Debabrata Talukdar & Charles Lindsey

To Buy or Not to Buy: Consumers'Demand Response Patterns forHealthy Versus Unhealthy Food

The authors integrate research on impulsivity from the psychology area with standard economic theories ofconsumer demand to make novel predictions about the effects of market price changes on consumers' foodconsumption behavior. The results from multiple studies confirm that consumers exhibit undesirable asymmetricpatterns of demand sensitivity to price changes for healthy and unhealthy food. For healthy food, demandsensitivity is greater for a price increase than for a price decrease. For unhealthy food, the opposite holds true. Theresearch further shows that the undesirable patterns are attenuated or magnified for key policy-relevant factors thathave been shown to decrease or increase impulsive purchase behavior, respectively. As the rising obesity trendbrings American consumers' food consumption behavior under increased scrutiny, the focal findings hold significantimplications for both public policy makers and food marketers.

Keywords: pricing, food, health, social networks, fear

Obesity has recently reached epidemic levels in theUnited States, and as a result, U.S. consumers' foodconsumption behavior has come under increasedscrutiny (Seiders and Petty 2004). The public health litera-ture suggests that a large percentage of food products pur-chased from grocery stores are unhealthy (Treuhaft andKarpyn 2010). Thus, a natural public policy imperative is toidentify avenues that may nudge consumers from the con-sumption of unhealthy food choices to healthier foodoptions (Thaler and Sunstein 2008). A potential avenue foraltering the composition of consumers' food baskets thathas received increased attention of late is the link betweenfood prices and consumers' food consumption behavior(Drewnowski and Darmon 2005). In particular, emphasishas been on the role of economic policy tools, in the formof surcharges and subsidies on food products, to promotehealthier food consumption behavior (Brownell and Frieden2009). An important prerequisite to effective design of sucheconomic policy tools is a clear understanding of the under-lying nature of consumers' demand response pattems toprice changes for healthy and unhealthy foods (Epstein etal. 2010).

In this context, imagine the following typical marketsetting for a hi-Io grocer. Assume that the price for a healthyand an unhealthy food item is $1 each in March, $.90 inApril, and $1 again in May. In addition, assume that the ini-

Debabrata Talukdar is Professor of Marketing (e-mail: dtalukda@buffalo.edu), and Charles Lindsey is Assistant Professor of Marketing (e-mail:clindsey@buffalo.edu), School of Management, State University of NewYork at Buffalo. Both authors contributed equally to this work. The authorsthank the three anonymous JM reviewers for their invaluable feedbackduring the review process. Barbara Kahn served as area editor for thisarticle.

tial quantity demanded in March is 100 units for each item.On the one hand, if the underlying nature of consumers'demand response pattem to price changes is symmetric(say, with a price elasticity of -.75 for a price increase/decrease for each item), the quantity demanded for eachitem in April and May will be 107.5 units and 99.99 units,respectively. On the other hand, if the underlying nature ofconsumers' demand response pattem to price changes isasymmetric (say, with a price elasticity of -.5 for a priceincrease and -1 for a price decrease for the healthy item, and-1 for a price increase and -.5 for a price decrease for theunhealthy item), the quantity demanded for the healthy itemin April and May will be 105 units and 93.45 units, respec-tively; the quantity demanded for the unhealthy item in Apriland May will be 110 units and 103.95 units, respectively.

Note that in the preceding example, the average priceelasticity values for both types of food items are the samebetween and across the symmetric and asymmetric scenar-ios, and yet the final quantities demanded are substantiallydifferent between the symmetric versus asymmetric scenar-ios for each food item. Equally noteworthy are the contrast-ing final demand quantities between the two food itemsunder the asymmetric scenario. Specifically, even thoughthe price in May is back to its initial value, quantitydemanded is lower than the original quantity demanded forthe healthy item but higher than the original quantitydemanded for the unhealthy item. In other words, from apublic health perspective, this illustrates a case of undesir-able asymmetry in consumers' demand response pattems.

The previous example highlights how a clear under-standing of the underlying nature of demand response pat-tems to price changeswhether they are symmetric orasymmetric and the specific direction of asymmetryisessential to evaluating how such price changes will affect

2013, American Marketing AssociationISSN: 0022-2429 (print), 1547-7165 (eiectronic) 124

Journal of MarketingVoiume 77 (iVIarch 2013), 124-138

consumers' food consumption behavior. Unfortunately,despite the recent research focus on food consumption inthe face of rising obesity, there is still a paucity of researchthat investigates demand response pattems to price changesfor both healthy and unhealthy food, employing groundedtheory and using primary and secondary data sourcesthereby limiting the current level of understanding fordeveloping effective economic policy tools to promotehealthier food consumption (Andreyeva, Long, andBrownell 2010; Epstein et al. 2010).

The current article integrates impulsivity research frompsychology with standard economic theories to make novelpredictions about the effect of price changes on demandresponse sensitivity for healthy and unhealthy food. We flrsttest these predictions by using extensive supermarket scan-ner data, consisting of consumers' demand responses toprice increases and decreases across multiple product cate-gories. A unique prediction and finding is that not only doconsumers exhibit asymmetric pattems of demand sensitiv-ity to price changes for both healthy and unhealthy food,but they do so in opposite and undesirable directions.Specifically, demand sensitivity for healthy food is greaterfor a price increase than for a price decrease, whereas thepattem is reversed for unhealthy food. An experimental sur-vey and a controlled experiment conceptually replicate thekey results from the supermarket study.

