Macroeconomic conditions influence consumers attitudes,

Transcription

1 Yu Ma, Kusum L. Ailawadi, Dines K. Gauri, & Druv Grewal An Empirical Investigation of te Impact of Gasoline Prices on Grocery Sopping Beavior Te autors empirically examine te effect of gas prices on grocery sopping beavior using Information Resources Inc. panel data from 2006 to 2008, wic track panelists purcases of almost 300 product categories across multiple retail formats. Te autors quantify te impact on consumers total spending and examine te potential avenues for savings wen consumers sift from one retail format to anoter, from national brands to private labels, from regular-priced to promotional products, and from iger to lower price tiers. Tey find a substantial negative effect on sopping frequency and purcase volume and sifts away from grocery and toward supercenter formats. A greater sift occurs from regular-priced national brands to promoted ones tan to private labels, and among national brand purcasers, bottom-tier brands lose sare, midtier brands gain sare, and toptier brand sare is relatively unaffected. Te analysis also controls for general economic conditions and sows tat gas prices ave a muc larger impact on grocery sopping beavior tan broad economic factors. Keywords: grocery expenditure, gas price effect, macroeconomic factors, retail format coice, promotions, private label Macroeconomic conditions influence consumers attitudes, sopping beavior, and consumption. Altoug tese conditions are not controllable by manufacturers and retailers, understanding ow tey affect consumers and ow consumers respond to tem is critical in guiding effective managerial actions. Tis issue currently occupies center stage as te world economy tries to emerge from te most severe economic crisis in decades. A small but ric body of literature in marketing examines te impact of macroeconomic factors. One stream of researc describes ow firms cange advertising, innovation, and oter proactive marketing activities during a recession and assesses te effectiveness of tese actions (Deleersnyder et al. 2009; Frankenberger and Graam 2003; Srinivasan, Rangaswamy, and Lilien 2005). Anoter stream studies te effect of business cycles and consumer confidence on sales of durable goods (Allenby, Jen, and Leone 1996; Deleersnyder et al. 2004; Kumar, Leone, and Gaskins 1995) and private labels (Lamey et al. 2007). Bot Yu Ma is Assistant Professor, Department of Marketing, Business Economics, and Law, Scool of Business, University of Alberta ( Yu.Ma@ualberta.ca). Kusum L. Ailawadi is Carles Jordan 1911 TU 12 Professor of Marketing, Tuck Scool of Business, Dartmout College ( kusum.ailawadi@dartmout.edu). Dines K. Gauri is Assistant Professor of Marketing, Witman Scool of Management, Syracuse University ( dkgauri@syr.edu). Druv Grewal is Toyota Cair of Commerce and Electronic Business and Professor of Marketing, Babson College ( dgrewal@babson.edu). Te autors tank Information Resources Inc. for providing te panel data used in tis researc and te Alberta Scool of Retailing for a researc grant awarded to te first autor. Tey also tank Scott Fay, Peter Golder, Kevin Keller, and Scott Neslin for teir elpful comments. streams of researc are typically conducted at an aggregate level, wit industry-, firm-, or product category level sales data. Yet cange in consumer beavior is at te root of wy tese macroeconomic variables affect sales and firm performance. Terefore, it is important to conduct a more disaggregate analysis (Deleersnyder et al. 2004) and understand ow consumers react to canges in macroeconomic factors (Grewal, Levy, and Kumar 2009). An underlying teme in te literature is te notion tat consumers response to macroeconomic factors is a function of not ust teir ability to buy (as measured by current and expected future income) but also teir willingness to buy (as measured by attitudes, sentiment, and so on) (Katona 1975). Conventional wisdom maintains tat macroeconomic factors and consumer sentiment ave an impact on durable goods sales because purcases of suc products are discretionary and can be postponed wen willingness to buy is low wereas nondurable products, suc as groceries, are less affected because tey cannot be postponed (Deleersnyder et al. 2004; Lamey et al. 2007). Te focus of te current researc is on a macroeconomic factor tat is qualitatively different from business cycles and consumer sentiment and as been prominent in recent years namely, te price of gasoline. Since 2006, te price of a gallon of regular gasoline as varied widely, from lows of ust over $2.00 to igs of more tan $4.00. Gasoline demand is fairly inelastic (Brons et al. 2008; Greening et al. 1995), so expenditure on gas goes up in accordance wit its price. A U.S. ouseold earning a median income spent 11.5% of tat income on gas in July 2008, up from 4.6% in 2003 (Te Wall Street Journal 2008). Wen te price of gas increases sarply, consumers ave less dispos- 2011, American Marketing Association ISSN: (print), (electronic) 18 Journal of Marketing Vol. 75 (Marc 2011), 18 35

2 able income, feel significant financial ardsip, become more price conscious, and must find ways to reduce spending in oter areas (Du and Kamakura 2008; Jacobe 2006). Consistent wit tis, Hamilton (2008) notes in is recent review of oil and te economy tat te key mecanism wereby energy price socks affect te economy is troug a disruption in spending on goods and services oter tan energy. Tus, te impact of gas prices on consumer sopping beavior derives not ust from psycological willingness to buy but also from te immediate economic ability to buy. Grocery products individually cost little relative to overall income. However, after ousing and transportation, tey form te largest percentage of U.S. ouseolds annual expenditures. For example, expenditure on food at ome for te average ouseold was 5.6% of total income after taxes in 2007, exceeding oter expense categories suc as apparel, entertainment, and ealt care (Bureau of Labor Statistics 2007). Furtermore, grocery sopping is done frequently, generally more tan once a week, so tere is plenty of opportunity to make adustments in purcases. Terefore, expenditures on grocery products provide a substantial and flexible means to adust spending in response to unexpected canges in discretionary income. In summary, grocery products may be relatively immune to general economic conditions and business cycles, but tis is not likely to be te case wit rising gas prices. However, tere is little systematic researc on te impact of gas prices on consumers sopping beavior. An exception is te work of Giceva, Hastings, and Villas-Boas (2010); using Consumer Expenditure Survey data (Bureau of Labor Statistics 2007), tey find tat expenditures on food outside te ome decrease by 56% and expenditures on food purcased at grocery stores increase sligtly wit a 100% increase in gas price. Tey also use sales data in four food categories from a California grocery cain to sow tat consumers substitute away from regular self-price products and toward promotional items to save money on overall grocery expenditures. Te obective of our researc is to provide a compreensive analysis of te impact of gas prices on consumers grocery sopping beavior. Consumers can alter ow muc tey purcase, ow often tey purcase, and ow muc tey spend on teir purcases, wic in turn is a function of wat tey buy and were tey buy it. We not only quantify te impact on ouseolds sopping frequency, total purcase volume, and spending but also examine te avenues tey use to save money on grocery sopping, suc as sifting from one retail format to anoter, from national brands to private labels, from regular-priced products to promotional purcases, and from iger-priced national brands to lower-priced ones. Suc an analysis is important not only for researcers and policy makers but also for manufacturers and retailers, wic must determine te best way to respond to and peraps preempt canges in sopping beavior in an era of peak oil and sustained volatility in energy prices. As we discuss in te next section, te answers are not obvious. We conduct our analysis using a ouseold panel data set from Information Resources Inc. Te data set captures grocery sopping information across multiple retail formats of approximately 1000 panelists from a maor U.S. metropolitan area. Te data are from January 2006 to October 2008 and span panelists purcases of almost 300 product categories. We supplement tese data wit gas prices in te same metropolitan area obtained from te Department of Energy Information Administration Web site. Some unique features of tese data make tem especially useful for our researc. First, we cover not ust a few product categories but rater te vast maority of consumer packaged goods products purcased by consumers. Second, we cover bot te traditional grocery store and drugstore cannels and te regular and supercenter stores of mass mercants (including Wal-Mart) and wareouse clubs. Tird, substantial variation in gas prices occurred during te period of our data. Fourt, we control for general economic conditions to isolate and contrast te impact of gas prices. We organize te remainder of tis article as follows: In te next section, we present te conceptual framework for our model and analysis, drawing on relevant literature wenever possible. Ten, we discuss our data and metodology. Following tis, we present our empirical results and discuss te implications of our findings for researcers and managers. Conceptual Development Overview Figure 1 depicts te framework tat guides our expectations and analysis. Gas prices affect consumers budget constraint because an increase in gas prices directly reduces te income available for oter purcases, given te relative inability to reduce gas consumption in te sort run. Te budget constraint requires consumers to reduce teir total spending. Tey can do tis by lowering teir purcase volume (consumption) and/or by reducing te cost of teir purcases. Cost can be reduced by sifting to less expensive retail formats, private label products or national brands on promotion, and/or lower-price-tier national brands (Griffit et al. 2009). In addition, consumers can adust te number of sopping trips tey make. However, consumers sopping utility is not ust a function of te quantity of products purcased and teir monetary cost. Altoug te monetary cost may be most salient in te face of gas price induced budget constraints, consumers also experience oter costs and benefits of sopping, suc as te opportunity cost of time spent in travel and searc (Bell, Ho, and Tang 1998; Blattberg et al. 1978; Marmorstein, Grewal, and Fise 1992), oter utilitarian benefits of quality and decision simplicity, and psycosocial benefits of self-expression and entertainment (Ailawadi, Neslin, and Gedenk 2001; Candon, Wansink, and Laurent 2000; Urbany, Dickson, and Kalapurakal 1996). Figure 1 includes te maor elements of total utility identified in prior researc but represents tem as costs for exposition simplicity. As gas prices increase and tigten consumers budget constraint, tere is pressure to cut monetary costs by searcing for lower prices. However, te reduction in monetary Empirical Investigation on Grocery Sopping Beavior / 19

