Beyond Engel's law - A cross-country analysis

B

WM

a

ARR3A

KEENLD

1

tiEsaac2at(Lri

da

BWtUana

1h

The Journal of Socio-Economics 47 (2013) 118– 134

Contents lists available at ScienceDirect

The Journal of Socio-Economics

j our na l ho me pa g e: www.elsev ier .com/ locate /soceco

eyond Engel’s law - A cross-country analysis�

olfhard Kaus ∗

ax Planck Institute of Economics, Evolutionary Economics Group, Kahlaische Straße 10, 07745 Jena, Germany

r t i c l e i n f o

rticle history:eceived 2 October 2012eceived in revised form0 September 2013ccepted 2 October 2013

a b s t r a c t

Engel’s law is known to be extraordinarily consistent across time and space. To substantiate the distinctionbetween necessities and luxuries, already Ernst Engel (1895) approached a behaviorally founded com-prehensive assessment of structural changes in consumer expenditures. To build upon Engel’s legacy andto complement the scarce empirical literature, a behavioral approach is applied. It is conjectured that dif-ferences in satiation patterns of universally-shared needs translate, on the aggregate level, into different

eywords:ngel’s lawngel curvesecessity

shapes of Engel curves and thus also into different income elasticities of demand. Utilizing nonparametricregression techniques, it is explored whether and which expenditure categories change systematicallywith rising income. In line with the theoretical expectations, a number of empirical regularities in con-sumer expenditure patterns can be identified that go well beyond Engel’s law.

© 2013 Elsevier Inc. All rights reserved.
uxuryifferential satiation
. Introduction

Since the work of Ernst Engel it is a well established fact thathe expenditure share dedicated to food consumption decreases asncome rises. Using a sample of only 199 Belgian family budgets,rnst Engel (1857, 1895) inferred the observed negative relation-hip between the expenditure share on food and income to ben economic law. Despite the meager database available to Engelt this time, this relationship turned out to hold true in mostountries and points in time (Houthakker, 1957; Seale and Regmi,006; Lewbel, 2008). The extraordinary consistency across timend space and the predictive character of the relationship led tohe use of Engel’s law to determine absolute household povertysee, e.g., Musgrove, 1985) and to define poverty lines (see, e.g.,anjouw and Lanjouw, 2001, and other references therein). Moreecently, Engel’s law is even used to correct purchasing power par-ties (Almås, 2012).

In contrast, evidence on systematic changes in other expen-iture categories is hardly available. Both theoretical conjecturesnd empirical evidence on other systematic changes in the

� For helpful discussions and valuable comments I would like to thank Elisabethublitz, Stephan Bruns, Leonhard Lades, Deirdre McCloskey, Alessio Moneta, Ulrichitt and the participants of a Jena Economic Research Workshop as well as the par-

icipants of the 14th Conference of the International Joseph A. Schumpeter Society,niversity of Queensland. Furthermore, I would like to thank the Journal’s editornd three anonymous reviewers for their comments and suggestions, which sig-ificantly contributed to improving the quality of the paper. The usual disclaimerpplies.∗ Tel.: +49 3641 686 835.

E-mail address: [email protected]

053-5357/$ – see front matter © 2013 Elsevier Inc. All rights reserved.ttp://dx.doi.org/10.1016/j.socec.2013.10.001

composition of consumer expenditures remain scarce. The lack ofsuch an assessment is even more remarkable as Ernst Engel alreadypursued a much more encompassing approach. In fact, his approachto consumption expenditures was never intended to focus solelyon the budget share for food. Engel’s original work was indeedan inquiry into the structural changes of consumption patternswith rising income, aiming at determining a measure of householdwelfare (Chai and Moneta, 2010). Therefore, he started out by cat-egorizing expenditure items according to the underlying “wants”they satisfy. In a second step, he assessed the importance of thesewants from the empirically derived expenditure patterns. In thissense, Engel was not only the first to empirically assess the dis-tinction between necessities and luxuries (instead of subjectivelyassuming) but also the first to attempt a behavioral explanation forthis taxonomy.

This paper aims to build upon Engel’s legacy. Therefore, itenriches Engel’s classification of needs with current scientificknowledge on the nature of consumer needs and how they aresatisfied. The paper offers a motivational approach to consumerbehavior, which facilitates formulating expectations about struc-tural changes of expenditure shares as income rises. The theorydistinguishes between innate needs on the one hand and bothwants and cognitively constructed motives that are acquired inculturally contingent ways on the other hand (Witt, 2001). Tothe extent to which consumption is driven by innate needs, itcan be expected that the common genetic basis induces patternsof behavior that are universally-shared by humans. This means
that consumer expenditure patterns should change interperson-ally and interculturally similarly with income. Engel’s law forexpenditures on food – for which a fairly rapidly satiable innateneed can be identified – is a case in point. For other expenditure
dx.doi.org/10.1016/j.socec.2013.10.001

http://www.sciencedirect.com/science/journal/10535357

http://www.elsevier.com/locate/soceco

http://crossmark.crossref.org/dialog/?doi=10.1016/j.socec.2013.10.001&domain=pdf

mailto:[email protected]

dx.doi.org/10.1016/j.socec.2013.10.001

io-Eco

cadin

siUigTic

Eracf

rdE

2

cwieelac

2

tIqtet(o

iiitecbe

2

ed2it

a reduction of their deprivation states, the likelihood that a par-ticular activity is chosen over another one depends on its relativecontribution to reducing deprivation. Individuals are accordingly

1 In behavioral and human sciences, research on motivational theories continuedand extended to focus on biological and evolutionary roots of behavior (Wilson,

W. Kaus / The Journal of Soc

ategories, the theory identifies other innate needs as driving forcesnd, corresponding to their differing satiation features, predictsifferent shapes for the respective Engel curves. The correspond-

ng consumption categories can be qualified accordingly as eitherecessities or luxuries.

In order to assess the evidence for the suggested human univer-als, the paper presents an empirical analysis of observable changesn consumption patterns across countries and time. Using thenited Nations National Account Statistics, a data set that contains

ndividual consumption expenditure by households in 12 cate-ories for more than 50 countries over up to 50 years is specified.he data at hand allows for explicitly describing the overall changesn consumption patterns with rising income (cross-country Engelurves).

Our findings broadly confirm the conjecture that the shape ofngel curves (inclusive of, but extending beyond, Engel’s law) rep-esents human universals. The study contributes to the literature by

structural assessment of general changes in budget shares acrossountries that helps to broaden our understanding of the behavioraloundations of the necessity – luxury taxonomy.

The paper is structured as follows. Section 2 outlines the theo-etical framework. In Section 3 the creation of a comparable incomeimension is described. The empirical results on cross-countryngel curves are shown in Section 4. The last section concludes.

. What determines the shape of Engel curves?

This section applies a behavioral approach to consumption toonnect differences in satiation patterns between innate needsith systematic changes in consumption expenditures when

ncome rises. It is conjectured that, at the level of aggregatexpenditure data, these differences translate into different incomelasticities of demand for groups of goods and services that areikely to be consumed to serve those needs. This way, it should beble to explain a good deal of the differences in the shapes of Engelurves of the underlying goods and services.

.1. Engel curves and the income elasticity of demand

The general connection between the shape of an Engel curve andhe income elasticity of the respective good can easily be illustrated.f an Engel curve for good i is expressed in terms of the expendituresi spent on i depending on the households’ income y, the slope ofhe fitted curve dqi/dy can indirectly be used to derive i′s incomelasticity of demand. As a good’s income elasticity is defined byhe relative change in qi (dqi/qi) divided by the relative change in ydy/y), it can be estimated by regressing logarithm of qi on logarithmf y (Lewbel, 2008).

