229

� 2011 by JOURNAL OF CONSUMER RESEARCH, Inc. ● Vol. 39 ● August 2012All rights reserved. 0093-5301/2012/3902-0002$10.00. DOI: 10.1086/662611

How Economic Contractions andExpansions Affect Expenditure Patterns

WAGNER A. KAMAKURAREX YUXING DU

In this study, we attempt to understand how household budget allocations acrossvarious expenditure categories change when the economy is in recession or ex-pansion. The common assumption is that a household’s tastes would not changeas a function of economic conditions and therefore any adjustments in expenditurepatterns during economic contractions/expansions would simply be due to changesin the consumption budget. Standard economic models translate these budgetaryeffects into lateral movements along a set of fixed Engel curves, which relatecategory expenditure shares to total expenditures. We propose and test a con-ceptual framework based on the notion of relative consumption, which prescribesthat, for any given total consumption budget, expenditure shares for positionalgoods/services will decrease during a recession, while shares for nonpositionalgoods/services will increase (i.e., shifting the entire Engel curve upward or down-ward, depending on the nature of the expenditure category and the economicconditions).

In 2009, personal consumption expenditures accounted for70% of GDP in the United States (Bureau of Economic

Analysis). Its importance to the economy aside, how con-sumers allocate expenditures across different categories ofcommodities (e.g., food, shelter, apparel, transportation, ed-ucation, health care, and recreation) reflects their life pri-orities and has been a fundamental issue in consumer re-search. Numerous studies in the marketing literature havebeen conducted in this area by, for example, Bellante andFoster (1984), Du and Kamakura (2008), Ferber (1956), Fisk(1959), Fritzsche (1981), Levedah (1980), Loeb (1954),Millican (1967), Ostheimer (1958), Rogers and Green(1978), Rubin, Riney, and Molina (1990), Soberon-Ferrerand Dardis (1991), Strober and Weinberg (1977), Wagnerand Hanna (1983), and Wilkes (1995). In fact, psychologicalanalysis of consumer spending under various economic con-

Wagner A. Kamakura (kamakura@duke.edu) is Ford Motor Co. GlobalMarketing Professor, Fuqua School of Business, Duke University, Box90120, Durham, NC 27708. Rex Yuxing Du (rexdu@bauer.uh.edu) is Hur-ley Associate Professor of Marketing, Bauer College of Business, Uni-versity of Houston, Houston, TX 77204-6021. The authors are thankfulfor the insightful and constructive input from the editor, associate editor,and reviewers.

Ann McGill served as editor and Brian Ratchford served as associate editorfor this article.

Electronically published October 24, 2011

ditions was the subject of the lead article in the inauguralissue of the Journal of Consumer Research (Katona 1974).

Recession, commonly defined as two or more consecutivequarters of negative GDP growth, can affect consumer ex-penditures in several ways. First, it can reduce disposableincome (through unemployment, pay cuts, lower investmentreturns, etc.), which, in turn, leads to a smaller budget forconsumption. Second, holding disposable income constant(e.g., for those who are not directly affected financially),people tend to save more or pay down debt during a re-cession, which again leads to less money spent on goodsand services. The focus of traditional economic analysis hasbeen on these budget-related issues, attempting to under-stand how total consumption expenditures change as a func-tion of economic conditions (e.g., Deaton 1992; Hall 1993;Jappelli and Pistaferri 2010; Magrabi et al. 1991; Parker andVissing-Jorgensen 2009).

A common assumption in traditional economic analysisis that the utilities a household derives from various com-modities at different levels of expenditure would be inde-pendent of economic conditions (e.g., Deaton and Muell-bauer 1980). Under this assumption, faced with a reducedconsumption budget in a recession, consumers cut expen-ditures disproportionately more in less essential categories(e.g., dining out), resulting in smaller shares for these cat-egories and larger shares for the more essential (e.g., foodat home). When the economy starts to grow again, consum-ers are expected to increase both total consumption budgetand shares allocated to the less essential categories.

230 JOURNAL OF CONSUMER RESEARCH

This stylized relationship between economic growth,household consumption budget, and expenditure share canbe readily represented through Engel curves (Aitchison andBrown 1954; Millican 1967), which depict how expenditureon a particular commodity varies with the total budget, hold-ing prices fixed. Often presented in a budget share form,the Y-axis of an Engel curve shows the proportion of totalexpenditures spent on a particular commodity, and the X-axis shows total expenditures. More essential commoditieswould have negative-slope Engel curves (i.e., increasedshare when the total budget shrinks), while less essentialcommodities would have positive-slope Engel curves (i.e.,increased share when the total budget expands).

We attempt to go beyond the above stylized facts in thisstudy. Our basic research question is: For any given levelof consumption budget, how would a household’s expen-diture pattern (i.e., category budget shares) differ, dependingon whether the economy is in recession or not? Standardeconomic models would suggest that for the same amountof total expenditures, category budget shares would remainunchanged. A key assumption of these standard models isthat the utilities a household derives from various com-modities at different levels of expenditure are independentof economic conditions. For example, consumers shouldenjoy jewelry in a recession as much as they do in normaleconomic times; it is just that in a recession, tighter budgetsforce them to cut more in spending on jewelry, so that theycan cut less in other more essential categories. This impliesthat Engel curves, shaped by a household’s underlying util-ity function, remain unchanged in a recession, and any ob-served adjustments in category budget shares reflect move-ments along the same set of fixed Engel curves, purely asa result of a shrunken consumption budget in a recession.

In this study, we depart from the standard economic as-sumption, arguing that the utilities a household derives fromvarious commodities could vary systematically, dependingon whether the economy is in recession or not. We postulatethat people care about their relative position in a societywhen it comes to expenditures in certain categories. In arecession, their desire to spend in these “positional” cate-gories will decrease, because there is no longer a need tospend as much to maintain the same social standing whenothers have reduced their expenditures. Consequently, forany given level of consumption budget, smaller shareswould be allocated to positional expenditures in a recession.This implies that Engel curves would not remain unchangedin a recession, in contrast to what standard economic modelswould suggest. Rather, they would shift downward or up-ward, depending on the positionality of the expenditure cat-egory involved.

Overview of Our Empirical Approach. We investigatethe above postulation empirically through a repeated cross-section data set containing information about expendituresacross more than 30 major commodities from 66,368 UShouseholds over 2 decades (1982–2003), wherein the USeconomy experienced three recessions (July 1981–Novem-

ber 1982, July 1990–March 1991, and March–November2001, according to the NBER). Such a rich data set lendsexternal validity to our findings, thanks to the large andrepresentative sample of households and the comprehensivecoverage of their real-life expenditures across a full spec-trum of goods and services.

As for internal validity, however, several challenges arise.First, the utilities a household derives from various com-modities at different levels of expenditure are not directlyobservable and therefore must be inferred from observedexpenditures, and related to economic conditions. Second,although it is tempting to equate differences in expenditurepatterns to differences in preferences, it is important to dis-tinguish between those caused by variations in consumptionbudget from those caused by variations in underlying util-ities. Because a household’s consumption budget cannot beheld constant in real life, one must disentangle the impactson expenditure patterns due to variations in consumptionbudget from those due to variations in underlying utilities.Finally, besides economic conditions, many other factors,both observable and unobservable, can cause householdpreferences to vary, and must therefore be accounted foradequately. To address these challenges in our empiricalinvestigation, we rely on a structural budget allocationmodel proposed by Du and Kamakura (2008), which canbe used to infer the underlying utilities that consumers derivefrom various commodities based on observed expenditurepatterns, controlling for total consumption budget, and var-ious observed and unobserved consumer heterogeneities.

In the rest of this study we proceed as follows. First, wereview the relevant literature that leads us to postulate thatthe utilities that consumers derive from various commoditiescan differ systematically, depending on whether they livein times of economic growth or decline. Based on our syn-thesis of the reviewed literature, we propose a conceptualframework that leads to empirically testable predictions re-garding how households may adjust their budget allocations(i.e., shifting the Engel curves vertically) as a function ofeconomic conditions, after accounting for budgetary effects(i.e., moving along the Engel curves laterally). As we dis-cuss next, this framework relies on the notion of position-ality of certain goods/services from the economics literature(Frank 1985a, 1985b, 1999; Hirsch 1976), which we directlyrelate to the concepts of essentiality and visibility, two keymoderators of reference-group influence in product deci-sions according to the consumer behavior literature (Beardenand Etzel 1982; Childers and Rao 1992). Then, we presentour structural budget allocation model, which extends themodel proposed by Du and Kamakura (2008), by includingper capita GDP growth rate as a proxy for economic con-ditions (along with sociodemographics) that can potentiallyinfluence consumers’ underlying utilities. This is followedby a discussion of our data and empirical findings. We con-clude with a discussion on the implications of this studyand directions for future research.

ECONOMIC CONTRACTIONS AND EXPANSIONS 231

PRODUCT/SERVICE POSITIONALITYAND CONSUMER EXPENDITURES

Standard economic theory assumes that the utilities a house-hold derives from a particular commodity are determinedsolely by the absolute level of consumption (e.g., Deatonand Muellbauer 1980). Accordingly, each household’s ex-penditure pattern is assumed to be independently shaped byits own preferences and the consumption budget available.From such a perspective, how a household allocates a givenbudget across various commodities should not be affectedby economic conditions. In this study, we take a more nu-anced view regarding this basic assumption. We postulatethat the utilities a household derives from certain commod-ities can vary systematically, depending on whether theylive in times of economic decline or growth. In the rest ofthis section, we provide an overview of the various litera-tures that have led us to such a postulation.