3 FIGURE 1 Guiding Framework Avenues for Reducing Cost of Purcases Sopping Costs Number of Trips Format Sare Brand/Promo Sare NB Tier Sare Monetary Cost Price Trips ( ) Drugstore, grocery ( ) Mass, supercenter, club (+) Regular NB ( ) PL, promotional NB (+) Top tier ( ) Mid-, bottom tier (+) Travel Cost Distance Frequency Trips ( ) Mass, club ( ) Grocery, supercenter (+) Promotional NB ( ) Regular NB, PL (+) Gasoline Price Quality Cost Downgrade PL ( ) Regular NB (+) Bottom tier ( ) Top, midtier (+) Budget Constraint (+) Searc Cost Assortment Deals Decision Cost Brand coice Trips ( ) Drugstore, mass, club ( ) Grocery, supercenter (+) Promotional NB ( ) Regular NB, PL (+) Regular NB ( ) PL, promotional NB (+) Purcase Volume ( ) Holding Cost Stockpiling Trips (+) Club ( ) Drugstore, grocery, mass, supercenter (+) Promotional NB ( ) Regular NB, PL (+) Total Expenditure ( ) Psycosocial Cost Variety Entertainment Trips (+) Mass, supercenter ( ) Drugstore, grocery, club (+) Regular NB, PL ( ) Promotional NB (+) Independent variable Constructs providing conceptual basis for our expectations Dependent variables Notes: NB = national brand, and PL = private label. 20 / Journal of Marketing, Marc 2011

4 costs must be traded off against possible increases in oter costs. Travel costs refer to te distance and ow frequently te consumer must travel for sopping, quality costs are driven by weter te consumer downgrades to a less preferred brand, searc costs are driven by ow easy it is to find preferred products and deals, decision costs refer to ow easy it is to decide wat to buy, olding costs are driven by weter sopping is done in bulk, and psycosocial costs refer to ow muc enoyment te sopping activity provides. We do not directly observe all tese costs, but tey provide an important conceptual basis for our work (for a similar conceptual development in oter contexts, see Geyskens, Gielens, and Dekimpe 2002; Narasiman, Neslin, and Sen 1996). In te following discussion, we consider te tradeoffs among tese costs in developing expectations about ow gas prices affect consumers overall sopping as well as te allocation of teir purcases to different formats, promotions, and brands. Note tat we control for general economic conditions troug te gross domestic product (GDP) growt variable, wic is a widely accepted measure of economic ealt. 1 Our expectation, based on prior researc and conventional wisdom, is tat te impact of tis variable is weaker tan tat of gas price. Effect of Gas Prices on Overall Sopping Total purcase volume. Te lower disposable income resulting from iger gas prices puts pressure on consumers to buy and consume less. Because consumers are also trying to save money by eating at ome more often tan at restaurants (Giceva, Hastings, and Villas-Boas 2010) and spending more time at ome (Mouawad and Navarro 2008), tere is a positive substitution effect for food products. Across all grocery products, owever, we expect a negative effect of gas prices on total purcase volume. Total dollar spending. Total spending comprises purcase volume and te cost of te purcases. To te extent tat consumers reduce purcase volume, spending sould decrease as well. Consumers sould also try to reduce te cost of teir purcases, especially because retail prices tend to go up as energy costs rise. We examine te avenues by wic te cost of purcases may be reduced subsequently. Sopping trips. Te most direct result of iger gas prices sould be to reduce travel costs as muc as possible. Tis implies a reduction in te number of sopping trips. However, consumers wit low searc costs can sop frequently to make better use of promotions, tus saving money (Gauri, Sudir, and Talukdar 2008; Putrevu and Ratcford 1997; Urbany, Dickson, and Kalapurakal 1996). Gauri, Sudir, and Talukdar (2008) find tat ouseolds can obtain savings of up to 68% if tey engage in eiter temporal (over time) or spatial (across stores) searc, and tey can increase tose savings to 76% if tey searc across bot stores and time. Furtermore, consumers must trade off te 1We obtained similar results wit te Conference Board s Composite Index of Coincident Indicators, wic combines four individual factors: payroll employment, personal income, industrial production, and manufacturing and trade sales. reduced travel cost against te psycosocial value from sopping and te iger inventory olding cost tey incur if tey sop less frequently. Tus, as Figure 1 sows, we cannot predict weter gas prices will ave a negative or positive effect on te number of sopping trips for te average ouseold. Avenues for Reducing Sopping Expenditure Effect of gas prices on retail format coice. Average price levels are lower in mass stores, supercenters, and wareouse clubs tan in drugstores and grocery stores, and tere is considerable, toug not complete, overlap in te product categories carried by different formats, making it feasible for consumers to sift teir spending from one format to anoter (Lucs, Inman, and Sankar 2007). Terefore, monetary costs sould drive consumers to sift from drugstores and grocery stores to te former formats. However, consumers must trade off tis benefit against oter costs. Mass stores, supercenters, and wareouse clubs are not as densely located as drugstores and grocery stores, and teir assortment is not as deep as grocery stores, tus increasing travel and searc costs. Apart from membersip fees, wareouse clubs require consumers to buy in bulk, increasing teir inventory olding costs. However, supercenters and wareouse clubs offer one-stop sopping, wic can reduce te number of sopping trips and travel costs. 2 Wit special displays and frequently canging layouts, especially in periperal parts of drugstores and grocery stores, and treasure unts for frequently canging assortment in some wareouse clubs, tese formats may offer more entertainment and exploration appeal tan mass store and supercenter formats. Prior researc as sown tat all tese costs are relevant to format coice. For example, Bell, Ho, and Tang (1998) sow tat consumers consider te sum of fixed (e.g., traveling to and from te store) and variable (product prices and quantities in te basket) costs of sopping wen making teir store coice. Similarly, Batnagar and Ratcford (2004) argue tat format coice is a function of consumers costs of travel, inventory olding, and so fort. Figure 1 sows te net impact of gas price on format coice. Given te countervailing effects of te different costs, te net impact of gas price on te sare of eac format is clearly an empirical question. 3 Effect of gas prices on brand and promotion coice. National brands are sold at retail prices tat are 20% 30% iger tan private labels (Ailawadi and Harlam 2004), and penetration of private labels as increased substantially in te past decade, wit most retailers offering private label products in a wide range of categories (Kumar and Steenkamp 2007). Tis makes sifting from national brands 2We distinguis between traditional mass stores, wic ave less square footage and carry a smaller assortment of categories, and te larger supercenter format, wose assortment is broader and includes perisable food products. 3Consumers could also sop more at convenience stores because tey are conveniently located, often togeter wit a gas station. However, we do not include tis format in our analysis, because it accounts for less tan 1% of total spending in our data. Empirical Investigation on Grocery Sopping Beavior / 21