Using expenditure shares (wi = qi/y), instead of the expend-tures spent on i, facilitates more directly an inference of thencome elasticity of demand from the curve’s slope. While anncreasing slope represents more than proportional expenditures,hus luxuries, a decreasing slope points to less than proportionalxpenditures with rising income, i.e., necessities. Although Engelurves usually show considerable nonlinearities (Lewbel, 2008),udget share Engel curves facilitate to readily observe the incomelasticity of demand for the underlying group of goods and services.

.2. A motivational approach to consumer behavior

Engel’s approach to consumption never solely focused on foodxpenditures. In fact, Engel’s original contribution was meant to
etermine and measure household welfare (Chai and Moneta,010). He started out by categorizing expenditure items accord-
ng to the underlying “wants” they satisfy. The importance ofhese wants was subsequently assessed by the empirically derived

nomics 47 (2013) 118– 134 119

expenditure patterns. Engel’s work thus essentially focused on abehavioral foundation of the necessity – luxury taxonomy. In arevised version of his original contribution, (Engel, 1895, p. 8)made even more explicit that the motivation of human action,and thus also consumption behavior, is rooted in the satisfactionof universally-shared needs (Chai and Moneta, 2010).

To build upon Engel’s legacy, this paper enriches the classifica-tion of needs with current scientific knowledge on the nature ofconsumer needs and how they are satisfied. However, motivationalunderpinnings of economic behavior in general and consumerbehavior in particular are rarely addressed in economics. Amongthe existing works, two different explanatory approaches can beidentified (Witt, 2010a). While in the utilitarian hedonic approach(Kahneman et al., 1997) the explanation refers to the motives ofseeking pleasure and avoiding pain, nonhedonistic variants insteadfocus on the motivating power that deprived needs and wants havefor consumption activities (see, e.g., Menger, 1871; Duesenberry,1949; Georgescu-Roegen, 1954; Ironmonger, 1972).1

This paper closely connects to the latter approach. A behavioral– need based – interpretation of the consumption motivation isoffered (Witt, 2001, 2010b) to actually theoretically explain thenecessity – luxury distinction that becomes visible through theshape of Engel curves. The theory postulates an intimate relation-ship between human biological and cultural evolution in the sensethat cultural development is based upon as well as constrainedby innate behavioral dispositions and cognitive learning abilities,which have emerged during human phylogeny. Hence, the theoryfocuses on the explanation of long-run economic change from abiological and psychological perspective (Witt, 2008).

The theory of the learning consumer (Witt, 2001, 2010b) empha-sizes the role of human needs as ultimate motives of consumptionbehavior. The theory distinguishes between genetically deter-mined – innate – needs and both culture and socialization specificacquired wants that result from processes of associative learning(classical and operant conditioning (Pavlov, 1927; Skinner, 1966))and social-cognitive learning (Bandura, 1986).2 The theory holdsthat the attempt to relieve or reduce deprivation of a limited num-ber of innate needs, consequently an increase in satiation level ofthese needs, is one major motivation to consume. Deprivation isthus seen to intrinsically motivate consumers to act which createsa rewarding experience.

Needs are the contingencies under which deprivation occurs.Although needs can be manifold, in this context only the subset ofuniversally-shared “basic” needs is relevant. Among these are thatfor sleep, for something to drink, for something to eat, for maintain-ing body temperature, for physical activity, for status recognition,or for sensory arousal (Witt, 2011).

An important characteristic of these needs is that their satis-faction by an action effects a primary reinforcement in the senseof instrumental or operant conditioning (Hull, 1943; Herrnstein,1990; Staddon and Cerutti, 2003). This, of course, has implica-tions for the allocation of resources. Imagine individuals to beequipped with a fixed set of innate needs that show different needdeprivation states. If individuals allocate their behavior to obtain

1975; Caplan, 1978), which draws attention to interpersonal commonalities thatcan be conjectured to be relevant also for consumer behavior.

2 As noted by Chai and Moneta (2012), a weakness of the needs-based approachto consumption is a missing consensus on the exact number of universally-sharedneeds.

1 io-Eco

attddcast

r2heaetdaitttwa

ecspiibaantiemE

giiaeabai

2

spctiptd

ac

income rises, two conceptual problems have to be mentioned first.The one is the problem of determining what expenditures servewhat need so that the predicted satiation features of the needscan be translated into predictions about changing consumption

3 For a discussion, please see Section 3.4 The data are freely available from the National Accounts Official Country Data

account of the United Nations.5 The change of standards with respect to Individual consumption expenditures

entails a methodological change from a domestic to national/residential conception.Unfortunately, the data provided by the United Nations does not allow the author tofacilitate a proper conversion from one into the other. Data which are only available

20 W. Kaus / The Journal of Soc

ssumed to shift their behavior to those alternatives which providehe highest average reward. The allocation of income to consump-ion categories results as proportional to the corresponding needeprivation states. Under reinforcement learning, the frequencyistribution over actions converges to a state satisfying the soalled “matching law” (Herrnstein, 1997), which seems to be a goodpproximation of behavior driven by mainly basic need deprivationtates and that is characterized by low levels of cognitive interven-ion.

Among the universally-shared needs, a further distinctionelates to the underlying satiation properties they show (Witt,010b, 2011). On the one hand, there are basic needs which followomeostatic features, i.e., deprivation can be reduced relativelyasily up to the temporary satiation point once rising incomellows for a sufficient increase in the corresponding consumptionxpenditures per period of time. The motivation to consume is,hen, temporarily reduced or removed. Satisfaction of these needsepends mainly on the intrinsic value of the corresponding goodsnd services. Examples are the homeostatic needs underlying eat-ng, drinking, sleeping, and maintenance of body temperature. Onhe other hand, there are also basic needs where homeostatic fea-ures are absent and where it is therefore difficult, if not impossible,o reduce the average deprivation to zero. Typically, these are needshose satiation level is defined in relative terms, like the need for

rousal and for social recognition.Despite interpersonal sources of variance, which can be

xpected due to individual cognitive and conditioning learning pro-esses, it can be conjectured that shared innate needs exert someystematic effects on behavior that are visible at the level of theopulation means, i.e., at the level of aggregate consumer expend-

tures. As needs differ with respect to the amount of spending thats necessary to reach satiation, this difference can be expected toecome relevant with rising real income (Witt, 2010b, 2011). Beingble to spend more, consumers should be able to approach the sati-tion level of some needs faster than the satiation level of othereeds. Their consumption motivation is not equally upheld andheir spending should thus not expand equally. Differences in thencome elasticity of demand for the products that serve the differ-nt needs should express this differential satiation effect. A formalodel of the differential satiation effects and its implications for

ngel curves has recently been put forward by Lades (2013).Taken together, the behavioral approach to consumption sug-

ests connecting the differences in satiation patterns betweennnate needs with systematic changes in consumption expend-tures when income rises. It is conjectured that, at the level ofggregate expenditure data, these differences translate into differ-nt income elasticities of demand for products or groups of goodsnd services that are likely to be consumed to serve those needs. Theehavioral approach to consumption should thus be able to explain

good deal of the differences in the shapes of Engel curves and thusn the income elasticities of the underlying goods and services.

.3. Applying the theoretical framework

Before hypotheses on the shape of Engel curves can be derived,ome attention should be given to the empirical framework. As thisaper, on the one hand, aims at inquiring into the generality of Engelurves across countries and, on the other hand, attempts to explainhe difference in shapes by universally-shared human dispositions,t is necessary to depart from the usual Engel curve framework thatlots cross sectional expenditure survey data in one region at a par-icular point in time on corresponding income or total expenditure
ata.
In order to derive a more general picture, it is rather desir-ble to use an empirical framework that facilitates exploringhanges in consumption patterns within and across countries with

nomics 47 (2013) 118– 134

different levels of income over time. Applying such a framework,however, necessarily entails some flaws. First, it is obvious that anincrease in the number of countries and observations across timewill challenge the comparability of the expenditure items. Findingcomparable expenditure categories is a natural task that will irrev-ocably lead to a higher level of aggregation. Dealing with variouscountries over a range of time moreover involves converting therelevant variables into a common denominator.3

2.3.1. The expenditure categoriesThe expenditure categories used in this paper are drawn from

the United Nations National Accounts Statistics. Since 1956, theUnited Nations have annually published the National AccountsStatistics: Main Aggregates and Detailed Tables.4 The series super-seded 10 issues of the Statistics of National Income and Expenditure.For the purpose of this analysis, Table III.2, i.e., Individual consump-tion expenditures, is used.