From a public policy perspective, the preceding findingimplies that the effectiveness of economic interventionsin inducing healthier food consumptionthrough priceincreases (decreases) from surcharges (subsidies) on un-healthy (healthy) foodwill be much more limited thantypically projected, on the basis of the default but misplacedpremise of symmetric pattems of demand response in thecurrent public policy discourse. From a food marketing per-spective, the preceding finding sheds important new lighton the relative efficacy of price promotions for healthy ver-sus unhealthy food. For example, demand for a healthyfood item is likely to receive a more limited boost thanexpected from a price promotion and may actually fallbelow its baseline demand level after the promotional pricereverts to its regular price.

Finally, our series of studies further predicts and showsthat the undesirable asymmetric demand response pattemsare moderated by several policy-relevant variables, whichhave been shown to either attenuate or exacerbate impulsivepurchase behavior. Perhaps the most notable one, from bothfood marketing and public policy perspectives, is that fooddemand response pattems may be nudged in a healthierdirection by merely priming participants to think about theenabling peer support aspects of social networks or byexposing participants to specifically constructed social net-work appeals. Next, we discuss our conceptual framework,followed by a delineation of the associated hypotheses.

Conceptual FrameworkRational DemandThe economic theory underlying consumers' demandresponse to price changes is well established and follows

the so-called law of demand (Marshall 1895). The lawstates that, except for a few products called "Giffen goods,"for which demand increases in response to a price increase,there is an inverse relationship in the directionality betweenprice change for products and consumers' demand response(Jensen and Miller 2008). This law implies that the rationalresponse for consumers is to demand less of a productwhenever its price increases, and vice versa. A fundamentalconcept related to the law is the notion of price elasticity,which measures the sensitivity of consumers' demandresponse to price changes (i.e., the percentage change indemand for a 1% change in price).

A typical premise in studies of price elasticity, includingthose focusing on food items, is that while consumers'demand response exhibits a negative relationship with pricechange, the sensitivity of such response is identical or sym-metric with respect to the directionality of the price change(Andreyeva, Long, and Brownell 2010; Bijmolt, VanHeerde, and Pieters 2005; Epstein et al. 2010). In otherwords, while the price elasticity value will always have anegative sign, the value itself will remain the same irrespec-tive of a price increase or decrease. However, there isempirical evidence from studies in the marketing area that,for some consumer packaged goods, consumers mayexhibit an asymmetric pattem of demand response sensitiv-ity with respect to the directionality of the price changes(Johnson and Meyer 1994). In the context of the presentstudy, this raises the following important questions: Shouldwe expect an asymmetric pattem of demand sensitivity toprice changes for common food items? If so, should suchpattems be the same for healthy and unhealthy foods? Ifnot, why and how should we expect them to differ? We nextdiscuss relevant theoretical perspectives fi^ om the psychol-ogy area, which, when integrated with those from the eco-nomics area, help address these questions.

Impulsive DemandPrior research in consumer psychology has shown that con-sumers exhibit natural consumption tendencies for bothunhealthy and healthy food, but in opposite directionsanoverconsumption impulse for unhealthy food and an under-consumption impulse for healthy food (e.g., Finkelstein andFishbach 2010; Loewenstein 1996; Raghunathan, Naylor,and Hoyer 2006; Wansink and Huckabee 2005). The under-lying reason for these natural impulses reflects fundamentalperceptual and sensory differences in palatability forunhealthy and healthy food. For example, researchers havefound systematic evidence suggesting that people operateunder the implicit intuition that "unhealthy food = tasty"(Raghunathan, Naylor, and Hoyer 2006, p. 170). In con-trast, healthy foods are intuitively assumed to be bland.Several studies have explored the sensory basis for suchperceptions and have demonstrated that unhealthy food istypically considered more palatable than healthy food, withpalatability ratings being positively correlated to fat contentand macronutrient intake (Stubbs and Whybrow 2004).

Researchers posit that these natural over- and undercon-sumption impulses for unhealthy and healthy foods, andtheir perceptual and sensory drivers, have an evolutionary

To Buy or Not to Buy/125

basis. For example, Wansink and Huckabee (2005, p. 8)note that "fatty foods helped our ancestors weather foodshortages,... [and] sugar and the sweetness associated withit helped them distinguish edible berries from poisonousones." Furthermore, studies show that energy-dense (i.e.,unhealthy) foods, while worse for long-term health, are bet-ter than non-energy-dense foods in providing short-termenergy stores, which human beings have been hardwired tofavor (Ostan et al. 2010).Exacerbating and Mitigating Factors for ImpuisiveDemandThe previously discussed natural impulses to overconsumeunhealthy food and underconsume healthy food runs counterto the desire of "most people to cherish long and healthylives" (Thomas, Desai, and Seenivasan 2011, p. 127). Thistension highlights the inherent conflict between the impul-sive and reflective or regulatory behavioral systems(Loewenstein 1996; Metcalfe and Mischel 1999), mostrecently denoted as Systems 1 and 2 by Kahneman (2011).System 1 is present oriented and driven by emotion anddesire, while System 2 is future based and driven by cogni-tion and willpower. For example, "urges to consume ...junk food occur in System 1 and lead to impulsive behaviorwhen System 2 is not able to control System 1" (Lades2012, p. 5). A key finding in this research domain is that thedegree of impulsive behavior for a person in any specificdecision scenario is contingent on the level of willpowerthat System 2 exerts on System 1 (Loewenstein 1996).