Outside the canonical economic theory on consumptionexpenditure, there is a substantial body of literature arguingthat, besides the absolute level of consumption, relative con-sumption (i.e., how much one consumes as compared toothers in the society) is also an important factor in shapinga household’s spending decisions. As early as in his days,Adam Smith observed that women in England required bet-ter clothing to appear in public without shame than womenin Scotland (Alpizar, Carlsson, and Johansson-Stenman2005). Similarly, in his seminal work, Duesenberry (1949)stressed the importance of concerns about relative standingin determining consumption patterns over time. He used theidea of “demonstration effect” to explain how a household’sexpenditure pattern can be influenced by the purchases of itsneighbors, arguing that the chief motivation behind higherexpenditure was not just the resulting higher standard of livingbut also the accompanied higher social status. Easterlin’s par-adox (1974) about the relationship between happiness andincome highlighted a stylized fact that, within a given coun-try, people with higher incomes are more likely to reportbeing happy; however, in international comparisons, the av-erage reported happiness does not vary much with nationalincome per person, suggesting a central role played by rel-ative consumption in determining not only consumers’spending but also their subjective well-being.

Veblen (1899) coined the term “conspicuous consump-tion” to refer to expenditures on socially visible consump-tion that is intended to signal an individual’s relative positionin society (i.e., the idea of “signaling-by-consuming”). Muchwork has been done ever since on the relationship betweensocial status and consumption. For example, Leibenstein(1950) advanced the idea that consumer purchases wereoften driven by their desire to either distinguish themselvesfrom other consumers or to imitate them, resulting in, re-spectively, a “snob” or a “bandwagon” effect in consump-tion behavior. Bagwell and Bernheim (1996) demonstratedthat the craving for social status associated with materialdisplays of wealth can explain why consumers are oftenwilling to pay higher prices for functionally equivalent prod-

ucts. More recently, in the marketing literature, Rucker andGalinsky (2008) showed that low power can increase con-sumers’ willingness to pay for status-related products; Chris-ten and Morgan (2005) showed that, as a result of conspicuousconsumption, rising income inequality forced householdswith smaller income gains to use debt to keep up theirspending relative to households with larger income gains;and Ordabayeva and Chandon (2011) showed that greaterequality can reduce conspicuous consumption when con-sumers focus on the narrower possession gap, but it increasesconspicuous consumption when consumers focus on thegreater position gain.

In her work on “competitive consumption,” sociologistSchor (1999) points out that, although it has always beenthe case that consumer spending is, in large part, driven bya comparative process through which individual householdstry to keep up with the norms of their reference group, whatconstitutes a reference group has changed over time. Schornotes that, in the 1950s and 1960s, Americans tried to “keepup with the Joneses” down the street and compared them-selves mainly to other households of similar means; by con-trast, nowadays, for households throughout the income dis-tribution, comparisons are more likely to be made againstthe lifestyles of the upper middle class and the rich. Schorargues that such a shift occurred as people spent less timewith their neighbors and friends, while television became amore important source of cues and information, which tendsto be skewed toward the lifestyles of the upper middle classand the rich, thereby inflating the viewer’s perception aboutthe norms of the society when it comes to consumption.Similar observations were made by Frank and Levine (2006)in their work on “expenditure cascades,” which argues thatincreased consumption by the rich can trigger increasedspending in the class directly below them, and the chainreaction continues down to the bottom. Based on personalexperience tracking consumer behavior in the past 20 years,Flatters and Willmott (2009) point out that, when less for-tunate consumers are forced to live a thriftier life during arecession, many affluent consumers start to economize aswell, not only due to lowered pressure for “competitiveconsumption” but also because the recession makes discre-tionary thrift acceptable or even fashionable, another man-ifestation of relative consumption.

Closely related to the literature on relative consumptionand status-driven, conspicuous, and competitive consumerspending is the idea of positional goods (Frank 1985a,1985b, 1999; Hirsch 1976), which argues that social com-parison and relative consumption are not equally importantfor all goods. For goods that are “nonpositional,” (e.g.,bread, insurance, leisure time, and safety devices), the util-ities that consumers derive from consuming them dependlittle on how one’s own consumption compares with con-sumption by others (Frank 1985a). For goods that are “po-sitional” (e.g., jewelry, cars, apparel, old paintings, andhouses), relative consumption carries far more weight. Thus,for a positional good, expenditures will depend on consum-ers’ beliefs about how much others are spending on it. To

232 JOURNAL OF CONSUMER RESEARCH

the extent that a consumer’s perception about societal normschanges systematically as a function of economic conditions,the utilities she derives from positional goods will also varysystematically, depending on whether the economy is inrecession or not. The same, however, cannot be said aboutnonpositional goods, thanks to the lack of relative concernsfor these goods.

In a recession, consumers tend to cut back on their spend-ing. For example, 62% of Americans reduced householdspending since the “Great Recession” began in December2007, and only 6% spent more (Pew Research Center 2010).Faced with such a reality, consumers would need to spendless to achieve the same level of relative consumption (Due-senberry 1949), knowing that society would, on average,spend less. In other words, the amount of expendituresneeded to credibly signal one’s social standing would belower as total consumption decreases (Heffetz 2010), be-cause a shrunken budget would make it more costly foranyone to engage in “signaling-by-consuming” type of be-havior (Veblen 1899), especially when there are more es-sential needs to meet. Consequently, in order to maximizetotal utility, it would be desirable for a household to re-allocate some of its expenditures away from positional goodsand toward nonpositional goods in a recession. Collectively,such reallocations by individual households would result ina new social equilibrium wherein positional expendituresaccount for a smaller share of total expenditures in society(Frank 1985b). However, when the economy starts to growagain, knowing that, on average, expenditure level wouldincrease for all kinds of products and services, consumerswould increase their spending more on positional goods tomaintain their social standing. Thus, in times of economicexpansion, consumers have the incentive to allocate a largershare of consumption budget toward positional goods. Thefollowing hypothesis formalizes the above line of argu-ments:

H1: For any given level of total consumption budget,a household’s allocation of that budget across var-ious expenditures will differ systematically, de-pending on whether or not the economy is in arecession. In a recession, positional goods arelikely to see decreased budget shares and non-positional goods are likely to see increased budgetshares; the reverse takes place in times of eco-nomic expansion.

To empirically test hypothesis 1 with actual household bud-get allocation data would require a direct measure of theextent to which a particular expenditure category is posi-tional, which unfortunately is not available from the existingliterature (Alpizar et al. 2005; Carlsson, Johansson-Stenman,and Martinsson 2007; Solnick and Hemenway 1998, 2005).However, because the idea of “relative consumption” canbe viewed as a variant of reference-group influence (Gasana2009), the positionality of a particular commodity shouldbe a function of the extent to which a household’s expen-diture is influenced by its reference group. In other words,

any measure of positionality should be highly correlatedwith the degree of reference-group influence.

Fortunately (we are grateful to a reviewer for pointing usto this literature), researchers in the consumer behavior lit-erature have established a substantial body of work regard-ing the boundary conditions of reference-group influence onconsumers’ product decisions (Escalas and Bettman 2003).In particular, Bourne (1957, 218) proposed that two elementsof “conspicuousness”—exclusiveness and visibility—canmoderate reference-group influence on consumer purchasedecisions. In their empirical investigation, Bearden and Etzel(1982) operationalized exclusiveness as the distinction be-tween luxury versus necessity (i.e., whether needed for or-dinary, day-to-day living), and visibility as the distinctionbetween publicly versus privately consumed (i.e., whetherother people outside of the immediate family are aware ofone’s consumption), which is related to the distinction Rich-ins (1994) makes between the public and private meaningsof possessions. Bearden and Etzel’s survey results indicatedthat, for product decisions, reference-group influence was(1) substantially stronger for public luxuries than for privatenecessities; (2) in large part equivalent between public lux-uries and public necessities; (3) significantly stronger forprivate luxuries than for private necessities; and (4) signif-icantly stronger for public necessities than for either privateluxuries or private necessities, which they argue may reflectstrong concerns about appearing in public without productsthat are considered necessary for ordinary living. Childersand Rao (1992) replicated and extended the work of Beardenand Etzel. Childers and Rao’s findings (of those based onsurvey participants in the United States) were directionallyconsistent with those of Bearden and Etzel, further sup-porting the idea that reference-group influence is substan-tially stronger for luxuries consumed in public as comparedto products that are not conspicuously consumed.

Given the above, we therefore reformulate hypothesis 1as follows:

H1a: For any given level of total consumption budget,a household’s allocation of that budget acrossvarious expenditures will differ systematically,depending on whether or not the economy is ina recession. In a recession, nonessential goodsthat are consumed in more visible circumstancesare likely to see decreased budget shares, andessential goods that are consumed in less visiblecircumstances are likely to see increased budgetshares; the reverse takes place in times of eco-nomic expansion.

Graphically, hypothesis 1a implies that in a recession, oneshould expect Engel curves of more visible nonessentialgoods to shift downward, and those of less visible essentialgoods to shift upward, and the reverse in times of economicexpansion. As for the Engel curves of more visible essentialgoods or less visible nonessential goods, hypothesis 1a isnondirectional.

