consumption self-control by rationing purchase quantities of ......marketing science/vol. 17, no. 4,...

0732-2399/98/1704/0317$05.00Copyright q 1998, Institute for Operations Researchand the Management Sciences

Marketing Science/Vol. 17, No. 4, 1998

pp. 317–337

Consumption Self-Control by RationingPurchase Quantities of Virtue and Vice

Klaus WertenbrochYale School of Management, Box 208200, New Haven, Connecticut 06520

[email protected]

AbstractConsumers’ attempts to control their unwanted consump-tion impulses influence many everyday purchases withbroad implications for marketers’ pricing policies. Address-ing theoreticians and practitioners alike, this paper uses mul-tiple empirical methods to show that consumers voluntarilyand strategically ration their purchase quantities of goodsthat are likely to be consumed on impulse and that thereforemay pose self-control problems. For example, many regularsmokers buy their cigarettes by the pack, although they couldeasily afford to buy 10-pack cartons. These smokers know-ingly forgo sizable per-unit savings from quantity discounts,which they could realize if they bought cartons; by rationingtheir purchase quantities, they also self-impose additionaltransactions costs on marginal consumption, which makesexcessive smoking overly difficult and costly.

Such strategic self-imposition of constraints is intuitivelyappealing yet theoretically problematic. The marketing lit-erature lacks operationalizations and empirical tests of suchconsumption self-control strategies and of their managerialimplications. This paper provides experimental evidence ofthe operation of consumer self-control and empirically illus-trates its direct implications for the pricing of consumergoods. Moreover, the paper develops a conceptual frame-work for the design of empirical tests of such self-imposedconstraints on consumption in consumer goods markets.Within matched pairs of products, we distinguish relative“virtue” and “vice” goods whose preference orderingchanges with whether consumers evaluate immediate or de-layed consumption consequences. For example, ignoringlong-term health effects, many smokers prefer regular (rela-tive vice) to light (relative virtue) cigarettes, because theyprefer the taste of the former. However, ignoring these short-term taste differences, the same smokers prefer light to reg-ular cigarettes when they consider the long-term health ef-fects of smoking. These preference orders can lead todynamically inconsistent consumption choices by consumerswhose tradeoffs between the immediate and delayed con-sequences of consumption depend on the time lag betweenpurchase and consumption. This creates a potential self-control problem, because these consumers will be temptedto overconsume the vices they have in stock at home. Pur-chase quantity rationing helps them solve the self-controlproblem by limiting their stock and hence their consumptionopportunities. Such rationing implies that, per purchase oc-

casion, vice consumers will be less likely than virtue consum-ers to buy larger quantities in response to unit price reduc-tions such as quantity discounts.

We first test this prediction in two laboratory experiments.We then examine the external validity of the results at theretail level with a field survey of quantity discounts and witha scanner data analysis of chain-wide store-level demandacross a variety of different pairs of matched vice (regular)and virtue (reduced fat, calorie, or caffeine, etc.) product cate-gories. The analyses of these experimental, field, and scannerdata provide strong convergent evidence of a characteristiccrossover in demand schedules for relative vices and virtuesfor categories as diverse as, among others, potato chips,chocolate chip cookies, cream cheese, beer, soft drinks, icecream and frozen yogurt, chewing gum, coffee, and beef andturkey bologna. Vice consumers’ demand increases less inresponse to price reductions than virtue consumers’ demand,although their preferences are not generally weaker for vicesthan for virtues. Constraints on vice purchases are self-imposed and strategic rather than driven by simple prefer-ences. We suggest that rationing their vice inventories at thepoint of purchase allows consumers to limit subsequent con-sumption. As a result of purchase quantity rationing, how-ever, vice buyers forgo savings from price reductionsthrough quantity discounts, effectively paying price premi-ums for the opportunity to engage in self-control. Thus, pur-chase quantity rationing vice consumers are relatively priceinsensitive.

From a managerial and public policy perspective, ourfindings should offer marketing practitioners in many con-sumer goods industries new opportunities to increase profitsthrough segmentation and price discrimination based onconsumer self-control. They can charge premium prices forsmall sizes of vices, relative to the corresponding quantitydiscounts for virtues. Virtue consumers, on the other hand,will buy larger amounts even when quantity discounts arerelatively shallow. A key conceptual contribution of this pa-per lies in showing how marketing researchers can investi-gate a whole class of strategic self-constraining consumer be-haviors empirically. Moreover, this research is the first toextend previous, theoretical work on impulse control by em-pirically demonstrating its broader implications for market-ing decision making.(Intertemporal Choice; Pricing Policy; Product Policy; Segmenta-tion; Self-Control)

CONSUMPTION SELF-CONTROL BY RATIONINGPURCHASE QUANTITIES OF VIRTUE AND VICE

318 Marketing Science/Vol. 17, No. 4, 1998

There are indeed some people not habitual drunkards, whocannot keep away from any alcohol they may have in thehouse: so they pay the retailer a very high wage for takingcharge of their stock of it, and serving it out to them a littleat a time. Alfred Marshall (1919, p. 814)

1. IntroductionWhy do many regular smokers buy single packs ofcigarettes instead of 10-pack cartons, even when theysmoke a pack every day and do not normally face li-quidity constraints? At, say, $2.75 a pack and $23 acarton, these consumers could save some $160 an-nually if they bought cartons. Many of them disputethat they behave irrationally when they regularlychoose single packs over cartons and knowingly forgosizable quantity discounts on cartons. They claim tocontrol their smoking by having only a small stock ofcigarettes available at any given time. This suggeststhat consumers may voluntarily ration their purchasequantities of certain coveted goods to control con-sumption by imposing transactions costs and perhapsassociated feelings of guilt on additional consumption.

Such strategic self-imposition of constraints is intu-itively appealing yet theoretically problematic. In par-ticular, we lack operationalizations and empirical testsof consumption control strategies and of their impli-cations for marketing decision making. This papershows not only that consumers engage in purchasequantity rationing but also what the direct implica-tions of this behavior are for the pricing of consumergoods. Moreover, this paper provides a conceptualframework for detecting self-control in consumergoods markets and for segmenting them accordingly.Note that we focus on consumption self-control viastrategic purchase behavior, not on controlling pur-chase impulses or compulsive buying (O’Guinn andFaber 1989, Rook and Fisher 1995). We distinguish be-tween relative vice and virtue goods whose preferenceordering changes with whether one evaluates imme-diate or delayed consumption consequences. Two ex-periments manipulate these perceived intertemporalconsequences to invoke and show self-control by pur-chase quantity rationing, and two subsequent fieldstudies suggest manifestations of such rationing in realmarkets. In sum, we provide a conceptual definition ofvices and virtues, and we show that demand is less

price sensitive for vices than for virtues, suggestingthat marketers can price discriminate on the basis ofconsumer self-control.

1.1. Virtue and Vice: The Causes of ConsumptionSelf-Control Problems

Self-control problems arise from impulsive behavior—we often find ourselves making tempting choicesagainst our own better judgment and self-interest. Inthese situations, maximizing the local, immediately re-alized utility of consumption conflicts with maximiz-ing some higher-order, long-term, or life-time utility(Elster 1984, Freud 1911, Loewenstein 1996, Rachlin1995, Thaler and Shefrin 1981). Consumers exerciseself-control to forestall this temptation (Hoch andLoewenstein 1991).

Impulsive or time-inconsistent behavior and self-control have long been discussed in the context of in-tertemporal choice (Loewenstein and Elster 1992). In aclassic paper, Strotz (1956) viewed impulsive behavioras reflecting dynamically inconsistent preferences dueto nonconstant discounting. Thus, a dieter’s choice (attime t) between chocolate cake or fresh fruit for dessert(at consumption time T), contributing to weight gainor continued weight reduction (realized at some futuretime T ` n), may vary depending on whether s/hechooses before dinner at the time of purchase at thegrocery store (at t , T) or during dinner at home whenthe cake and fruit are immediately available (at t 4 T).Hence, the rate at which the time interval [T, T ` n] isdiscounted is not constant and depends on the timelag between decision (at t) and consumption (at T).

Such dynamically inconsistent preferences typicallycharacterize choices among goods and activities whoseconsumption consequences occur over multiple timeperiods, requiring consumers to make intertemporaltradeoffs between these consequences (Thaler 1980).To describe these goods and activities, let X sI Y de-note a strict preference for good X over a comparablegood Y when the consumer considers only concurrent,or immediate, consequences of consumption (e.g., tasteat T) and ignores long-term, or delayed, consequences(e.g., health effects at T ` n). Let X sD Y denote a strictpreference for X over Y when s/he considers only de-layed consequences and ignores immediate ones.

DEFINITION. Call X a vice relative to Y, and Y avirtue relative to X, if and only if, at the margin, X sI

WERTENBROCHConsumption Self-Control

Marketing Science/Vol. 17, No. 4, 1998 319

Y (maximizing immediate pleasure) and Y sD X (max-imizing delayed utility).