Thus, factors that have been shown to moderate reflec-tive or regulative behavior can be expected to influence theintensity of consumers' natural impulses to overconsumeunhealthy food and underconsume healthy food. Such fac-tors may be attributable to individual differences or be con-textual in origin. For example, research has shown thatincome and age can influence regulatory behavior and thusimpulsive consumption (Killgore and Yurgelun-Todd 2005).Recent research has also shown that the method of paymentat the time of a purchase can make it either viscerally moredifflcult to resist or easier to give in to impulsive urges(Thomas, Desai, and Seenivasan 2011).Summary of Key ElementsOur focus is on demand response to price changes in themarketplace for healthy and unhealthy food. Taking intoaccount only rational economic forces, demand responseshould have a negative relationship with price change, butthe sensitivity of such response should be identical or sym-metric in terms of directionality of the change (Andreyeva,Long, and Brownell 2010; Bijmolt, Van Heerde, and Pieters2005; Epstein et al. 2010). However, consumers are alsosubject to psychological impulses when it comes to foodconsumption behaviorto overconsume unhealthy foodand underconsume healthy food (e.g., Finkelstein and Fish-bach 2010). In addition, given the presence of dual or com-peting (impulsive vs. regulatory) behavioral systems (Kah-neman 2011; Loewenstein 1996; Metcalfe and Mischel1999), factors that influence willpower can exacerbate ormitigate these natural consumption impulses. We next inte-

grate the aforementioned three key elements of our concep-tual framework to develop a set of empirically testablehypotheses in terms of consumers' demand response pat-tems to price changes for healthy and unhealthy foods.

HypothesesMain Asymmetric EffectAs we noted previously, people's sensory- and perceptual-based preferences for unhealthy versus healthy food resultin an impulsive tendency to overconsume the former andunderconsume the latter (e.g., Hausman 2012). When thisimpulsive behavioral tendency is considered in light of therational economic behavioral tendency as per the law ofdemand, the following suppositions unfold. For healthyfood, the rational economic response to decrease quantitydemanded for a price increase is reinforced by the naturalimpulse to underconsume. Moreover, the rational economicresponse to increase quantity demanded for a price decreaseis counteracted by the natural impulse to underconsume.Consequently, for healthy food, we expect the rationaldemand response to reduce quantity when faced with aprice increase to be stronger than the rational demandresponse to increase quantity when faced with a pricedecrease. In contrast, because of the natural impulse tooverconsume unhealthy food, analogous reasoning impliesthe opposite outcome for unhealthy food.

In this context, it is pertinent to note that prior researchhas indicated that consumers perceive price changes in themarketplace in relation to some intemal reference prices(IRPs; Mazumdar, Raj, and Sinha 2005). For example, asWiner (1988, p. 35) states, "defining p" to be the observedretail price and p' to be the individual's intemal referenceprice, the underlying assumption of this (behavioral pric-ing) literature is that positive values of (p" - p') are per-ceived negatively ... while negative values of (p" - p') areviewed positively." Previous research has measured IRPeither through the most recent price at which a product waspurchased or through an extrapolative model based on agiven number of past prices encountered (Janiszewski andLichtenstein 1999). In our study, we operationalize IRP interms of the most recent or last purchase price.

The preceding discussion implies that for unhealthy(healthy) food, a given percentage increase in price relativeto the last purchase price will elicit a smaller (larger)decrease in demand quantity from consumers than theincrease in demand quantity triggered by the same percent-age decrease in price. In other words, although the quantitythat consumers demand for both unhealthy and healthy foodshould rise or fall as price decreases or increases (as per thefirst law of demand), we expect each food type to exhibit anasymmetric pattem of demand sensitivity to price changes,but in opposite directions. Stated formally:

H,: For unhealthy food, consumers' demand response sensi-tivity is greater for a price decrease than tliat for a priceincrease, relative to the last purchase price.

H2: For iiealthy food, consumers' demand response sensitivityis greater for a price increase than that for a pricedecrease, reiative to the last purchase price.

126 / Journal of Marketing, March 2013

Exacerbating FactorsNaturally, any factor that exacerbates consumers' impulsivefood consumption behavior will accentuate the relativeeffect of impulsive demand on rational demand, therebymagnifying the main asymmetric patterns in demandresponse to price changes noted previously. Here, we dis-cuss three specific exacerbating factors highlighted inimpulsivity research and delineate the related hypotheses.

Bank card usage. When consumers pay with a creditcard, the pleasure derived from acquiring goods is tempo-rally divorced from the "pain of payment," making it easierto give in to impulsive urges (Soman and Gourville 2001).In terms of food choice, recent research has shown that thepain of paying in cash decreases the likelihood that con-sumers will purchase vice foods, whereas paying with abank card (credit and debit) increases the likelihood of suchpurchases (Thomas, Desai, and Seenivasan 2011). The rea-son cited for this is that both credit and debit card paymentsare less "vivid and emotionally more inert" than paying incash and viscerally less painful. Thus, bank card purchasesreinforce consumers' impulsive tendency for overconsump-tion already present for unhealthy foods. The upshot will bemore resistance against the rational economic response todecrease quantity demanded for a price increase but a re-inforcement of the rational response to increase quantitydemanded for a price decrease. Stated formally:

H3: The undesirable asymmetdc pattern of demand responsesensitivity for unhealthy food is magnified for bank cardpurchases by consumers.