In order to classify the more than 30 major household

ECONOMIC CONTRACTIONS AND EXPANSIONS 233

TABLE 1

PREDICTIONS OF ENGEL CURVE SHAPE AND SHIFT IN A RECESSION

Public (higher social-cultural visibility) Private (lower social-cultural visibility)

Less essential (upward-sloping Engel curve) Positional (Engel curve shifts down dur-ing recessions):

Apparel (3) Charity (23)Apparel services (3) Alcohol at home (8)Home furnishings and appliances (4) Personal businessJewelry and watches (5) Personal insurance (30)Recreation (6) Medical nonprescription (22)Food away from home (7) Medical doctors (22)Personal care (8) Motor fuel (21)Alcohol away from home (9) Motor insurance (29, 30)Lodging away from home (18) Motor services and parts (20)Household operations/maintenanceEducation (13)Airfare (17)

More essential (downward-sloping Engel curve) Nonpositional (Engel curve shifts upduring recessions):

Public transportation (19) Food at home (14)Tobacco (1) Housing rental or equivalent (15)Telephone (16) Medical prescriptions (22)

Electricity (25)Health insurance (29, 30)Water, sewer, and trash (25)Home gas, fuel, coal (25)

NOTE.—Values in parentheses are visibility rankings reported by Heffetz (2011, table 3) out of 31 categories, based on a survey of 480respondents in the United States; the classification between less vs. more essential categories rely in large part on Du and Kamakura (2008,table 3).

expenditure categories in our data (i.e., the Consumer Ex-penditure Survey from the Bureau of Labor Statistics) intoessential versus nonessential, we rely conceptually on the“hierarchy of needs” (e.g., Maslow 1943), which posits thatconsumers first attempt to satisfy “physiological” needs, andonce these needs are met, move on to satisfy safety andsocial needs, and finally to self-actualization needs. Becausethe satisfaction of the more basic needs cannot be easilypostponed, substituted, or forsaken, expenditure shares foressential goods should be higher among households withsmaller consumption budgets (i.e., negative-slope Engelcurves). Conversely, one should expect higher expenditureshares for nonessential goods among households with largerconsumption budgets (i.e., positive-slope Engel curves).Empirically, we follow Du and Kamakura (2008, table 3)in our classification of essential versus nonessential expen-diture categories. Their findings suggest that the followingare essential consumption needs: food at home; tobacco andsmoking products; health insurance; telephone services;electricity; water, sewer, and trash collection services; andgas, heating oil, and coal. Based on conventional wisdom,we also classify public transportation, housing rental orequivalent, and medical prescriptions as essential goods. Fi-nally, we categorize expenditures on medical doctors andthose associated with the utilization of an automobile asnonessential because most poor households in the UnitedStates can substitute them with other publicly available ser-vices.

In order to classify all the household expenditure cate-gories in our data into high visibility versus low visibility,

we rely heavily on the work of Heffetz (2011), which createda survey-based, social-cultural (as opposed to just physical)visibility index of consumer expenditures, roughly definedas the speed with which members of society notice a house-hold’s expenditures on different commodities. A commodityis considered more visible if, in the cultural context in whichit is consumed, society has more means to correctly assessthe expenditures involved. Heffetz (2011) found that, forUS consumers, expenditures on goods such as cigarettes,cars, clothes, furniture, appliances, jewelry, and equipment(TV, video, audio, music, and sports) are the most visible,while household expenditures on insurance policies, legaland accounting fees, and utilities bills are the least visible.

Table 1 summarizes our classification of all the majorhousehold expenditure categories in our data into essentialversus nonessential and high versus low visibility (we ex-cluded medical hospital due to the extremely low incidenceof this expense in our sample). The numbers shown withinparentheses in table 1 are visibility rankings reported byHeffetz (2011, table 3). However, because our focus is onthe visibility of expenditures during an economic cycle, ourhigh-low visibility classification departs from Heffetz’ vis-ibility rankings in six incidences (out of a total 31 cate-gories), five of which are near the median of the rankings(i.e., 14th, 15th, 17th, 18th and 19th), where there is moreuncertainty in the underlying visibility index.

We categorize expenditures on telephone, with a medianHeffetz’ visibility ranking of 16, as visible because it is amedium used to communicate with one’s social network,and would therefore be noticeable if one’s usage changed

234 JOURNAL OF CONSUMER RESEARCH

substantially. We classify expenditures mostly related tofamily vacations (airfare, lodging away from home) as vis-ible, because being away on a vacation would be fairlynoticeable to one’s neighbors, friends and colleagues; Hef-fetz’ survey ranked these two categories just below the me-dian in visibility (17th and 18th, respectively). We also clas-sify expenditures on public transportation (19th) as visible,because consumers are likely to be able to notice heavier/lighter usage of public transportation by others. We classifyhousing rental or equivalent (15th) as less visible becauseexpenditures on shelter are unlikely to change substantiallyin the short term, except in extreme circumstances such asforeclosures or evictions. Finally, we classify expenditureson food at home (14th) and alcohol at home (8th) as lessvisible, because they are much less likely to be noticed byothers beyond the immediate family, as compared to foodand beverages consumed away from home.

Also presented in table 1 are, for all the expenditure cat-egories, our expectations regarding the slopes of their re-spective Engel curves and more importantly, the shifts ofthe Engel curves themselves in a recession, which are pre-dicted according to hypothesis 1a (i.e., to have an Engelcurve shifting upward/downward is equivalent to seeing in-creased/decreased category budget share at all levels of totalbudget). We test these predictions empirically through thestructural model of household expenditure described next.

MEASURING THE IMPACT OFECONOMIC CONTRACTIONS/EXPANSIONS ON CONSUMER

EXPENDITURES

Obviously, shifts in relative desirability between positionaland nonpositional commodities would result in changes inbudget shares. Again, it is important to distinguish budgetshare changes due to shifts in relative desirability fromchanges due to variations in the total budget. The formermeans moving the entire Engel curve up or down, whilethe latter means moving from one point to another alongthe same Engel curve. For example, faced with a smallerbudget during a recession, consumers may cut more heavilyin less essential categories (which may include more posi-tional goods), and less so in more essential categories (whichmay include more nonpositional goods), leading to changesin budget shares, even though the underlying desirability ofeach category remains the same. In other words, shifts incategory desirability will lead to changes in category shares,but category share changes do not necessarily imply shiftsin category desirability, unless the total budget is held con-stant. Of course, this brings a real empirical challenge, be-cause (1) a household’s total budget cannot be held constantover time, and (2) the relative desirability of each categoryis not directly observable and can only be inferred frompatterns of budget allocation. In the rest of this section, wepresent our empirical strategy to disentangle budget sharevariation due to relative category desirability shifts frombudget share variation due to total budget changes. By doing

so, we set out to investigate the impact of economic growthrate on consumers’ underlying preferences.

Given the main goal of our empirical investigation (i.e.,to quantify the impact that economic contractions/expan-sions have on how households allocate their total expen-ditures across various commodities), we abstract away frommodeling the total consumption budget decision for tworeasons. First, the intertemporal trade-off between more con-sumption now versus more savings for consumption in thefuture has been the focus of traditional economic analyses,and a vast literature can be found elsewhere (e.g., Gour-inchas and Parker 2002). Second, each household in oursample reported its expenditures for only one year, pre-cluding any type of dynamic analysis necessary for the studyof intertemporal decisions about consumption versus sav-ings. Instead, we focus on examining how households al-locate a given consumption budget (i.e., the cross-categorytrade-off between more consumption in one category versusmore in another). For this purpose, we need a model thatconsiders how households allocate a consumption budgetacross all expenditure categories to maximize the total utilitythey accrue from this budget. Such a model allows us toinfer the relative desirability of each expenditure category.Applying the model to household-level expenditure dataover time allows us to investigate how shifts in categorydesirability may occur as the economy goes through variousperiods of expansion and contraction.

To accomplish this goal, we extend the budget allocationmodel recently applied by Du and Kamakura (2008) in astudy of expenditure patterns among American households.In particular, our model incorporates observed heterogene-ity, captured by (1) sociodemographic variables describingeach household and (2) an indicator of economic growthrate during the year when the expenditures were observed.Because our main focus is on quantifying the shifts in cat-egory desirability during economic recessions/expansions,and because the model we apply is a direct extension of apublished model, we describe the model only briefly next,and refer readers to Du and Kamakura (2008) for moredetails.

We assume that household h maximizes a continuouslydifferentiable quasi-concave utility function G(xh) over aset of J nonnegative quantities xh p (x1h, x2h, . . ., xJh),subject to a budget constraint , where p p′p x ≤ mh h

(p1, p2, . . ., pj)′ 1 0, pi is the price of good in category i,

and mh is household h’s total consumption budget. Follow-ing Du and Kamakura (2008), we use the Stone-Geary utilityfunction, which has the form

J

G(x ) p a ln (x � b ), (1)�h ih ih iip1

where , , and J is the number of avail-a 1 0 (x � b ) 1 0ih ih i

able categories. Note the h subscript in , which impliesa ih

that the utility derived from spending in category i, andtherefore its desirability, is household specific. This budgetallocation problem implies that the household incrementally

ECONOMIC CONTRACTIONS AND EXPANSIONS 235

allocates its disposable income to the consumption categorythat produces the highest marginal utility per dollar,

�G(x ) 1 ah ihp ,�x p (p x � p b )ih i i ih i i

given the current expenditure levels xh, until the budget isreached, . Solving this optimization problem

J� p x p mi ih hip1

leads to an expenditure system that is linear in total budgetand prices,

∗J

∗ ∗p x p p b � v m � p b , i p 1, 2, … , J , (2)�( )i ih i i ih h j jjp1

where , and J* is the set of categories with∗J∗v p a / � aih ih jhjp1

positive expenditures.In order to account for unobserved heterogeneity (i.e.,

diversity in underlying preferences not accounted by ob-served predictors) in the taste parameter (aih) for each cat-egory i and still have a model of feasible size, Du andKamakura (2008) use a factor-analytic random-effectsmodel by extracting the principal factors of the covariancematrix of the individual-specific terms, a formulation similarin purpose to traditional factor analysis, which attempts tocapture most of the covariance in observed variables througha few common components, except that in our model, thefactor decomposition is applied to unobservable random ef-fects. This formulation leads to

a p exp (g � l Z � � ), and (3)ih i i h ih

b p min (x ) � exp (h ), to ensure thati i i (4)x � b 1 0 for Gh,ih i

where

is the geometric mean of the taste parameter forgie a ih

category i across the sample;Zh is a p-dimensional vector of i.i.d standard normal

factor scores for household h;is a p-dimensional vector of factor loadings for cat-l i

egory i;is a random disturbance normally distributed with�ih

mean zero and standard deviation ji.