5 to private labels an easy way for ouseolds to save money on teir grocery sopping. Value-conscious consumers can also save money by searcing for promotions, and bot private labels and promotions reduce decision costs by making it easy for consumers to decide wat to buy (Ailawadi, Neslin, and Gedenk 2001; Candon, Wansink, and Laurent 2000). However, te temporal and spatial searc for promotions and te pressure to stock up on deals increase travel, searc, and inventory olding costs. (Private labels do not increase tese costs because of teir everyday low prices.) However, despite te emergence of premium private labels, in general U.S. consumers perceive a quality cost in downgrading to private labels, wereas tey may be able to buy preferred brands on promotion. In addition, promotions offer psycosocial benefits, wile private labels do not. Overall, terefore, we expect a negative effect of gas prices on regular-priced national brand sare and a positive effect on private label and promotional purcases of national brands. An empirical question, owever, is weter te positive impact is greater for promotions or private labels. Effect of gas prices on national brand price tier sare. Despite private labels price advantage, national brands still ave a unit market sare of more tan 75% across consumer packaged goods categories (Private Label Manufacturers Association 2009), in support of Seturaman s (2000) finding tat consumers are willing to pay a significant premium for national brands even if a private label is of equivalent quality. Consumers can save money by switcing from ig- to mid- and low-price-tier national brands even if tey are not willing to switc to private labels. Tis suggests a positive effect of gas prices on lower-tier sare and a negative effect on top-tier sare. However, consumers incur a quality cost in switcing away from ig-tier national brands to low-tier ones. Tose to wom monetary savings are important and quality cost is not are likely to switc to private labels, leaving te more quality conscious consumers to buy national brands. Te literature on asymmetric price and context effects sows tat consumers of low-tier national brands are more likely to switc to private labels wile iger-tier national brands are more insulated (Blattberg and Wisniewski 1989; Geyskens, Gielens, and Gisbrect 2010; Pauwels and Srinivasan 2004; Seturaman, Srinivasan, and Kim 1999). Tus, despite te conventional wisdom tat top-tier brands urt wen times are tigt, we cannot predict te effect of gas prices on bottom-, mid-, and top-tier sares among national brand purcasers. Role of Demograpic Variables Altoug economic costs are likely to be more salient tan psycosocial ones in te face of a significant budget constraint, it is reasonable to expect eterogeneity in ow consumers make trade-offs among te various costs in Figure 1. Two overarcing consumer caracteristics tat determine tese costs and consequent sopping beaviors are financial constraints and time constraints (Ailawadi, Neslin, and Gedenk 2001; Blattberg et al. 1978; Fox, Montgomery, and Lodis 2004; Lucs, Inman, and Sankar 2007; Mar- morstein, Grewal, and Fise 1992; Urbany, Dickson, and Kalapurakal 1996). Financially constrained consumers are more likely to empasize monetary and travel costs, and time-constrained consumers are more likely to empasize travel, searc, and decision costs. To preserve parsimony and strong teoretical grounding, we select tree key demograpic variables tat are directly relevant to financial and time constraints: ouseold income, presence of cildren, and presence of at least one ouseold ead wo does not work outside te ome. Income drives financial constraints, and te latter two variables drive time constraints. 4 We expect tat tese demograpic variables ave main effects on te various aspects of purcase beavior we study and also may moderate te impact of gas prices. For example, lower-income ouseolds are more likely to engage in savings beaviors (e.g., smaller supermarket and drugstore sares, greater private label and promotion sares, smaller top-tier national brand sare) and also may be more sensitive to gas price increases. In contrast, time-constrained ouseolds may be less likely to engage in searc and more attracted to one-stop sopping (i.e., lower promotion sare, iger supercenter sare). We follow Ailawadi, Neslin, and Gedenk (2001) and Fox, Montgomery, and Lodis (2004) in including demograpics to account for eterogeneity but do not develop explicit ypoteses about teir effects. Data We obtained an Information Resources Inc. panel data set from a maor metropolitan area for tis study. Te data capture ouseold-level sopping and spending of 1389 panelists across stores and formats, including all items bougt in 297 categories tracked by te firm. Purcases are tracked over 147 weeks between 2006 and For eac ouseold, we also obtained information on key demograpic variables, including ouseold income, ouseold size, age, and employment status. Finally, we obtained gasoline prices in te metropolitan area during te same period from te Department of Energy Information Administration Web site, and we obtained quarterly GDP growt rate figures from te Bureau of Economic Analysis Web site. Figure 2 depicts bot gas prices and GDP growt during te period of our study. Table 1 provides te descriptive statistics. Our unit of analysis is a ouseold and mont (e.g., Fox, Montgomery, and Lodis 2004). Te average ouseold spends $270 per mont across 9.31 sopping trips. Note tat total purcase volume is also measured in dollars. Tis is because te vastly different units across categories (e.g., pounds, gallons, square feet) cannot be aggregated in a meaningful way. We use an average category price per unit volume to aggregate purcase volume, so te resultant variable is in dollar units. However, variations in tis variable occur only because of volume canges, not price canges, so we can 4Tese demograpics are also related to oter caracteristics. For example, ouseolds wit cildren ave greater needs, and tose wo do not work outside te ome are more likely to be older and retired. 22 / Journal of Marketing, Marc 2011

6 FIGURE 2 Gas Prices and GDP Growt Rates $ % $ % $ % 0.0% $ % $2.50 Gas price GDP growt 4.0% $2.00 Jan-06 Apr-06 Jul-06 Oct-06 Jan-07 Apr-07 Jul-07 Oct-07 Jan-08 Apr-08 Jul-08 Oct % Notes: Mean (SD) of gas price is $3.07 ($.58), and mean (SD) of GDP growt is 1.19% (2.29%). assess te impact of gas prices on purcase volume by modeling variation in tis variable. Table 1 also summarizes marketing-mix differences across formats and ow ouseolds allocate teir purcases across formats, brands, and promotions. Metod In line wit Figure 1, we specify and estimate models for four sets of sopping decisions. Te first set pertains to overall sopping and includes tree dependent variables number of sopping trips, purcase volume, and total expenditure per mont. Te second set pertains to ow consumers allocate teir total purcase volume across five different retail formats, te tird set pertains to te sare of regular versus promotional national brands and private label in teir total purcase volume, and te fourt set pertains to teir sare of top-, mid-, and bottom-price-tier national brands. Eac model contains tree groups of explanatory variables. Te first group accounts for eterogeneity in preferences among ouseolds using demograpic variables and te ouseold s value of te dependent variable during a two-mont initialization period (Briesc, Cintagunta, and Fox 2009; Bucklin, Gupta and Han 1995). Te second group includes te macroeconomic variables of central interest to our researc: gasoline price and GDP growt rate. We allow for eterogeneity in response to tese variables by interacting tem wit te demograpic variables. Te final group contains te control variables tat also drive ouseolds sopping, including distance traveled for sopping and te key retailer marketing-mix variables (i.e., net price, assortment size, and percentage of assortment devoted to private label). We compute te variables in tis group at different levels of aggregation as appropriate for eac set of models, using ouseold-specific weigts obtained from an initialization period. Detailed definitions of te variables for eac set of models appear in te Appendix. Because retailers may adust teir marketing mix according to local demand socks and gas prices, we control for potential endogeneity in te tree marketing-mix variables by using teir values from markets oter tan te focal market as instruments (for examples of similar instruments, see Cintagunta, Kadiyali, and Vilcassim 2006; Nevo 2001). Total Trips, Purcase Volume, and Dollar Spending We provide te model specifications for te tree total montly sopping variables subsequently. We specify all tree equations in log-log form because ouseolds vary widely in te magnitudes of tese dependent variables. and tis specification provides coefficients in percentages rater tan absolute terms. 5 Te only variables not in log form are te two dummy variables (AtHome and Kids) and te GDP variable, wic can take on negative values. 5Te fit of te log-log specification was as good as or better tan te linear specification, particularly in oldout sample comparisons. Empirical Investigation on Grocery Sopping Beavior / 23

7 TABlE 1 Descriptive Statistics Grocery Mass Variable Overall Drugstore Store Store Supercenter Club Price index Assortment index Private label % of assortment Top-tier % of national brand assortment Midtier % of national brand assortment Bottom-tier % of national brand assortment Regular-priced national brand sare (%) Promoted national brand sare (%) Private label sare (%) Top-tier national brand sare (%) Midtier national brand sare (%) Bottom-tier national brand sare (%) Montly trips (n) Montly spending ($) Montly purcase volume ($) Distance to ouseolds (miles) Overall sare (%) Sare: ouseolds wit no ead at ome Sare: ouseolds wit 1 ead at ome Sare: ouseolds wit > average income Sare: ouseolds wit < average income Sare: ouseolds wit no cildren at ome Sare: ouseolds wit cildren at ome Mean ( ) = β + β + β ( ) + β1 4AtHome + β1 + β1 + β1 5Kids [ 6 7 ln( Inc ) + β1athome + β1kids ] ln( GPrice ) ( ) ln Numtrps ln( Numtrps ) ln Inc 1 t ( ) = β + β + β ( ) + β2 4AtHome + β2 5Kids + [ β6 2 + β7 2 ln( Inc ) + β AtHome + β Kids ( GPrice ) ( ) ln Dolspnd ln( Dolspnd ) ln Inc 2 t t ] ln [ β β 2 ln Inc β 2 AtHome ( ) + β13 2 Kids ] GDP β 14 2 ln Dist + t + ( ) + + β2 15 ln( NPrice 16 2 t ) + β ln( AssrtSize t ) + β 2 ln( PctPL t ) + ε 2 t, and t + [ β + β ln( Inc ) + β12athome + β1 13Kids GDP β1 ] t + 14 ln( Dist ) + β ln( NPrice t ) + β1 16 ln( AssrtSize t ) + β1 ln( PctPL ) + ε t 17 ( ) ln Purvol β β ln( Purvol ) β ln Inc 3 t t ( ) = + + ( ) + β3 4AtHome + β3 + β3 + β3 5Kids [ 6 7 ln( Inc ) + β AtHome + β3kids ] ln( GPrice ) 8 3 1, [ β β ln Inc β t ( ) Kids] GDPt β14 3 ln Dist 15 3 AtHome + β + ( ) + + β ln( ) + β 3 ln( AssrtSize ) NPrice t + β17 3 ln( PctPLt ) + ε t 16 3, t were Numtrps t = number of sopping trips made by ouseold in mont t, Dolspnd t = total grocery spending in dollars by ouseold in mont t, Purvol t = total purcase volume by ouseold in mont t measured in constant dollars, Numtrps 0 = average number of trips per mont by ouseold in initialization period, Purvol 0 = average purcase volume per mont by ouseold in initialization period, Dolspnd 0 = average dollar spending per mont by ouseold in initialization period, Inc = annual income of ouseold, AtHome = 1 if at least one ouseold ead is at ome (not working) and 0 if oterwise, Kids = 1 if ouseold as cildren less tan 18 years of age at ome and 0 if oterwise, GPrice t = average price per gallon of regular gas in mont t, GDP t = annualized real GDP growt rate in te quarter of mont t, Dist = distance traveled by ouseold for sopping, NPrice t = net price facing ouseold in mont t, AssrtSize t = assortment size facing ouseold in mont t, and PctPL = percentage of assortment size facing ouseold in mont t tat is private label. 24 / Journal of Marketing, Marc 2011