Younger consumers. Research has shown that the ten-dency to make impulsive food selections is greater for chil-dren, teenagers, and young adults than for older people(Killgore and Yurgelun-Todd 2005). One reason for suchage-based differences in food preference is that the regionof the brain that governs impulsive and risky behavior, thelimbic system, is hypersensitive during adolescence,whereas the prefrontal cortex, the rational part of the brain,continues to develop throughout the third decade of life(Dumontheil et al. 2010). In the context of food choice,lower age will thus accentuate the relative effect of impul-sive demand (i.e., to overconsume unhealthy and undercon-sume healthy food) on rational demand, leading to morepronounced asymmetric patterns of demand response forunhealthy and healthy food for younger consumers. Statedformally:

H4: The undesirable asymmetdc pattern of demand responsesensitivity is magnified for younger (vs. older) consumers,with regard to (a) unhealthy food and (b) healthy food.

Lower-income consumers. From the famous StanfordUniversity marshmallow study (Mischel et al. 1989), whichtested the ability of 600 hungry four- to six-year-old chil-dren to delay gratification (and then tracked them as adults)to more methodologically rigorous neurological studies(e.g., Matthews et al. 2000), research has shown an inversecorrelation between impulsiveness and income. Thus, wehypothesize that the relative effect of impulsive demand(i.e., to overconsume unhealthy and underconsume healthy

food) on rational demand will be higher for lower-incomepeople. Furthermore, low-income populations have fewer"slack" resources (Bertrand, Mullainathan, and Shafir2006), and healthy food is more expensive than unhealthyfood on a per-calorie basis (see Drewnowski and Darmon2005). This implies that the natural impulse to undercon-sume healthy foods will be reinforced even more for lower-income consumers. Thus, the negative link between incomeand impulsivity as well as the reality of a tighter resourceconstraint versus food cost will further accentuate the asym-metric patterns of demand response for unhealthy andhealthy food for lower income consumers. Stated formally:

H5: The undesirable asymmetric pattern of demand responsesensitivity is magnified for lower- (vs. higher-) income con-sumers, with regard to (a) unhealthy food and (b) healthyfood.

Mitigating FactorsAny factor that enhances consumers' willpower shouldreduce the relative impact of impulsive demand on rationaldemandthereby mitigating the main asymmetric patterns indemand response to price changes noted in Hj and H2. Here,we discuss two specific mitigating factors that have beenhighlighted in impulsivity research and in a key explanatoryhealth behavior model (protection motivation theory; Rogers1983) and offer corresponding hypotheses. Specifically, pro-tection motivation theory identifies four variables critical tothe success of any health behavior change (Witte 1992). Thefirst two variables, perceived susceptibility/vulnerabilityand severity, involve threat appraisal, and the second twovariables, self-efficacy and response-efficacy, involve cop-ing appraisal. Both sets of variables have been shown toregulate impulsive behavior. We discuss each in turn.

Fear. The first set of variables advanced here, "threatappraisal," is defined as a person's assessment of the levelof his or her own personal vulnerability and susceptibilityto experiencing a particular negative outcome (Boer andSeydel 1996) and is typically activated through a viscerallyaversive cue (Witte 1992). Prior research has suggested thatviscerally aversive cues can be effective in the regulation ofvice-related products, which tend to be impulsively pur-chased and consumed (Metcalfe and Mischel 1999;Thomas, Desai, and Seenivasan 2011). One such viscerallyaversive cue is fear (Witte 1992). "Fear salience" is definedas the momentary increase in a person's awareness of beingafraid of a particular activity or event (Gore et al. 1998). Acredible fear cue leads to increased threat appraisal percep-tions (Witte and Allen 2000). This has important implica-tions in the food domain, in that the short-term health-related costs of eating unhealthy are not always obvious(Schlosser 2001). Thus, the use of a credible fear cueshould increase threat appraisal perceptions surroundingunhealthy eating.

When threat appraisal perceptions increase (and self-efficacy is adequate), people are more likely to adopt adap-tive health protection behaviors (Witte and Allen 2000).Given that self-efflcacy perceptions related to nutrition areat moderate levels (e.g., Bartfield et al. 2010), the use of afear cue should lead to positive and adaptive healthy behav-

To Buy or Not to Buy /127

ior. Thus, threat appraisal perceptions (through fear) shouldcause consumers to muster more willpower to resist thenatural impulsive urge to overconsume (underconsume)unhealthy (healthy) food. Even though the use of fear isarguably the most widely used public service announce-ment archetype (Witte and Allen 2000), little research hasexamined its effect on food selection.