National account statistics represent an account of a country’sannual economic statistics covering GDP related financial flowsamong corporations, households, nonprofit institutions and thegovernment. It can be used to map flows within a country fromthe origin of GDP (production of goods and services by corpora-tions) to its allocation (spending by households and government,savings, and investment). Data on all subcategories are provided bythe respective national statistical agencies. Individual consumptionexpenditures are compiled annually using representative house-hold surveys, in which detailed information on income, assets, andexpenditures is recorded over a certain period.

Data availability is restricted to member countries of the UnitedNations and some other countries or areas which report to theUnited Nations. However, only after joining the United Nations, allmembers are required to report such statistics based on commonstandards. The sample used in this paper can be found in Table A.2in the appendix. It is restricted to data that complies with the Sys-tem of National Accounts 1993 standards (SNA 93).5 To the best ofthe author’s knowledge, this is the most valuable source that allowsfor an exploration of structural changes of consumption patterns asincome rises in a long term cross-country framework.6

As mentioned before, the long term cross-country framework isnecessarily accompanied by aggregation of the expenditure cate-gories. In fact, this is the case with the national account statisticdata. A big advantage, however, is that the 12 categories (cf.Table A.1) are consistently defined for all observations. The data arecategorized according to the UN Statistics Division’s Classificationof Individual Consumption According to Purpose (COICOP).

2.3.2. Connecting innate needs and expenditure categoriesWhen connecting differences in satiation patterns of innate

needs with systematic changes in consumption expenditures as

according to SNA 68 standards are therefore omitted from the analysis to maintaina comparable frame of analysis.

6 Especially among the currently less affluent countries, there is hardly any, letalone long term, data available. A valuable exception is India, where consumerexpenditure surveys have been collected for more than 60 years by now.

io-Eco

ethdm

fastap

umtaagaotb

ifidshwpCraf(abEfbMeiftntpnbwsssi

gawmssno

status consumption and insatiability of social recognition as a primeexample to explain the shape of the Engel curve for status relatedgoods.


xpenditure. The second is the aggregation problem. The reactionso increasing satiation explained at the individual consumer levelave to be aggregated to the level of aggregate consumption expen-iture data, at which structural changes in both consumption andarket shares are usually recorded (Witt, 2010a).The second problem is of a more general nature and lies in micro-

ounded theories of consumption. The present approach provides natural solution with the assumption of innate, need-based con-umption motives and the corresponding satiation patterns. Sincehese individual features are shared with the usual genetic vari-nce in human populations, an aggregate level – of the humanopulation – is already implied.

The first problem of associating goods and services with partic-lar needs is more critical. In the theoretically ideal case, one needotivates expenditures on one good. This is, however, unlikely

o be the case. As products usually have several characteristicsnd the different characteristics can appeal to different needst the same time. Accordingly, the consumption of a particularood can be motivated in multiple ways. The differential satiationpproach generates clear propositions about the deprivation naturef needs related to a specific good. Thus the need-based approacho preferences can only explain a limited subset of consumerehavior.

The use of 12 aggregate expenditure categories furthermorencreases the likelihood of including goods with different satiationeatures. Nevertheless, it is not impossible, to merge the differ-ng satiation patterns into a compound prediction for the satiationynamics that occur with rising income. To be as accurate as pos-ible, it is necessary to either make sure that a category is asomogeneous as possible in their deprivation nature or to find aay to measure the relative contribution of the different need com-onents that motivate the demand in one expenditure category.oncerning the later, a very convenient and helpful approach wasecently suggested by Barigozzi and Moneta (2012). The authorsre using a factor model approach to identify common functionalactors spanning the space of Engel curves. Barigozzi and Moneta2012) utilize U.K. Family Expenditure Survey data between 1968nd 2006 to find three fundamental sources driving consumerehavior identified as decreasing, increasing, and almost constantngel curve components. This approach is exceptionally help-ul when thinking about the demand for goods being motivatedy a composition of a number of different needs. Barigozzi andoneta (2012) suggest understanding the Engel curve for a certain

xpenditure category as a linear combination of three underly-ng fundamental sources. The approach in this paper connects theundamental factors identified by Barigozzi and Moneta (2012) tohe differing need deprivation structure across universally-sharedeeds. Although the number of universally-shared needs alludedo before are not restricted to three, the approaches are very com-atible, as the differentiation between easily and hardly satiableeed can directly be connected to the first two factors identifiedy Barigozzi and Moneta (2012). Utilizing this perspective, it mightell be that expenditures on, say, food are partly related to a hardly

atiable need as social recognition (e.g., in the case of caviar con-umption). On average it is, however, very likely that the easilyatiable component turns out to be the dominant factor character-zing the Engel curve for food.

In this sense, we have to distinct between expenditure cate-ories that allow formulating strong hypotheses on the one handnd expenditure categories that are too diverse such that onlyeak, say rather explorative, hypotheses can be derived. As alreadyentioned above, a certain set of needs could be identified of which

ome belong to rather physiological needs that follow homeostaticatiation features and are relatively easy to satiate, while othereeds are rather psychological needs that show dynamic nonhome-static satiation features which are less easily, if at all, satiable.

nomics 47 (2013) 118– 134 121

In the case of physiological needs, the homeostatic featuresimply that the motivation for taking actions decreases the closer tothe satiation level of the need the consumer gets. The motivationto expand consumption further is then reduced or vanishes suchthat consumption stagnates, if nothing else happens. As consump-tion expenditures grow with rising disposable income, satiabilityof needs results in a saturated demand for corresponding goods.

A very plausible example for an easily satiable need is foodconsumption. Satiation in this case depends on the satiating charac-teristics of the items consumed such as caloric or caffeine content.Demand is thus naturally constrained by an upper quantity boundthat is the absorbing capacity of the human body. On the aggregatelevel, the strength and the universality of this effect should be dom-inant and therefore allow for a decrease of the proportion spent onfood with rising income (Witt, 2011).

In the case of psychological needs that lack homeostatic fea-tures, the motivation to expand consumption, once increasingincome allows for it, is not easily withheld and, thus, will not leadto a stagnating demand. Therefore, the budget share on expen-diture categories that can be related to satisfying such needs areexpected to increase once the satisfaction of easily satiable needsand increases of income allow for this.

An illustrative example is the need for sensory and cognitivearousal (Witt, 2011). As already argued by Scitovsky (1981), sat-isfaction can hardly be achieved, as the arousal drawn from whatis currently consumed is subject to a stupefaction effect. Satiabil-ity is thus endogenously changing, such that ever stronger stimuliare needed to uphold arousal. Although related expenditures grow,the average deprivation level is hardly decreasing. The insatiabilityof the need for sensory and cognitive arousal should empirically bereflected in the various expenditures related to entertainment (e.g.,electronic equipment, ballet, art, theater, sporting events, and thelike) and media services. Another vivid expenditure category moti-vated by the need for arousal is recreational traveling (Chai, 2011).As these items broadly match the expenditure categories recreationand culture and restaurants and hotels, it is conjectured that upwardsloping Engel curves can be expected.7

Explorative hypothesesDue to the very heterogeneous composition of some of

the remaining expenditure categories it is hardly possible tostraightforwardly connect innate universally-shared needs to theseexpenditure categories. Nevertheless, there are additional sharedneeds that one would expect to exert remarkable effects on expend-itures on a less aggregated expenditure level.