We extend the model by incorporating observed hetero-geneity, captured by the sociodemographic characteristicsof each household (Wh) and by variables capturing the eco-nomic environment (Qt) in year t. With this simple exten-sion, equation 3 becomes

a p exp(g � l Z � dW � t Q � � ). (3a)iht i i h i h i t ih

While gi provides insights into the average taste or desirefor category i across households over time, the product ofthe factor loadings (li) and factor scores (Zh) will show howmuch higher or lower the taste of household h is, relativeto the average, thereby capturing differences in the desira-bility of category i for different households due to unob-served sources. In addition to these differences in tastes due

to unobserved sources, our extension of the Du and Ka-makura (2008) model also captures the impact of the ob-served sociodemographic profile of a household (throughdi) and of shifts in the economic environment (through ti),the latter of which being the focus of the current study. Foridentification purposes, tastes for one expenditure categoryare set to unit ( ) without loss of generality.a p 11ht

The main purpose for our application of the above budgetallocation model is to quantify relative category desirabil-ities in order to assess how the Engel curves for differentconsumption categories may shift up or down, in responseto changes in the economic environment. As discussed ear-lier, we expect that in periods of economic expansion, evenconsumers who do not face an increase in their consumptionbudget will tend to devote more of their budgets to positionalgoods, because they perceive a general increase in con-sumption of these goods, and therefore, a need to spendmore in order to maintain their relative standing. In periodsof economic contraction, the reverse is expected.

Given the budget allocation model defined by equations1–4, the budget share for category i in household h and yeart, Siht, is determined by

∗Jp pit jt∗s p b � v 1 � b , (5)�iht i iht j( )m mjp1ht ht

where

exp (g � l z � dW � t Q )i i h i h i t∗v p ,∗iht J� exp (g � l z � d W � t Q )j j h j h j tjp1

and J∗ the set of categories with positive expenditures. Po-sitionality effects due to changing economic conditions aredirectly captured by in equation 5. Because economicv ∗iht

growth rate, Qt, can potentially affect all categories via thecoefficient t, its impact on the budget share of category idepends on the relative magnitude of ti compared to all theother categories. More specifically, a shift of �Qt in theeconomic growth rate will result in an upward/downwardshift of

∗J∗�viht ∗ ∗p v t � t v�( )iht i j jht�Q jp1t

in the Engel curve.Once the model parameters are estimated for a sample of

households, counterfactual analyses can be easily performedon that same sample, by simulating the budget allocationprocess for each household under different economic growthrates, for any given level of consumption budget, so thatchanges in expenditure patterns can be solely attributed tochanges in category desirability under different economicconditions. For this simulation, one needs only to apply asimple allocation heuristic implied by the model in equations1–4 for each household:

1. Start with zero allocations to all categories (miht p$0) and a full budget.

2. Calculate all categories’ marginal utilities �Uiht /�miht

p exp(aiht)/(miht � pibi).

236 JOURNAL OF CONSUMER RESEARCH

3. Allocate a dollar to the category i with the highestmarginal utility (miht p miht � $1).

4. Deduct a dollar from the total budget.5. Repeat steps 2–4 until the budget is totally depleted.

EMPIRICAL RESULTS

The data for our empirical investigation comes from theConsumer Expenditure Survey (CEX) family-level extracts,which are made available by the National Bureau of Eco-nomic Research (NBER) for the 1982–2003 period (seehttp://www.nber.org/data/ces_cbo.html for more details onhow NBER made the extracts using raw data from the Bu-reau of Labor Statistics). Unfortunately, the NBER has notreleased more recent family-level extracts (after 2003) as ofyet. Even though raw CEX data from 2004 through 2009are now available to the public through the Bureau of LaborStatistics (http://www.bls.gov/cex/), these raw data must besubjected to NBER’s proprietary extraction procedure; oth-erwise, these new data will not be compatible with NBER’shistorical extracts (1982–2003) used in the current study.

The CEX Survey is collected from different samples eachyear, so that each of the 66,368 households in our studyreports its consumption expenditures for only one year, andtherefore cannot be treated as a longitudinal panel (rather,it is a repeated cross-section). The CEX-NBER extracts con-tain the dollar amounts allocated by each sample householdduring a one-year window across 32 consumption categories(we excluded the purchase of new or used automobiles,because they usually involve long-term savings or loan pay-ments, and therefore do not represent discretionary expen-ditures within a single year). For each of the 32 categoriesand 22 years in the CEX-NBER extracts, we collected therelevant price index from the Bureau of Labor Statistics,which we normalized with 2003 as the base year. Asidefrom the unobserved heterogeneity in the random interceptsof the taste parameters, we use the following exogenousvariables to explain variations in preferences across house-holds and economic conditions:

education: number of years of schooling by the house-hold head

age: age of the household headkid ! 7: number of children under age 7kid7–18: number of children ages 7–18kid_college: number of children attending collegeother adults: number of adults in the household (other

than the household head)married: indicator variable equal to 1 if the household

head is married, 0 otherwisemale: 1 if the household head is male, 0 otherwisewhite: 1 if the household head is white, zero otherwiseemployed: 1 if the household head is in the labor force,

zero otherwisegdp growth: percentage growth in GDP per capita rel-

ative to the previous year (we use this measure be-cause it is often viewed as the single best proxy

available for average standard of living in a particulareconomy).

Summary statistics for these variables in our sample arereported in table 2.

Using these same expenditure and price data, Du andKamakura (2008) found that a six-factor solution was suf-ficient to capture the unobserved heterogeneity in consumerpreferences. Because we incorporate observed heterogeneitywith the exogenous variables discussed above, there is lessvariance in preferences to be explained by the latent factors,and therefore, we need a smaller latent space (four factors),producing the estimates reported in table 3.

Interpretation of the results reported in table 3 becomeseasier if one considers their implications for the predictedbudget share, shown in equation 5. From this equation, onecan see that all predictors (Wh and Qt) affect the budgetshares through the multinomial logistic function , which∗viht

determines whether allocated shares shall shift upward ordownward for any given budget. Therefore, the impact ofthese predictors must be interpreted within the context of amultinomial logistic function. For example, because of theidentification restrictions in the multinomial logistic function( ), the response parameters (d, t) can onlyg p d p t p 01 1 1

be interpreted in relation to the baseline category, set asfood at home. Thus, the results in the “GDP growth” columnof table 3 suggest that in a recession (i.e., negative growthin per capita GDP), consumer tastes for charity; tobacco;medical prescriptions; motor insurance; and water, sewer,and trash increase relative to food at home, so that more ofthe consumption budget is allocated to the former categoriesthan to food at home, compared to normal economic con-ditions. Similarly, parameter estimates in the “Education”column of table 3 suggest that taste for tobacco is higher,relative to food at home, for households with a less educatedhousehold head, so that a higher share of the consumptionbudget is allocated to the former, compared to householdswith more educated heads.

As discussed earlier, the main role of the four latent factors(Zh) is to capture individual differences in tastes that cannotbe explained by the observed predictors (Wh and Qt). There-fore, their respective loadings (li) serve only to capturepotential correlations in preferences across categories, andare not necessarily interpretable. Nevertheless, the loadingsreported in table 3 reveal some interesting correlation pat-terns. For example, the loadings for the second factor sug-gest strong correlations between alcohol at home and alcoholaway from home, indicating that households reportinghigher budget allocations for alcohol consumed at home alsotend to report higher allocations for alcohol consumed awayfrom home. Similarly, loadings for the third factor indicatea negative correlation between allocated shares for motorfuel and public transportation, suggesting that householdsallocating higher than average shares to one tend to reportlower than average shares to the other. These intuitive resultslend face validity to this component of our budget allocationmodel.

More meaningful and interpretable insights are obtained

TA

BL

E2

SU

MM

AR

YO

FS

OC

IOD

EM

OG

RA

PH

ICA

ND

EC

ON

OM

ICP

RE

DIC

TO

RS

,M

EA

NS

AN

DS

TA

ND

AR

DD

EV

IAT

ION

Yea

rP

erca

pita

GD

Pgr

owth

Yea

rsof

educ

atio

n

Age

ofth

ehe

adof

hous

ehol

d

Num

ber

ofki

dsag

es1–

6

Num

ber

ofki

dsag

es7–

18

Num

ber

ofki

dsin

colle

geN

umbe

rof

othe

rad

ults

Mar

ried

head

ofho

useh

old

(%)

Mal

ehe

adof

hous

ehol

d(%

)

Whi

tehe

adof

hous

ehol

d(%

)

Em

ploy

edhe

adof

hous

ehol

d(%

)T

otal

budg

et(2

003

pric

es)

Sam

ple

size

1982

�2.

012

.748

.5.5

.7.1

.462

6887

7132

,108

2,62

4.0

2.8

17.3

1.0

1.2

.3.9

4947

3445

16,8

7319

834.

512

.649

.3.5

.7.1

.460

6886

7033

,300

2,81

8.0

2.9

17.4

1.0

1.2

.3.9

4947

3546

18,0

3419

847.

212

.549

.8.5

.7.1

.462

6886

7132

,781

3,26

5.0

2.9

17.5

1.0

1.2

.3.9

4947

3446

18,9

1619

854.

112

.749

.2.5

.7.1

.464

6987

7333

,292

1,57

0.0

2.8

17.0

1.1

1.2

.3.9

4846

3444

18,9

9019

863.

412

.749

.5.3

.4.0

.362

6787

7133

,499

3,23

5.0

2.8

17.5

.7.9

.2.7

4947

3446

19,0

7219

873.

312

.650

.2.3

.4.0

.362

6887

6933

,467

3,09

9.0

2.2

17.2

.8.8

.2.7

4947

3446

18,9

1319

884.

112

.750

.1.3

.4.1

.361

6787

7034

,478

3,11

5.0

2.2

17.2

.8.8

.2.8

4947

3346

19,4

1819

893.

512

.949

.8.3

.4.0

.362

6688

7335

,580

3,07

1.0

2.8

17.0

.7.9

.2.8

4847

3245

20,5

5219

901.