Peers (social networks). The second set of variablesadvanced here, coping appraisal (e.g., efficacy), is fre-quently activated through peer or social support (Brissette,Scheier, and Carver 2002). While the negative effects ofsocial influence on health outcomes such as obesity, ciga-rette smoking, and drinking have been well documented(ChristakJs and Fowler 2008), recent research has begun toexamine the positive effects of social influence on such out-comes through the modiflcation of people's social networks(Umberson, Crosnoe, and Reczek 2010). A social networkis deflned as a "social structure made up of a set of actors(such as individuals or organizations)" (Wasserman andFaust 1994, p. 2). Social networks can be either positive ornegative, depending on the type of influence they wield(i.e., good or bad). Such modifications can take the form of"smoking- and alcohol-cessation programs and weight-lossinterventions that provide peer supportthat is, that modifythe person's social network" (Christakis and Fowler 2008.p. 377). Prior research has demonstrated that people whohave access to supportive social networks show a greaterlikelihood of engaging in social support-based coping,through potential interactions with groups and people witha like-minded or common objective (Umberson, Crosnoe,and Reczek 2010).

Such support-based coping has been positively linked toincreased feelings of self-efficacy (Brissette, Scheier, andCarver 2002). Self-efficacy is defined as "the belief in one'sability to execute a recommended course of action success-fully" (Boer and Seydel 1996, p. 98). In particular, self-efficacy has been shown to be one of the most impactfulpredictors of healthy eating (Luszczynska, Tryburcy, andSchwarzer 2007). Indeed, "enhancing self-efficacy resultsin nutritional change" (Luszczynska, Tryburcy, andSchwarzer 2007, p. 630). Thus, enhanced feelings of self-efficacy (induced by social network salience) should lead toincreased confidence in a person's ability to regulate ordecrease consumption of unhealthy foods and increase con-sumption of healthy foods. Notably, although several stud-ies have examined the effect of social network membershipor modification (e.g., Christakis and Fowler 2008) on healthbehaviors, no research to date has examined the effect ofmerely making social networks salient on health behavior.The preceding discussion on fear and peer sources leads tothe following predictions:

H^ a: The undesirable asymmetric pattem of demand responsesensitivity is attenuated for consumers when fear is madesalient, with regard to (i) unhealthy food and (ii) healthyfood.

Hgt,: The undesirable asymmetric pattem of demand responsesensitivity is attenuated for consumers when positivesocial networks are made salient, with regard to (i)unhealthy food and (ii) healthy food.

H7a: The effect of making fear salient on demand responsesensitivity for food is mediated by consumers' threatappraisal perceptions.

H7b: The effect of making positive social networks salient ondemand response sensitivity for food is mediated by con-sumers' self-efficacy perceptions.

We next present our first empirical study, which usessupermarket scanner data. The nature of the scanner dataallows us to test the fu-st three hypotheses: the main asym-metric effects and the first exacerbating factor (mode ofpayment). We subsequently conduct two other studies thatuse primary data to test our other hypotheses.

Supermarket Scanner Data: Study 1DataThe supermarket study uses individual household-leveltransaction scanner data in eight food categories, from onestore of a large regional supermarket chain in the northeast-em United States. The transaction data cover 52 consecu-tive weeks during 2003-2004. The categories included fourrelatively healthy (fresh broccoli, grapes, raisins, andwholegrain bread) and four relatively unhealthy (fresh non-lean beef, potato chips, nondiet soft drink, and white bread)food categories. These categories are generally consideredrelatively unhealthy or healthy in the existing literature(e.g., Martikainen, Brunner, and Marmot 2003). Healthyfoods are deflned as those that are "low [in] fat,... low [in]saturated fat,... and contain at least 10% of daily value ... forvitamins A, C, calcium, iron, protien or fiber," and are lim-ited in amount of sodium and cholesterol (U.S. Food andDrug Administration 2012). Unhealthy food is defined asthose foods not meeting these standards. Nonetheless, it isrelevant to note that our conceptual framework is based onperceived or subjective rather than objective groupings ofhealthy and unhealthy food categories. At the same time, asmight be expected, groupings based on subjective percep-tions and objective criteria are highly correlated in thisdomain (e.g., Stubbs and Whybrow 2004).

For each category, we obtained the sample of house-holds from the participating store's customer database, andeach sample household met the following criteria: (1) Itused the participating store as its primary store, and (2) itmade at least one purchase in the category each month dur-ing the year. Table 1 presents key sample descriptive statis-tics. In addition to category-speciflc transaction-level datafor each sample household, we had access to infonnation onthe mode of payment (e.g., bank card, cash), householdsize, and census block group code of the household's resi-dence. In absence of direct income information for eachhousehold, we used mean household income at the censusblock group level to capture relative income levels for oursample households (Talukdar, Gauri, and Grewal 2010).Finally, for our participating store, which has a weeklyprice cycle, we also had information on the overall storeprice index (OSPI) on a weekly basis. This index is a sales-share-weighted composite price covering top-selling items(using stockkeeping units [SKUs]) across key product cate-

128 / Journal of Marketing, March 2013

TABLE 1Study 1 : Scanner Data Summary Statistics

A: Healthy Food Categories

Descriptive Variable Broccoli Grapes Raisins Whole-Grain Bread

Number of all households% of single householdsMean number of transactions per householdTotal number of transactionsNumber of SKUs involved in the transactions% of households using bank card at least once% of transactions on bank card% of transactions with same last purchase price% of transactions with higher last purchase price% of transactions with lower last purchase price

9933122

21,8461

10048423028

10292921

21,6092

10051442927

11072927

29,8898

10050432829

10943031

33,9141310052453025

B: Unhealthy Food Categories

Descriptive Variable Beef Potato Chips Soft Drinks White Bread

Number of all households% of single householdsMean number of transactions per householdTotal number of transactionsNumber of SKUs involved in the transactions% of households using bank card at least once% of transactions on bank card% of transactions with same last purchase price% of transactions with higher last purchase price% of transactions with lower last purchase price

10382834

35,2928

10052453025

10593034

36,0061610049422830

11272742

47,3342310047433126

10983133

36,2341410051443026

Notes: The SKUs for beef used in the study exclude all fresh beef labeled as "lean," and the SKUs for soft drinks used in the study exclude allsoft drinks labeled as "diet."

gories and reflects relative differences in overall store pricepromotion intensity across weekly price cycles.