In the case of status recognition, insatiability results from therelative nature of this kind of expenditures. Certain consumptionitems that are suitable to signal status allow for the distinction ofoneself from others. However, the strength of this signal dimin-ishes if rising income enables individuals from lower income groupsto purchase similar items. To restore the status quo ante, moreintense status signals are required. However, additional expend-itures are repeatedly neutralized by expenditures of others suchthat deprivation of status can, if at all, only be avoided by contin-uously rising expenditures (Witt, 2011). The instability of statusraces that leads to rising expenditures on status goods has repeat-edly been illustrated by Hirsch (1978), Frank (2011), and manyothers. Additionally, Lades (2013) uses the connection between the

7 Similarly, the connection between expenditures on recreation and culture and thehardly satiable need for sensory and cognitive arousal has been empirically exploitedby Volland (2012) to explain differences in the persistence of habits relative toexpenditures on food.

1 io-Eco

aoaaaieFsfsmnlaseteu

tabfiaaimes

stsusrriapgli

iimcbiilb

tcipiep


In a recent study, Charles et al. (2009) determine which goodsre signaling better economic circumstances, “. . .are readilybservable in anonymous social interactions, and [. . .] portablecross those interactions” (Charles et al., 2009, p. 426). Accordingly,

basket of goods, including spending on apparel, accessories, suchs watches and jewelry, personal care, and vehicles have beendentified. When abstracting from the requirement of portability,xpenditures on housing additionally qualify (Charles et al., 2009;rank, 1999). Among the consumption categories in the COICOPystem, these items can be found in transport, housing and utilities,urnishings and household equipment, and miscellaneous goods andervices (see Table A.1). All of these categories, however, containuch more items that may or may not be related to social recog-

ition. When thinking about Engel curves as being composed of ainear combination of three fundamental components (Barigozzind Moneta, 2012), it is well possible to expect the differentialatiation effect to become visible in the overall Engel curve of thesexpenditure categories. It is up to a explorative empirical exerciseo determine which of the three factors dominates the Engel curvesventually. It can, therefore, only tentatively be concluded thatpward sloping Engel curves can be expected in these categories.

Another example is the expenditure category of clothing. One ofhe underlying innate needs that is satisfied by the use of clothingnd footwear is the maintenance of body temperature. As every-ody can only wear a limited number of items at a time, the purelyunctional demand for clothing should be easily saturated once ris-ng disposable income allows for this. However, this category is also

prime example for good composed of different characteristics thatppeal to different needs at the same time. Consumption of cloth-ng and footwear is, of course, not only motivated by the desire to

aintain body temperature. The study by Charles et al. (2009) forxample identifies apparel as one of the main goods used to signaltatus.

In a time-series analysis of the U.S., Frenzel Baudisch (2006)hows that the demand for footwear mainly followed basic func-ional uses up to the beginning of the 1970s. In fact, the expenditurehare on footwear was steadily decreasing and thus a “necessity”p to this time. Frenzel Baudisch (2006) is able to show that theubsequent increase in the expenditure share for footwear, albeitelatively small at the aggregate level, was driven by the producer’seaction to the saturated demand for footwear. Producers increas-ngly released functionally differentiated product innovations anddded new characteristics (cognitively mediated motives) to theirroducts to escape functional satiation. These supply-side strate-ies also gave rise to social comparison and group dynamics thated to a decoupling of the item’s consumption and the initial purelynstrumental use (maintenance of body temperature).

Despite the fact that multiple motivations to consume itemsn the clothing category might prevail, the example is suitable tollustrate that, the satisfaction of the universally-shared needs for

aintaining body temperature and the avoidance of bodily harman still be expected to explain the main structural pattern in theudget share Engel curve before purely functional satiation occurs,

.e., supply-side driven reactions to escape saturated markets setn. It is therefore tentatively suggested that, as long as income isow to moderate, the budget share Engel curve for clothing shoulde downward sloping.

Institutionally constrained expenditure categoriesWhen inspecting the expenditure categories in Table A.1, a fur-

her limitation needs to be taken into account. The allocation ofonsumption expenditures is not unlikely to be constrained bynstitutional settings for a number of reasons, such as regulated
rices in the form of lower or upper price boundaries, taxes, tar-
ffs, or the like. Among the aggregate COICOP categories, health andducation expenditures appear to be the most outstanding exam-les. Cross-country expenditure patters in these categories are thus

nomics 47 (2013) 118– 134

likely to show much less consistency. Accordingly, hypotheses onstructural changes in expenditures as income rises are missing.

In the case of communication, network externalities are likely toprohibit a consistent pattern in cross-country Engel curves. Indi-vidual consumption expenditures on communication very muchdepends on the provision of information and telecommunicationnetworks, which likely differs between countries.

Lastly, in the case of alcohol and tobacco, cognitively constructedmotives to consume are likely to dominate. Such “acquired wants”(Witt, 2001) are socially constructed and culturally contingent.Accordingly, cross-country Engel curves are likely to show someheterogeneity.

3. Constructing a comparable income dimension

The empirical setting of the paper requires transforming thevariables of interest such that comparisons across countries andtime are permitted. In the case of the expenditure categories, thiscan easily be facilitated by transforming the expenditures, given incurrent values and local currency, into expenditure shares. Expen-diture shares are created by summing the nominal expenditurevalues and dividing the expenditures of each of the categories bythe composite. Without any further adjustments, the shares arealready comparable.

To make the income variable comparable across countries andtime, several treatments have to be undertaken. First of all, GDPper capita is calculated using population data. Second, using annualexchange rates, the national currency values are transformed intocurrent U.S. $. Third, current nominal income per capita in U.S. $ istransformed into real income per capita. Therefore, price changeswithin countries are accounted for by utilizing implicit price defla-tor data for each country. Eventually, the resulting income variable,constant U.S. $ at 2005 prices, is comparable across time andcountries.

Although a common currency is created and national inflationis controlled for, the income variable still has a natural limitation.Comparing expenditure shares across a standardized income vari-able implicitly assumes that one 2005 U.S. $ buys approximatelythe same in country x and y. That would imply that the law ofone price holds across space for all kinds of expenditure items.Price levels, however, are consistently found to be higher in moreaffluent countries. The systematic difference is usually hypothe-sized to stem from productivity differences in the tradable sectorthat induce prices in the nontradable sector, for which similar pro-ductivity across countries is assumed, to be biased (Harrod, 1933;Balassa, 1964; Samuelson, 1964). A conversion of income measuresat market prices (exchange rate) does therefore not accurately rep-resent the true income differentials.

As the law of one price is unlikely to apply to internationally non-tradable goods, these goods can be purchased at markedly lowerprices in less affluent countries. Such systematic differences thusunderestimate the purchasing power in less affluent countries.Accounting for the regional prices of nontradable goods within astandardized consumption basket consequently adjusts for “real”purchasing power. Purchasing power parity conversion rates thushelp to alleviate this kind of bias.

The conversion of nominal GDP at current prices in nationalcurrency YNC,t into a common currency at constant prices that iscomparable over time and space YPPP,2005, i.e., Geary–Khamis inter-national dollar (I$), can be conducted as follows:

YNC,2005 YNC,t/IPD2005
YPPP,2005(I$) =
PPPt(NC/I$)=

PPPt(NC/I$),

where PPPt(NC/I $) denotes the implicit purchasing power par-ity conversion rate that enables the comparison of a similar

io-Eco

cia

suEtcs

fopicAtciIePmGl

iiNiGr

4

tce1bStgeteu

4

titacs

iips

cases.11

Overall, the charts depicted in Fig. 1 suggest that there aremuch more regularities in cross-country consumption patterns

8 Using national account statistics it is, however, only possible to show changesin the consumption patterns of a mean consumer in each country as income rises.

9 The Nadaraya–Watson kernel estimator also solves a locally weighted leastsquares problem but uses local mean values instead of predicted values for imput-ing the smoothed values. Moreover, the Nadaraya–Watson kernel estimator uses a


onsumption basket at regional prices and IPD2005 denotes themplicit price deflator referring to a fixed base year (2005) toccount for national inflation rates.