912

.950

.3.3

.3.1

.361

6686

7034

,517

3,10

9.0

2.8

17.3

.8.8

.3.8

4947

3546

20,0

4519

91�

.212

.950

.1.3

.5.1

.361

6687

7033

,939

3,18

2.0

2.8

17.2

.8.9

.3.7

4947

3446

20,0

5019

923.

312

.950

.5.3

.5.1

.459

6486

6933

,647

3,15

0.0

2.8

17.2

.7.9

.3.7

4948

3547

19,3

9619

932.

713

.149

.8.3

.5.1

.458

6587

7133

,929

3,25

1.0

2.8

17.2

.7.9

.3.7

4948

3446

19,9

6319

944.

113

.150

.5.3

.5.1

.458

6387

7034

,419

2,98

8.0

2.6

17.0

.7.9

.3.7

4948

3446

19,8

7019

952.

512

.950

.5.3

.5.1

.459

6386

7033

,581

1,35

0.0

2.5

16.8

.7.9

.2.7

4948

3546

18,5

1319

963.

713

.250

.3.3

.6.1

.458

6286

6834

,310

2,64

3.0

2.6

17.3

.71.

0.3

.749

4935

4719

,722

1997

4.5

13.2

51.3

.3.5

.1.3

5761

8567

34,7

392,

950

.02.

617

.1.7

.9.2

.750

4936

4720

,872

1998

4.2

13.3

51.0

.3.5

.1.3

5960

8767

34,3

632,

762

.02.

617

.1.7

.9.2

.749

4933

4719

,741

1999

4.5

13.3

51.2

.3.5

.1.4

5856

8569

36,2

393,

654

.02.

517

.1.7

.9.3

.749

5036

4626

,581

2000

3.7

13.3

51.3

.3.5

.1.4

5853

8469

35,1

683,

797

.02.

517

.0.7

.9.3

.749

5037

4620

,572

2001

.813

.451

.1.3

.5.1

.458

5285

7036

,250

4,03

7.0

2.5

16.8

.6.9

.3.8

4950

3646

21,2

6820

021.

613

.451

.8.3

.5.1

.457

5285

6836

,344

4,46

3.0

2.6

16.7

.6.9

.3.7

5050

3647

21,5

2020

032.

513

.451

.5.3

.5.1

.457

5185

7036

,217

2,34

5.0

2.5

16.9

.7.9

.3.7

5050

3646

23,2

44T

otal

3.1

13.0

50.4

.3.5

.1.4

6062

8670

34,4

9966

,478

1.9

2.7

17.1

.81.

0.3

.849

4935

4620

,387

TA

BL

E3

PA

RA

ME

TE

RE

ST

IMA

TE

SF

OR

TH

EB

UD

GE

TA

LLO

CA

TIO

NM

OD

EL

gt

d2

jl

b

Exp

endi

ture

sIn

terc

ept

GD

Pgr

owth

Edu

catio

nA

geK

id!7

Kid

7–18

Kid

_co

llege

Oth

erad

ults

Mar

ried

Mal

eW

hite

Em

ploy

edV

aria

nce

Load

1Lo

ad2

Load

3Lo

ad4

Bet

a

Foo

daw

ayfr

omho

me

-2.8

3*.0

3*.2

9*-.

09*

-.09

*-.

06*

.02*

.00

.00

.11*

.13*

-.06

*.5

0*-.

44*

�.3

1.0

2.2

9-.

010*

Tob

acco

-1.4

1*-.

02*

-.09

*-.

22*

-.10

*-.

06*

-.05

*.1

5*�

.01

.05*

.05*

.00

1.02

*�

.04

�.3

1.0

0.0

5-.

009*

Alc

ohol

atho

me

-4.0

6*.0

0.2

5*-.

19*

-.06

*-.

12*

�.0

1.0

4*-.

02*

.16*

.11*

-.05

*.6

7*�

.19

-.61

*.0

5.1

7-.

005*

Alc

ohol

away

from

hom

e-6

.34*

.01*

.50*

-.39

*-.

13*

-.20

*.0

1*.0

7*-.

10*

.24*

.28*

-.11

*.8

8*�

.53

-1.0

5*.2

0.1

9-.

001*

App

arel

-2.1

6*.0

3*.2

1*-.

14*

-.01

*.0

0.0

3*.0

0.0

7*-.

04*

.04*

-.05

*.4

6*-.

47*

�.1

5.0

2.1

8-.

011*

App

arel

serv

ices

-4.6

5*.0

3*.3

2*-.

15*

-.05

*-.

12*

.02*

-.01

*-.

10*

.06*

-.10

*-.

07*

.99*

�.5

4�

.24

.13

.01

-.00

1*Je

wel

ry&

wat

ches

-6.5

6*.0

6*.4

1*-.

24*

-.10

*-.

03*

.04*

.08*

.24*

.01

.16*

-.09

*1.

18*

-.94

*�

.28

.04

.26

-.00

2*P

erso

nalc

are

-3.7

6*.0

3*.2

0*.0

7*-.

05*

-.07

*.0

2*.0

0.0

6*-.

02*

.00

-.03

*.5

4*�

.30

�.1

5.0

3.2

6*-.

004*

Per

sona

lbus

ines

s-5

.71*

.02*

.38*

-.03

*-.

08*

-.06

*.0

0.0

7*.1

3*.0

6*.1

6*-.

07*

1.21

*�

.49

�.2

9�

.05

.53

-.00

1*P

erso

nali

nsur

ance

-3.5

6*.0

1*.2

3*.0

9*-.

06*

-.04

*.0

1*.0

2*.1

9*.0

3*-.

02*

-.03

*.6

8*�

.27

�.1

0.0

0.4

4*-.

014*

Med

ical

pres

crip

tions

-6.5

9*-.

02*

.13*

.59*

-.01

*-.

11*

-.02

*�

.01

.28*

-.14

*.1

8*.0

01.

26*

�.3

6.1

3�

.05

.57

-.00

1*M

edic

alno

npre

scrip

tion

-6.1

3*.0

2*.2

9*.1

9*-.

09*

-.03

*.0

3*.0

1.1

5*-.

02*

.14*

-.03

*.9

0*�

.45

�.0

8.0

5.4

3-.

004*

Med

ical

doct

ors

-5.1

6*.0

2*.2

8*.2

1*-.

02*

-.03

*.0

2*-.

02*

.20*

-.03

*.2

0*-.

04*

.91*

�.4

5�

.02

�.0

1.4

4-.

003*

Med

ical

hosp

ital

-8.5

7*�

.01

.19*

.19*

.11*

-.03

*�

.01

.07*

.45*

-.06

*.2

2*.0

02.

24*

�.4

9.2

1.0

0.6

7-.

002*

Hea

lthin

sura

nce

-3.8

4*�

.01

.17*

.45*

-.03

*-.

10*

�.0

1-.

04*

.09*

-.05

*.0

6*�

.01

.78*

�.1

2�

.05

.05

.50*

-.01

0*M

otor

serv

ices

&pa

rts

-3.8

9*.0

2*.3

3*-.

08*

-.08

*-.

07*

.03*

.07*

.12*

.10*

.13*

-.07

*.7

3*�

.44

�.3

2�

.30

.48*

-.00

5*M

otor

fuel

-1.6

8*.0

0.1

1*-.

11*

-.05

*-.

05*

.02*

.05*

.08*

.08*

.09*

-.03

*.3

8*�

.21

�.2

0-.

23*

.34*

-.02

4*M

otor

insu

ranc

e-3

.02*

-.01

*.2

3*-.

01*

-.09

*-.

09*

.03*

.05*

.11*

.04*

.11*

-.04

*.5

5*�

.24

�.2

3�

.19

.50*

-.01

0*T

rans

port

atio

npu

blic

-6.2

5*.0

2*.4

2*.0

7*-.

09*

-.06

*.0

3*.0

9*-.

12*

-.07

*-.

23*

-.06

*1.

42*

�.5

5�

.43

1.00

*�

.13

-.00

1*T

rans

port

airf

are

-6.3

4*.0

3*.5

9*.0

8*-.

10*

-.12

*.0

4*.0

5*.0

8*.0

0.0

8*-.

08*

.86*

�.5

9�

.52

.49

.39

-.00

9*R

ecre

atio

n-2

.56*

.02*

.30*

-.11

*-.

06*

-.04

*.0

1*-.

01*

.04*

.05*

.13*

-.07

*.4

9*-.

38*

�.2

3.0

5.3

2*-.

010*

Edu

catio

n-8

.09*

.02*

.93*

-.43

*.0

7*.1

3*.3

1*.0

3*.3

1*-.

10*

.16*

-.16

*1.

64*

�.7

8�

.17

.11

.44

-.00

1*C

harit

y-6

.77*

-.05

*.5

1*.4

1*-.

03*

-.04

*.0

4*-.

05*

.23*

�.0

1.0

7*-.

07*

1.22

*�

.60

�.0

6.0

8.5

7-.

005*

Tel

epho

ne-1

.84*

.01*

.10*

-.03

*-.

04*

-.07

*.0

1*.0

0-.

05*

-.05

*-.

02*

-.02

*.5

0*�

.11

�.1

0�

.02

.28*

-.00

5*Lo

dgin

gaw

ayfr

omho

me

-6.6

4*.0

3*.5

4*.0

4*-.

12*

-.08

*.0

5*.0

3*.2

2*.0

5*.2

0*-.

09*

.78*

-.64

*�

.47

.21

.43*

-.00

4*H

ome

furn

ishi

ngs

and

appl

ianc

es-3

.68*

.05*

.25*

-.09

*-.

03*

-.08

*.0

1*.0

3*.1

9*.0

0.1

4*-.

06*

.90*

�.5

7�

.16

.00

.31

-.00

6*H

ouse

hold

oper

atio

ns/m

aint

enan

ce-4

.67*

.02*

.30*

.22*

.05*

-.07

*.0

1-.

05*

.11*

-.02

*.1

3*-.