8 Sare Allocation Models Te sare models are of te following form: 4 t ( ) ln( Sare ) = β + β ln( Sare ) + β ln( Inc ) + β4athome + β5kids + [ β6 + β7 ln( Inc) + β AtHome + β Kids ] ln( GPrice ) Inc 12 13Kids GDPt β14 15 RelNPrice t + β16 β1 7 ln( RelPctPL t ) + εt, + [ β + β ln( ) + β were subscript refers to te t alternative witin eac set (one of five retail formats, regular or promotional national brand or private label, one of tree national brand price tiers) and distance and te marketing-mix variables for an alternative are computed relative to te weigted average across all alternatives in te set to account for cross-effects parsimoniously. We categorize national brands as bottom, mid-, or top tier depending on weter teir average retail prices are in te lowest, middle, or top tird of te national brand price distribution. All oter variables are as defined previously. Format sares ave a significant number of zero values in our data because not all ouseolds sop at all five formats every mont. Terefore, we use a two-tiered model in wic a probit governs te zero nonzero format coice and a regression of log sare determines te magnitude of nonzero format sare (Ailawadi and Harlam 2009; Wooldridge 2002). Because te percentages of zero values for te brand/ promotion and national brand tier sares are small (generally between.5% and 5% of te observations), a two-tiered model is not needed for tese models. Consistent wit te overall sopping models, we use a log-log formulation. Te independent variables in all sare models are as sown in Equation 3. Note tat te relative marketing-mix variables are defined appropriately in eac case tat is, relative to te weigted average of all formats for te format sare models, relative to te weigted average of national brands and private label for te brand/promotion sare models, and relative to te weigted average of te tree price tiers for te national brand tier sare models. We include te RelPctPL variable only in te format sare models because it is not relevant in te oters, and te distance variable is relative only for format sare models because it does not vary across alternatives for te oter sare models. Results We specify some overarcing points and ten report specific results. First, we account for potential endogeneity of marketing-mix variables in all te models, using te instrumental variables noted previously. Te first-stage regressions confirm tat te instruments are strong; te R-squares are in te range of.40 to.80, and te F-statistics far exceed cutoffs recommended in te econometrics literature. Second, we mean-center gas price, GDP growt, and income so tat AtHome + β ] + ln( RelDist ) + β + ln( ) ln( RelAssrtSize ) t t we can interpret te coefficients of te gas price and GDP growt variables as teir respective effects on te dependent variable for ouseolds wit no ead at ome, no cildren, and average income. Tird, we cecked for multicollinearity and did not find it to be a concern. Table 2 sows te correlation matrix for te variables in te overall sopping models. Te igest correlation among te independent variables is between AssrtSize and PctPL. Correlations of main variables wit teir interactions are substantial, as we expected, but none are ig enoug to be of concern. We also cecked variance inflation factors, and none are greater tan 5. Fourt, we perform tree F-tests for te role of demograpics in eac model to test te oint significance of teir main effects, interactions wit gas price, and interactions wit GDP growt, respectively. 6 We include tese effects only wen te corresponding F-tests are significant. Fift, te initialization period values of dependent variables tat capture unobserved eterogeneity are igly significant and positive in all te models, tus confirming tat preferences are relatively stable. Effect of Gas Price on Sopping Beavior Overall sopping. Table 3 presents te estimates of our overall sopping models. For te coefficient for gas price, we find tat montly number of sopping trips, expenditure, and purcase volume all decrease significantly as gas prices increase. Te coefficient can be interpreted as an elasticity; tat is, for a 100% increase in gas prices, te average ouseold reduces tese tree variables by approximately 20%, 6%, and 14%, respectively. Retail format sares. Table 4 summarizes te results of te format sare models. Because grocery store format sare is zero for less tan 3% of te observations, it is not meaningful to estimate a probit visit model for tat format. For all oter formats, we report bot probit and log-sare model results. As gas prices increase, consumers sift to one-stop sopping formats; tat is, tey visit drugstores and mass stores less often and supercenters more often. Hig-income ouseolds are particularly likely to sift from mass stores to supercenters. Because te probit model coefficients cannot be directly interpreted as effects on visit probability, we use tem to compute te cange in predicted visit probability wen gas price increases by 100% from $2.00 to $4.00 per gallon; all oter model variables are eld at teir means. We find tat te predicted visit probability decreases from 53.8% to 49% for drugstores and from 58.8% to 54% for mass stores, wile it increases from 13.9% to 18.5% for supercenters. Te impact of gas price on te sare of spending at eac format, given tat te format is visited, is also consistent wit consolidation of sopping to offset travel costs. As gas prices increase, ouseolds visit drugstores and mass stores less often, but wen tey do visit tese formats, te sare of teir total spending at tese formats goes up by 38.2% and 6We also tested for interactions of gas price wit distance and net price. Because tese interactions were significant and negative only in te sopping trip model, we do not include tem ere. Empirical Investigation on Grocery Sopping Beavior / 25

9 26 / Journal of Marketing, Marc 2011 TABlE 2 Correlations Between Sopping Model Variables GPrice GDP GPrice GDP GPrice GDP Num- Dol- Assrt- At At At trps spnd Purvol Dist NPrice Size PctPl GPrice GDP Income Kids Home Income Income Kids Kids Home Home Numtrps 1 Dolspnd.51 1 Purvol Dist NPrice AssrtSize PctPL GP GDP Income Kids At ome GP income GDP income GP kids GDP kids GP at ome GDP at ome Notes: All variables are in log form except GDP and dummy variables.

10 TABlE 3 Overall Sopping Model Estimates log Total log Number log Total Purcase Variable of Trips Spending Volume Intercept 3.870** (6.70) (1.41) (.53) Log gas price.201**.057*.144** ( 8.82) ( 2.37) ( 5.81) GDP growt **.005** (1.29) ( 5.52) ( 3.04) Log distance.026**.009*.006 ( 7.13) ( 2.49) ( 1.48) Log net price.265**.105**.189** ( 7.38) ( 2.80) ( 4.89) Log assortment.203**.094*.122** ( 5.65) (2.43) (3.06) Log % private label.681**.324*.484** (5.18) ( 2.28) ( 3.30) Log income.022**.021**.004 ( 5.23) (4.76) (.83) At ome.020**.016**.020** (2.62) (3.12) (3.30) Cildren.087**.033**.059** ( 13.12) (4.79) (8.33) Heterogeneity variable.690**.569**.561** (151.38) (114.58) (108.99) N Adusted R *p <.05. **p <.01. Notes: Coefficient estimates are reported wit t-statistics in parenteses. 21.8%, respectively. Houseolds wit cildren reduce teir sare of spending at grocery stores by 8.7% and sop more often at supercenters, increasing teir sare of spending tere by 50%. Higer-income ouseolds also increase teir sare of spending at supercenters conditional on a visit. To compute te net effect of gas price on te unconditional sare of eac format, we compute bot te predicted visit probability and te predicted conditional sare at a gas price of $2.00, olding all oter model variables at teir means. Te product of te two provides te predicted unconditional sare at a gas price of $2.00. By doing te same ting at a gas price of $4.00, we can compute te cange in predicted unconditional sare of eac format wen gas price increases by 100% from $2.00 to $4.00. As a percentage of te average sare of te format (Table 1), tese canges are 3.6%, +10%, +5%, and +24.9% for grocery store, drugstore, mass store, and supercenter formats, respectively. 7 Brand and promotion sares. Te first tree columns of Table 5 sow estimates of te national brand/private label/promotion sare models. Consistent wit our expectations, an increase in gas price decreases te sare of regular-priced national brands. For te average income ouse- 7In sare points, te 3.6% decrease in grocery sare is larger tan te 24.9% increase in supercenter sare, given te muc larger average sare of te grocery format. old wit no ead at ome, wen gas price increases by 100%, te sare of regular-priced national brands decreases by 11.9%, te sare of promoted national brands increases by 28.8%, and tere is no significant cange in private label sare. Tat te private label does not make gains in tis group may be partly due to te sift of ouseolds from grocery stores, wic ave more private label assortments, to supercenters, wic ave fewer private label assortments (see Table 1). We do find an almost 10% increase ( =.097) in private label sare among ouseolds wit at least one ead at ome, and tese ouseolds constitute 48% of te panel. Suc ouseolds sop more at drugstores, wic ave a fairly large private label assortment (see Table 1). Tese ouseolds also are more prone to buy private labels anyway (see te main effect of At ome on private label sare in Table 5), and iger gas prices seem to reinforce tis savings beavior. Surprisingly, ig-income ouseolds are more likely tan oters to sift from regular-priced national brands to promotional purcases. Tis is contrary to wat we would expect from teir financial constraints, but it is important to note tat low-income ouseolds already ave lower sares of regular-priced national brands and iger sares of promotional national brands and private labels (see te main effect of income), so wen gas prices increase, ig-income ouseolds ave more room to sift. National brand tier sares. Te last tree columns of Table 5 sow estimates of te national brand tier sare models. As we expected, iger gas prices increase te sare of middle-tier national brands. Furtermore, tey do not affect te sare of top-tier brands, and tey actually decrease te sare of bottom-tier brands. For a 100% increase in gas price, bottom-tier sare of te average ouseold s national brand purcases decreases by 9.6%, wile midtier sare increases by 5.7%. Heterogeneity in gas price effect across demograpic groups. Interactions of demograpic variables wit gas price are not significant for te most part, sowing tat te gas price effect is not dramatically different across demograpic groups. Wen tere is a difference, it is generally due to te presence of cildren and ouseold income. Te effect of cildren is largely consistent wit te notion tat suc ouseolds ave iger requirements but are more constrained by time. Terefore, te negative effect of gas price on sopping trips is more pronounced for tese ouseolds. Tese ouseolds also switc more tan oters to supercenters because tey are attracted by one-stop sopping. Te effect of income is largely consistent wit fewer financial constraints. Hig-income ouseolds can afford to make larger dollar outlays, so tey sop less frequently, spend more, spend more at wareouse clubs, and are more likely to sift to supercenters wen gas price increases. Tey are also more likely to sift from regular-priced national brands to promotional ones as gas price increases, but as we noted previously, tey ave more room to sift because tey buy more regular-priced and fewer promotional national brands overall. Tey are also eavier buyers of top-tier national brands, so it makes sense tat tey take Empirical Investigation on Grocery Sopping Beavior / 27