If both X sI Y and X sD Y, dynamic inconsistencybetween X and Y is impossible, because both imme-diate and delayed considerations prescribe the samechoice, X s Y. Even if there is intertemporal conflict,however, a utility-maximizing consumer will also al-ways prefer one good over the other, because his orher tradeoff between delayed (at T ` n) and immedi-ate consequences of consumption (at T) is constant nomatter how long the time lag is between the choice andthe onset of the immediate consequences (T 1 t). Incontrast, Strotz (1956) showed that nonconstant dis-counting can lead to time-inconsistent preferences asdescribed in the dessert example above. Thus, it is pos-sible that Y sD X at t , T, when the consequences ofone’s choice are all delayed, but that X sI Y at t 4 T,when at least some of the consequences are immediate.Such nonconstant discounting and the ensuing time-inconsistent choices are well documented (e.g., Ainslie1975, Benzion et al. 1989, Kirby 1997), suggesting thatvices as defined here are relatively more likely to beconsumed on impulse than comparable virtues. Hence,they impose a greater potential need for self-control.1

1.2. Testing for Self-Control by Purchase QuantityRationing

According to the above definition, impulsive consum-ers’ preferences between relative vices and virtues de-pend on the temporal perspective, which they take tochoose between them. Without constant discounting,there is no simple, consistent preference relationshipof the form X s Y (Strotz 1956). Vices are both covetedand spurned. Self-control has typically been viewed asforestalling the impulsive preferences in deference tothe long-term ones (e.g., Ainslie 1975, Schelling 1984).Consumers strategically forgo at least some of the pre-ferred immediate benefits of vice consumption to max-imize delayed utility. This can be achieved by alteringone’s incentives by restricting consumption opportu-nities through precommitment or otherwise raising the

1We focus on nonconstant discounting as a well-established formaldescription of impulsive, time-inconsistent behavior. Other concep-tualizations, e.g., in terms of visceral influences (Loewenstein 1996),would also predict that vices are more tempting yet subject to agreater need for self-control than virtues.

immediate cost of impulsive behavior (Rachlin 1995,Strotz 1956). Consumers can also reduce temptationthrough substitution and avoidance of, or distractionfrom, an impulse good (Hoch and Loewenstein 1991)and through mental accounting-based rules of behav-ior (Thaler 1985, Thaler and Shefrin 1981).

We suggest a hybrid mechanism. Consumers self-impose a constraint on their vice consumption by ra-tioning their purchase quantities (relative to virtues).The rationing rule says “Never buy more of a vice thanrX units at a time.” This precommits them to consumevices at no more than rate rX during an interpurchaseinterval. Consumption at higher rates can only occurat the expense of incurring additional transactionscosts. In the extreme, rX is the rate that maximizes thedelayed utility of current consumption. At the sametime, the rationing rule allows consumers to partiallygive in to temptation to get some immediate utility aswell (for all rX . 0). This is both a strength and weak-ness of purchase quantity rationing—if rX . 0, con-sumers can bend the rule by accelerating their inter-purchase intervals and simply buying vices moreoften. In contrast, if rX 4 0, consumers don’t ration butconstrain themselves so severely that they don’t buythe vice at all. This corner solution technically consti-tutes the most stringent form of self-control,renunciation.

More formally, let the utility of consuming k unitsof a relative vice X or of a relative virtue Y per inter-purchase interval be a function of the immediate utility[uX,I(•) or uY,I(•)] and the delayed utility [uX,D(•) oruY,D(•)] of consumption. Immediate marginal utility isgreater for vices than for virtues [ (k) (k)], whileu8 .u8X,I Y,I

delayed marginal utility is greater for virtues than forvices [ (k) . (k)]. We also assume that at leastu8 u8Y,D X,D

(k) and (k) are positive; otherwise, there wouldu8 u8X,I Y,D

be no reason at all to consume X or Y. Consumers ra-tion their purchase quantities of relative vices at rX sub-ject to the rule-based constraint # rX , wherek* k*X,D X,I

and are the consumption rates that maximizek* k*X,D X,I

uX,D(•) and uX,I(•). This limits vice consumption at ratek # rX. Relative virtues are rationed less or not at all.

The self-imposed purchase quantity rationing con-straint keeps vice consumers from increasing their de-mand in response to price reductions. They simply



can’t buy a larger amount of vices without violatingtheir self-imposed rule. Rule violations are possible(Thaler 1985) but psychologically costly (e.g., Hochand Loewenstein 1991), imposing a tax on purchasequantity increases which, in turn, mitigates the posi-tive impact of any given price cut. Thus, our centralhypothesis throughout the empirical work to follow is:

Hypothesis 1. All else equal (e.g., consumers’ inventoryholding costs), consumers are less price sensitive for vicesthan for virtues.

That is, demand for vices is increasingly constrainedwhen the price constraint is relaxed. Notice, though,that a smaller increase in consumers’ demand for vicesthan for matched virtues in response to a price reduc-tion might also imply that consumers simply prefervirtues to vices. To rule out this alternative explana-tion, we simultaneously predict the following:

Hypothesis 2. Consumers do not prefer virtues overvices at all prices.

Thus, there should be at least a crossover in demandschedules rather than an upward shift for virtues (cf.Boulding et al. 1994).

We test these predictions in two experiments andexamine their external validity in two field studies. Ex-periment 1 shows that demand for a relative vice ac-celerates less in response to an increase in quantity dis-counts than demand for a relative virtue. Yet, overallpreferences are stronger for the vice than for the virtue.Experiment 2 shows that the predicted crossover in de-mand schedules for a relative vice and a relative virtuedepends on how impulsive consumers are, that is, onwhether they have a heightened need for self-control.The experimental results imply that impulsive con-sumers do not prefer the virtue to the vice, yet theirmarginal valuations at each quantity are neverthelesslower for the vice. Field Study 1 illustrates this impli-cation by showing that retail quantity discounts aredeeper for relative vices than virtues, assuming thatretailers set prices equal to these marginal valuations.Field Study 2 employs scanner data to show that store-level aggregate demand for regular (relative vice)products is less price sensitive than for light (relativevirtue) products.

2. Experiment 1: Do We ForgoQuantity Discounts to RationOur Vice Purchases?

Using an experimental market approach, we testwhether (relative) vice consumers are less price sensi-tive than (relative) virtue consumers by examiningbuyers’ demand for potato chips at two different quan-tity discount depths offered for a large purchase quan-tity. We manipulate the conditions of intertemporalpreference inconsistency and potential need for self-control by framing the chips as either 25% fat (relativevice) or 75% fat-free (relative virtue).

Quantity discounts provide an interesting and rele-vant experimental context, because marketers oftenuse them to induce consumers to accelerate purchasequantities (Wansink 1996). In addition, they allowmarketers to price discriminate against homogeneousconsumers by setting the unit price at each purchasequantity equal to consumers’ marginal valuations atthat quantity (Dolan 1987). Larger quantities carrylower unit prices because these marginal valuationsare diminishing.

If vice consumers self-impose constraints on theirpurchase quantities, their marginal valuations of largerpurchase quantities should be less than virtue consum-ers’ marginal valuations. Hence, vice buyers should beless likely to accelerate purchase quantities than virtuebuyers in response to a given increase in quantity dis-count depth, that is, when the exogenous price con-straint is relaxed. One needs to rule out, however, thatforgoing a quantity discount and choosing a smallerquantity of the vice merely indicates weaker prefer-ences for vices. So buyers’ demand for virtues shouldnot exceed that for vices at both discount depths.

2.1. Method

Subjects and Procedure. Subjects were MBA stu-dents (N 4 304) who participated in this experimentas part of a class requirement. They were first showna 6-oz. bag of an existing brand of potato chips as areference package size. A questionnaire then offeredthem the opportunity to buy zero, one, or three 6-oz.bags of a new brand of potato chips at different pricesper bag. The new brand was described as having aninnovative mix of ingredients and as currently beingtest marketed. We used potato chips because they area prototypical impulse product as illustrated by,



Table 1 ANOVA Least-Square Means of Manipulation Check Variables in Experiment 1

Vice Frame Virtue Frame

Quantity Discount Deep Shallow Deep Shallow F Value p ,

Expensive 2.03 4.06 2.36 3.89 108.19a 0.0001Safety 3.23 3.22 4.07 4.13 15.63b 0.0001Concern 6.12 6.15 5.42 5.28 8.92b 0.01Taste after eating 6 oz.d 4.67 4.58 4.16 4.05 4.02c 0.05Taste after eating 18 oz.d 2.82 2.85 2.81 2.50

aMain effect of quantity DISCOUNT, N 4 291.bMain effect of FRAME, N 4 291.cBetween-subjects main effect of FRAME in repeated-measures ANOVA of taste ratings across both package sizes (6 and 18 oz.), N 4 290.dWithin-subjects main effect of 6- versus 18-oz. package size in repeated-measures ANOVA of taste ratings, F 4 216.46, p , 0.0001; interaction with

FRAME: F 4 2.40, p , 1; N 4 290.

for instance, Frito Lay’s slogan “No One Can Eat JustOne!” Subjects were informed that approximately onein 10 of those who completed their questionnaireswould be randomly selected in a lottery to receive $10as compensation.2 So that subjects would accuratelyreveal their demand for the chips, they learned that thelottery winners would have to buy that amount of po-tato chips at the given price that they had chosen inthe questionnaire. This procedure is incentive compat-ible because a subject who understates her true de-mand at a given price will forgo some potato chips thatshe would prefer to buy at this price, if she is selected.Similarly, if a subject overstates her true demand at agiven price, she will have to buy potato chips that shewould prefer not to have at that price. The lottery win-ners ultimately bought chips of a popular existingbrand.