05*

.72*

�.3

0�

.15

.04

.51*

-.00

8*E

lect

ricity

-1.6

6*.0

2*.0

6*.0

4*-.

03*

-.04

*.0

0.0

0*.0

2*.0

0.0

3*-.

01*

.40*

�.0

6�

.07

�.0

7.3

3*-.

017*

Wat

er,

sew

er,

and

tras

h-3

.85*

-.01

*.1

9*.1

5*-.

03*

-.02

*.0

1*.0

2*.1

1*-.

02*

.06*

-.02

*.5

2*�

.07

�.1

2.0

3.5

5*-.

007*

Hom

ega

s,fu

el,

coal

-2.6

6*.0

1*.1

0*.1

2*-.

01*

-.03

*.0

1.0

0.0

3*.0

0.0

5*.0

0.5

8*�

.10

�.1

4.0

1.3

6*-.

018*

Hou

sing

rent

alor

equi

vale

nt-.

16*

.00

.15*

.08*

-.03

*-.

09*

.00

-.04

*-.

06*

.00

.04*

-.03

*.3

8*�

.08

�.1

5.0

5.3

0*-.

045*

Foo

dat

hom

e.0

0*.0

0*.0

0*.0

0*.0

0*.0

0*.0

0*.0

0*.0

0*.0

0*.0

0*.0

0*.0

0*.0

0.0

0.0

0.0

0-.

028*

NO

TE.—

GD

Pp

gros

sdo

mes

ticpr

oduc

t.*V

alue

sar

est

atis

tical

lysi

gnifi

cant

atth

e.0

1le

vel.

ECONOMIC CONTRACTIONS AND EXPANSIONS 239

by combining the parameter estimates in table 3 with theobserved sociodemographic and the latent factor scores forthe 66,368 households, along with the per capita GDPgrowth rates over the 22 years, to produce theoretical Engelcurves for the 32 consumption categories. Given that ourmain purpose is to quantify how category desirabilities shiftduring economic recessions and expansions, we producethree Engel curves for each category, using the observedbudget and sociodemographic profile, and the estimated la-tent scores for each household, but assuming three differentlevels of per capita GDP growth: the historical average(3.1% growth) as the baseline, a 2% decline (equivalent tothe 1982 recession), and a 6% growth (close to the highestgrowth of 7.3% observed in 1984). These Engel curves areshown in figures 1 and 2 for the categories where we foundthe most significant shifts from the baseline.

Comparing these three sets of Engel curves is insightfulbecause it illustrates how economic contraction versus ex-pansion can affect household expenditure patterns in a sys-tematic fashion, after partialling out any effects due to bud-get changes. Note that because budget effects can only movea household from one point to another along the same Engelcurve, any shifting up or down of the Engel curve itselfmust be a result from shifts in category desirabilities. Inother words, whenever such shifts in consumer tastes occur(as our results clearly indicate), they will cause the Engelcurves to move up or down, depending on economic ex-pansion/contraction and the particular category involved,which is exactly what figure 1 shows for the positional andnonpositional commodities for which we have predictionsin table 1. (Note that due to the wide range of expenditureshares, different scales are used for each expenditure cate-gory.)

When developing our hypotheses we conjectured that theEngel curve for a nonessential category would have a pos-itive slope with the budget, because as the budget increases,households can afford to spend larger shares on nonessen-tials. Among all the categories listed in table 1, we foundonly two Engel curves that did not match our predictions.We expected water, sewer, and trash to be essential, whenempirically its Engel curve shows a positive slope with thebudget, suggesting it is nonessential. This seemingly oddresult might be due to the fact that renters generally do notincur these expenses directly, and therefore these expensesare associated with households who own their homes andtend to have higher total consumption budgets. Another cat-egory that somewhat contradicts our expectation is apparelservices, which was expected to be nonessential, but doesnot have a positive-slope Engel curve. However, as shownin figure 1, the Engel curve for apparel services suggeststhat it is not a clear-cut essential category either. Overall,out of the 31 predictions regarding the slope of the Engelcurve (we excluded medical hospital because of the verylow incidence of this expense in our sample), we found onlytwo exceptions. One of them (water, sewer, and trash) hasa plausible explanation, while the other (apparel services)

shows an Engel curve that does not clearly reflect essenti-ality.

Regarding our hypothesis 1a, figure 1 shows the Engelcurves for the expenditure categories we classify as bothnonessential and visible. These are positional goods or ser-vices for which we expect the Engel curves to have a pos-itive slope with the total budget, and to shift downwardduring recessions. In all of the 12 categories, we see theexpected shift (for education, the shift is in the right direc-tion, but minor); that is, for any given budget, during arecession (dashed line), the shares devoted to these cate-gories decrease relative to the baseline scenario (solid line),while the opposite happens during economic booms (dottedline), as one would expect for positional goods (i.e., peoplecare more about their relative standing when it comes todining out, dressing up, being pampered, buying new fur-nishings and appliances, being entertained, and flying around).However, as discussed earlier, the slope of the Engel curvefor apparel services does not confirm this expenditure cate-gory as nonessential.

Our hypothesis 1a also posited that expenditures that areessential and have lower social-cultural visibility are non-positional, and therefore, their shares, for any given budget,would increase during recessions. Figure 2 shows the Engelcurves for the seven categories identified as nonpositional.Again the dashed line shows the Engel curve under eco-nomic contraction, while the dotted line shows the Engelcurve under economic expansion. Out of the seven non-positional categories, we find two related categories (homegas, fuel, and coal; and electricity) for which expenditureshares do not grow, for any given total budget, during re-cessions. Unfortunately, it is not immediately clear to uswhy in a recession the relative desirability for home gas,fuel, and coal would decrease, especially given that we ac-counted for total consumption budget and various observedand unobserved heterogeneities that could affect prefer-ences. But one might argue that electricity is an essential,but visible expenditure category, because a substantial por-tion of it is for illumination, which is visible to one’s neigh-bors, which would place this category out of the nonposi-tional group. In sum, out of all the 19 predictions we maderegarding positionality and Engel curve shifts, we foundonly two (or possibly one) instances (home gas, fuel, andcoal; and electricity) in which our empirical result did notconfirm our prediction.

COUNTERFACTUAL ANALYSESBesides estimating shifts in consumer taste parameters re-sulting from changing economic conditions, which was themain purpose of the above analyses, our budget allocationmodel also provides a tool for assessing how economic con-traction/expansion may affect household expenditure pat-terns, using the simulation procedure described earlier. Here,the goal is not to test our hypotheses (something alreadyaccomplished through the results presented earlier), but togain additional insights into the impact of economic con-traction/expansion on consumption through counterfactual

240

FIG

UR

E1

EN

GE

LC

UR

VE

S(B

AS

ELI

NE

,E

CO

NO

MIC

CO

NT

RA

CT

ION

AN

DE

XP

AN

SIO

N)

FO

RP

OS

ITIO

NA

LC

AT

EG

OR

IES

241

242 JOURNAL OF CONSUMER RESEARCH

FIGURE 2

ENGEL CURVES (BASELINE, ECONOMIC CONTRACTION AND EXPANSION) FOR NONPOSITIONAL CATEGORIES

simulations. For these analyses, we consider five scenarios,which will provide us with additional insights into howconsumption patterns change during economic expansionsversus recessions. The first scenario establishes the baseline,where we set per capita GDP growth rate at the historicalaverage (3.1%), and use actual household budgets, socio-demographic profiles and factor scores. The second scenarioconsiders a reduction of 10% in the total consumption bud-get for every household in our sample, while holding percapita GDP growth rate at 3.1%, the historical average. At

first glance, a 10% budget reduction seems quite high foran across-the-board shift, but this reduction is realistic forindividuals whose incomes are directly affected by a reces-sion. An across-the-board reduction of 10% in consumptionwhile holding GDP growth constant will not happen in reallife, because consumer expenditures account for about 70%of GDP, and any major reduction in consumption is boundto cause the economy to contract. However, comparing thiscounterfactual to the baseline scenario allows us to isolatethe effects of shrunken budgets on expenditure patterns from

ECONOMIC CONTRACTIONS AND EXPANSIONS 243

FIGURE 2 (CONTINUED)

the effects of changing consumer tastes in a recession.Graphically, this second scenario captures lateral movementsalong the same set of baseline Engel curves.

The third and fourth scenarios assume a 2% drop and 6%increase in per capita GDP (which are all within the rangeof our historical data), respectively, while keeping householdconsumption budgets at their actual levels. These counter-factuals allow us to isolate the effects of changing consumertastes from the effects of changing consumption budgets,under different economic conditions. Graphically, these two

scenarios capture upward or downward shifts in the Engelcurves themselves due to shifts in consumer tastes.

Finally, the fifth scenario considers a 10% cut in the con-sumption budget along with a 2% drop in per capita GDP,to demonstrate the total effects on expenditure patterns dueto a combination of shrunken budgets and changing tastesunder a more realistic recessionary scenario. The resultsfrom these counterfactuals are shown in table 4, as per-centage changes in expenditure relative to the baseline sce-nario.

