow tired of their favorite item more rapidly thanthey actually did.

We have already speculated that one mecha-nism underlying this overprediction of satiation istime contraction, or the tendency to psychologi-cally shrink interconsumption intervals. Time con-traction could be due to the low salience of theinterconsumption interval to subjects who attendonly to the order in which they will consume goodsor to subjects' use of a choice process in which theyfirst anchor on what they would choose if they weregoing to consume immediately and then adjustinsufficiently for the duration of the interconsump-tion interval. Both of these mechanisms imply thatincreasing the salience of the interconsumptioninterval will decrease the degree of variety seekingin simultaneous choice, which is what we tested inExperiments 2 and 3.

Experiments 2 and 3: Time Contraction

The classical model of variety seeking, discussedabove, holds that variety seeking is a function ofdiminishing marginal returns for additional goodsof the same type. However, simultaneous choice isdetermined by imagined rates of marginal return.Time contraction can be interpreted as a factorthat influences these imagined marginal rates ofreturn, thereby altering desire for variety. Timecontraction is the tendency for people to contracttime psychologically when making choices aboutfuture activities and therefore to overpredict satia-tion. If people contract time, they will choose lessvariety when making single choices at the time ofconsumption than when choosing concurrently forthe same consumption occasions.

In the equation for the CES indifference func-tion, time contraction can be modeled by transform-ing the expression for r (Equation 2) into r+ by


incorporating an additional parameter, X, thateffectively shrinks the interconsumption interval:

(3)

Time contraction, X, can vary between 0 and 1. ForX = 0 (complete time contraction) the individualtreats consumption as simultaneous regardless ofthe interconsumption interval. For X = 1 (no timecontraction) the individual accurately responds tothe impact of satiation, given a particular intercon-sumption interval. Time contraction might reflectan anchoring and insufficient adjustment processwhereby people use the convexity factor for imme-diate consumption (r0) as the anchor, but insuffi-ciently adjust for the effect of the interconsump-tion interval (X < 1).

Overview

In these experiments, we tested the time contrac-tion hypothesis with a method similar to that inExperiment 1, but with an attempt to reduce theinfluence of anchoring. We did this by havingparticipants contrast choices made for sequentialdays with those made for sequential weeks. Weassumed that the contrast between the two inter-vals would still highlight the length of the weekinterval, but that the extremity of the anchor wouldbe reduced if the initial interconsumption intervalwas a day. Experiment 2 involved hypotheticalchoices and Experiment 3 involved real choices. InExperiment 2, participants either chose which ofsix snacks they would want on 3 consecutive daysfollowed by the same choice for consecutive weeks(day -» week condition), or else chose for consecu-tive weeks followed by days (week —> day condi-tion). We reasoned that subjects would treat thefirst question more or less identically in bothconditions. Specifically, they would contract the1-week interconsumption interval so that choicesfor both the 1-week and 1-day intervals wouldshow similar levels of variety seeking when theinterval was not made salient. For the secondquestion, we expected that time would be salientand therefore that participants would considerhow variations in the interconsumption intervalmight influence their preference. Week -*• daysubjects will have already contracted the intervalwhen making week judgments and would thereforenot ask for greater variety when making day

judgments. On the other hand, participants in theday -»• week condition will, in effect, expand theinterval when asked about weekly consumption.Experiment 3 was similar to Experiment 2, exceptthat participants made real choices, and the week—> day condition was replaced with a week-onlycondition because the class in which data wascollected met weekly, so it was not practical to givesubjects snacks on sequential days. For both experi-ments, we predicted that variety seeking would besimilar for all judgments except for the weekjudgments made in the day -> week condition,which would show less variety seeking.

Method

Participants. Experiment 2 was conducted with122 graduate students in two classes at CarnegieMellon University, one in the Heinz School ofPublic Policy and Management and another inSocial and Decision Sciences. They were eachgiven a single chocolate bar as a reward forparticipation. Experiment 3 involved 61 under-graduate business students at Indiana Universityof Pennsylvania. This experiment involved actualchoices, so all students received three snacks oftheir choice at the appointed times.