AnaiysisFor each product category, we analyzed the data using ahousehold-level random-effects regression model in log-logform (for the model, see Table 2). The price movement indi-cator (PMI) variable captures the three cases of price move-ment in terms of the current price paid compared with thelast purchase price paid by a household: (1) no change, (2)price went up, and (3) price went down. As in many exist-ing studies (Mazumdar, Raj, and Sinha 2005), we used thelast purchase price to operationalize consumers' IRP.Specifically, we measured the last purchase price as theprice paid per relevant standardized unit (PR) (e.g., ounces,pounds) on the last purchase occasion by a household in agiven category.

It is relevant to note here that although IRP is oftenmodeled at the brand level (e.g., Winer 1986), it is alsomodeled at the category level (Mazumdar, Raj, and Sinha2005). The latter approach is used in this study and is justi-fied in our context for two reasons. First, the brand-levelapproach is not applicable in nonbranded categoriessuchas the fresh broccoli, grapes, and beef used in our study.Second, even for branded categories, the category-levelapproach is especially reasonable when relatively smallprice differentials exist across brands such that the cognitivecost of attempting to retain price data for various brands isperceived to be greater than the benefits derived (Mazum-dar, Raj, and Sinha 2005). In our one-year data for the five

branded categories, the differentials in mean prices amongthe most purchased brands (i.e., those accounting for 90%or more of category sales) in each category ranged betweenonly 1% and 3% of the mean category price.

We used the bank card usage indicator (BCUI) variableto capture the following four distinct situations of bank cardusage: (1) not used, (2) used when price did not change, (3)used when price went up, and (4) used when price wentdown. In addition to our three focal independent variables(PR, PMI and BCUI), we also used several relevant controlvariables-BCCI, LQ, TLP, HS, HI, WD, and OSPI (fordefinitions of these variables, see Table 2). We used theBCCI control variable only for the analysis in the brandedcategories to control for individual-level differences in brandloyalty propensity within a given category. It is measured asa Herfindahl concentration index in terms of a household'schoice shares across the brands in a category (Sudhir andTalukdar 2004). A higher value of the index reflects ahigher level of brand loyalty in the household's choicebehavior. In addition, for white bread and whole-grainbread, which are close substitute product categories and forwhich we have price data, we explicitly use respectivecross-category prices to control for any direct substitutioneffect between these two categories.

The household- and trip-specific control variables captureobserved heterogeneity across households, and the random-effects term helps our regression model account for unob-served household heterogeneity. For each category, we esti-mate the regression model using the generalized estimatingequation technique with our transaction panel data across

To Buy or Not to Buy /129

TABLE 2Study 1 : Regression Model for Each Product Category

Variable DefinitionQj Quantity purchased in category relevant standardized unit (e.g., oz. or Ib.) by household i on purchase occasion jPR Price paid per relevant standardized unitPMI Price movement indicatorBCUl Bank card usage indicatorBCCI Brand choice concentration index (only for a branded category)LQ Quantity purchased in relevant standardized unit on the last purchase occasionTLP Time (in days) since the last purchase occasionHS Househojd sizeHI Household income levelWD Whether the current purchase occurred on a weekendOSPI Overall store price index on the current purchase occasionM-i Household-specific random effectsEij Random error termo ... io Model coefficients, where i estimates consumers' base level of own price elasticity; g estimates how the base

level elasticity is affected by the price movement direction since the last purchase occasion; and % estimates howthe base level elasticity is affected by the price movement direction and by its interaction with bank card usage

households (Liang and Zeger 1986). The technique allowsfor heteroskedasticity as well as clustering across house-holds. The clustering enables within-household serial corre-lation in the error term to account for the expected noninde-pendence in within-household observations across differentpurchase occasions. The log-log functional form of ourregression model is consistent with the dominant approachin the literature for empirically estimating a consumerdemand function as a multiplicative model, especially whenthe primary focus is on estimating consumers' price elastic-ity (Bijmolt, Van Heerde, and Pieters 2005). The log-logfunctional form of our regression model means that theparameters, |-3, are of focal interest to us because theycapture consumers' demand response sensitivities to priceand bank card usage. Per the first law of demand, we expecti < 0that is, own price elasticity should be negative. Theparameters 2 and 3 test H,, H2, and H3. For example, anegative (positive) sign for 2 indicates that a specificdirection of price movement since last purchase increases(decreases) consumers' demand sensitivity to price.ResuitsTable 3 presents the results of our regression analyses. Thecoefficient of current price paid in our model estimates thebase level of consumers' own price elasticity, and it is nega-tive and significant {p < .01) for all eight categories. Next,the coefficient of the indicator for price change since lastpurchase is negative and significant (p < .01) when priceswent up but insignificant (/? > .1) when prices went downfor all the healthy food categories. In contrast, for all theunhealthy food categories, the coefficient is negative andsignificant (p < .01) when prices went down but insignifi-cant (/?>.!) when prices went up. Thus, for healthy food,consumers' demand sensitivity (i.e., absolute value of ownprice elasticity) exhibits a greater relative increase inresponse to a price increase than a price decrease since last

purchase. However, we found the opposite for unhealthyfoods. Thus, Hj and H2 are supported.