Recent research on purchasing power adjustment howeverhows that purchasing power parities are biased as well. Drawingpon micro data from a number of countries, Almås (2012) utilizesngel curves for food to assess the magnitude of this bias whichurns out to be systematic. The results indicate the income of poorerountries to be overestimated. Consequently, PPP adjusted incomeeems to be overadjusted.

The degree of purchasing power adjustment can be relevantor the graphical assessment of Engel curves in Section 4.1. Whilebservations of low income countries tend to move right afterurchasing power adjustments are applied, observations of high

ncome countries move to the left. The shape of the Engel curvean thus be expected to change as well. As the results presented bylmås (2012) indicate that PPP adjusted income is overadjusted,

he “true” income adjustment should lie between real income atonstant prices (constant U.S. $ at 2005 prices) and PPP adjustedncome at constant prices (Geary–Khamis international dollar, i.e.,$). Therefore, the charts depicted in the following sections plotxpenditure share observations on real GDP (maroon circles) andPP adjusted real GDP (gray triangles). The resulting nonpara-etrically derived curves in Fig. 1 is color coded as well: realDP (red dashed line) and PPP adjusted real GDP (black solid

ine).To construct a comparable income per capita indicator, annual

nformation on gross domestic product, population, deflatorndices, and exchange rates are drawn from the United Nationsational Accounts and Main Aggregates database. Data on purchas-

ng power adjusted GDP per capita at constant 2005 prices, i.e.,eary–Khamis international dollar (I$), are drawn from the most

ecent version of the Penn World Tables (Heston et al., 2011).

. Beyond Engel’s law

A natural limitation of the approach chosen in this paper is therade off between breath of the study in terms of the number ofountries and years covered and the level of aggregation of thexpenditure categories. Therefore, it is not surprising that not all the2 internationally standardized consumption categories can easilye matched with their underlying universally-shared needs (seeection 2.3). However, the hypotheses put forward already implyhat from a behavioral consumption approach perspective there areood reasons to presuppose the existence of more than one robustmpirical regularity (“Engel’s law”) in the analysis of demand pat-erns. To validate the hypotheses and to explore the remainingxpenditure patterns, a graphical & nonparametric assessment issed.

.1. Assessing the shape of cross-country Engel curves

This section aims at contributing an empirical account of sys-ematic changes in consumption patterns across countries asncome rises. While it is a well established stylized fact thathe expenditure share dedicated to food consumption decreasess income rises (Engel’s law), evidence on other systemati-al changes in the composition of expenditure shares remainscarce.

The chosen empirical framework facilitates looking at changesn expenditure patterns from a different perspective. Deviat-ng from the standard approach of considering a particularopulation at a certain point in time, it allows for more conclu-ive insights about whether observable changes in expenditure

nomics 47 (2013) 118– 134 123

shares can be generalized across a range of countries.8 Utilizingnonparametric regression techniques, it is explored whetherexpenditure categories, inclusive of, but extending beyond, food,change systematically with rising income. Applying this methodmoreover facilitates verifying the hypotheses outlined in Sec-tion 2.3.

To properly assess the shape of cross-country Engel curves,a nonparametric approach is used. In contrast to parametricapproaches that require an a priori assumption about the functionalform, nonparametric approaches allow for the data themselves toreveal the actual shape of the estimate (Engel and Kneip, 1996). Var-ious nonparametric regression techniques have been developed inrecent years.

An early but still very appealing nonparametric regressionapproach is local linear regression. The idea of local linear regres-sion is to fit straight lines locally to the data whereby only a limitednumber of neighboring observations, determined by the band-width, are used which are weighted with decreasing weights thefurther away the observations are (Bowman and Azzalini, 1997).The locally fitted lines are then used to predict a smoothed valueys

ifor each local regression at covariate point xi. The smoothed

curve, thus results from repeated separate local linear regressionsfor every covariate point.

Local linear regression is used in this paper for its appealingsimplicity. Moreover, the technique is, under certain conditions,equivalent to the well known Nadaraya–Watson kernel estimator(Nadaraya, 1964; Watson, 1964) but behaves superior at the edgesof the covariate space as it employs a variable bandwidth reflectingthe density of the observations through a nearest neighbor distance(Harrod, 1997).9

Locally weighted scatterplot smoothing (LOWESS, Cleveland,1979) is a local linear estimator that uses the tricube kernel func-tion to easily evaluate adequately smooth weights for neighboringobservations. The smoothing parameter can straightforwardly beinterpreted as the share of the sample that enters the local regres-sion with positive weights (Bowman and Azzalini, 1997).

Figs. 1 and 2 show Engel curves for all 12 expenditure categoriesderived via locally weighted scatterplot smoothing. All charts usethe sample described in Table A.2. Parametric as well as nonpara-metric regression methods are susceptible to outliers. Both figurestherefore draw on a slightly reduced sample.10

Regarding the potential income adjustment problem outlinedin Section 3, it can be found that the shapes of the local linearregression estimates for real GDP and PPP adjusted real GDP arenot too different. The curve for PPP adjusted real GDP is usu-ally smoother than in the case of real GDP. Differences resultmostly at the upper tail of the income distribution. However,the curves are used to indicate an either positive or negativeslope in the overall sample of the data. Therefore, it can beconcluded that the qualitative relationship between the expen-diture share category and income remains unchanged in all

constant weighting function.10 The empirical setting implies the data to be clustered countrywise. Omitting

outliers to ensure a less volatile course of the curve is thus more controversial thanin a cross sectional setting. For a discussion on the outlier treatment please seeappendix A.

11 All estimations in the following sections use PPP adjusted real GDP.

124 W. Kaus / The Journal of Socio-Economics 47 (2013) 118– 134

F e chars d linet

teodo

ig. 1. Changes in expenditure shares with rising income – need based. Notes: Thhare on real GDP (red dashed line) as well as on PPP adjusted real GDP (black solihis figure legend, the reader is referred to the web version of the article.)

han Engel’s law. The tremendous consistency in the changing
xpenditure shares as income rises is unlikely to be the resultf pure chance. In fact, it calls for a rigorous theoretical foun-ation. As outlined in Section 2, the behavioral interpretationf the consumption motivation offers such an explanation by
ts plot the locally weighted smoothing scatterplots of the respective expenditure). The smoothing parameter is 0.7. (For interpretation of the references to color in

connecting differences in satiation patterns between innate needs
with systematic changes in consumption expenditures as incomerises.
The charts depicted in Fig. 1 have been conjectured to relate touniversally-shared human contingencies. The Engel curves of the

W. Kaus / The Journal of Socio-Economics 47 (2013) 118– 134 125

F arts pr oothin

rtccs

npttcsSgn

ipgsatnfttsh

dd

ig. 2. Changes in expenditure shares with rising income – remaining. Notes: The chelative to expenditures on food on PPP adjusted real GDP (black solid line). The sm

emaining expenditure categories are shown in Fig. 2. Althoughhe slopes are most probably different from zero, compared to thehanges in Engel’s law their magnitude is small. In contrast to theharts depicted in Fig. 2, all expenditure categories in Fig. 1 showystematic changes as income rises.

Among the changes observable in Fig. 1, the steadily and expo-entially decreasing shape of the Engel curve for food is a uniqueattern. Although the expenditure share for clothing is decreasingoo, the slope is much less steep. The downward sloping rela-ionship of the Engel curves indicates the underlying expenditureategories to be necessities. Both findings are in line with the, eithertrong or in the case of clothing more explorative, hypotheses inection 2.3 that relate the instrumental use of the expenditure cate-ories food and clothing to easily satiable basic needs of homeostaticature.