11 28 / Journal of Marketing, Marc 2011 TABlE 4 Format Sare Model Grocery Store Drugstore Mass Store Supercenter Club Variable log Sare Probit log Sare Probit log Sare Probit log Sare Probit log Sare Intercept.192*** 1.432***.985*** 1.334*** 1.059***.918***.973*** 1.170***.185** ( 32.87) (15.75) ( 11.04) (43.15) ( 35.33) (20.53) ( 13.59) (14.23) ( 2.43) Log gas price (GPrice) **.382***.209***.218***.178** (.36) ( 2.34) (5.19) ( 2.87) (2.71) (2.04) (1.17) (0) (.04) GDP growt.005*** ** **.014***.026*** *** (3.38) ( 1.18) ( 2.29) (1.04) ( 1.96) ( 2.87) ( 3.09) (1.50) ( 2.73) Log relative distance **.060***.088***.051***.194***.107***.077***.023 (.90) ( 2.42) (6.25) ( 9.06) (4.62) ( 20.64) ( 6.48) ( 6.57) (1.63) Log relative net price 1.023*** **.352* 1.214***.512** 1.628*** ( 4.77) ( 1.51) ( 2.53) ( 1.92) (5.80) (2.47) (5.17) ( 1.17) (1.50) Log relative assortment.129*** *** ** *** (3.36) (.41) (5.85) (.75) ( 2.30) (.07) (.13) (.82) (7.18) Log relative % private label 2.169*** *** *** * ( 13.02) ( 1.49) ( 6.81) (1.23) (3.51) (.14) (.90) (1.44) ( 1.87) Log income ***.069***.024*.038***.038***.050**.254***.176*** (.55) ( 11.23) ( 5.39) ( 1.95) ( 2.86) (2.64) ( 2.00) (16.69) (9.91) At ome.037*** *** *** ***.000 ( 5.98) (1.14) (2.63) (.36) ( 5.06) (.78) ( 1.49) (4.00) (.02) Kids.033***.096***.173***.052*** ***.058*.099***.009 (4.75) ( 5.47) ( 8.23) (2.98) (.49) (3.24) (1.73) ( 5.00) (.45) Log GPrice log income ** ** (.01) ( 2.34) (.19) (.01) (1.85) Log GPrice at ome ( 1.27) (.64) (.22) (.39) (.77) Log GPrice kids.087** **.501*** ( 2.33) (.23) (.96) (2.39) (2.75) Log eterogeneity variable.164***.419***.286***.349***.261***.326***.198***.403***.141*** (13.62) (56.71) (32.17) (59.68) (36.69) (45.88) (18.29) (64.50) (19.61) N 31,286 32,178 16,274 32,178 17,725 32, , Adusted R *p <.10. **p <.05. ***p <.01. Notes: All sares are sares of total purcase volume. Coefficient estimates are reported wit t-statistics in parenteses.

12 TABlE 5 Brand, Promotion, and Tier Sare Models National Brand National Brand Bottom-Tier Middle-Tier Top-Tier Regular Promo Private label National Brand National Brand National Brand Variable log Sare log Sare log Sare log Sare log Sare log Sare IIntercept.313** 1.054**.404**.269**.809**.371** ( 67.44) ( 70.98) ( 12.08) ( 49.56) ( 97.02) ( 7.16) Log gas price (GPrice).119**.288** **.057**.017 ( 6.21) (5.88) (.88) ( 8.80) (3.48) (.54) GDP growt.003** * (2.77) (.37) (.64) (.02) (.52) (2.64) Log distance.005*.062**.022**.013**.009**.028** ( 2.01) (10.16) ( 5.09) (6.39) ( 3.04) ( 4.97) Log relative net price.423** **.365**.513**.047 (2.87) (.90) (4.14) (17.60) ( 13.03) (1.34) Log relative assortment size.574**.507*.370**.183** ** ( 6.63) ( 2.24) (8.43) (2.78) ( 1.14) (22.51) Log income.046**.032**.085**.028**.046**.101** (14.18) ( 3.85) ( 14.35) ( 10.26) (11.42) (13.46) At ome.015** ** ( 3.52) (.84) (7.19) (1.20) (.99) (.10) Kids **.049**.029**.023**.062** (0) ( 5.95) (5.81) (7.35) ( 4.09) ( 5.72) Log GPrice log income.036*.197**.041 ( 2.01) (4.28) (1.23) Log GPrice at ome ** (.38) ( 1.15) (2.93) Log GPrice kids (1.02) ( 1.45) ( 1.56) Log eterogeneity variable.424**.401**.477**.402**.265**.281** (53.78) (72.86) (54.10) (40.01) (41.39) (32.79) N 32,137 29,937 31,553 32,092 32,014 30,748 Adusted R *p <.05. **p <.01. Notes: All sares are sares of purcase volume. Coefficient estimates are reported wit t-statistics in parenteses. advantage of promotions on tese top brands to reduce teir cost. 8 Effects of Oter Variables on Sopping Beavior GDP growt rate. Te overall sopping results in Table 3 sow tat te GDP growt coefficient is not significant for sopping frequency and is negative for expenditure and purcase volume. Te negative sign is surprising but may reflect a small substitution effect as consumers travel and eat out more wen economic conditions are good and terefore buy less for ome consumption. Recall tat tis variable is not in log form, so te coefficient is te percentage cange in te dependent variable for a unit increase in GDP growt. For a one percentage point increase in GDP growt, te decreases in expenditure and purcase volume are small (.9% and.5%, respectively). In elasticity terms, a 100% increase in GDP growt from its average of 1.29% (see Table 1) is associated wit 1.1% and.65% decreases in expenditure and volume, respectively. 8Our key results are robust and remain substantively uncanged wit a linear formulation. Two differences wort noting are tat, in te linear formulation, te small gas price effect on total spending becomes insignificant and te insignificant gas price effect on wareouse club visit probability becomes a small, but significant, positive effect. In line wit our expectations, te impact of tis variable is substantially smaller tan te impact of gas price across all te oter models. Indeed, tere are only a few significant effects in Tables 4 and 5. As GDP growt rises, ouseolds increase teir sare of purcases at grocery stores and reduce sares at oter formats, especially supercenters. Consistent wit economic teory, as GDP growt increases, ouseolds sligtly increase teir sare of regular-priced and top-tier national brands. Control variables. In general, te effects of te control variables, wen significant, are intuitive. We begin wit te impact of distance. It as a negative effect on sopping frequency and spending (Table 3). Tat is, ouseolds consolidate teir sopping into fewer trips wen tey must travel farter, and tey conserve spending to offset iger travel costs. Table 4 confirms tat te farter te relative distance to a format, te less likely a ouseold is to visit tat format (Ailawadi, Pauwels, and Steenkamp 2008; Fox, Montgomery, and Lodis 2004). Distance also induces some sopping consolidation in tat te farter te drugstore or mass store format, te greater is te sare of spending in tose formats conditional on a visit. Te supercenter format suffers because of its locational disadvantage in tat ouseolds tat are farter away not only visit it less often but also spend less tere. Finally, Table 5 sows a negative Empirical Investigation on Grocery Sopping Beavior / 29