Procedure. Response sheets were distributedto participants just before the end of their classes.The first page of all two-page forms began with anexhortation to answer the questions one page at atime and not to change any answers once they hadturned the page. Experiment 2 involved two hypo-thetical choice conditions. In the day -» week(week -» day) condition, participants were firsttold to "imagine that you are going to eat onesnack per day (week) for the next 3 days (weeks)"and were then asked to choose which of the threesnacks they wanted on each occasion. On the nextpage, they were told to "imagine that instead ofeating one snack per day (week), you are going tobe eating one snack per week (day) for the next 3weeks (days)." (Boldface in the original.) Again,they chose the three items they wanted for thethree occasions.

Experiment 3 was identical except that partici-pants in the day -» week condition made hypotheti-cal choices for an interconsumption interval of 1day and then made real choices for a 1-weekinterval, while participants in the week-only condi-tion chose only for the 1-week interval.


Results

Table 3 depicts the results of both experiments.For the hypothetical choices of Experiment 2,variety seeking varied in the manner predicted bythe time contraction hypothesis. We conducted a2 x 2 ANOVA with interconsumption interval(day or week) as a within-subjects factor and order(day ->• week versus week -» day) as a between-subjects factor, and with level of variety seeking asthe dependent measure. The main effects of bothorder, F(l, 23) = 4.34, p < .05, and time, F(l,123) = 5.83, p < .05, were significant, as was theinteraction between them, F(l, 123) = 18.00, p <.001. Inspection of the data suggests that the twomain effects were the result of much lower varietyseeking for week judgments in the day -> weekcondition, and this was confirmed by separateANOVAs. Day and week judgments did not differin the week -» day condition, F(l, 61) = 1.58,whereas there was significantly more variety seek-ing in the day judgments of the day —» weekcondition, F( 1,61) = 23.53,p < .001.

For Experiment 3, there was numerically morevariety seeking for week judgments in the day —>week condition than in the week-only condition,but this result did not achieve conventional levelsof significance, f(59) = 1.44, p < .15. However,the results of other analyses also support ourhypothesis. There was no difference between vari-ety seeking for day choices than for week-onlychoices, f(59) < 1 (in fact, week-only judgmentsshowed marginally more variety seeking), but therewas a difference between day choices and week

choices within the subjects in the day —» weekcondition, t(29) = 3.25, p < .005.

The results of Experiments 2 and 3 along withthe prominence of ABA choices in the simulta-neous choice conditions of Experiment 1 suggestthat time contraction is one cause for the highlevels of variety seeking observed during simulta-neous choice. At least under the circumstancestested by Simonson and ourselves, it appears thatpeople underweight the interconsumption inter-val. Our participants gave little evidence of sensitiv-ity to the difference between the two interconsump-tion intervals unless the long interval was madesalient by explicitly contrasting it with the first,short interval.

Discussion

Time contraction has implications that extendbeyond the domain of variety seeking; it applies toany situation in which there is an interconsump-tion interval that affects choice in some way. Toillustrate, we demonstrated the effects of timecontraction on intertemporal choice, drawing on astudy conducted by Loewenstein and Prelec (1993).Their subjects imagined that they had to take twoweekend trips, one pleasant (a visit to friends) andthe other unpleasant (a visit to a disliked aunt).Participants first specified in which order theywould like to take the visits if the intervisit intervalwas 1 week and then what they would prefer if theinterval was 6 months. They found that subjectspreferred to get the aunt over with quickly when

Table 3Variety Seeking in All Conditions of Experiments 2 and 3: Tests ofTime Contraction Hypothesis

Number of different kinds chosen byinterconsumption interval

Condition

Day

One Two Three

% n % n % n

Week

One Two

M % n % n

Three

% n M

Experiment 2Day -»week 13 8 42 26 45 28 2.32 35 22 39 24 26 16 1.90Week -»day 25 15 28 17 47 28 2.22 27 16 30 18 43 26 2.17

Experiment 3Day -> week 20 6 33 10 47 14 2.27 33 10 33 10 33• week 20Week only 16

10 2.005 38 12 45 14 2.29


the interval was 1 week, but were much less likelyto do so when the interval was 6 months.