Design. The design was a 2 2 2 between-subjectsfull factorial in which a new brand of potato chips wasdescribed as 25% fat to one group of subjects and as75% fat-free to another (FRAME). This framing manip-ulation was adopted from Levin and Gaeth (1988) to

2The $10 were provided to eliminate budget constraints. WhileThaler and Johnson (1990) suggest that consumers may be less re-luctant to risk spending windfall gains than existing assets, this $10windfall should not affect subjects in the 25% fat and the 75% fat-free conditions differently depending on discount depth.

vary the perceived intertemporal consequences of con-sumption while holding the objective stimulus infor-mation fixed. Chips that are 25% fat should appearmore preferable when considering only immediateconsequences of consumption, while 75% fat-free chipsshould appear more preferable when considering onlydelayed consequences.

The other between-subjects factor was the depth ofthe quantity discount offered (DISCOUNT). All sub-jects were offered a choice between a small purchasequantity of one 6-oz. bag of potato chips for $1 or alarge purchase quantity of three 6-oz. bags of chips,which were separately sealed to maintain freshness.This large size was offered to one group at a shallowquantity discount at $2.80 and to the other group at adeep discount at $1.80. Alternatively, subjects couldchoose not to buy any potato chips and keep the $10,if they were among the lottery winners.

As manipulation checks, subjects rated on 9-pointscales how good they predicted the new brand of po-tato chips would taste when eating 6 ounces and wheneating 18 ounces (both from 1 4 bad tasting to 9 4 goodtasting), how much of this new brand was safe to eat(from 1 4 a little to 9 4 a lot), how concerned theywould be about eating too much of the new brand(from 1 4 not at all concerned to 9 4 very much con-cerned), and how the large size was priced relative tothe small one (from 1 4 inexpensive to 9 4 expensive).



Finally, the dependent measure was subjects’ choice ofzero, one, or three bags of potato chips.

2.2. Results and DiscussionThirteen subjects were eliminated from the analysesbecause they had indicated that they never ate potatochips and hence had no relevant consumptionexperience.

Manipulation Checks. Least-square cell means forthe manipulation check variables are displayed in Ta-ble 1. As predicted, analyses of variance (ANOVAs)revealed that subjects saw the large package as moreexpensive under the shallow (M 4 3.97) than underthe deep (M 4 2.20) quantity discount (F(1, 287) 4

108.19, p , 0.0001), and that they rated the potato chipsunder the 25% frame as less safe to eat (M 4 3.23) thanunder the 75% fat-free frame (M 4 4.10, F(1, 287) 4

15.63, p , 0.0001). Similarly, subjects were also moreconcerned about eating too many chips under the 25%fat frame (M 4 6.14) than under the 75% fat-free frame(M 4 5.35, F(1, 287) 4 8.92, p , 0.01]. A repeated-measures ANOVA showed that they also regarded thechips as better tasting under the 25% fat frame, inde-pendent of package size (M 4 4.63 for 6 oz. and M 4

2.84 for 18 oz. under the vice frame, compared to M 4

4.11 and M 4 2.66 under the virtue frame; F(1, 286) 4

4.02, p , 0.05), and that taste was rated worse aftereating 18 ounces of chips (M 4 2.75) than after eating6 ounces (M 4 4.36) across frames (F(1, 286) 4 216.46,p , 0.0001). No other main effects or interactions ap-proached statistical significance. These results indicatethat subjects preferred the 25% fat chips over the 75%fat-free chips when considering only the immediateconsequences of consumption (taste), but that theirpreferences reversed when considering only delayedconsequences (concern and safety). Thus, the potatochips were successfully framed as a relative vice anda relative virtue.

Purchase Quantity Choices. To detect rationingeffects, we excluded nonbuyers from the analysis ofpurchase quantities because one cannot distinguishwhether those subjects exercised self-control or simplydidn’t like potato chips. Thus, we focus on the pur-chase quantity decision, given that a (probabilistic)purchase occurred, and not on the decision of whether

or not to buy. We ran the following logistic regressionmodel to predict purchase quantity probabilities forthe 151 subjects who bought one or three bags of potatochips (see Maddala 1983):

x8bePr(3 bags) 4 andx8b1 ` e

x8be 1Pr(1 bag) 4 1 1 4 , (1)x8b x8b1 ` e 1 ` e

where x8 is a row vector with 1 as the first element andFRAME (coded as 0 4 75% fat-free virtue frame, 1 4

25% fat vice frame), DISCOUNT (0 4 shallow quantitydiscount, 1 4 deep quantity discount), and FRAME 2

DISCOUNT as the remaining elements, and where b isa vector of parameters to be estimated. Figure 1 showsthe observed probabilities of choosing three bags,given that a purchase occurred, and Table 2 shows theresults of the logistic regression.3

Overall, subjects who bought potato chips weremore likely to prefer the large size when the chips wereframed as 25% fat (Pr 4 0.49) than as 75% fat-free (Pr4 0.42; bFRAME 4 1.07, p , 0.10). Buyers also showeda stronger preference for the large package size whenthe quantity discount was deep (Pr 4 0.59) rather thanshallow (Pr 4 0.28; bDISCOUNT 4 2.00, p , 0.0001). Aspredicted, this effect of quantity discount depth wasmitigated under the 25% fat vice frame (bFRAME 2 DIS-

COUNT 4 11.57, p , 0.05). The probability of buyingthe large size under the virtue frame increased from Pr4 0.20 under the shallow discount to Pr 4 0.65 underthe deep discount, an increase of 225%, while underthe vice frame the corresponding increase was from Pr4 0.41 to Pr 4 0.53, or merely 29% (see Figure 1).

As predicted, these findings show that buyers’

3To contrast rationing effects among buyers with the impact of theexperimental variables on overall demand by both buyers and non-buyers, we also jointly analyzed subjects’ decisions of whether andhow much to buy. Both a three-category (0, 1, or 3 bags bought)ordered logit model (Maddala 1983) and a two-part model, combin-ing a logit analysis of whether or not subjects bought with an ordi-nary least squares (OLS) regression of buyers’ logged purchasequantities (Manning et al. 1987), failed to reveal an interaction ofFRAME and DISCOUNT. The OLS regression of buyers’ logged pur-chase quantities in the two-part model, however, showed the sameeffects as the logistic regression reported in Table 2, also confirmingthe rationing hypothesis.



Figure 1 Observed Probability of Choosing Large Purchase Quantity (Three Bags of Potato Chips), Given a Purchase, inExperiment 1

Table 2 Effects of FRAME and DISCOUNT on Purchase Quantities ofPotato Chips in Experiment 1

Logistic RegressionParameter Estimates SE

Intercept 11.45*** 0.39FRAME (25% fat) 1.07* 0.55DISCOUNT (deep) 2.00**** 0.51FRAME 2 DISCOUNT 11.57** 0.71log likelihood 194.03Chi square 18.87***df 3N 151

Note. SE 4 standard error.

*p , 0.10, **p , 0.05, ***p , 0.001, ****p , 0.0001.

slower acceleration of purchase quantities under thevice frame was not due to weaker overall preferencesfor the potato chips under that frame. To the contrary,subjects had a slight overall preference for the 25% fatchips as indicated by their greater likelihood of buyingthe large purchase quantity. Yet, increasing the depthof the quantity discount was less effective in enticingvice buyers to increase their purchase quantities, sug-gesting that they self-imposed a rationing constraint as

external price constraints were relaxed. Nonetheless,the vice-virtue manipulation merely created condi-tions for impulsive behavior without measuring theresulting need for self-control directly. Thus, vice buy-ers may have capped their purchase quantities not tocontrol their temptation to consume but simply be-cause they actually preferred to consume at lower ratesas consumption at higher rates causes increasinglynegative delayed consequences. Consumption at lowerrates allows vice buyers to enjoy the taste without hav-ing to worry about these delayed consequences. More-over, vice consumption may have a bigger impact thanvirtue consumption on some nonprice dimension (e.g.,health) so that price is relatively less important inchoosing purchase quantities of vices, leading to lessprice-sensitive demand. These rivals are ruled outnext.

3. Experiment 2: Do PurchaseQuantity Preferences Depend onNeed for Self-Control?

Experiment 1 showed that a deep quantity discount isless effective in inducing buyers to purchase largequantities of a relative vice than of a relative virtue,even though the virtue is not preferred over the vice.



Experiment 2 replicates this finding and shows that itis the result of purchase quantity rationing, not of dif-ferences in consumption rates or relative importanceof price. Experiments 1 and 2 differ in that we nowexamine purchase quantity preferences in anotherproduct category. Second, purchase quantity and unitprice are orthogonal; both small and large quantitiesare available to all subjects at all unit prices. Third, weinclude a measure of the need for self-control, subjects’scores on Puri’s (1996) Consumer Impulsiveness Scale(CIS). In several experiments, Puri (1996) found thatconsumers with high impulsiveness scores (hedonics)were more likely to behave impulsively than consum-ers with low impulsiveness scores (prudents). Hedon-ics are thus faced with a greater potential need thanprudents to self-impose external constraints on theirvice consumption because they are more likely to givein to temptation when they have an opportunity to doso. Prudents, in contrast, are intrinsically controlled.Unlike purchase quantity rationing, neither of the ri-vals raised above predicts that hedonics require deeperquantity discounts as an incentive to buy larger quan-tities than the farsighted, highly controlled prudents.