13 effect of distance on private label sare and a positive effect on promotional national brand sare. Consumers also reduce mid- and top-tier brand sares and increase bottomtier brand sares wen tey must travel farter to sop. Tis may be to offset driving cost, but it could also be because te retail formats tat are fartest away (wareouse clubs) ave a lower-tan-average top-tier assortment and a igertan-average bottom-tier assortment (Table 1). Te effect of net price is not always negative. Net price as te expected negative effect on sopping trips, purcase volume, and expenditure. Wit a couple of exceptions, it also as te expected negative effect in te format sare models, toug it is not always significant. Tis is consistent wit te mixed effect of price in prior researc (e.g., Fox, Montgomery, and Lodis 2004). Te positive effect for regular national brand and private label sares is consistent wit previous findings tat te price differential between private labels and national brands may be too big and tat private labels can benefit from reducing te differential (Ailawadi, Pauwels, and Steenkamp 2008; Hoc and Baneri 1993). Finally, te impact of assortment is noteworty. It is negative for sopping frequency; we presume tat consumers need fewer sopping trips to find wat tey need wen assortment is large. Bot spending and purcase volume increase wit te variety of coices offered by a bigger assortment. In addition, we find tat te more te private label is empasized in te total assortment, te more frequently ouseolds sop, and te lower is teir purcase volume and spending. Consistent wit prior researc (e.g., Ailawadi, Pauwels, and Steenkamp 2008), tis suggests tat empasizing private labels too muc is not good for retailers. As we expected, in general te relative size of a format s assortment increases its sare, wit te exception being te mass store format. As grocery stores and drugstores increase teir empasis on private labels, ouseolds lower teir visits and sare of spending tere, but te opposite is true for mass stores. Tis is consistent wit different expectations and obectives in sopping trips made to different formats (Fox and Seturaman 2006). Consumers want variety at te local grocery store and drugstores and lower-priced private labels wen tey visit a mass store. Table 5 sows tat assortment size as te expected positive effect for bottom- and top-tier national brands and for private labels, but surprisingly, it as a negative effect for regular and promoted national brand sare. Te latter is consistent wit prior findings tat sales actually improve wen retailers prune assortment strategically (Broniarczyk, Hoyer, and McAlister 1998). Demograpic variables. Te main effects of demograpics on sopping beavior are largely in line wit intuition. Table 3 sows tat income as a negative effect on te number of sopping trips and a positive effect on spending. Bot effects make sense because te financial constraints of lowincome ouseolds encourage tem to searc more and terefore make more trips; according to basic economic teory, ig-income ouseolds can afford to spend more. Table 4 sows tat ig-income ouseolds make fewer visits to drugstores and mass stores and also ave lower sares in tese formats conditional on a visit. Tey also visit supercenters and wareouse clubs more often and ave iger wareouse club sare conditional on a visit, peraps because tey can afford te budget outlay required for bulk sopping. Finally, Table 5 sows tat, consistent wit basic economic teory, ig-income ouseolds ave iger sares of regular-priced national brands and middle- and top-tier national brands and lower sares of promotional and bottom-tier national brands and private labels. Houseolds wit at least one nonworking ouseold ead make sligtly more trips, consistent wit fewer time constraints, and teir volume and spending are also sligtly greater (Table 3). Tey spend less at grocery and mass stores and more at drugstores (Table 4). Tis may ave less to do wit teir time constraints (or lack tereof) and more to do wit te probability tat tey are older and/or retired, a prime target market for drugstores. Tey buy fewer regularpriced national brands, but surprisingly, tey buy more private labels, not more promotional national brands, despite teir fewer time constraints. Tis finding may be related to teir preference for drugstores, wic carry a iger percentage of private label products tan supercenters and wareouse clubs (Table 1). In general, te impact of cildren is intuitive. Houseolds wit cildren make fewer sopping trips, presumably because of time constraints, but teir total expenditure and purcase volume are iger because of te greater needs of large families. Tey visit and spend less at drugstores, wic is consistent wit te older, female target market of drugstore cains. In contrast, tey spend more at grocery stores, mass stores, and supercenter formats. Tey are time constrained, and tey need to save money on teir large sopping baskets, so tey prefer formats tat balance convenience (grocery store) wit affordability (mass store and supercenter). Also in line wit intuition, time-constrained ouseolds wit cildren buy more private labels and fewer national brands on promotion. Discussion Macroeconomic conditions ave maor effects on consumer beavior and, terefore, on firm performance. Tis researc provides a compreensive, disaggregate analysis of ow and ow muc consumers cange teir grocery sopping beavior in response to gas price, a macroeconomic variable tat is increasing in importance. In quantifying te effect of gas price, we control for general macroeconomic conditions as reflected in te GDP growt rate. On te one and, our work complements aggregate researc on te impact of macro factors, and on te oter and, it builds on te large body of researc on consumer grocery sopping beavior, wic examines a ost of variables, suc as price, promotions, assortment, and competitive factors, but in general does not incorporate macroeconomic factors. Gas Price Effect We summarize te estimated average magnitude of te gas price effect on eac component of sopping beavior in Figure 3. For easy interpretation, te magnitudes are in terms of elasticities, tat is, te percentage cange in a com- 30 / Journal of Marketing, Marc 2011

14 FIGURE 3 Summary of Average Gas Price Effect 100% increase in gas price from $2.00 to $ % decrease in montly sopping trips 14% decrease in montly purcase volume 6% decrease in montly expenditure Format Sares 3.6% decrease in grocery format sare, driven by ouseolds wit cildren 10% increase in drugstore format sare, despite lower visit probability 5% increase in mass format sare, despite lower visit probability 24.9% increase in supercenter format sare, driven especially by ouseolds wit cildren No significant cange in wareouse club format sare Brand/Promotion Sares 12% decrease in regular-priced national brand sare 29% increase in promotional national brand sare 3% increase in private label sare, driven by ouseolds wit one or more ead at ome National Brand Tier Sares 10% decrease in sare of bottom-tier brands 6% increase in sare of midtier brands No significant cange in sare of top-tier brands Notes: Percentage canges in sares are from te average sare of eac option. Terefore, in sare points, te 3.6% decrease in grocery sare is larger tan te 24.9% increase in supercenter sare, given te muc larger average sare of te grocery format. ponent of sopping beavior for te average ouseold, attributable to a 100% increase in gas price (e.g., from $2.00 per gallon to $4.00 per gallon). We summarize te effect of gas price as follows: Houseolds consume less and consolidate teir sopping as rising gas prices take a bigger bite out of teir wallet, but tey preserve teir preference for ig-quality brands, searcing for tem on promotion to save money. Next, we igligt te aspects of tese results tat are eiter surprising or contrary to conventional wisdom and discuss teir implications. We found tat an increase in gas prices reduces sopping frequency and purcase volume, and toug tis may not seem surprising, it is important for at least two reasons. First, altoug prior researc as establised a strong effect of macroeconomic factors, suc as recessions and lower consumer confidence, on sales of durable goods, consumer packaged goods products are considered more abitual and necessary, less conducive to purcase postponement, and, terefore, less vulnerable (Deleersnyder et al. 2004; Katona 1975). Second, evidence sows tat consumers travel less and eat at ome more as gas prices rise, so tere is a positive substitution effect tat sould increase food purcases. Indeed, Giceva, Hastings, and Villas-Boas (2010) report a positive effect of gas price on expenditure in four food categories. Despite tese penomena, we document a substantial reduction especially in overall purcase volume wen we include a compreensive array of grocery products and control for oter variables. Apart from te general decrease in purcase volume, wic urts bot manufacturers and retailers, manufacturers of impulse products may be especially urt by lower sopping frequency as impulse purcase opportunities decrease. In addition, consumers will stockpile, so manufacturers and retailers sould offer fre- quent promotions to generate sopping trips and increase te opportunities for consumers to select teir offering. In general, te impact of gas prices on were consumers sop is intuitive: Tey consolidate teir sopping. Driven largely by ouseolds wit cildren, supercenters, te quintessential one-stop sopping format, benefit at te expense of traditional grocery stores. Altoug consumers visit drugstores and mass stores less, tey buy a larger sare of teir requirements tere wen tey do visit, and te net impact is sligtly positive. Manufacturers need to consider te sift in consumer coice as tey negotiate prices and trade deals wit te different formats. Tey may need to offer more promotional funds to te traditional formats to keep tem competitive wile engaging te supercenter cannel wit different, possibly larger-size stockkeeping units tat iger-income and larger ouseolds are willing to buy. Consistent wit economic teory, iger gas prices make ouseolds sift away from regular- to promotionalpriced national brands. Te sift toward private labels is muc smaller. Indeed, tere is no significant impact on private label sare except among ouseolds wit one ead at ome. Tis is a surprising finding given te attention in te business press to private label growt, and its implications are important. Retailers sould realize tat continuing to empasize private labels at te expense of national brands, unless it is accompanied by credible quality improvements and strong marketing and differentiation (Ailawadi, Pauwels, and Steenkamp 2008), may not increase sare even in tigt times. Promotions are an effective retention tool as gas prices increase, so balancing a robust private label wit attractive promotions on national brands is more likely to be effective. Manufacturers sould also realize tat Empirical Investigation on Grocery Sopping Beavior / 31