We suspected that the week -> 6 months orderof the two questions made the long interval verysalient, but that if the questions had been asked inreverse order (6 months -» week) the long intervalwould not have been so salient. Therefore, basedon the time contraction hypothesis, we predictedthat participants in the 6 months -» week condi-tion would contract the long interval and want tovisit the aunt first. As predicted, we found thatwhereas only a minority (39%) of week -> 6 monthsubjects wanted to visit the aunt first when theintervisit interval was 6 months, a sizable majority(75%) of the 6 month -> week subjects made thischoice. Moreover, there were no effects of ques-tion ordering on the number of people who wantedto visit the aunt first if the intervisit interval wasonly 1 week. These results parallel those of Experi-ments 2 and 3.

Experiment 4: Choice Bracketing

Choice bracketing refers to the impact on deci-sion making of the way sets of choices are subjec-tively partitioned. We suggest that subdividing aset of integral choices can cause the separatedchoices to be treated independently. Likewise,presenting several normatively independent choicesas a package can cause them to be treated in anartificially interdependent fashion. Insofar as vari-ety seeking is concerned, the bracketing, or simul-taneous presentation, of several similar choices islikely to evoke a diversification heuristic.

A crucial difference between simultaneous andsequential choices is the great discrepancy in timebetween choices—simultaneous choices are madeall at once, whereas sequential choices are sepa-rated by several days. According to the choice-bracketing hypothesis, the interval between choicesis only one of many ways that choices can bepsychologically combined or separated. We arguethat even if the interval between choices is heldconstant, choices that are bracketed separately willresult in less variety seeking than will those thatare bracketed together. Experiment 4 tests thechoice-bracketing hypothesis by having partici-pants choose a pair of items with the task framedas either two separate choices or as a combinedchoice of two items.

Overview

Experiment 4 was conducted with young trick-or-treaters visiting two neighboring homes on Hallow-een. Some children chose two candy bars at onehouse (combined choice), whereas others choseone candy bar at each of the two houses (separatechoice). In both conditions, choices were made atvirtually the same time: the houses were separatedby a small distance, and the time separatingchoices was less than 1 min. Moreover, the chil-dren had no opportunity to consume one candy barbefore receiving another, and we assumed that thetime between receipt of a candy bar and itsconsumption would be randomly distributed acrossconditions. In both situations, the outcomes of thechoices had the same material consequences: twocandy bars in each child's bucket. If the childrenbracket their choices by the houses they arevisiting, then we would predict less variety seekingwhen they get one candy bar at two houses thanwhen they get two candy bars at one house.

Method

Participants. Thirty-eight young trick-or-treat-ers, with approximate ages ranging from 3 to 14years, participated in the experiment. No formalrecord of ages was kept because the experimentwas done unobtrusively. Because of an asymmetryin the procedure by which children were assignedto groups, 13 participated in the simultaneouschoice condition and 25 in the sequential choicecondition.

Procedure. The experiment was conducted onHalloween 1993 in two neighboring houses. Thetrick-or-treaters petitioning both houses were of-fered a choice between two snack-size candy barsplaced in stacks on a tray. The two candy bars weresimilar in both appearance and content—ThreeMusketeers and Milky Way. Children in the com-bined choice condition were told they could "choosewhichever two candy bars you like"—instructionsdesigned to minimize potential demand effects tochoose two different bars. We also placed verylarge stacks of candy on the tray (and there werefew trick-or-treaters because of bad weather), so itis unlikely that the children chose different barsbecause they believed that by taking two bars ofthe best kind they would be depriving others.Children in the separate choice condition chose


one candy bar at each house. To coordinate databetween houses, we recorded a brief description ofeach child and the time at which they arrived.