The most pervasive pattern characterizing half of the categories,.e., housing and utilities, furnishings and household equipment, trans-ort, recreation and culture, restaurants and hotels, and miscellaneousoods and services, is a positive relationship which increases at ateadily decreasing rate. In some cases, the curve seems to reachn upper ceiling.12 However, at the high end of the income dis-ribution, the slope of the curves is determined only by a smallumber of countries. Kinks in the curves in this range should there-

ore not distract from the overall pattern. The positive slope ofhe Engel curves indicates the underlying expenditure categories
o be luxuries. This result complies favorably with the hypothe-es in Section 2.3 which relate these expenditure categories withardly satiable basic needs of nonhomeostatic nature. Considering
12 As these estimates draw on expenditure share instead of nominal expenditureata, the concavity of the Engel curves does not directly allow to infer saturatedemand of the expenditure categories in the sense of Moneta and Chai (2010).

lot the locally weighted smoothing scatterplots of the respective expenditure shareg parameter is 0.7.

the expenditures for which a neat connection between needs andexpenditures could be drawn, very consistent structural changeswith rising income can be observed. The most striking patternis obtained for recreation and culture. Notwithstanding a certainheterogeneity, the other four expenditure categories, for whichonly tentative hypothesis could have been formulated, show struc-turally increasing budget shares as income rises.

Overall, the visible inspection of Figs. 1 and 2 shows that beyondEngel’s law a number of consistent relationships between expen-diture categories and income exist. In eight out of the 12 COICOPexpenditure categories defined by the United Nations statisticsdivision, steady and consistent changes could empirically be identi-fied as income rises. The theoretical explanation offered facilitatesaccounting for some of the systematically changing expenditurecategories. Given the above described limitations of the theo-retical approach and the widespread reluctance of the economicliterature to explain structural changes in expenditure patterns,this is already a lot. To complement the findings in this section,cross-country income elasticities of demand are estimated in thefollowing section.

4.2. Estimating cross-country income elasticities

When estimating Engel curves at any aggregate level cer-
tain assumptions are implied. Pooling individual level data in across section at a given point in time t implicitly assumes thatthey are independently drawn from an identical distribution. Inother words, the are assumed to behave structurally similar.13 In
13 Only in the very restrictive and obviously implausible case of exact linear aggre-gation this necessitates the marginal propensity to consume to be similar for all

1 io-Economics 47 (2013) 118– 134

tadtcabstiatafpccipttfta

pairqaWsbs˛aımfeissd

Tocotttitaεct

(w

o

ε

Fig. 3. Double-log estimation of the income elasticity of demand for food. Notes:The chart plots the double-log model for expenditures on food. The thin solid blackline represents proportionally increasing expenditures with rising income. The reddashed line depicts a linear fit and thus the income elasticity of demand of the whole

nonlinear fitting technique which represents the change in elas-ticities with rising income. While the gradient is rather steep forless affluent countries it gets less steep with increasing income. To

15 The double-log approach can, however, also be applied to the budget share Engelcurves presented in Figs. 1 and 2. This can be shown by the following transformation.As a starting point, the double-log specification of the budget share Engel curve ischosen. Disassembling the left hand side and rearranging the income component tothe right hand side yields

log(wi) = log

(qi

y

)= log(qi) − log(y) = ˛i + ˇi ∗ log(y)


his paper, country level (average) observations are pooled.14 Thepproach again assumes that each of the observations is indepen-ently drawn from an identical distribution. It is thus assumedhat, in expectations, behavior follows structural similarities acrossountries, too. Before estimating cross-country income elasticities,

methodological shortcoming of this empirical approach has toe mentioned. When estimating income elasticities in a dynamicetting, potentially changing factors have to be taken into account,oo. It would, for example, be desirable to account for changes inncome inequality and in the aging structure of the countries. Bothre closely related and have been shown to influence the estima-ion of food demand in individual based cross sectional data (Cirerand Masset, 2010). Moreover, it would be desirable to accountor relative price changes over time. If, for example, the relativerice for food would consistently decrease over time and acrossountries, this could translate into structurally similar changes inonsumption patterns across countries and time, say a decreasen the expenditure share for food. While changes in inequality andopulation dynamics are likely to be absorbed in this empirical set-ing (unit of observation is the average of the sum of individuals),he later is a more serious concern. In the absence of consistent dataor all countries and all time periods, this remains a major caveato the analysis. Given the limitations of the data, the estimationccuracy is naturally constraint.

In Section 2, it was conjectured that differences in satiationatterns of universally-shared human needs translate, on theggregate level, into different shapes of Engel curves and thus alsonto different income elasticities of demand. To back up the explo-ative graphical approach used in Section 4.1, this section aims touantify the income elasticities of demand. Theoretically, therere different ways to estimate the income elasticities of demand.hen dealing with single item equations (as compared to demand

ystems), probably the most prominent model was developedy Working (1943). The Working model relates the expenditurehare of a certain expenditure i to the logarithm of income (wi =i + ıi ∗ log(y)). Income elasticities of demand can be derived by

reformulation of the Working model dqi/d(y) = ˛i + ıi ∗ log(y) +i = wi + ıi such that (dqi/d(y))/(qi/yi) = εi = 1 + ıi/wi. The enor-ous appeal of the Working model lies in the explicit nonlinear

ormulation of the elasticities. As the budget share spent on thexpenditure category wi changes with rising income, the elastic-ty εi changes accordingly. This is consistent with various empiricaltudies such as Banks et al. (1997) or Blundell et al. (1993) thathow evidence for changing income elasticities of demand whenisaggregated for groups of varying affluence within a cross section.

The special appeal of the double-log model lies in its simplicity.he model, however, suffers from modeling the income elasticityf demand as a constant. Estimating a single elasticity in a cross-ountry setting could thus mask the potentially nonlinear structuref the data. This paper utilizes the double-log approach while athe same time applying a nonlinear average derivative approacho cross check the validity of the chosen approach. In contrast tohe Working model, the appeal of the double-log method can eas-ly be illustrated in a graphical representation. Moreover, unlike inhe Working model, this approach does not necessitate to utilize

budget share wi to determine the income elasticity of demandi which in this setting would have to be formulated either across
ountries or time. Both would hardly do justice to the structure inhe data.
rich and poor) consumers (Deaton and Muellbauer, 1980). Accordingly, preferencesould then restricted to be either homothetic, or quasi-homothetic.

14 This approach is generally feasible as regressions are expected values and a sumf expected values equals the expected value of a sum.

sample. The black solid line shows the locally weighted smoothing scatterplot of logfood expenditures on log PPP adjusted real GDP. The vertical dotted lines mark thedata restrictions for nonlinear estimation technique.

Utilizing a double-log approach, the income elasticity of demandcan be derived by

εi = d(log(qi))d(log(y))

. (1)

The15 expenditures on food are an illustrative example for thesuitability of the combination of double-log method and a nonpara-metric plausibility check. Fig. 3 plots the logarithm of expenditureson food on the logarithm of PPP adjusted GDP per capita. Whilethe thin solid black line depicts a stylized proportional increase inincome, thus an income elasticity of demand of unity, the dashedline represents a linear fit to the full data set and thus representsthe double-log estimation of the income elasticity of demand ofthe whole sample. Additionally, Fig. 3 depicts the locally weightedsmoothing scatterplot for food expenditures (black thick solid line).The vertical dashed lines mark the data restrictions for nonlinearestimation technique.16

The example vividly illustrates the changing steepness of the

log(qi) = ˛i + (ˇi + 1) ∗ log(y).

The slope of the double-log budget share Engel curve ˇi can be derived by taking thefirst derivative d(log(qi))/d(log(y)) = ˇi + 1. Using (1) it can be shown that εi = ˇi + 1.

16 The charts displayed in Figs. A.4 and A.5 nicely visualize the necessity to restrictthe data when dealing with nonlinear estimation techniques. As the data is usuallymore scarce in the tails of the distribution, the nonlinear estimation will yield shiftsthat are not representative. The chosen restriction includes only observations largerthan 0.25 and smaller than 2 of the mean normalized PPP adjusted GDP per capitadimension.

W. Kaus / The Journal of Socio-Eco

Table 1Cross-country income elasticities of demand.