As for the coefficient of the indicator for bank cardusage, we found it to be insignificant (p > .1) for the healthyfood categories. In contrast, for the unhealthy food cate-gories, we found the coefficient to be positive and signifi-cant (/? < .01) when bank cards are used in conjunction witha price increase since last purchase and negative and sig-nificant ip < .01) when bank cards are used in conjunctionwith a price decrease since last purchase. Thus, for theunhealthy food categories, consumers' demand sensitivity isaccentuated when bank cards are used in conjunction with aprice decrease since last purchase but mitigated when bankcards are used in conjunction with a price increase since lastpurchase. Thus, H3 is supported.

Finally, although we advanced no formal predictionswith regard to the control variables, we briefly note the fol-lowing findings. The results show that the coefficients of thecross-category prices between white and whole-grain breadsare positive and significant (p < .05 or less), indicating theexpected demand substitution effects between the two cate-gories. In addition, for each of the branded categories, thecoefficient of the brand choice concentration index is posi-tive and significant (p < .05 or less). This finding is againconsistent with the expectation that a higher level of a con-sumer's brand-loyal behavior in a category will lower theabsolute value of his or her price elasticity for the category.

DiscussionThe supermarket study confirms our predictions regardingasymmetric and opposite pattems of demand response forunhealthy and healthy food. From a nutritional standpoint,the asymmetries in demand response are in undesirabledirections for both types of food. Thus, an important nextstep is to understand the various factors that might mitigateor accentuate the observed asymmetric pattems. Towardthis end, the supermarket study already demonstrates that

130 / Journal of Marketing, March 2013

TABLE 3Study 1 : Scanner Data-Based Regression Analysis Resuits for Quantity Purchased

A: Healthy Food Categories

Coefficient Estimate (SE)

Independent Variables

Current price paidIndicators for Price Change Since Last Purchase

Remained same (base)Went upWent down

Indicators for Bank Card Usage ContextNo usage (base)Used and price remained sameUsed and price went upUsed and price went down

Brand choice concentration indexCurrent price for white breadLast purchase quantityTime since last purchaseHousehold sizeHousehold incomeIf purchased on weekendOverall store price index in current purchase weekInterceptNPseudo R2

Broccoli

-.593 (.113)**

-.384 (.077)**-.107 (.087)

.026 (.027)

.019 (.018)-.008 (.006)

N.A.N.A.

-.119 (.060)*.188 (.091)*.114 (.075).271 (.094)**.009 (.010)

-.441 (.127)**.182 (.021)**

20,853.64

Grapes

-.624 (.133)**

-.297 (.056)**-.134 (.101)

.

.012 (.011)

.016 (.012)-.011 (.010)

N.A.N.A.

-.126 (.062)*.133 (.067)*.101 (.078).212 (.081)**.109 (.107)

-.204 (.055)**.095 (.010)**

20,580.69

Raisins

-.609

-.326-.121

.029

.021-.006

.044N

-.114.098.076.186.008

-.164.326

28

(.125)**

(.071)**(.098)

.

(.033)(017)(.005)(.020)*

I.A.(.052)*(.048)*(.049)(.066)**(.006)(.057)**(.038)**

,78272

Whole-GrainBread

-.686

-.298-.124

.038

.023-.002

.051

.127-.164

.116

.131

.242

.002-.184

.37632

(.098)**

(.052)**(.097)

(.042)(.017)(.002)(.025)*(.063)*(.059)**(.045)**(.050)**(.108)*(.002)(.059)**(.040)**

,82077

B: Unhealthy Food Categories

Coefficient Estimate (SE)Independent Variables

Current price paidIndicators for Price Change Since Last Purchase

Remained same (base)Went upWent down

Indicators for Bank Card Usage ContextNo usage (base)Used and price remained sameUsed and price went upUsed and price went down

Brand choice concentration indexCurrent price for whole-grain breadLast purchase quantityTime since last purchaseHousehold sizeHousehold incomeIf purchased on weekendOverall store price index in current purchase weekInterceptNPseudo R2

Beef

-.681 (.123)**

-.012 (.013)-.329 (.059)***

.019 (.016)

.221 (.083)**-.276 (.092)**

N.A.N.A.