Initially, we assigned children to groups basedon the direction in which they were moving downthe street. In this way, there needed to be nocommunication between houses to decide to whichgroup a child should be assigned. However, be-cause few children were moving in the combinedchoice direction, we reversed directions after ap-proximately 1 hr. However, the change did notcome soon enough to equalize the number ofparticipants in each group.

Results and Discussion

All 13 children in the combined choice conditionchose two different candy bars, compared withonly 48% (12 of 25) in the separate choice condi-tion, x2(l) = 10.27, p < .001. Notice that in bothcases, the children were selecting the same numberof candy bars from the same assortment at virtuallythe same time for future consumption. However,in one case they invariably chose two differentcandy bars, while in another case they were morelikely to choose the same candy bar.

Besides variety seeking, there are many situa-tions for which people treat packages of choicesdifferently from individual choices. For example,research on decision making under uncertainty hasshown that people may choose to play a gamble ifthey can do so repeatedly, but not if they can play itonly once (Lopes, 1981; Redelmeier & Tversky,1992; Samuelson, 1963). Research on intertempo-ral choice shows that people prefer to experienceoutcomes starting with the worst and ending withthe best when the outcomes are expressed assequences, but that they often reveal the preciseopposite order when presented with a series ofseparate decisions concerning when to receiveeach component of the series.

Another example of choice bracketing is foundin the literature on self-control. Ainslie (1975)posits a difference between viewing an individualact of consumption in isolation and "bundling" itwith other choices made over the long term.Indeed, Read, Loewenstein, and Camerer (1994)show that the rate at which people plan to indulgein a fairly "sinful" dessert is strongly related to theperiod over which they plan their consumption.Their subjects imagined that they were attending aconference for a week and had to decide before-

hand how many tasty but highly caloric breadpuddings to order as desserts. In one condition,subjects checked off the days of the week fromMonday through Sunday on which they would liketo consume bread puddings. The other conditionwas equivalent, except that the week was dividedinto two intervals, weekdays and the weekend.Subjects in the latter condition ordered 73% morebread puddings (2.3 versus 4) than those in theformer. It appears that people ration their indul-gences by permitting themselves a certain numberper period. Perhaps because of time contraction,they are insensitive to the length of those periods,and so bracketing a single interval into two canvirtually double the number of indulgences (pud-dings, in our case) permitted.

General Discussion

The preceding research confirms that, contraryto the traditional notion that combined choices arebest, there are occasions when consumers arebetter off choosing separately. Indeed, what wehave characterized as the diversification bias isaptly named: when people choose many goods incombination they commonly choose more varietythan they end up wanting. Experiment 1 containedtwo demonstrations of this. The first demonstra-tion was a test of the cognitive capacity limitationhypothesis. We contrasted the variety seeking oftwo groups of subjects. One group was informedthat they would be choosing three snacks andreminded of their earlier choices, while anothergroup was left uninformed and was not reminded.We reasoned that if subjects really wanted as muchvariety as they chose during simultaneous choicebut were hindered by capacity limitations, thensequential choices would elicit more variety seek-ing if those limitations were reduced. However,there was no evidence that informed participantswanted more variety than uninformed ones. In asecond demonstration that combined choices elicitan unwanted degree of variety seeking, partici-pants who chose simultaneously were later asked ifthey wanted to change their minds on the days ofconsumption. These participants commonly statedthat they wished to shift their choices in thedirection of less variety. Subsequent experimentsare consistent with the conclusion that surplusvariety seeking during combined choices is attribut-able to time contraction and choice bracketing. In


two sequels to this article, we provide many moreexamples of choice bracketing (Read et al., 1994).