Specifically, participants in an experimental marketstate how many packages of Oreo chocolate chip cook-ies they want to buy at each of 20 different packageprices. If subjects use purchase quantity rationing as aself-control mechanism, then hedonics (i.e., those witha high need for self-control) will be more likely thanprudents (i.e., those with a low need for self-control)to ration their purchase quantities of vices. Accord-ingly, we predict that individual demand is less pricesensitive for regular-fat Oreos (i.e., the relative vice)than for reduced-fat Oreos (i.e., the relative virtue) forhedonic subjects but not for prudent subjects. Further,hedonics do not generally prefer reduced-fat Oreos,that is, their virtue demand does not exceed their vicedemand at all prices.

3.1. Method

Subjects and Procedure. Subjects were MBA stu-dents (N 4 310) who completed a brief questionnaireduring an orientation session before beginning theircoursework. They stated how many packages of Oreosthey wanted to buy at each of 20 different packageprices. Subjects were told that, after filling out the

questionnaire, 10% of them would be selected at ran-dom to receive $10 as compensation. The experimenterwould also randomly pick one of the 20 package pricesin the questionnaire, and the lottery winners wouldhave to buy as many Oreo packages at that price asindicated by their questionnaire responses. Like Ex-periment 1, this procedure is incentive compatible be-cause a subject who overstates her true demand at agiven package price will end up having to buy morepackages of Oreos than she wants to if she is selectedto buy at that price. Similarly, if she understates hertrue demand, she will receive fewer packages than shewould like for her money.

Design. The design included three independentvariables. First, in a two-level between-subjects manip-ulation, one group of subjects was shown a 1 lb. 4 oz.package of regular-fat Nabisco Oreo chocolate chipcookies, while another group saw a package of 25%reduced-fat cookies of the same brand and size (FAT).Subjects were asked to predict how much they “wouldlike the taste” (from 1 4 would not at all like the taste to9 4 would like the taste very much) and to “evaluate anydelayed consequences” (from 1 4 bad delayed conse-quences to 9 4 good delayed consequences) of eating oneand two packages of these Oreos during the week afterthe experiment. These measures tested whether thefactor FAT successfully manipulated the perceived in-tertemporal consequences of consumption.

Second, we measured subjects’ chronic tendencies toact impulsively as a continuous between-subjects vari-able, using the 12-item Consumer Impulsiveness Scaledeveloped and validated by Puri (1996). The CIS con-sists of two independent subscales. One measures sub-jects’ hedonic orientation, while the other reflects amore cognitive dimension called prudence. Puri (1996)scored the items such that lower scores reflected higherimpulsiveness. She then classified respondents withimpulsiveness scores below the median on both di-mensions as impulsive or hedonic. Those with scoresabove the median on both dimensions were classifiedas nonimpulsive or prudent. In contrast, for ease ofinterpretability, we scored the items such that higherscores reflected greater impulsiveness. That is, we re-verse scored the five items that make up the hedonicsubscale (impulsive, careless, extravagant, easily



Table 3 Repeated-Measures ANOVA Least-Square Means ofManipulation Check Variables in Experiment 2

FAT

QUANTITY Regular Reduced F Valuea p ,

Tasteb One pack 6.13 5.19 5.02 0.05Two packs 4.49 4.33

Delayed consequencesc One pack 3.75 4.14 6.22 0.05Two packs 2.41 2.96

Note. N 4 310.aBetween-subjects main effect of FAT across both Quantities.bWithin-subjects main effect of QUANTITY: F 4 174.37, p , 0.0001;

interaction with FAT: F 4 16.95, p , 0.0001.cWithin-subjects main effect of QUANTITY: F 4 302.65, p , 0.0001; in-

teraction with FAT: F 4 1.31, p , 1.

tempted, enjoy spending) and used subjects’ actual rat-ings of the seven items that make up the prudent sub-scale (self-controlled, farsighted, responsible, re-strained, rational, methodical, a planner). Then, weadded subjects’ scores on both subscales, characteriz-ing each subject by a continuous impulsiveness score(IMPULSE). Applying Puri’s (1996) original categoricalclassification scheme did not affect the results.

Third, subjects stated for each of 20 package prices(ranging from $5 to $0.25 in steps of $0.25) whetherthey wanted to buy zero, one, or two packages of theparticular Oreos they were being shown at that price.This was counterbalanced with the elicitation of CISscores. Subjects’ responses yielded the lowest per-unitprice at which they would buy one pack of Oreos andthe lowest per-unit price at which they would buy twopacks, that is, their reservation prices for buying onepack and two packs, respectively. This served as a two-level within-subjects factor (QUANTITY). The reser-vation prices at these two levels were the dependentvariable.

3.2. Results and Discussion

Manipulation Checks. The effect of the FAT manip-ulation was tested in a repeated-measures ANOVA ofthe manipulation check variables. Least-square cellmeans are given in Table 3. As predicted, subjects ratedthe regular-fat Oreos as tasting better (M 4 5.31 acrossboth quantities) than the reduced-fat Oreos (M 4 4.76;F(1, 308) 4 5.02, p , 0.05), and they evaluated the de-layed consequences of consuming regular-fat Oreos asworse (M 4 3.08) than those of consuming reduced-fat Oreos (M 4 3.55; F(1, 308) 4 6.22, p , 0.05). Sub-jects thus preferred regular-fat Oreos to reduced-fatOreos when they considered the immediate conse-quences of consumption (i.e., taste), but they preferredthe reduced-fat Oreos with respect to delayed conse-quences. That is, they regarded regular-fat Oreos as avice relative to reduced-fat Oreos.

Subjects rated the regular-fat Oreos as better tastingonly when eating one pack but not two in one week(Table 3). While this result suggests that the taste ofregular-fat Oreos satiates when they are consumed inlarge amounts, subjects should stockpile regular-fatOreos at least as much as reduced-fat Oreos, simply to

take advantage of a low price, as they are free to delayconsumption of the second pack by one week (givensealed packaging and assuming negligible inventoryand opportunity costs). After one week, the regular-fatOreos would again taste better than the reduced-fatOreos. Satiation, therefore, cannot explain any differ-ences in purchase quantity preferences.

Reservation Prices. For descriptive purposes, wefirst conducted a median split of IMPULSE (median 4

35, mean 4 36.32, standard deviation 4 8.01, mini-mum 4 19, maximum 4 62). Adopting Puri’s (1996)terminology, we refer to the 161 respondents with highimpulsiveness scores (.35) as hedonics and to the 146respondents with low impulsiveness scores (#35) asprudents. The key result is as predicted and shown inFigure 2. Hedonics’ mean reservation prices showed asteeper decline for regular Oreos, dropping from $1.95for one pack to $0.90 per unit for two packs, than forreduced-fat Oreos, for which they dropped from $1.86for one pack to $0.92 per unit for two packs. For pru-dents, the pattern was reversed. Their mean reserva-tion prices per unit dropped from $1.77 for one packto $0.86 for two packs of regular Oreos but showed asteeper decline from $1.81 to $0.77 for reduced-fat Or-eos. A repeated-measures ANOVA of these reserva-tion prices with QUANTITY (within subjects), FAT,and IMPULSE (median split into hedonics and pru-dents) as independent variables supported our hy-



Figure 2 Mean Willingness to Pay for Oreo Cookies in Experiment 2 (Based on Median Split of IMPULSE into “Hedonics”and “Prudents”)

pothesis. Due to missing values, only 307 observationscould be considered. The order of CIS score and pur-chase quantity preference elicitation had no effect andwas dropped from the analysis. The ANOVA con-firmed that hedonics required a deeper quantity dis-count for the regular-fat Oreos than for the reduced-fat Oreos to induce them to buy a larger quantity.Prudents showed the opposite pattern (FQUANT-

ITY2FAT2IMPULSE (1, 303) 4 3.02; p , 0.10). Neverthe-less, hedonics were not willing to pay more for therelative virtue than the relative vice (FFAT2IMPULSE (1,303) 4 0.00; p , 1). Not surprisingly, subjects’ per-unitreservation prices were generally lower for buying twopacks than for buying one (FQUANTITY (1, 303) 4 735.36;p , 0.0001). No other effects approached statisticalsignificance.

To take into account the full variance of subjects’ CISresponses, we conducted a second ANOVA with theoriginal impulsiveness scores as a continuous indepen-dent variable. The results confirmed the predictedthree-way interaction (FQUANTITY2FAT2IMPULSE (1, 303)4 5.86; p , 0.05). With increasing impulsivenessscores, the decline in subjects’ per-unit reservation

prices for two packs relative to one pack became sig-nificantly steeper for regular-fat Oreos than forreduced-fat Oreos. In spite of this, more impulsive sub-jects did not value reduced-fat Oreos more thanregular-fat Oreos (FFAT2IMPULSE (1, 303) 4 0.43; p , 1).Again, subjects’ reservation prices were generallylower for two packs than for one pack (FQUANTITY (1,303) 4 28.88; p , 0.0001). Controlling for impulsive-ness, this decline was less steep for regular-fat than forreduced-fat Oreos (FQUANTITY2FAT (1, 303) 4 5.83; p ,0.05). No other effects approached significance.