244 JOURNAL OF CONSUMER RESEARCH

TABLE 4

RESULTS FROM COUNTERFACTUAL ANALYSES: PERCENTAGE CHANGES IN EXPENDITURES,RELATIVE TO THE BASELINE SCENARIO

Scenarios

Expenditures 10% budget cut (%)2% drop in per capita

GDP (%)6% growth in per cap-

ita GDP (%)

10% budget cut and2% drop in per capita

GDP (%)

Food away from home �10.1 �7.5 4.3 �16.7Tobacco �9.5 15.9 �8.6 4.3Alcohol at home �10.0 4.6 �2.8 �5.9Alcohol away from home �10.0 �3.8 2.1 �13.4Apparel �10.1 �11.0 6.5 �19.9Apparel services �10.2 �13.4 8.0 �22.0Jewelry and watches �9.8 �34.9 23.7 �41.3Personal care �10.2 �10.5 6.1 �19.4Personal business �10.1 �3.7 2.0 �13.3Personal insurance �10.1 �6.3 3.4 �15.7Medical prescriptions �9.8 7.6 �4.2 �3.1Medical nonprescription �9.8 �18.5 11.2 �26.6Medical doctors �10.1 �5.0 2.9 �14.5Medical hospital �15.4 13.2 �7.4 2.0Health insurance �9.9 4.9 �2.8 �5.6Motor services and parts �10.1 �4.3 2.4 �13.9Motor fuel �10.0 3.2 �2.0 �7.1Motor insurance �9.9 4.4 �2.6 �6.0Transportation public �9.9 �12.3 7.5 �20.9Transport airfare �10.2 �24.0 15.2 �31.6Recreation �10.1 �5.4 3.1 �14.9Education �8.7 �2.8 1.6 �11.2Charity �9.2 31.6 �15.7 18.4Telephone �10.0 �0.8 0.4 �10.7Lodging away from home �10.3 �12.2 7.2 �20.9Home furnishings and appliances �10.1 �15.7 9.6 �24.1Household operations/maintenance �10.2 �7.7 4.4 �16.9Electricity �10.1 �6.3 3.6 �15.7Water, sewer, and trash �9.8 9.1 �5.1 �1.8Home gas, fuel, coal �10.2 �8.0 4.4 �17.2Housing rental or equivalent �10.0 2.5 �1.5 �7.7Food at home �10.0 1.8 �1.1 �8.4

The first column of table 4 suggests that a reduction inthe consumption budget, either due to a drop in income ora shift from consumption to savings does not affect budgetallocation patterns substantially. In response to a substantialbudget cut of 10%, most categories see a drop of about 10%in expenditures, with education, charity and tobacco show-ing the smallest reductions, and medical hospital showingthe largest reductions (although this particular category hasvery low incidence across the sample, and therefore thesubstantial change in this category might be more of a re-flection of shifts in incidence).

The changes in expenditure patterns in a recession (secondcolumn of table 4) are more interesting. Our counterfactualssuggest that even with the consumption budget unchanged,expenditures in nonessentials, such as jewelry and watches,public transportation, airfare, and home furnishings and ap-pliances, see substantial drops (�35%, �24%, and �16%,respectively, in response to a 2% drop in per capita GDP).One result raises a particular concern regarding consumerwelfare, that is, the 18% reduction in nonprescription drugs,although this predicted change is in line with survey results

reported by A. C. Nielsen in 2009 (http://www.blog.nielsen.com), which show that 32% of American respondents wouldreduce their consumption of nonprescription drugs in a re-cession. However, holding the consumption budget unchan-ged, expenditures on charity increase by 32% in responseto a 2% drop in per capita GDP, while expenditures ontobacco increase by 16%. The 13% increase in medical hos-pital, again, might be due to the very low incidence of thiscategory in our sample. These results suggest that, ceterisparibus, consumers who do not change their consumptionbudget during a recession become more charitable, as man-ifested in larger shares of their budgets allocated to charity,which is consistent with the notion that charity, being lowin visibility/positionality, would gain in relative desirabilityas compared to expenditures that are more visible/positional.Notice, however, that an increase in the budget share forcharity might also happen because the need is clearly greaterin recessions, leading charity organizations to be more ag-gressive in their collection efforts.

Our counterfactuals also suggest that in hard economictimes, smokers spend a higher percentage of their budget

ECONOMIC CONTRACTIONS AND EXPANSIONS 245

on tobacco, confirming survey-based results showing thatsmoking onset and relapses increase during recessions (Bar-clay 2009), and the possible role of tobacco as an “affordablepleasure” in tough economic times (Shafey et al. 2009).

The last column of table 4 provides a more completepicture of what one may expect for a household facing a10% budget cut during a recession with a 2% drop in percapita GDP, combining the effects of a shrunken budget andchanging tastes. In such a scenario, one observes substantialreductions in expenditures on most categories, except forcharity (18% increase), tobacco (4% increase), and medicalhospital (2% increase). The most dramatic drops in expen-ditures under this recessionary scenario are found in jewelryand watches (�41%), airfare (�32%), nonprescription drugs(�27%), and home furnishings and appliances (�24%).

DISCUSSIONIn this study, we examined if and how GDP growth rate, abroad indicator of economic conditions, may be related toa household’s consumption budget allocation, an individualdecision. In light of the Great Recession that the UnitedStates has just gone through, we were particularly interestedin understanding how consumers’ taste or desire for variouscommodities may have shifted in difficult times like this,as compared to their preferences in more normal times.

Traditional economic analyses of consumer expenditurehave assumed, with limited empirical evidence, that con-sumer tastes will remain unchanged, regardless of economicconditions. Under such a standard assumption, the impactof economic conditions on household budget allocation willonly come through changing consumption budgets. Mean-while, the underlying desirability of each category will re-main stable. Faced with smaller budgets in a recession, con-sumers have less to spend on luxuries, and therefore allocatinglarger shares to more essential categories. In other words,according to traditional economic models, changes in ex-penditures during recessions are simply due to budgetaryeffects, because the utility which consumers draw fromgoods and services does not change with the economic en-vironment.

Drawing on the literatures on relative consumption (Due-senberry 1949; Frank and Levine 2006; Schor 1999), po-sitional goods (Frank 1985a, 1985b, 1999; Hirsch 1976),and reference-group influences on consumer behavior (Bear-den and Etzel 1982; Childers and Rao 1992; Escalas andBettman 2003; Gasana 2009; Richins 1994), we postulatedthat for visible and nonessential commodities, consumersderive utility not only from consumption but also from theirpositional value, either because these expenditures signal ahigher status, or because consumers derive indirect utilityfrom simply being able to spend more than their peers.During a recession, as consumers see others spending lessin positional categories (due to a more limited budget), theyfigure that they can also spend less and still maintain theirrelative standing, which in turn leads to reduced budgetshares for these categories across society.

The notion that consumers derive utility not only from

the direct consumption of a good but also from how muchothers are spending on the same good (through the positionalvalue) is an unorthodox departure from the traditional eco-nomic paradigm, because it implies that consumers will drawmore value from consuming the same quantity of a certaingood, if they see others consuming less of it. This notionwas recently tested by Heffetz (2011), who developed astylized game-theoretical model with households making ra-tional budget allocation decisions between two goods, onevisible to the public and the other invisible. When house-holds care about not only the direct utility from consumptionbut also its signaling value, Heffetz’ model shows that, inequilibrium, higher income households would spend largershares on the visible good, and smaller share on the invisiblegood, thereby explaining why the Engel curves for invisiblegoods decrease with income while those for visible goodsincrease with income. While we rely on the same notion of“positional value,” our budget allocation model is able toempirically isolate two distinct effects: (a) the budget effect,which, similarly to Heffetz (2011), explains why the Engelcurve for essential goods (e.g., food at home) has a negativeslope while the slope is positive for nonessential goods (e.g.,food at restaurants); and (b) the positional effect, whichexplains why, during a recession, consumers reduce theirexpenditures on nonessential and visible goods even whenthey do not experience a reduction in consumption budget.

Among the significant shifts in expenditure patterns thatwe are able to empirically identify, the vast majority are ina direction that is consistent with our hypothesis about po-sitionality and expenditures during economic contractions/expansions. More visible nonessential (positional) goods(e.g., food away from home, apparel, apparel services, jew-elry and watches, personal care, home furnishings and ap-pliances, recreation, and airfare) become relatively lessdesirable in a recession, while less visible essential (non-positional) goods (e.g., food at home; housing; prescriptiondrugs; water, sewage, and trash; health insurance) gain inrelative desirability during a recession. Our Engel curvesfor nonessential visible goods/services clearly show that in-dustries tied to those commodities could suffer a doublewhammy during a recession. First, consumers generally re-duce their consumption budget either because their incomeis lower, or because they become more risk averse, allocatingmore of their income toward savings, which forces them tosatisfy essential needs first (the budget effect). Second, con-sumers derive less relative utility from, and thus less desirefor, visible nonessentials (the positional effect), leading tofurther lowered shares of shrunken budgets for these posi-tional goods/services.

In sum, we see our key contribution to the marketingliterature on household expenditures as twofold. First, wepresent strong empirical evidence showing that consumertastes, and thereby their consumption budget allocation pat-terns, do shift systematically as a function of GDP growthrates, even after controlling for the budget effect. Second,our study provides a field test of the theory that relativestanding in a society plays an important role in household

246 JOURNAL OF CONSUMER RESEARCH

spending decisions and that this role is stronger for goodsthat are more visible and nonessential. The modeling frame-work used in our study can also be readily adapted forexamining how other environmental variables may affectindividual consumption preferences.

Taken together, the empirical findings of our study high-light the need in consumer research to further examine thelink between household-level spending decisions and macro-level economic conditions. Our study suggests that such alink exists because spending decisions are not made by eachhousehold in isolation. Instead, each household’s decisionsare made within a broader social context, wherein certainrelative standards, jointly shaped by spending decisionsmade by all the households in the society, play a significantrole. To the extent that macro-level economic conditions canchange those relative standards (or the perception of them),they can change household-level spending decisions.

DIRECTIONS FOR EXTENSIONS

Theoretically, the positional effect we managed to detectand isolate (from the budget effect) with our modelingframework is driven by consumers’ comparison to their re-spective reference groups. Therefore, a potentially fruitfulextension would be to identify these reference groups forindividual consumers, in order to further explore the mech-anisms behind the positional effect. For example, Bourdieu(1984) theorizes that consumer preferences are determinedby social class, suggesting socioeconomic status as a po-tentially useful construct in understanding the positional ef-fect. Within our empirical modeling framework, once eachhousehold in the CEX-NBER database is classified into asocioeconomic stratum, their status can be included as oneof the predictors in our model, allowing one to test foranother form of positional effect, and obtain a better un-derstanding of the impact of social class on consumptionpreferences.