It should be noted that in the majority of cases,combined choices are superior to separate ones.Combining choices allows us to take into accountcomplementarity (imagining choosing a tie withoutknowing what jacket to wear) and substitutabilityof goods (if you are having clam chowder as anappetizer, you probably don't want clam sauce onyour pasta). Combining choices may also help us tochoose what is best in the long run rather thanfalling prey to momentary desires. As Ainslie andHaslam (1992b) propose, people can increase theirincentive to make optimal choices if they combinetogether, in imagination, all their choices of onekind or another. For example, one can more easilyavoid taking a single cigarette by considering thechoice as between a policy of smoking and one ofnot smoking.

Dynamic Consistency

In addition to the question of whether peopleare better off making combined or separate choices,our research bears on another normative issue.Economic models generally demand that peoplebe dynamically consistent. This means that so longas no economically relevant changes occur in theperiod intervening between planning choices andtheir execution, the choices themselves should notchange (Machina, 1989). A failure of such consis-tency can lead (theoretically) to people beingtransformed into money pumps because a dynami-cally inconsistent person can be induced to pay toexchange A for B at one time and then later paymoney to change back. Although in reality peopledo not appear to be money pumps, there is agrowing catalogue of evidence suggesting thatneither are they dynamically consistent. To illus-trate, a well-established finding is that people showhyperbolic rather than exponential discounting,which results in changes in the ordering of prefer-ences as objects approach in time (Ainslie &Haslam, 1992a; Rachlin & Raineri, 1992).

Our studies show a hitherto unrecognized sourceof dynamic inconsistency. People's preferences forsnacks, for example, can change depending on howthey bracket the choices they are making. If theyplan a whole series of consumption occasions inadvance they will choose more variety than theywant when the time for consumption arises. Inprinciple, as in all cases of dynamic inconsistency,

choice combination and separation can turn peopleinto money pumps. Because trade-offs betweencomplements and substitutes are made locally (inreference to a single bundle of choices), consumerscan in theory be induced to pay to change back andforth between two bundles of goods by changingthe set with which the bundle is combined.

This can be illustrated by considering the con-sumer depicted in Figure 1. Imagine offering herthe choice between a combination of two orangesor candy bars. Recall that this consumer prefersoranges to candy bars but also prefers one of each(when they are bracketed together) to two or-anges. Let us say that she first buys a combinationof one orange and one candy bar. If we now induceher to consider the candy bar as a separate good,she will voluntarily pay us a small amount toexchange it for a more desirable orange. If wereframe the two items as a combined bundle ofgoods, she should then be willing to trade the twooranges for one of each good plus a premium paidto us.

Machina (1989) observed that there are practi-cal reasons why people cannot be so readily trans-formed into money pumps. First, since consumersremember what their previous choices were, theywill not be tricked into going back. Second, andspecific to the bracketing problem as formulated,the endowment effect (Kahneman, Knetsch, &Thaler, 1991) may act as a countervailing force toprevent consumers from too easily parting withwhat they already possess, even if in a direct choicethey would prefer the item they do not have to theone they have. These mechanisms prevent theconsumer from experiencing fully the conse-quences of dynamic inconsistency.

The Pennies-a-Day Phenomenon

Time contraction may make payment scheduleswith short interpayment intervals more attractivethan those with long intervals. Rent-to-own compa-nies are able to persuade people to pay remarkablyhigh total prices through installment plans. One oftheir marketing strategies is to advertise weeklypayment rates rather than traditional monthlyones, a technique based on what Gourville (1994)has referred to as the pennies-a-day phenomenon.These weekly rates are much more than onequarter of the monthly rates offered by more


legitimate companies (Swagler & Wheeler, 1989).We suggest that people's tendency to contract timeincreases the total amount they are willing to pay ifthe payment schedule is described in weekly ratherthan monthly terms. Note that if the intervalbetween payments is contracted, the psychologicalinterpayment interval for monthly payments willbe much shorter than four times the interval forweekly payments. The time contraction hypothesispredicts that, within some normal limits, as thepayment interval is increased, with the (objective)cost per unit time held constant, the likelihoodthat a payment schedule will be acceptable willdecrease. Presumably, framing the purchase interms of final cost will lead to the fewest takers—those people willing to make 78 weekly paymentsof $15 for a 19-inch television (Swagler & Wheeler,1989) may be disinclined to make those paymentsif they are informed that the final cost will be$1,170. Thus, those who concern themselves witheducating consumers to avoid the lure of pooreconomic decisions may wish to encourage them torecede all candidate payment schedules into astandard form such as final cost or cost per year.