-.196 (.076)**.242 (.118)*.438 (.121)**.225 (.075)**.184 (.068)**

-.152 (.073)*.474 (.044)**

34,254.73

Potato Chips

-.688 (.112)**

-.006 (.004)-.414 (.076)***

-.006 (.006).155 (.058)**

-.299 (.086)**.093 (.040)*

N.A.-.177 (.067)**

.328 (.144)*

.251 (.103)*

.153 (.106)

.181 (.052)**-.018 (.016)

.407 (.041)**34,947

.71

Soft

-.707

-.011-.476

-.008.183

-.301.106

Drinks

(.102)**

(.008)(.082)***

(.006)(.066)**(.090)**(.037)**

N.A.-.211

.384

.295

.123

.148-.007

.5764

(.076)**(.163)*(.123)*(.085)(.049)**(.005)(.049)**

1,207.76

White

-.679

-.142-.352

-.024.103

-.246.032.188

-.187.162.138.197.012

-.129.376

35,

Bread

(.090)**

(.117)(.060)***

(.023)(.038)**(.080)**(.016)*(.067)**(.063)**(.061)**(.052)**(.125)(.006)*(.081)(.040)**13674

'p < .05.**p

Experimental Data:Studies 2a and 2b

We conducted two experimental studies that complementone another by incorporating competing alternatives on twoimportant aspects of the experimental design. The firstaspect focuses on how different forms of salience for fearand social network stimuli can moderate the observedasymmetrical patterns of demand found in the supermarketstudy. Research suggests that salience can be manipulatedthrough environmental primes or through specifically con-structed advertisements/appeals (e.g., Obermiller and Span-genberg 2000). Second, in situations in which a topic is per-ceived to be of a sensitive nature, as may be the case withdiet, existing literature suggests that people may responddifferently to questions depending on whether they or oth-ers are the focus of inquiry (Mick 1996). Thus, anotheraspect of our experimental studies is to examine bothpeople's own demand estimates in response to changes infood prices as well as their cognitions about how others willrespond to such price changes.

Accordingly, we designed one experimental study(Study 2a) to examine participants' cognitions about howothers will react to price changes for food items and theeffect of fear and social network environmental primes (inthe form of actual newspaper articles) on those cognitions.The other experimental study (Study 2b) involves a con-trolled experiment that assesses participants' own estimatesof demand in response to price changes for food items andexamines the effect of specifically constructed fear andsocial network appeals on those demand estimates. Takentogether, the two experimental studies thus provide astronger basis on which to draw our empirical conclusionsthan would be the case with either one on its own.

Experimental Survey: Study 2aMethod. Participants for the experimental survey were

employees of a large nonprofit organization in the northeast-em United States (male = 52%, Mggg = 36 years, Mncome =$47.5K/year). Experimental surveys were distributed throughthe organization's internal mail system and/or were droppedoff to respondents. The survey was administered in a stag-gered manner over a five-week period in 2010. Participa-tion was based on a pre-request, comprising 613 respon-dents, or 73% of those initially surveyed. In-personfollow-up resulted in a 100% response rate for the subse-quent survey. The survey was a 3 (newspaper article: pro-unhealthy food prime [Fernandez 2010], fear prime [Leeder2009], and peer prime [Landro 2006]) x 2 (food category:healthy, unhealthy) x 2 (price change: increase, decrease)mixed design, with newspaper article serving as thebetween-subjects factor. All participants responded to thesame question pertaining to a 20% price increase and 20%price decrease for a healthy (broccoli) and unhealthy (soda)food category (for article excerpts and links to full articles,article pretesting, and experimental survey procedures, seethe Web Appendix at www.marketingpower.coni/jm_webappendix).

Results (demand asymmetry). Statistical tests show nosignificant difference {p> .\) across the three experimentalconditions in terms of age, income, or general food attitudescores. As such, we dropped these variables from furtheranalysis. We calculated the degree of demand responseasymmetry, ^c, for respondent k in product category c as6kc = |e^f- |ej,|, where |ej^| is the absolute value of theestimated price elasticity in response to a price decreasesince last purchase and |ej|, | is the absolute value of theestimated elasticity in response to a price increase since lastpurchase. A nonzero value of S^ c indicates an asymmetry indemand sensitivity in response to a price decrease versus aprice increase since last purchase in a product category.Note that from a public policy perspective, it is desirablefor kc to exhibit positive or higher values for healthy foodand negative or lower values for unhealthy food categories.As Table 4, Panel A, indicates, the experimental surveyresults show that the mean values of ^ c for the pro-unhealthy food condition (i.e., the typical media/advertisingenvironment consumers face) are significantly differentthan zero in the predicted undesirable directions (^ c Healthy =-.46, t(207) = 3.29, p < .01; ^ e unhealthy = .68, t(207) =3.29, p < .01). Thus, the intention data in Study 2a providea replication of the revealed data in Study 1 for H] and H2.

Results (primes). Table 4, Panel A, also demonstratesthat the fear and peer primes attenuate the undesirableasymmetry of demand sensitivity for unhealthy/healthyfood. For healthy food, an analysis of variance (ANOVA)shows a significant difference across values of ^ c for thethree conditions (F(2, 610) = 12.94, p < .01). Furthermore,Dunnett tests show that relative to the pro-unhealthy foodcondition (5^^ = -.46), the value of i^c is higher for both thefear prime ( c^ = .05,p < .01) and the peer prime {b\^c. = .29,p < .01). For unhealthy food, an ANOVA also shows a sig-nificant difference across values of ^ c for the three condi-tions (F(2, 610) =J4.27,p < .01). Moreover, Dunnett testsagain show that relative to the pro-unhealthy food condi-tion (ojjc = 68), the value of ^ c is lower for both the fearprime (];c = .16,p< .01) and the peer prime (kc = - 1 Up