15 unkering down in toug times by cutting promotions and allowing prices to rise will lead to sare losses (see Deleersnyder et al. 2004). Furtermore, given prior researc on te asymmetry of consumer sifts (Lamey et al. 2007), consumers lost during toug times may not return wen financial constraints are eased. Equally important is our finding tat iger gas prices do not urt te sare of top-tier brands among consumers wo continue to purcase national brands. Rater, it is te sare of bottom-tier brands tat drops, wile midtier brands gain. Tis finding is contrary to te conventional wisdom tat top-tier brands suffer wen times are toug but is consistent wit researc on context effects; private labels are muc more likely to take sare away from bottom-tier brands tan from top-tier ones. Tus, manufacturers sould tread carefully in introducing lower-priced extensions of teir ig-equity brands. Unless tey can significantly cut costs and preserve margins at te substantially lower price needed to combat private labels, tey may find tat te lower-tier introductions are not effective in retaining customers and may end up urting teir overall brand equity. It would be wortwile to monitor te performance of basic versions of national brands tat companies suc as Procter & Gamble ave begun introducing (Bryon 2009). In summary, te most direct way consumers can offset iger gas prices is by using less gas, but te economics literature as convincingly sown tat gasoline demand is fairly inelastic. Our results sow tat travel cost plays a role in consumer sopping sifts, but it is far from te sole determinant. On te one and, sopping frequency decreases substantially as gas prices increase. Distance is an important control variable in our models and sows te expected negative sign. Furtermore, te one-stop sopping supercenter format increases sare as gas prices rise. On te oter and, supercenters are generally farter away, and consumers buy more on promotion as gas prices increase, despite te need to searc at different times and in different stores. In addition, sensitivity to distance does not become stronger wit gas price increases, except in te sopping trip model. Overall, terefore, our results underscore te importance of considering not ust monetary cost but also te full spectrum of oter economic costs and, to a lesser extent, te psycosocial aspects of sopping in understanding consumer sopping response to macroeconomic factors suc as gas price. Limitations We note some limitations of our work tat we ope researcers can address. First, we estimate te impact of gas price on eac component of sopping beavior separately. However, it is likely tat consumers sopping decisions are interdependent or follow a ierarcical structure (e.g., tey first decide on teir budget and were to sop and ten decide wat to buy). We leave tis to furter researc to test more integrated models. Second, altoug we allow for eterogeneity in te gas price effect across key demograpic groups, tere may be eterogeneity in response among consumers tat tese demograpic variables do not capture. For example, consumers wit different psycograpic and sopping profiles may react differently. Generating suc profiles and studying differences in teir response are fruitful directions for additional researc. Tird, altoug we control for economic conditions wit te well-accepted GDP growt variable and find similar results to te Conference Board Composite Index of Coincident Indicators, oter macroeconomic variables may affect sopping beavior and sould be examined, especially wen te data span longer periods. Similarly, we account for key marketing-mix variables, but oter retail factors, suc as service and convenience (Ailawadi and Keller 2004; Berry, Seiders, and Grewal 2002; Gauri, Trivedi, and Grewal 2008), influence consumers format and store coice. To te extent tat tese factors are stable, we control for tem troug te eterogeneity variable. However, furter researc sould examine weter te influence of tese factors canges wit te macroeconomic environment. Conclusion Te significant effect of gas price we document erein makes it important to incorporate tis factor explicitly into consumer sopping beavior models wen it exibits substantial variance. We also contrast te effects of gas price and general economic ealt on grocery sopping and find tat te former is muc stronger. Altoug we examine te sort-term impact, it is important to understand weter tese canges persist over te long run and weter tey reverse wen gas price goes down again. Furtermore, it is important to understand te extent to wic sopping beavior canges are driven by psycological factors rater tan te direct budgetary constraints tat iger gas prices impose. Gas prices and general economic conditions are likely to ave very different effects on consumer sentiment and attitudes, wic in turn affect consumers psycological willingness to buy. Finally, we sow tat gas price is ust one important macroeconomic variable outside te control of managers tat influences consumer sopping. Oter macroeconomic variables, suc as ome values and tax rates, migt also affect consumers willingness and/or ability to buy. We ope tat our work stimulates furter researc in tis important domain. Appendix Variable Definitions We aggregate all variables from te category to te format and market level using ouseold-specific category and format sares in te initialization period as weigts. 9 We use te first two monts as te initialization period. Variables for Format Sare Models Distance. Distance to retail format for ouseold is calculated as 9Te exception is net price of a category, wic we aggregate up from stockkeeping unit and brand levels using market sares in te initialization period as weigts. 32 / Journal of Marketing, Marc 2011

16 were dist n is te distance from te closest store of retailer n in format to ouseold and ts nt0 is te sare of retailer n in retail format for ouseold in te initialization period. RelDist. Distance to format for ouseold relative to weigted average distance to all formats is calculated as 5 = 1 Distance Distance were ts t0 is ouseold s sare of format in te initialization period. NPrice t. Net price is calculated as C were Nprice tc is net price of category c in format in mont t and cs t0c is sare of total spending by ouseold in te initialization period on category c. RelNPrice t. Net price of format in mont t for ouseold, relative to weigted average net price of all formats, is calculated as AssrtSize t. Assortment size of format for ouseold in mont t is calculated as were AssrtSize tc is te number of distinct stockkeeping units of category c in te quarter of mont t in retail format. RelAssrtSize t. Assortment size of format for ouseold relative to weigted average assortment size of all formats is calculated as PctPL t. Percentage private label in assortment of format for ouseold in quarter of mont t is calculated as tst0 NPrice tc cs NPrice t 0 c, c = 1 5 = 1 C c = 1 5 distntsnt0, n = 1 = 1 C N c = 1 NPrice NPrice t0c AssrtSize t AssrtSize AssrtSize PLPct tts t0 tc tc t t cs cs ts., t0 c, t0 t0 c,. were PLPct tc is te percentage of te total assortment tat is private label in category c of format in te uarter of mont t. RelPctPL t. Percentage private label in assortment size of format for ouseold relative to weigted average percentage private label in all formats is calculated as 5 = 1 PLPct PLPct Variables for Brand/Promotion Sare Models Distance. Average distance to stores for ouseold is calculated as N n = 1 were Distance is as defined previously and fs t0 is te sare of total trips to retail format by ouseold in te initialization period. NPrice kt. Net price is calculated as C were NPrice kct is net price of brand type k in category c in mont t and cs t0c is as defined previously. Note tat k = 1 is national brands and k = 2 is private label. RelNPrice kt. Relative net price is calculated as were ts kt0 is ouseold s sare of brand type k in te initialization period. AssrtSize kt. Assortment size of brand type k for ouseold in mont t is calculated as were AssrtSize ktc is te number of distinct stockkeeping units of brand type k in category c in te quarter of mont t. RelAssrtSize kt. Assortment size of brand type k for ouseold relative to weigted average assortment size of bot brand types is calculated as t t Distance ts 0 t. fs t 0, NPrice kct cs NPrice t 0 c, c = 1 2 k = 1 C c = 1 2 k = 1 NPrice NPrice ct0 kt kt AssrtSize ktc AssrtSize AssrtSize ts 0 kt, cs t 0, c kt kt ts 0 kt. Empirical Investigation on Grocery Sopping Beavior / 33