Applications to Marketing and Education

The clearest applications of choice bracketingare to marketing. Based on our findings, we sus-pect that some products will benefit from market-ing strategies that invoke the diversification bias,while other products could benefit from attempt-ing to eliminate the bias. Products that alreadyhave the lion's share of the market will benefitfrom encouraging people to purchase the samegood all the time. Our research suggests that thismight be accomplished by emphasizing the timeinterval between consumption (thereby reducingthe impact of time dilation), or by encouragingpeople to think about what they will want onseparate consumption occasions before they maketheir purchases. For instance, an advertiser mightpoint to the disappointment of the consumer whogoes to his or her refrigerator and finds that afavorite soft drink is not there and all that remainsis a less desirable brand. On the other hand, lesspopular brands can position themselves as changesof pace by emphasizing the possibility of satiationalong with the pleasures of diversity. Indeed, this isprecisely the strategy used by 7-Up, which empha-sizes its "uncola" as an alternative to a typical colasoft drink. Because unusual brands are more likely

to be chosen by consumers buying in bulk (Simon-son & Winer, 1992), those marketing unusualbrands might wish to expend most advertisingresources to promote sales where quantity pur-chases are made, such as in supermarkets ratherthan corner stores.

Consumers prefer greater diversity when theirchoices are bracketed together than when they aremade separately. In the simple cases examinedhere, there probably was no good reason to choosediversity (i.e., people are not better off havingeaten many different snacks rather than many ofthe same snack) and there is some reason for themnot to (i.e., they will likely prefer the same snackevery time at the specified times of consumption).However, on many occasions more diversity is tothe advantage of the individual, and under thesecircumstances encouraging people to bracketchoices together may well induce them to act intheir own best interest. For example, nutritionistsrecommend a varied diet, and a proper exerciseregime is composed of a good mix of aerobic andanaerobic activity. Practitioners who wish to facili-tate such diversification may find that encouragingpeople to simultaneously plan many meals orexercise occasions may be an effective and unobtru-sive way of inducing them to meet these goals. Ofcourse, our research also suggests that the activi-ties planned will likely deviate from what is wantedwhen the plan is to be executed, and therefore itremains an empirical question to what extent suchlong-term planning is beneficial.

Variety Seeking for Others

Recently, one of us attended a party where thehost had thoughtfully provided a good quantity ofseveral varieties of premium beer. Quite quickly,all the bottles of one brand were gone and whatremained was a large quantity of several margin-ally less desirable beers. This illustrates that thediversification bias operates interpersonally as wellas intrapersonally. Apparently, when the hostpurchased a quantity of beer for other people sheused the same reasoning about their tastes as otherpeople use when they buy quantities for them-selves.5 She inferred that other people prefer vari-

5 Interestingly, the host's preferred beer was the onewhich went most quickly. She would have been better offpurchasing her favorite beer only.


ety more than they actually do. Her guests did whatindividual consumers do after they have purchaseda variety for themselves. They first consume thegoods they prefer the most and then turn to theless desirable items that they had purchased whenunder the influence of a goal (diversification) thatis no longer operative.

Clearly, whether we are deciding for ourselvesor for others, we should select the item that eachindividual (or ourself at different points in time)will want to consume and not the bundle of itemsthat appears most attractive from our own presentperspective.

References

Ainslie, G. (1975). Specious reward: A behavioral theoryof impulsiveness and impulse control. PsychologicalBulletin, 82, 463^*96.