Expenditure category εnonpari

εpari

R2pari

Food 0.35 0.49 0.79Alcohol and tobacco 0.86 0.94 0.75Clothing 0.82 0.92 0.83Housing and utilities 1.16 1.23 0.95Furnishings and household equipment 1.12 1.18 0.93Health 1.06 1.17 0.71Transport 1.13 1.14 0.95Communication 1.23 1.17 0.72Recreation and culture 1.56 1.55 0.95Education 0.62 0.67 0.32Restaurants and hotels 1.11 1.33 0.84Miscellaneous goods and services 1.47 1.40 0.89

Notes: The parametrically derived income elasticities of demand are estimatedusing Eq. (1). These estimations use the full sample described in Table A.2. Thenonparametrical estimations are derived using an average derivative estimationtechnique that decomposes the sorted income data in 100 bins and calculates anaverage derivative from the estimated bin derivatives. As nonparametric estima-tion techniques are susceptible to particularities at the tails of the distributions, theerp

etbtttiiwloaedcettfidegac

itcihcehcbe

nconb

stimations are restricted to observations within a [0.25:2] normalized range of theespective expenditure category’s mean. The sample thus shrinks by about 13%. Therocedure is illustrated in Figs. A.4 and A.5.

valuate the elasticity more thoroughly it is necessary to comparehe gradient with the stylized proportional elasticity representedy the thin solid black line. Across the whole range, the slope ofhe nonlinear average derivative estimation is less steep than thehin black line. Accordingly, irrespective of the level of affluence,he income elasticity of demand is lower than unity. As incomencreases, however, the slope successively gets less steep. Thisndicates, that the income elasticity of demand for food decreases

ith rising income and again validates the robustness of Engel’saw. Moreover the illustration in Fig. 3 illustrates the usefulnessf combining a (graphical) double-log estimation approach with

nonlinear robustness check. In this case, the application of lin-ar double-log estimation overestimates the income elasticity ofemand. This can also be found in Table 1. Table 1 reports theross-country income elasticities of demand for all 12 COICOPxpenditure categories. Alongside the linear double-log estimates,he table presents the nonlinear estimates of the average deriva-ive estimations and the coefficient of determination of the lineart. Strikingly, in the cross-country setting, the results of the linearouble-log estimations resemble the nonlinear results to a largextent. Despite the obvious counterexample presented in Fig. 3, theraphical representation in Figs. A.4 and A.5 largely points to thedequacy of the linear double-log technique for most expenditureategories.17

Overall, the presented results support the findings of the visualnspection of cross-country Engel curves in Section 4.1. In all caseshe estimation of cross-country income elasticities of demandonform with the identified structural changes in the correspond-ng cross-country Engel curves. While expenditure categories thatave been related to hardly satiable needs show increasing cross-ountry Engel curve slopes, they are also characterized by incomelasticities of demand above unity. Expenditure categories thatave been related to easily satiable needs show decreasing cross-
ountry Engel curve slopes and income elasticities of demandelow unity. Among the potentially institutionally constrainedxpenditure categories, the estimates for alcohol and tobacco and
17 This does, however, not imply that these expenditure categories might not showonlinear properties in individual based cross sectional data. Most likely this is thease. Given that the approach in this paper utilizes data that represent the averagef the sum of individual behavior in different countries, the degree of variation isaturally smaller than the degree of variation that can be obtained using individualased cross sectional data.

nomics 47 (2013) 118– 134 127

health are at least close to unity. In the case of communication andeducation the estimates are clearly different from unity. The chartsdisplayed in Fig. A.5, however, indicate a comparatively high degreeof country specific heterogeneity in these two categories.

The qualitative findings are also consistent with earlier resultson cross-country data. A study that is at least comparable in scopeand heterogeneity of the sample is Theil et al. (1989). In theirapproach, Theil et al. (1989) visualize cross-country Engel curvesat three different points in time and utilize a Working-PI model toestimate income elasticities of demand for each of the 51 countriesat a given point in time (1980). Similarly they find food to be “themost basic necessity” while recreation turns out to be “the mostluxurious” (Theil et al., 1989, p. 108). Moreover, consistently withour result alcohol and tobacco and education show flat cross-countryEngel curves for which income elasticities of demand close to unityare estimated. Clothing turns out to be a necessity, while transport,furnishings, gross rent and fuel, other are luxuries. A remarkable dif-ference are expenditures on health that show income elasticities ofdemand well above unity in their data.

5. Conclusions

Since the work of Ernst Engel it is a well established fact thatthe expenditure share dedicated to food consumption decreasesas income rises (Engel’s law). The extraordinary consistency acrosstime and space and the predictive character of the relationship ledto its use to determine absolute household poverty and the defini-tion of poverty lines. In contrast, evidence on systematic changesin other consumer expenditures categories is hardly available. Thelack of such an assessment is even more remarkable as Ernst Engelalready pursued a much more encompassing approach. In fact,Engel’s original work was an inquiry into the structural changesof consumption patterns with rising income aiming to determinea measure of household welfare. With his pioneering work on adistinction between necessities and luxuries, Engel was not onlythe first to empirically assess this taxonomy but also the first toattempt a behaviorally founded – need based – explanation.

To build upon Engel’s legacy, this paper, on the one hand,improves on part of the theoretical assumptions by enrichingEngel’s original approach with current scientific knowledge on thenature of consumer needs and how they are satisfied. The behav-ioral approach to consumer behavior offered in Section 2 allows forthe formulation of expectations about structural changes of expen-diture shares as income rises. While expenditure categories that canbe associated with hardly satiable needs of nonhomeostatic natureare conjectured to show increasing cross-country budget shareEngel curves that reflect an income elasticity of demand above unity(“luxury”), expenditure categories that can be associated with eas-ily satiable needs of homeostatic nature are conjectured to showdecreasing slopes, reflecting income elasticities of demand belowunity (“necessity”). The hypotheses put forward already imply thatfrom a behavioral consumption approach perspective there aregood reasons to expect the existence of more than one robustempirical regularity in the analysis of demand patterns.

On the other hand, this paper assesses the systematic changesof consumer expenditure shares with rising income. The paperpresents an empirical analysis of observable changes in con-sumption patterns across countries and time, using a data setthat contains individual consumption expenditure in 12 cate-gories for more than 50 countries over up to 50 years. Ourfindings confirm the conjecture that the distinction between lux-
uries and necessities rests upon behavioral foundations that canbe traced back to the differing satiation features of the underlyinguniversally-shared needs. Utilizing nonparametric regression tech-niques, it is explored whether expenditure categories, inclusive

1 io-Economics 47 (2013) 118– 134

oieE

A

easBn

cvttocs

iceat

smsowiwa

TC

Ns

Table A.2Availability of data by countries

# Country First year Last year N

1 Australia 1959 2007 492 Belarus 1994 2007 143 Botswana 1994 2001 84 British Virgin Islands 1995 1999 55 Bulgaria 1995 2006 126 Cameroon 1996 2004 97 Canada 1970 2007 388 Hong Kong, ChinaJ 1966 2004 239 Macao, ChinaJ 1993 2007 15

10 Colombia 1994 2005 1211 Croatia 1999 2006 812 Czech Republic 1992 2007 1613 Denmark 1966 2007 4214 Dominican Republic 1991 1996 615 Estonia 1994 2007 1416 Finland 1975 2007 3317 France 1959 2008 5018 Germany 1991 2007 1719 Greece 1960 2004 4520 Guatemala 2001 2006 621 Honduras 2000 2005 622 Hungary 1995 2008 1423 Iceland 1990 2006 1724 India 1999 2007 925 IranD 1992 2007 1626 Ireland 1990 2007 1827 Israel 1995 2007 1328 Italy 1970 2007 3829 Japan 1980 2007 2830 KenyaD 1996 2007 1231 Korea, Rep. 1970 2007 3832 Kyrgyzstan 2000 2005 633 Latvia 1995 2007 1334 Lithuania 1995 2007 1335 Luxembourg 1985 2007 23


f, but extending beyond, food, change systematically with risingncome. This way a number of empirical regularities in consumerxpenditure patterns could be identified that go well beyondngel’s law.

ppendix A. Outlier treatment

In a cross sectional setting, it is usually convenient to cutxpenditures on the category of interest as well as the income vari-ble at a specific boundary. As a cutoff point, the mean plus threetandard deviations is a commonly used rule of thumb (see, e.g.,anks et al., 1997). This kind of standard procedure however doesot fit in with the data structure in this paper.