In summary, these results show that impulsive con-sumers, or hedonics, are less price sensitive and re-quire a deeper quantity discount for a relative vicethan for a relative virtue to induce them to buy a largerquantity. The reluctance to buy larger quantities is notdue to weaker preferences for the vice because hedon-ics do not generally prefer the virtue over the vice. In-stead, they impose purchase quantity constraints onthemselves when buying vices. Prudents, in contrast,do not need to self-impose such constraints becausethey are less impulsive and hence less likely to give into a temptation to consume a larger quantity of the vice



when an opportunity arises. They buy a larger quan-tity even at a relatively shallow quantity discount, dis-playing greater price sensitivity. In sum, consumerswith a need for self-control require deeper price cutsas an incentive to overcome self-imposed purchasequantity constraints.

4. Field Study 1: Are RetailQuantity Discounts Consistentwith Purchase QuantityRationing?

Experiments 1 and 2 have provided evidence of pur-chase quantity rationing by examining purchase quan-tity preferences under different quantity discounts andimpulsive consumers’ per-unit reservation prices fordifferent quantities of relative vices and virtues. In con-trast to these experiments, however, consumers in realmarkets may not only ration purchase quantities ofvices to control their consumption, but they may alsobuy relative virtues (e.g., low fat) instead of vices (e.g.,regular fat; Hoch and Loewenstein 1991). Such substi-tution of relative virtues reduces the need to rationpurchase quantities because virtue consumption en-tails fewer delayed costs. We now search for sugges-tive evidence of purchase quantity rationing in realmarkets where demand for vices and virtues may besubject to such substitution effects as well as manyother influences that were controlled in theexperiments.

We assume that firms set retail prices in response tothe demand they observe, as a function of revealedconsumer preferences for different purchase quanti-ties. The experiments have shown that vices requiredeeper price reductions than virtues to induce impul-sive consumers to purchase larger amounts. Purchasequantity rationing would thus manifest itself bydeeper quantity discounts for vices than for relativevirtues at the retail level. To test this prediction, wecreated a set of matched pairs of (mostly nonaddictive)product categories for which consumers should differin their need for self-control. If consumers engage inpurchase quantity rationing, one should detect corre-sponding firm pricing behavior for those pairs of cate-gories within which they express conflicting temporalpreference orders (i.e., for pairs of relative vices and

virtues) but not for pairs with consistent temporal pref-erence orders. We first conducted a consumer surveyof temporal preference orders to classify the productcategories as vices and virtues and then tested our pre-diction with field data on retail-level quantity dis-counts for these categories.

4.1. Procedure

Design. To control supply- and demand-based ef-fects on retail prices that are unrelated to self-control,we created 30 pairs of consumer product categoriesthat were approximately matched within each pairwith respect to production technology, retailer andconsumer inventory holding and handling costs, andfrequency and expandability of consumption. To varythe relative need for self-control, the categories wereselected to be ordered substitutes within each pair (Ta-ble 4). One category (the relative vice) was hypothe-sized to be preferable to the other category (the relativevirtue) if the delayed consequences of consumptionwere assumed to be identical for both categories (X sI

Y), while the relative virtue was hypothesized to bepreferable to the relative vice if the immediate conse-quences were assumed to be identical (Y sD X).

Data Collection. Two sets of data were collected.To test the pairwise temporal preference ordering be-tween the categories, the pairs and the categorieswithin them were put in random order and then pre-sented to a group of 136 MBA students naive to thehypothesis. These subjects rated for each pair whichcategory they preferred to consume, first consideringfor all pairs only the immediate consequences of con-sumption (assuming identical delayed consequences),and second considering only the delayed conse-quences (assuming identical immediate conse-quences). Specifically, when evaluating the immediateconsequences (i), subjects were asked: “If you were aconsumer of each of the following products and con-sumption of these products entailed identical long-term consequences (such as long-term health or socialeffects or any other long-term costs and benefits),which of the two products in each pair would yourather consume? To evaluate these short-term effects,think about taste, ease of use, fun, temptation, or any-thing else that would make consuming the products



Table 4 Relative Vice and Virtue Product Categories in Field Study 1 (as hypothesized)

Relative Vices Relative VirtuesMean TemporalReversal Score N

Regular salad dressing Light salad dressing 2.95*** 130Regular fat cream cheese Light cream cheese 2.93**** 122Regular processed cheese Light processed cheese 2.77**** 125Regular mayonnaise Light mayonnaise 2.74**** 125Ice cream Frozen yogurt 2.69**** 134Regular yogurt Light yogurt 2.37**** 125Alcoholic beer Nonalcoholic beer 2.20**** 107Regular ice tea Low-calorie ice tea 1.71**** 105Sugared cereal Low-sugar cereal 1.64**** 134Regular chewing gum Sugarless chewing gum 1.50**** 131

Dunkin’ Donuts munchkins Dunkin’ Donuts muffins 1.44**** 115Regular soft drinks Diet soft drinks 1.35**** 127Regular coffee Decaffeinated coffee 1.34**** 98Whole milk Low-fat milk 1.18**** 133Butter Margarine 1.17**** 133Beef bologna Turkey bologna 0.96*** 95Regular tea Decaffeinated tea 0.91*** 116Regular cigarettes Light cigarettes 0.68 28Hair spray (aerosol) Hair spray (pump) 0.53 75Dexatrim Slimfast 0.53 30

Snacks with preservatives Snacks without preservatives 0.51** 136White rice Brown rice 0.43**** 134Sugared fruit drinks Fruit juice 0.40** 134Bleached flour Whole wheat flour 0.23 128Pornographic magazines News magazines 0.22* 128White bread Whole grain bread 0.17 128Deodorant (aerosol spray) Deodorant (roll-on) 0.16 123Seltzer water Natural spring water 0.14 125Sugar Brown sugar 0.11 133Vegetable shortening Vegetable oil 10.15† 126

*p , 0.05, **p , 0.01, ***p , 0.001, ****p , 0.0001 in two-sided t test.

†Mean difference score was counter to hypothesis (p , 1);

enjoyable.” When evaluating the delayed conse-quences (d), they were asked: “If you were a consumerof each of the following products and consumption ofthese products entailed identical short-term effects(such as taste, ease of use, fun, temptation, or anythingelse that would affect how much you enjoy consump-tion in the short run), which of the two products ineach pair would you rather consume? To evaluatethese long-term effects, think about health or social ef-fects of consumption, etc.” The ratings for each pairwere given on a 9-point scale anchored at 1 and 9 by

the two categories such that the midpoint representedindifference. This procedure served to identify condi-tions for time-inconsistent preferences, i.e., which, ifany, of the categories in a pair was perceived as therelative vice and which as the relative virtue.

The second set of data was a convenience sample ofregular (nonpromotion) retail package prices andpackage sizes for the 30 pairs. These data were col-lected in five stores of two major supermarket chains,two stores of a major drug store chain, one store of amajor bookstore chain, a Dunkin’ Donuts outlet, and a



high-volume liquor store in the Chicago metropolitanarea. Up to five different package sizes and 15 differentbrands per product category were available in thosestores and included in our convenience sample.

4.2. Results and Discussion

Category Classifications. As a prerequisite for thequantity discount analysis, the categories in each pairwere first classified depending on whether or not sub-jects perceived them as relative vices and virtues. Notethat we had randomized whether the lower or upperanchors of the i and d preference scales marked thecategory that was hypothesized to be the relative vir-tue in each pair. Therefore, we first rescaled subjects’ iand d ratings from 14 to `4, such that a preferencefor the hypothesized virtue (vice) in a pair would al-ways be indicated by a positive (negative) rating. Wethen sorted responses depending on whether a respon-dent at least weakly preferred the same category in apair from both temporal perspectives (i.e., d $ 0 and i$ 0, or d # 0 and i # 0) or whether a respondent pre-ferred one category in a pair to the other from onetemporal perspective (e.g., d . 0) but expressed a re-verse preference from the other temporal perspective(e.g., i , 0).

If preferences reversed (i.e., d . 0 and i , 0, or d ,0 and i . 0), the rating of the immediate consequenceswas subtracted from the rating of the delayed conse-quences (d 1 i). Positive temporal reversal scores (d 1

i . 0) meant that d . 0 and i , 0, while negative re-versal scores (d 1 i , 0) meant that d , 0 and i . 0.Thus, positive temporal reversal scores implied thatthe category that was preferred from the delayed per-spective (Y sD X) was not preferred from the imme-diate perspective (Y aI X), making it a relative virtueand the other category in the pair a relative vice ashypothesized. Negative temporal reversal scores im-plied that the category not preferred from the delayedperspective (X aD Y) was preferred from the imme-diate perspective (X sI Y), making it a relative vice andthe other category a relative virtue. For instance, con-sider light and regular cigarettes. If a subject markeda 2 on the immediate scale (indicating a preference forone anchor, e.g., regular cigarettes) and a 9 on the de-layed scale (indicating a preference for the other an-chor, e.g., light cigarettes), these values were rescaled

to 13 (immediate) and `4 (delayed). The temporalreversal score (delayed minus immediate) would thenbe `7, so that light cigarettes would be classified as arelative virtue and regular cigarettes as a relative vice.

Pairs in which one category was at least weakly pre-ferred over another from both the immediate and thedelayed perspectives (so that after rescaling both rat-ings were either nonnegative or nonpositive) receiveda zero temporal reversal score, indicating that they didnot induce time-inconsistent preferences. Thus, forevery respondent, each pair was characterized by atemporal reversal score that identified which category,if any, was the relative vice and which the virtue.