17 Variables for National Brand Tier Sare Models Tese are calculated te same as previously, except tat k = 1, 2, and 3, respectively, for bottom-, mid-, and top-tier national brands. Variables for Overall Sopping Models Distance. Tis is calculated as defined previously. NPrice t, AssrtSize t, and PctPL t. Tese are calculated as follows: 5 = 1 5 = 1 NPrice t AssrtSize ts t0 t ts, t0, and 5 = 1 PctPL t were NPrice t, AssrtSize t, PctPL t, and ts t0 are as defined in te format sare models. Purcase Volume t. Total purcase volume by ouseold in mont t is calculated as C c = 1 were q tc is total equivalent units of category c purcased by ouseold in mont t and NPrice t0c is net price of category c in te initialization period. ts t0, q N Pr ice, tc t0c REFERENCES Ailawadi, Kusum L. and Bari Harlam (2004), An Empirical Analysis of te Determinants of Retail Margins: Te Role of Store-Brand Sare, Journal of Marketing, 68 (January), and (2009), Retailer Promotion Pass-Troug: A Measure, Its Magnitude, and Its Determinants, Marketing Science, 28 (4), and Kevin Keller (2004), Understanding Retail Branding: Conceptual Insigts and Researc Priorities, Journal of Retailing, 80 (4), , Scott A. Neslin, and Karen Gedenk (2001), Pursuing te Value-Conscious Consumer: Store Brands Versus National Brand Promotions, Journal of Marketing, 65 (January), , Koen Pauwels, and Jan-Benedict E.M. Steenkamp (2008), Private-Label Use and Store Loyalty, Journal of Marketing, 72 (November), Allenby, Gregory, L. Jen, and Robert Leone (1996), Economic Trends and Being Trendy: Te Influence of Consumer Confidence on Retail Fasion Sales, Journal of Business & Economic Statistics, 14 (1), Bell, David R., Teck-Hua Ho, and Cristoper S. Tang (1998), Determining Were to Sop: Fixed and Variable Costs of Sopping, Journal of Marketing Researc, 35 (August), Berry, Leonard L., Katleen Seiders, and Druv Grewal (2002), Understanding Service Convenience, Journal of Marketing, 66 (July), Batnagar, Amit and Brian T. Ratcford (2004), A Model of Retail Format Competition for Non-Durable Goods, International Journal of Researc in Marketing, 21 (1), Blattberg, Robert, Tomas Buesing, Peter Peacock, and Subrata Sen (1978), Identifying Deal Prone Segment, Journal of Marketing Researc, 15 (August), and Kennet J. Wisniewski (1989), Price-Induced Patterns of Competition, Marketing Science, 8 (4), Briesc, Ricard A., Pradeep K. Cintagunta, and Edward J. Fox (2009), How Does Assortment Affect Grocery Store Coice? Journal of Marketing Researc, 46 (April), Broniarczyk, Susan M., Wayne D. Hoyer, and Leig McAlister (1998), Consumers Perceptions of te Assortment Offered in a Grocery Category: Te Impact of Item Reduction, Journal of Marketing Researc, 35 (May), Brons, Martin, Peter Nikamp, Eric Pels, and Piet Rietveld (2008), A Meta-Analysis of te Price Elasticity of Gasoline Demand: A SUR Approac, Energy Economics, 30 (5), Bucklin, Randolp E., Sunil Gupta, and S. Han (1995), A Brand s Eye View of Response Segmentation in Consumer Brand Coice Beavior, Journal of Marketing Researc, 32 (February), Bryon, Ellen (2009), Tide Turns Basic for P&G in Slump, (August 6), (accessed December 15, 2010), [available at ttp:// online.ws.com/article/sb tml]. Bureau of Labor Statistics (2007), Consumer Expenditure Survey, (accessed November 4, 2010), [available at ttp://www. bls.gov/cex/csxstnd.tm#2007]. Candon, Pierre, Brian Wansink, and Gilles Laurent (2000), A Benefit Congruency Framework of Sales Promotion Effectiveness, Journal of Marketing, 64 (October), Cintagunta, Pradeep K., Vrinda Kadiyali, and Naufel J. Vilcassim (2006), Endogeneity and Simultaneity in Competitive Pricing and Advertising: A Logit Demand Analysis, Journal of Business, 79 (6), Deleersnyder, Barbara, Marnik G. Dekimpe, Miklos Sarvary, and Pilip M. Parker (2004), Weatering Tigt Economic Times: Te Sales Evolution of Consumer Durables Over te Business Cycle, Quantitative Marketing & Economics, 2 (4), ,, Jan-Benedict E.M. Steenkamp, and Peter S.H. Leeflang (2009), Te Role of National Culture in Advertising s Sensitivity to Business Cycles: An Investigation Across Continents, Journal of Marketing Researc, 46 (October), Du, Rex Y. and Wagner A. Kamakura (2008), Were Did All Tat Money Go? Understanding How Consumers Allocate Teir Consumption Budget, Journal of Marketing, 72 (November), Fox, Edward J., Alan L. Montgomery, and Leonard M. Lodis (2004), Consumer Sopping and Spending Across Retail Formats, Journal of Business, 77 (2), S25 S60. and Ra Seturaman (2006), Retail Competition, in Retailing in te 21st Century: Current and Emerging Trends, Manfred Krafft and Murali K. Mantrala, eds. Berlin: Springer- Verlag, Frankenberger, Kristina and Roger Graam (2003), Sould Firms Increase Advertising Expenditures During Recessions? Report No , Marketing Science Institute. Gauri, Dines K., K. Sudir, and Debabrata Talukdar (2008), Te Temporal and Spatial Dimensions of Price Searc: Insigts from Matcing Houseold Survey and Purcase Data, Journal of Marketing Researc, 45 (April), / Journal of Marketing, Marc 2011

18 , Minaksi Trivedi, and Druv Grewal (2008), Understanding te Determinants of Retail Strategy, Journal of Retailing, 84 (3), Geyskens, Inga, Katrin Gielens, and Marnik Dekimpe (2002), Te Market Valuation of Internet Cannel Additions, Journal of Marketing, 66 (April), ,, and Els Gisbrect (2010), Proliferating Private- Label Portfolios: How Introducing Economy and Premium Private Labels Influences Brand Coice, Journal of Marketing Researc, 47 (October), Giceva, Dora, Justine Hastings, and Sofia Villas-Boas (2010), Revisiting te Income Effect: Gasoline Prices and Grocery Purcases, American Economic Review, 100 (2), Greening, Lorna A., Han Tarn Jeng, Jon P. Formby, and David C. Ceng (1995), Use of Region, Life-Cycle and Role Variables in te Sort-Run Estimation of te Demand for Gasoline and Miles Traveled, Applied Economics, 27 (7), Grewal, Druv, Micael Levy, and V. Kumar (2009), Customer Experience Management in Retailing: An Organizing Framework, Journal of Retailing, 85 (1), Griffit, Racel, Epraim Leibtag, Andrew Leicester, and Aviv Nevo (2009), Consumer Sopping Beavior: How Muc Do Consumers Save? Journal of Economic Perspectives, 23 (2), Hamilton, James D (2008), Oil and te Macroeconomy, in New Palgrave Dictionary of Economics, 2d ed., Steven Durlauf and Lawrence Blume, eds. New York: Palgrave McMillan. Hoc, Stepen J. and Sumeet Baneri (1993), Wen Do Private Labels Succeed? Sloan Management Review, 34 (4), Jacobe, Dennis (2006), Hig Gas Prices Causing Consumer Spending Cuts, Gallup News Service, (May 24), (accessed November 22, 2010), [available at ttp:// poll/22918/ig-gas-prices-causing-consumer-spendingcuts.aspx]. Katona, George (1975), Psycological Economics. New York: Elsevier Scientific Publising. Kumar, Nirmalya and Jan-Benedict E.M. Steenkamp (2007), Private Label Strategy: How to Meet te Store Brand Callenge. Boston: Harvard Business Scool Press. Kumar, V., Robert P. Leone, and Jon N. Gaskins (1995), Aggregate and Disaggregate Sector Forecasting Using Consumer Confidence Measures, International Journal of Forecasting, 11 (3), Lamey, Lien, Barbara Deleersnyder, Marnik G. Dekimpe, and Jan- Benedict E.M. Steenkamp (2007), How Business Cycles Contribute to Private-Label Success: Evidence from te United States and Europe, Journal of Marketing, 71 (January), Lucs, Ryan J., Jeffrey Inman, and Venkates Sankar (2007), Cannel Blurring: A Study of Cross-Retail Format Sopping Among U.S. Houseolds, working paper, Katz Graduate Scool of Business, University of Pittsburg. Marmorstein, Howard, Druv Grewal, and Raymond Fise (1992), Te Value of Time Spent in Price-Comparison Sopping: Survey and Experimental Evidence, Journal of Consumer Researc, 19 (1), Mouawad, Jad and Mireya Navarro (2008), Teet Gritted, Drivers Adust to $4 Gasoline, Te New York Times, (May 24), (accessed November 22, 2010), [available at ttp:// com/2008/05/24/business/24gas.tml?_r=1&scp=1&sq=teet% 20Gritted,%20Drivers%20Adust%20to%20%244%20Gasoline%20&st=cse]. Narasiman, Cakravarti, Scott A. Neslin, and Subrata Sen (1996), Promotional Elasticities and Category Caracteristics, Journal of Marketing, 60 (April), Nevo, Aviv (2001), Measuring Market Power in te Ready-to-Eat Cereal Industry, Econometrica, 69 (2), Pauwels, Koen and Suba Srinivasan (2004), Wo Benefits from Store Brand Entry? Marketing Science, 23 (3), Private Label Manufacturers Association (2009), PLMA Yearbook Documents Private Label Sales Surge of More Tan 105, (accessed November 3, 2010), [available at ttp://www. plma.com/pressupdate/pressupdate.asp#id39]. Putrevu, Sanay and Brian T. Ratcford (1997), A Model of Searc Beavior wit an Application to Grocery Sopping, Journal of Retailing, 73 (4), Seturaman, Ra (2000), Wat Makes Consumers Pay More for National Brands Tan for Store Brands: Image or Quality? Report No Marketing Science Institute., V. Srinivasan, and Doyle Kim (1999), Asymmetric and Neigborood Cross-Price Effects: Some Empirical Generalizations, Marketing Science, 18 (1), Srinivasan, Rai, Arvind Rangaswamy, and Gary Lilien (2005), Turning Adversity into Advantage: Does Proactive Marketing During a Recession Pay Off? International Journal of Researc in Marketing, 22 (2), Urbany, Joel E., Peter R. Dickson, and Rosemary Kalapurakal (1996), Price Searc in te Retail Grocery Market, Journal of Marketing, 60 (April), Te Wall Street Journal (2008), Income vs. Gas Prices, an Update, (August 4), (accessed November 22, 2010), [available at ttp://blogs.ws.com/economics/2008/08/04/ income-vs-gasprices-an-update/]. Wooldridge, Jeffrey M. (2002), Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press. Empirical Investigation on Grocery Sopping Beavior / 35

19 Copyrigt of Journal of Marketing is te property of American Marketing Association and its content may not be copied or ed to multiple sites or posted to a listserv witout te copyrigt older's express written permission. However, users may print, download, or articles for individual use.