Ainslie, G., & Haslam, N. (1992a). Hyperbolic discount-ing. In G. Loewenstein & J. Elster (Eds.), Choice overtime (pp. 57-92). New York: Russell Sage.

Ainslie, G., & Haslam, N. (1992b). Self-control. In G.Loewenstein & J. Elster (Eds.), Choice over time (pp.177-211). New York: Russell Sage.

Baddeley, A. (1976). The psychology of memory. NewYork: Basic Books.

Frederickson, B. L., & Kahneman, D. (1993). Durationneglect in retrospective evaluations of affective epi-sodes. Journal of Personality and Social Psychology, 65,45-55.

Gourville, J. T. (1994, May). The pennies-a-day phenom-enon: How the temporal framing of a consumer requesteffects compliance. Paper presented at the BehavioralDecision Research and Management Conference,MIT Sloan School, Cambridge, MA.

Kahn, B., & Lehmann, D. R. (1991). Modelling choiceamong assortments. Journal of Retailing, 3, 274-298.

Kahneman, D. (1973). Attention and effort. EnglewoodCliffs, NJ: Prentice-Hall.

Kahneman, D., Frederickson, B. L., Schreiber, C. A., &Redelmeier, D. A. (1993). When more pain is pre-ferred to less: Adding a better end. PsychologicalScience, 4, 401-405.

Kahneman, D., Knetsch, J., & Thaler, D. (1991). Theendowment effect, loss aversion, and status quo bias.Journal of Economic Perspectives, 5, 193-206.

Kahneman, D., & Snell, J. (1992). Predicting a changingtaste: Do people know what they will like? Journal ofBehavioral Decision Making, 5, 187-200.

Loewenstein, G., & Prelec, D. (1993). Preferences forsequences of outcomes. Psychological Review, 100, 91-108.

Lopes, L. (1981). Decision making in the short run.Journal of Experimental Psychology: Human Learningand Memory, 7, 377-385.

Machina, M. J. (1989). Dynamic consistency and non-expected utility models of choice under uncertainty.Journal of Economic Literature, 27, 1622-1668.

McAlister, L. (1982). A dynamic attribute satiationmodel of variety seeking behavior. Journal of Con-sumer Behavior, 9, 141-150.

Miller, G. A. (1957). The magical number seven, plus orminus two. Psychological Review, 63, 81-97.

Pessemier, E., & Handelsman, M. (1984). Temporalvariety in consumer behavior. Journal of MarketingBehavior, 21, 435^444.

Rachlin, H., & Raineri, A. (1992). Irrationality, impul-siveness and selfishness and discount reversal effects.In G. Loewenstein & J. Elster (Eds.), Choice overtime(pp. 93-118). New York: Russell Sage.

Read, D., Loewenstein, G., & Camerer, C. (1994).Choice bracketing: A neglected aspect of problem fram-ing. Unpublished manuscript.

Redelmeier, D. A., & Tversky, A. (1992). On theframing of multiple prospects. Psychological Science,3, 191-193.

Samuelson, P. (1963). Risk and uncertainty: A fallacy oflarge numbers. Scientia, 98, 108-113.

Simon, H. (1957). Models of man: Social and rational.New York: Wiley.

Simonson, I. (1990). The effect of purchase quantity andtiming on variety seeking behavior. Journal of Market-ing Research, 32, 150-162.

Simonson, L, & Winer, R. S. (1992). The influence ofpurchase quantity and display format on consumerpreference for variety. Journal of Consumer Research,19, 133-138.

Swagler, R. M., & Wheeler, P. (1989). Rental-purchaseagreements: A preliminary investigation of consumerattitudes and behaviors. Journal of Consumer Affairs,23, 145-160.

Tversky, A., & Kahneman, D. (1981). The framing of decis-ions and the psychology of choice. Science, 211,453-458.

Received March 15,1994Revision received June 30,1994

Accepted July 5,1994 •

diversification bias: explaining the discrepancy in …diversification bias 35 deed, large quantity...

Documents