The mean temporal reversal score for each pair wasthen taken across all subjects who had stated that theyhad consumption experience with both categories inthe pair, including those who at least weakly preferredthe same category in a pair from both temporal per-spectives (i.e., d $ 0 and i $ 0, or d # 0 and i # 0) andwho had therefore been assigned a temporal reversalscore of d 1 i 4 0. Table 4 lists the mean temporalreversal scores for each pair as well as the number ofsubjects with consumption experience. These meanscores differed from zero for 21 pairs (at p , 0.05 orless), indicating that the product categories in thesepairs created intertemporal conflict between prefer-ences and can thus be characterized as relative vicesand virtues.

Quantity Discount Analysis. Several separateOLS regressions were run to determine quantity dis-count depths for subsets of m category pairs for whichpackage price and size data had been collected. All re-gressions were log-log models of the form

ln(p) 4 a ` a R ` b ln(q)Diff

` b R ln(q) ` Rc C ` e, (2)Diff i i

wherep 4 unit price,R 4 indicator variable for perceived vice categories (14 vice, 0 4 virtue),q 4 number of units per package (between 1 and 360;10th percentile 4 6 units, 90th percentile 4 64 units),Ci 4 indicator variables for m 1 1 category pairs tocontrol for pair-specific differences in unit price levels,e 4 error term (iid N ; 0, r2).



Figure 3 Predicted Unit Price as a Function of Package Size for the 21 Pairs of Relative Vices and Virtues (p , 0.05) inField Study 1

Number of units per package

The first analysis included the m 4 21 pairs of rela-tive vices and virtues with mean temporal reversalscores greater than zero at p , 0.05. Figure 3 and thefirst column of Table 5 show the results (ignoring thepair-specific intercept terms ci). Due to the logarithmictransformation of the data, the slope parameter b 4

10.451 (p , 0.0001) is a constant elasticity. It showsthat doubling package sizes for the virtues in these 21pairs results in a 45% decrease in unit price. As pre-dicted, unit prices drop even more rapidly, by 12%more, for vices when package size doubles—vicescarry deeper quantity discounts than matched virtues(bDiff 4 10.120; p , 0.01). At the same time, smallpackage sizes of relative vices carry higher unit pricesthan small sizes of relative virtues (aDiff 4 0.232; p ,0.05).4

4As an approximation, we assume that consumers’ underlying utilityfunctions are defined with respect to the measure of package size qthat is used by the manufacturer of the product. For example, pack-age sizes are measured in ounces for most grocery categories in oursample but in number of items for cigarettes, muffins, and maga-zines. As a result, the package size scale varies across product cate-gories. However, matching relative vices and virtues ensures that

Table 5 also shows the results of an additional set ofanalyses, performed separately for each third of m 4

10 pairs of the categories in Table 4. The 10 categorieswith the highest mean temporal reversal scores exhibitthe predicted pattern of deeper quantity discounts forvices than for matched virtues (bDiff 4 10.225; p ,0.01) and higher unit prices for small package sizes ofrelative vices than for small sizes of relative virtues(aDiff 4 0.424; p , 0.05). Also consistent with our the-ory, this difference in the signs of aDiff and bDiff dis-appears as temporal reversal scores decrease and thecategories within pairs cannot be characterized as rela-tive vices and virtues anymore (see the four right-mostcolumns of Table 5 for the estimates for the middle andbottom thirds of pairs in Table 4).

The pricing structure of the product categories inthis convenience sample is consistent with the premisethat firms behave as if vice consumers are less pricesensitive than virtue consumers: Relative vices are

the scale is the same within each category pair so that any remainingspecification error affects all observations within a pair equally.



Table 5 Unit Price Estimation as a Function of Package Size (Number of Units per Pack) for Pairs of Relative Vices and Relative Virtues in FieldStudy 1

21 Pairs With Vice-VirtueDistinction at p , 0.05 SE

Top Third(10 Pairs) SE

Middle Third(10 Pairs) SE

Bottom Third(10 Pairs) SE

a (virtue intercept) 4.238*** 0.131 4.291*** 0.238 4.217*** 0.148 4.729*** 0.223aDiff (intercept difference for vices) 0.232* 0.113 0.424* 0.206 10.013 0.180 10.046 0.209b (virtue slope) 10.451*** 0.033 10.428*** 0.088 10.411*** 0.048 10.691*** 0.052bDiff (slope difference for vices) 10.120** 0.038 10.225** 0.077 10.056 0.060 0.019 0.069R2 0.92*** 0.89*** 0.93*** 0.91***df 23 12 12 12n 307 135 124 108

*p , 0.05, **p , 0.01, ***p , 0.0001.

more expensive than relative virtues when bought insmall packages, but they also carry deeper quantitydiscounts. That is, sellers have to offer deeper dis-counts to induce consumers to increase their vice pur-chases. This result, however, does not imply awarenessamong retailers of consumer self-control processes. Itmerely illustrates a market-level pricing structure thatis consistent with our experimental results.

This pricing structure cannot be explained by differ-ences in overall preferences for vices and virtues. Forexample, if consumers had stronger preferences for thevices in this sample, perhaps due to more differenti-ated competition among vices, one would observe anoutward shift in the pricing schedule instead of a cross-over (cf. Boulding et al. 1994). That is, vices should bemore expensive for all package sizes (aDiff . 0) andshould exhibit equal or shallower quantity discounts(bDiff $ 0), contrary to our findings. Put differently, ifvices simply carried greater utility, consumers wouldtend to buy them in large quantities even when theseare not heavily discounted. The different quantity dis-count depths observed here are also not just a functionof how health-oriented the categories are, because thedifference in discount depth disappears in pairs forwhich subjects expressed weaker, if any, intertemporalpreference inconsistencies, although these pairs showsimilar differences in health orientation (see the middleand bottom thirds of category pairs in Table 4). Finally,one might expect more educated and wealthier con-sumers to prefer the relative virtues and less educatedand poorer consumers to prefer the relative vices, per-haps leading to deeper quantity discounts for vices. If

that were true, however, relative vices should becheaper in general—again in contrast to the findings.

5. Field Study 2: Is Store-LevelDemand Consistent withPurchase Quantity Rationing?

Aggregate demand does not allow for an experimentaltest of rationing. Nonetheless, price-discriminating re-tailers need to determine the net effect of purchasequantity rationing on the specific aggregate demandschedules they are facing. Hence, we now illustrate theimplications of the experimental results further by us-ing store-level scanner data to estimate the price elas-ticity of aggregate demand in retail markets for a sub-set of the categories that were characterized as relativevices and virtues in the previous study. Less price-sensitive demand for vices than for virtues representssuggestive evidence of the presence of purchase quan-tity rationing. Since the operation of self-control inshaping aggregate demand is not directly observable,we control for the effects of several other, normativevariables on demand.

5.1. Procedure

Data. The data came from Dominick’s Finer Foods, amajor supermarket chain with 86 stores and a 20%market share in metropolitan Chicago. The databasecontains store-specific, weekly retail unit sales volumeand unit price data as well as descriptive data for allbrands and UPCs sold throughout the chain during an



entire calendar year, plus demographic informationabout the store trading areas (from 1990 Census dataprovided by Market Metrics). Price was exogenous tothe extent that the retailer did not adjust prices everyweek according to elasticity estimates derived fromprevious observations. Vice (regular) and virtue (light)UPCs were matched by brand, store, week, packagesize, unit price, by whether they were offered on deal,and whether they were featured in newspaper insertsor in-store displays (based on IRI InfoScan data onchain-wide promotional activity). Thus, every lightUPC specified by these variables was matched by atleast one corresponding regular UPC to ensure that theelasticity comparison would be over identical rangesof variation in price, packaging, promotional support,and store demographics.

Cream cheese, processed cheese, soft drinks, andbeer were the only available categories for which UPCscould unambiguously be classified as regular or light,for which the previous study had empirically verifiedthe perceived vice-virtue characterization, and forwhich enough regular and light UPCs were offered tomeet the matching criterion. Due to a lack of obser-vations for nonalcoholic beer, we contrasted regularbeer jointly with light and nonalcoholic beer (availablefor 70 stores). The resulting data set had 574,217 ob-servations across the four categories.

Model Specification. We estimated the demandfor regular and light UPCs with the following OLS re-gression model:

ln(q) 4 a ` a R ` b ln(p) ` b R ln(p)Diff Diff

` Rc C ln(s) ` h ln(v) ` h ln(c)j j 1 2

` h ln(w) ` h D ` h F ` e, (3)3 4 5

where, for each UPC,q 4 unit sales per store and week in ounces,R 4 indicator for the regular subcategory (1 4 regular,0 4 light),p 4 price per ounce per store and week,Cj 4 category indicator,s 4 percent unit volume market share of each UPC iwithin category j across stores and weeks (100•qi/Rqi,v 4 total unit volume sold per store and categoryacross weeks,

c 4 total number of UPCs available per store, week,and regular or light subcategory,w 4 proportion of homes in individual store tradingarea with property values . $150,000,D 4 deal indicator for sale, bonus buy, or coupon pricepromotion per week,F 4 indicator for newspaper feature insert and/or in-store display per week, ande 4 error term (iid N ; 0, r2).