In its present version, our modeling framework assumesa symmetric effect of economic contraction/expansion onconsumption preferences. In future research, one might con-sider the possibility that a household’s tastes may shift asym-metrically in response to perceived spending changes by itsreference group, depending on whether the total consump-tion budget of the reference group is increasing (in economicexpansions) or decreasing (in economic contractions). How-ever, empirical tests for such asymmetric effects would re-quire true longitudinal data, where a panel of householdsare tracked over time, as opposed to repeated cross-sectionssuch as the CEX data used in this study.

REFERENCES

Aitchison, J., and J. A. C. Brown (1954), “A Synthesis of EngelCurve Theory,” Review of Economic Studies, 22 (1), 35–46.

Alpizar, Francisco, Frederik Carlsson, and Olof Johansson-Sten-man (2005), “How Much Do We Care about Absolute versusRelative Income and Consumption?” Journal of EconomicBehavior and Organization, 56 (3), 405–21.

Bagwell, Laurie S., and B. Douglas Bernheim (1996), “VeblenEffects in a Theory of Conspicuous Consumption,” AmericanEconomic Review, 86 (June), 349–73.

Barclay, Eliza (2009) “Outlook Hazy for Smoking Rates in theUSA,” Lancet, 373 (9673), 1415.

Bearden, William O., and Michael J. Etzel (1982), “ReferenceGroup Influence on Product and Brand Purchase Decisions,”Journal of Consumer Research, 9 (September), 183–94.

Bellante, Don, and Ann C. Foster (1984), “Working Wives andExpenditure on Service,” Journal of Consumer Research, 11(2), 700–707.

Bourdieu, Pierre (1984), Distinctions: A Social Critique of theJudgment of Taste, trans. Richard Nice, Cambridge, MA: Har-vard University Press.

Bourne, Francis S. (1957), “Group Influence in Marketing andPublic Relations,” in Some Applications of Behavioral Re-search, ed. R. Likert and S. P. Hayes, Basel: UNESCO.

Carlsson, Fredrik, Olof Johansson-Stenman, and Peter Martinsson(2007), “Do You Enjoy Having More than Others? SurveyEvidence of Positional Goods,” Economica, 74 (296), 586–98.

Childers, Terry L., and Akshay R. Rao (1992), “The Influence ofFamilial and Peer-Based Reference Groups on Consumer De-cisions,” Journal of Consumer Research, 19 (September),198–211.

Christen, Markus, and Ruskin M. Morgan (2005), “Keeping Upwith the Joneses: Analyzing the Effect of Income Inequalityon Consumer Borrowing,” Quantitative Marketing and Eco-nomics, 3 (2), 145–73.

Deaton, Angus (1992), Understanding Consumption, New York:Oxford University Press.

Deaton, Angus, and John Muellbauer (1980), Economics and Con-sumer Behavior, New York: Cambridge University Press.

Du, Rex Y., and Wagner A. Kamakura (2008), “Where Did AllThat Money Go? Understanding How Consumers AllocateTheir Consumption Budget,” Journal of Marketing, 72 (6),109–31.

Duesenberry, James S. (1949), Income, Saving, and the Theory ofConsumer Behavior, Cambridge, MA: Harvard UniversityPress.

Easterlin, Richard A. (1974), “Does Economic Growth Improvethe Human Lot? Some Empirical Evidence,” in Nations andHouseholds in Economic Growth: Essays in Honor of MosesAbramovitz, ed. Paul A. David and Melvin W. Reder, NewYork: Academic Press.

Escalas, Jennifer E., and James R. Bettman (2003) “You Are WhatYou Eat: The Influence of Reference Groups on Consumer’sConnections to Brand,” Journal of Consumer Psychology, 13(3), 339–48.

Ferber, Robert (1956), “Consumer Expenditures for Services in theUnited States,” Journal of Marketing, 21 (1), 24–35.

Fisk, George (1959), “Toward a Theory of Leisure-Spending Be-havior,” Journal of Marketing, 24 (2), 51–57.

Flatters, Paul, and Michael Willmott (2009) “Understanding thePost-recession Consumer,” Harvard Business Review, 87(7–8), 106–12.

Frank, Robert H. (1985a), Choosing the Right Pond: Human Be-havior and the Quest for Status, New York: Oxford UniversityPress.

——— (1985b), “The Demand for Unobservable and Other Non-positional Goods,” American Economic Review, 75 (1),101–16

——— (1999), “Positional Externalities Cause Large and Pre-

ECONOMIC CONTRACTIONS AND EXPANSIONS 247

ventable Welfare Losses,” American Economic Review, 95 (2),137–41.

Frank, Robert H., and Adam Seth Levine (2006), “ExpenditureCascades,” mimeograph, Cornell University.

Fritzsche, David J. (1981), “An Analysis of Energy ConsumptionPatterns by Stage of Family Life Cycle,” Journal of MarketingResearch, 18 (2), 227–32.

Gasana, Parfait U. (2009), “Relative Status and InterdependentEffects in Consumer Behavior,” Journal of Socio-Economics,38 (1), 52–59.

Gourinchas, Pierre-Olivier, and J. A. Parker (2002), “Consumptionover the Lifecycle,” Econometrica, 70 (1), 47–89.

Hall, Robert E. (1993), “Macro Theory and the Recession of1990–1991,” American Economic Review, 83 (2), 275–79.

Heffetz, Ori (2011), “A Test of Conspicuous Consumption: Visi-bility and Income Elasticities,” Review of Economics and Sta-tistics, 93 (4), 1101–17.

Hirsch, Fred (1976), Social Limits to Growth, Cambridge, MA:Harvard University Press.

Jappelli, Tullio, and Luigi Pistaferri (2010), “The ConsumptionResponse to Income Changes,” NBER Working Paper No.w15739, NBER, Cambridge, MA.

Katona, George (1974) “Psychology and Economics,” Journal ofConsumer Research, 1 (1), 1–8.

Leibenstein, H. (1950), “Bandwagon, Snob, and Veblen Effects inthe Theory of Consumer Demand,” Quarterly Journal of Eco-nomics, 64 (May), 183–207.

Levedahl, J. William (1980), “The Impact of Permanent and Tran-sitory Income on Household Automobile Expenditure,” Jour-nal of Consumer Research, 7 (1), 55–66.

Loeb, Benjamin S. (1954), “The Use of Engel’s Laws as a Basisfor Predicting Consumer Expenditures,” Journal of Market-ing, 20 (1), 20–27.

Magrabi, Frances M., Young Sook Chung, S. Sohn Cha, and Se-Jeong Yang (1991), The Economics of Household Consump-tion, New York: Praeger.

Maslow, Abraham H. (1943), “A Theory of Human Motivation,”Psychological Review, 50 (4), 370–96.

Millican, Richard D. (1967), “A Re-examination of Engel’s Lawsusing BLS Data (1960–1961),” Journal of Marketing, 31 (4),18–21.

Ordabayeva, Nailya, and Pierre Chandon (2011), “Getting Aheadof the Joneses: When Equality Increases Conspicuous Con-sumption among Bottom-Tier Consumers,” Journal of Con-sumer Research, 38 (June), 27–41.

Ostheimer, Richard (1958), “Who Buys What? Life’s Study of

Consumer Expenditures,” Journal of Marketing, 22 (3),260–72.

Parker, Jonathan, and Annette Vissing-Jorgensen (2009), “WhoBears Aggregate Fluctuations and How?” NBER WorkingPaper No. W14665, NBER, Cambridge, MA.

Pew Research Center (2010), “A Balance Sheet at 30 Months:How the Great Recession Has Changed Life in America,”Pew Social and Demographic Trends report, Pew ResearchCenter, Washington, DC.

Richins, Marsha (1994), “Valuing Things: The Public and PrivateMeanings of Possessions,” Journal of Consumer Research,21 (December), 504–21.

Rogers, David, and Howard L. Green (1978), “Changes in Con-sumer Food Expenditure Patterns,” Journal of Marketing, 42(2), 14–19, 135.

Rubin, Rose M., Bobye J. Riney, and David J. Molina (1990),“Expenditure Pattern Differentials between One-Earner andDual-Earner Households: 1972–1973 and 1984,” Journal ofConsumer Research, 17 (1), 43–52.

Rucker, Derek D., and Adam D. Galinsky (2008), “Desire to Ac-quire: Powerlessness and Compensatory Consumption,” 35(August), 257–67.

Schor, Juliet (1999), The Overspent American: Why We Want WhatWe Don’t Need, New York: Harper Perennial.

Shafey, Omar, Michael Eriksen, Hana Ross, and Judith Mackay(2009), The Tobacco Atlas, 3rd ed., Atlanta: American CancerSociety.

Soberon-Ferrer, Horacio, and Rachel Dardis (1991), “Determinantsof Household Expenditures for Services,” Journal of Con-sumer Research, 17 (4), 385–97.

Solnick, Sara J., and David Hemenway (1998), “Is More AlwaysBetter? A Survey on Positional Concerns,” Journal of Eco-nomic Behavior and Organization, 37 (3), 373–83.

——— (2005), “Are Positional Concerns Stronger in Some Do-mains than in Others?” American Economic Review, 95 (2),147–51.

Strober, Myra H., and Charles B. Weinberg (1977), “WorkingWives and Major Family Expenditures,” Journal of ConsumerResearch, 4 (3), 141–47.

Veblen, Thorstein (1899/1965), The Theory of the Leisure Class,New York: Macmillan.

Wagner, Janet, and Sherman Hanna (1983), “The Effectiveness ofFamily Life Cycle Variables in Consumer Expenditure Re-search,” Journal of Consumer Research, 10 (3), 281–91.

Wilkes, Robert E. (1995), “Household Life-Cycle States, Transi-tions, and Product Expenditures,” Journal of Consumer Re-search, 22 (1), 27–42.