The intercept of the logged demand schedule forlight products is a, while aDiff is the difference in in-tercepts between the regular and light subcategories.Due to the conventional log-log demand specification,b represents the constant own-price elasticity of thelight subcategory, while bDiff is the price elasticity dif-ferential between regular and light, the parameter ofinterest. These parameters are adjusted for the effectsof several control variables Cj, s, v, c, w, D, and F thatcapture normative influences on demand. These mayco-vary with prices and create differential demand forlight and regular UPCs independently of self-control.Some of the control variables measure consumer andmarket characteristics (Cj • s, v, w); others control formanufacturer and/or retailer merchandising and pro-motion policies (c, D, F). Thus, c1 to h5 are scaling pa-rameters that convert a given value of these variablesinto a specific level of demand.

First, differences in demand may be category- andUPC-dependent due to variations in taste and otherproduct characteristics. We captured these demand id-iosyncrasies through the interaction of a category in-dicator with UPC volume market shares per categoryacross stores and weeks (Cj • s). This measured the im-pact of consumers’ strength of preference for each UPC(as an annual baseline relative to the other UPCs in thecategory), conditional on overall category demand andindependently of whether a UPC was light or regular.We used this measure instead of including separateUPC-specific intercept terms, which caused multicol-linearity. Second, effects of store-specific demand foreach category were controlled by including store vol-ume, computed as the total number of ounces sold percategory and store over the 52 weeks (v). Third, differ-ences in price elasticity should arise if there were dif-ferences in the numbers of perceived substitutes forlight and regular UPCs. Hence, we included the total



Table 6 Descriptive Statistics of Database in Field Study 2

Mean SD Minimum Maximum

Number of ounces sold per week and store 1,226 4,054 5 770,544Indicator for regular subcategory 0.54 0.50 0 1Price per ounce in cents 6.74 7.11 1.24 37.80Volume market share in category (%) 2.05 3.85 7.85 2 1014 55.12Total unit volume sold per store and category (in ounces) 4,108,574 2,591,980 19,916 10,015,848Number of competing UPCs (per store, week, and regular or light) 13.33 7.92 1 37Proportion of homes worth more than $150,000 0.35 0.24 2.51 2 1013 0.92Deal (sale, bonus buy, or coupon) 0.26 0.44 0 1Feature and/or in-store display 0.12 0.32 0 1

Note. N 4 574,217

number of UPCs on the shelf in each regular or lightsubcategory per store and week as a measure of theeffect of competitive intensity and level of substituta-bility on demand (c). Fourth, consumers’ strength ofpreference for light and regular may vary with theireducation and wealth. Therefore, we used the propor-tion of homes in each store trading area whose prop-erty values exceeded $150,000 as a proxy for aggregatewealth in the store trading area (w). Finally, marketingand promotional support may create differential de-mand for light and regular products. So we controlledfor the effects of weekly sale, bonus buy, and couponprice promotions (D) and of newspaper inserts and in-store displays (F) on demand by allowing the interceptto vary with changes in these promotional activities.

5.2. Results and DiscussionDescriptive statistics are shown in Table 6, while theOLS parameter estimates are in Table 7. There was nomulticollinearity between the variables. Tolerances,i.e., the proportion of variance in an independent vari-able that cannot be accounted for by a linear combi-nation of the other independent variables in the OLSmodel (1 1 R2), ranged from 12.2% for store volumeto 99.0% for store trading area wealth. The control vari-ables have normatively correct and expected signs.UPC-level demand was an increasing function of mar-ket share (c1 to c4), store volume (h1), customer wealth(h3), promotional activity (h4), and newspaper or in-store display advertising (h5), and a decreasing func-tion of competitive intensity (h2). For example, on av-erage and conditional on all other variables equaling

zero, a 1% increase (decrease) in within-category UPCmarket share, or relative preference, led to an increase(decrease) in demand of c1 4 .6% for beer.

Figure 4 and the top of Table 7 show the key result.Demand for regular products (relative vices) wasweaker than demand for light products (relative vir-tues) only at low unit prices (aDiff 4 10.0871). As pre-dicted, demand for regular products was also less pricesensitive (bDiff 4 0.0996), resulting in a crossover ofdemand as unit price rises. In other words, demandfor regular products was increasingly constrainedwhen price constraints were relaxed, although neithersubcategory was preferred over the other across theentire price range as there was no outward shift of de-mand (see Boulding et al. 1994). The constraint on vicedemand appears to be self-imposed, since several keynormative effects on demand were directly controlledas described above. Excluding beer from the analysis(because light and nonalcoholic beer were jointly clas-sified as a relative virtue, in contrast to the empiricalclassification in Field Study 1) led to comparableresults.

The price elasticity differential between light andregular products is consistent with our hypothesis ofself-control through purchase quantity rationing,given that the model adjusted aggregate demand forthe effects of these key control variables and light andregular UPCs were carefully matched. For example,because light and regular UPCs were matched in pack-age size, inventory requirements were identical forboth subcategories. Hence, the elasticity differential



Table 7 Demand Estimation for Light and Regular Products in Field Study 2

OLS Parameter Estimates (Includes Beer) SE

a(intercept) 12.3791 0.0358aDiff(Indicator for regular subcategory) 10.0871 0.0049b[ln(price)] 10.3408 0.0039bDiff[Indicator for regular subcategory 2 ln(price)] 0.0996 0.0029

c1 [beer indicator 2 ln(share)] 0.6000 0.0012c2 [cream cheese indicator 2 ln(share)] 0.5723 0.0027c3 [processed cheese indicator 2 ln(share)] 0.5621 0.0022c4 [soft drinks indicator 2 ln(share)] 0.7270 0.0013h1 [ln(store volume)] 0.6139 0.0021h2 [ln(number of competing UPCs)] 10.1377 0.0016h3 [ln(proportion of housing value . $150K)] 0.0276 0.0008h4 (deal indicator) 0.6806 0.0028h5 (feature indicator) 0.4680 0.0036R 2 0.69df 12N 568,487

Note. All estimates significant at p , 0.0001.

Figure 4 Predicted Store-Level Demand in Ounces for Regular and Light UPCs in Field Study 2

cannot be explained as a result of differences in inven-tory constraints, unless regular consumers had less in-ventory space available than light consumers. Thismight have been the case if light consumers were

wealthier and lived in larger homes, but this was con-trolled in the model by including housing value as acovariate. Overall, the results suggest the presence ofpurchase quantity rationing in actual retail markets



and are useful to retailers for segmentation purposes.We interpret these results in light of the convergentexperimental evidence provided above. Without suchevidence, it would be difficult to draw inferences aboutindividual consumer behavior from a demand modelthat is based on aggregate store-level data because con-struct operationalizations are necessarily impreciseand one can never rule out the presence of other, un-observed effects on demand.

6. General Discussion: Scope,Limitations, Future Research, andManagerial Implications

Consumers’ attempts to control their impulsive con-sumption influence many everyday purchase deci-sions. This research is the first to extend previous, theo-retical work on impulse control by empiricallyshowing its broader implications for marketing deci-sion making. The consistent finding across multiplemethods and data sources is a characteristic crossoverin demand schedules for vices and virtues. Vice de-mand increases less in response to price reductionsthan virtue demand, although consumers do not gen-erally prefer virtues over vices. Hence, inventory con-straints on vices appear self-imposed and strategicrather than driven by simple preferences. By restrictingtheir inventory of vices at the time of purchase, con-sumers can limit subsequent consumption. As a resultof purchase quantity rationing, vice buyers forgo sav-ings from price reductions through quantity discounts,effectively paying price premiums in order to engagein self-control.

Consumers may reason that purchase quantity ra-tioning imposes transactions costs on additional con-sumption once their inventory is depleted or they maytry to avoid feelings of guilt that may be associatedwith buying large amounts of vices. From a behavioralperspective, the reasoning on which the rationing de-cision is based is not relevant (Rachlin 1995). Thus, Ex-periment 1 induces purchase quantity rationing byvarying need for self-control indirectly, merely manip-ulating subjects’ perceptions of the intertemporal con-sequences of consumption without examining the cog-nitive processes underlying their behavior.Experiment 2 shows that the relationship between

these perceptions and rationing depends directly onsubjects’ impulsiveness and consequently on theirneed for self-control. Both experiments utilize realproducts and real choices to assess subjects’ demandunder controlled laboratory conditions, but are limitedto two single, nonaddictive product categories.

Field Studies 1 and 2, in turn, rely on statistical in-stead of experimental control to examine ecological va-lidity and managerial implications through aggregatemarket data from multiple categories. Field Study 1shows the implications of Experiment 1 by reversingthe dependent (quantity) and independent (price) vari-ables. Quantity discounts as a function of actual pack-age sizes offered by manufacturers are deeper for vicesthan for matched virtues. Field Study 2 builds on Ex-periment 2 by tracing out actual demand schedules forregular and light products faced by a large retailer.Consistent with the findings in Experiments 1 and 2,regular (vice) demand is less price sensitive than, andcrosses with, light (virtue) demand when other, nor-mative effects on demand are controlled. Neither fieldstudy allows for experimental control in a test of pur-chase quantity rationing. Both have to be interpretedin conjunction with the preceding experiments. How-ever, the consistent pattern of results across all studiessuggests strong convergent evidence of purchasequantity rationing in consumer markets.

The findings raise interesting follow-up questions.For example, does purchase quantity rationing indeedaffect consumption rates? The transactions costs of ad-ditional shopping trips, which rationing imposes onmarginal consumption, as well as recent findings oninventory effects on consumption suggest that ration-ing should reduce consumption rates (Folkes et al.1993; Wansink 1996). If it does, vice manufacturersmay want to counteract it. Second, how else do wecontrol our consumption? For example, can we resolvedynamic inconsistencies by manipulating our tastes(Gibbs 1998)? Some smokers seem to acquire a taste forlight cigarettes, perhaps to reduce an otherwise contin-ued temptation to smoke regular cigarettes. How canmarketers support this change in tastes? Third, howdo marketing mix variables affect purchase quantityrationing? Building on Experiment 1, research couldexamine ways to influence rationing through different



advertising message appeals or different promotionalvehicles. Perhaps the mere provision of a quantity dis-count makes the need for self-control more salient toconsumers because they can only obtain the cheaperprice if they buy more. So couponing and other formsof price promotion that are not explicitly directed atpackage size may be more effective in accelerating vicepurchases and consumption.

More generally, this research on purchase quantityrationing opens a whole class of strategic self-constraining behaviors to empirical investigation bymarketing researchers (cf. Wertenbroch and Carmon1997). The crossover of demand schedules for relativevices and virtues shown here provides a conceptualbasis for a test of self-control in any domain. For ex-ample, we can apply the analysis to consumer re-sponses to different pricing schedules and payment ve-hicles. Thus, self-controlling consumers may prefer topay for video rentals on a per-unit basis to make ex-cessive watching costly. In contrast, health clubsshould charge fixed up-front fees to minimize the costof the marginal workout. If fixed-fee video club or pay-as-you-go health club memberships became availableat cheaper average prices, hedonics might be less likelythan prudents to switch to these. Easily tempted he-donic shoppers may prefer carrying cash instead ofcredit cards (see Prelec and Loewenstein 1998), evenwhen they can pay off credit card debt in full at theend of the month. Constraining liquidity by carryingless cash than their available credit line may allowthese consumers to curb spending. Increasing theircredit line or lowering interest rates may not raise theirdemand for credit card usage as much as it might raisemore prudent shoppers’ demand. They willingly forgointerest earnings from credit card use (which theycould realize by paying off the balance at the end ofthe grace period), paying the price of self-control.

Our findings suggest that marketers can segmentand price discriminate based on consumer self-control.By offering a variety of package sizes, including rathersmall ones (e.g., M&Ms sold in vending machines),vice manufacturers can best appeal to both rationingand nonrationing consumers. For example, Michelobbeer is sold not only in 6 packs of 12-oz. bottles butalso of 7-oz. bottles. Similarly, cigarettes may well be

sold in packs of 10 in addition to packs of 20 and car-tons.5 Purchase quantity rationing allows marketers tocharge premium prices for small sizes of vices relativeto the corresponding quantity discounts for virtues. Onthe other hand, firms that want consumers to pantry-load vices to increase consumption (Wansink 1996) orpreempt competitors will find this costly. To moveconsumers beyond the self-imposed constraint on con-sumption and to induce them to build up larger in-ventories of vices, firms will have to offer deep quan-tity discounts. In contrast, consumers should find evenrelatively shallow quantity discounts sufficient to stockup on virtues. Field Study 1 shows that, at least to somedegree, some sellers follow these pricing strategies al-ready. For example, for the top 10 pairs with the stron-gest vice-virtue distinction shown in Table 4, consum-ers receive an average quantity discount of only 25.7%for doubling the purchase quantity of the virtue, com-pared to a much deeper 36.4% quantity discount fordoubling that of a corresponding vice. The price elas-ticity estimates in Field Study 2 provide additional evi-dence that smaller sizes of regular products can bepriced relatively higher per unit than those of lightproducts. Thus, the experimental- and market-level re-sults show that consumers’ attempts to control theirconsumption have important segmentation and pric-ing implications for many packaged goods sellers. If afirm can successfully price discriminate, it will be bet-ter off than if it cannot (Pigou 1948). Purchase quantityrationing provides an opportunity for doing just that,although public policy makers may, or perhapsshould, show concern about firms charging consumers

5Note again that the formal vice-virtue distinction is not absoluteand denotes merely which in a pair of goods is preferable when oneconsiders the immediate and delayed consumption consequencesseparately. As a managerial heuristic, however, it seems plausible tocategorize known impulsive or addictive goods such as cigarettes,alcohol, or candy and so forth as vices. More generally, Brian Gibbs(personal communication) suggests the following definition of ab-solute vices and virtues: Call W at price pW a vice if and only if wI .

pW and pW . wD, where wI is the immediately realized value of Wand wD is the delayed value. Call W at pW a virtue if and only if pW

. wI and wD . pW. This suggests that an effective retail tactic maybe to promote virtues by redressing their deficiency in immediatebenefits, e.g., by attaching small indulgences (a small of piece ofchocolate) to healthy but less tempting products (fat-free cheese) toincrease wI.



a premium for “virtuous,” or self-constrained, con-sumption behavior.6

ReferencesAinslie, G. 1975. Specious reward: A behavioral theory of impul-

siveness and impulse control. Psychological Bulletin 82 463–496.Benzion, U., A. Rapoport, J. Yagil. 1989. Discount rates inferred from

decisions: An experimental study. Management Sci. 35 (March)270–284.

Boulding, W., E. Lee, R. Staelin. 1994. Mastering the mix: Do adver-tising, promotion, and sales force activities lead to differentia-tion? J. Marketing Res. 31 (May) 159–172.

Dolan, R. J. 1987. Quantity discounts: Managerial issues and researchopportunities. Marketing Sci. 6 (Winter) 1–22.

Elster, J. 1984. Ulysses and the Sirens, 2nd ed. Cambridge UniversityPress, New York.

Folkes, V., I. Martin, K. Gupta. 1993. When to say when: Effects ofsupply on usage. J. Consumer Res. 20 (December) 467–477.

Freud, S. 1911, 1959. Formulations regarding the two principles ofmental functioning, in Collected Papers, Vol. 4. Basic Books, NewYork, 13–21.

Gibbs, B. J. 1998. Direction theory: Inward versus outward decisionmaking, chameleonic consumers, and the self-manipulation oftastes. Research Paper #1239R, Graduate School of Business,Stanford University, Stanford, CA.

Hoch, S. J., G. F. Loewenstein. 1991. Time-inconsistent preferencesand consumer self-control. J. Consumer Res. 17 (March) 492–507.

Kirby, K. N. 1997. Bidding on the future: Evidence against normativediscounting of delayed rewards. J. Experimental Psychology: Gen-eral 126 54–70.

Levin, I. P., G. J. Gaeth. 1988. How consumers are affected by theframing of attribute information before and after consumingthe product. J. Consumer. Res. 15 (December) 374–378.

6This paper has benefited from comments by Bill Boulding, BrianGibbs, Chip Heath, Wes Hutchinson as acting editor, John Lynch,Michael Moore, Rick Staelin, and particularly Steve Hoch and theanonymous reviewers.

Loewenstein, G. F. 1996. Out of control: Visceral influences on be-havior. Organizational Behavior and Human Decision Processes 3(March) 272–292.

——, J. Elster. 1992. Choice over Time, Russell Sage Foundation, NewYork.

Maddala, G. S. 1983. Limited Dependent and Qualitative Variables inEconometrics. Cambridge University Press, Cambridge, U.K.

Manning, W. G., N. Duan, W. H. Rogers. 1987. Monte Carlo evidenceon the choice between sample selection and two-part models.J. Econometrics 35 59–82.

Marshall, A. 1919. Industry and Trade, 2nd ed. Macmillan, London.O’Guinn, T. C., R. J. Faber. 1989. Compulsive buying: A phenome-

nological exploration. J. Consumer Res. 16 (September) 147–157.Pigou, A. C. 1948. Economics of Welfare, 4th ed. Macmillan, London.Prelec, D., G. F. Loewenstein. 1998. The red and the black: Mental

accounting of savings and debt. Marketing Sci. 17 4–28.Puri, R. 1996. Measuring and modifying consumer impulsiveness: A

cost-benefit accessibility framework. J. Consumer Psychology 587–113.

Rachlin, H. 1995. Self-control: Beyond commitment. Behavioral andBrain Sci. 18 109–121.

Rook, D. W., R. J. Fisher. 1995. Normative influences on impulsebuying behavior. J. Consumer Res. 22 (December) 305–313.

Schelling, T. C. 1984. Self-command in practice, in policy and in atheory of rational choice. Amer. Econom. Rev. 74 (May) 1–11.

Strotz, R. H. 1956. Myopia and inconsistency in dynamic utility max-imization. Rev. Econom. Studies 23 165–180.

Thaler, R. H. 1980. Toward a positive theory of consumer choice. J.Econom. Behavior and Organization 1 39–60.

——. 1985. Mental accounting and consumer choice. Marketing Sci.4 199–214.

——, E. J. Johnson. 1990. Gambling with the house money and tryingto break even: The effects of prior outcomes on risky choice.Management Sci. 36 (June) 643–660.

——, H. M. Shefrin. 1981. An economic theory of self-control. J. Po-litical Econom. 89 392–406.

Wansink, B. 1996. Can package size accelerate usage volume? J. Mar-keting 60 (July) 1–14.

Wertenbroch, K., Z. Carmon. 1997. Dynamic preference mainte-nance. Marketing Letters 8 145–152.

This paper was received December 15, 1996, and has been with the author 14 months for 2 revisions; processed by J. Wesley Hutchinson.

consumption self-control by rationing purchase quantities of ......marketing science/vol. 17, no. 4,...

Documents