The empirical setting applied in this paper combines annualountry level data for more than 50 countries. Pooling the data ofarious years implies obtaining a cloud of data points for each coun-ry. Depending on the number of years for which data is available,he respective data cloud varies in size. Combining the data of vari-us countries then results in more or less overlapping countrywiselusters. Outliers thus usually do not take the form of displacedingle points but separate clusters.

Being interested in assessing and describing structural changesn consumption patterns as income rises, the particularities in aertain country should not determine the overall shape of thestimated Engel curve. An outlying cluster of observations shouldccordingly be eliminated. Figs. A.1–A.3 present the applied outlierreatment.

The left hand side charts plot the locally weighted smoothingcatterplots without any outlier treatment. In some cases, the esti-ated local linear regression curve thus shows unrepresentative

hifts. The right hand side charts in Figs. A.1–A.3 additionally markutliers with ellipses and/or green color. For comparison, the locally
eighted smoothing scatterplots are estimated without the outly-
ng observations. Moreover, another local linear regression curve,hich applies the standard procedure of cutting off observations

bove the mean plus three standard deviations, is added.

able A.1OICOP expenditure categories

Expenditure category Description

1 Food All kinds of food and nonalcoholic beveragesconsumed at home

2 Alcohol and tobacco Spirits, wine, beer, tobacco, narcotics3 Clothing Clothing and footwear incl. related services

(cleaning, repair, hire)4 Housing and utilities Rentals, maintenance and repair of dwelling,

electricity, water, sewerage, gas, other fuels(incl. heat energy)

5 Furnishings andhousehold equipment

Furniture and furnishings, carpets, householdappliances, glass- and tableware, routinemaintenance of dwelling, related services

6 Health Pharmaceutical and therapeuticalproducts/equipment, outpatient services(medical and dental), hospital services

7 Transport Purchase of vehicles, fuels, maintenance,services by air, rail, road, waterways

8 Communication Postal services, telephone/fax services andequipment

9 Recreation and culture Recreation equipment and services, culturalservices, package holidays

10 Education Expenses on primary, secondary, tertiary, andother education

11 Restaurants and hotels Expenses on catering, restaurants, cafés,canteens, accommodation services

12 Miscellaneous goodsand services

Personal care, prostitution, personal effects(incl. jewelry), social protection, insurances(incl. live, dwelling, health, transport), financialservices, other services

otes: The expenditure categories are defined according to the UN Statistics Divi-ion’s Classification of Individual Consumption According to Purpose (COICOP).

36 Malaysia 2000 2007 837 Malta 1996 2007 1238 Mexico 1988 2004 1739 Mongolia 1997 2007 1140 NamibiaD 2000 2007 841 Netherlands 1987 2007 2142 Netherlands AntillesF 1997 2006 1043 New ZealandD 1987 2007 2144 Nicaragua 1994 2005 1245 NigerF 2003 2008 646 Norway 1980 2006 2747 Papua New GuineaD 1992 2003 1248 Poland 1995 2007 1349 Serbia 2002 2007 650 Sierra LeoneF 2001 2007 751 Singapore 1996 2008 1352 Slovakia 1993 2007 1553 Slovenia 1995 2007 1354 Solomon Islands 2003 2007 555 South AfricaJ 1946 2008 6356 Spain 1995 2007 1357 Sri Lanka 1998 2007 1058 Sweden 1993 2007 1559 Switzerland 1990 2006 1760 Yugoslavia, former (Macedonia) 1996 2007 1261 Ukraine 2001 2007 762 United Kingdom 1970 2007 3663 United States 1970 2007 3864 YemenD 1996 2007 12

Notes: Countries indicated with D are dropped from the analysis due to missings inmore than 2 expenditure share categories. Countries indicated with F are droppedfrom the analysis as they reported no changes in any expenditure share over thesample period. Countries indicated with J are dropped from the analysis due to huge– implausible – jumps in expenditure shares at a certain point in time.


Fig. A.1. Changes in expenditure shares with rising income – Outlier correction. Notes: The left hand side charts plot the locally weighted smoothing scatterplots of therespective expenditure share on real GDP (red dashed line) as well as on PPP adjusted real GDP (black solid line). The right hand side charts additionally mark the identifiedoutliers with ellipses and/or green color. For comparison the locally weighted smoothing scatterplots are run without the outlying observations. Moreover another locallinear regression curve for PPP adjusted real GDP is added that applies the standard procedure of cutting off observations above the mean plus three standard deviations(brown solid line). In all cases the smoothing parameter is 0.7. (For interpretation of the references to color in this figure legend, the reader is referred to the web version ofthe article.)






Fig. A.4. Changes in log expenditure shares with rising log income. Notes: The charts plot the locally weighted smoothing scatterplots of log expenditures on log PPPadjusted real GDP (black solid line). A linearly fitted regression line (red, dashed) is added to each graph. Additionally, the thin black line represents proportionally increasinge

rtea

xpenditures with rising income.

Without any outlier treatment, the estimated nonparamet-
ic curves would yield unrepresentative shifts particularly inhe higher income range. Most remarkable examples are thexpenditure categories alcohol and tobacco, health, and recre-tion and culture. A comparison of both outlier treatments yields
remarkable differences too. At least for alcohol and tobacco and fur-
nishings and household equipment the data cloud for Luxembourglies above the overall pattern. Applying the standard outlier proce-dure omits only a part of the data cloud. The estimated local linearregression curve thus remains affected. The comparison of the left


7 8 9 10 11

45

67

89

log(GDP_PPP_cap_2005)

log(

Exp

_tra

nspo

rt)

local linear fitlinear fitε=1

8 9 10 11

23

45

67


log(

Exp

_com

mun

icat

ion)


7 8 9 10 11

23

45

67

89


log(

Exp

_rec

reat

ion_

cultu

re)


7 8 9 10 11

23

45

67


log(

Exp

_edu

catio

n)


7 8 9 10 11

23

45

67

89


log(

Exp

_res

taur

ants

_hot

els)


7 8 9 10 11

23

45

67

89


log(

Exp

_mis

cella

neou

s)


F tes: Tl shed)i

was

ctot

ig. A.5. Changes in log expenditure shares with rising log income - continued. Noog PPP adjusted real GDP (black solid line). A linearly fitted regression line (red, dancreasing expenditures with rising income.

ith the right hand side charts in Figs. A.1–A.3 shows the need forn outlier treatment that accounts for the countrywise clusteredtructure of the data.

Overall, it can be summarized that especially Luxembourg, the
ountry with the highest income, causes changes in the slope ofhe locally weighted smoothing scatterplots. Other readily visibleutliers are much less disturbing. In fact, only in one case (health)he shape of the estimated curves is affected by countries other
he charts plot the locally weighted smoothing scatterplots of log expenditures on is added to each graph. Additionally, the thin black line represents proportionally

than Luxembourg. Although the outlier treatment proposed heremay appear arbitrary at the first glance, it can be concluded that itreasonably fits the structure of the data.

References

Almås, I., 2012. International income inequality: measuring PPP bias by estimatingEngel curves for food. American Economic Review 102 (2), 1093–1117.

http://refhub.elsevier.com/S1053-5357(13)00141-8/sbref0005




















1 io-Eco

B

B

B

B

B

B

CC

C

C

C

C

C

D

D

E

E

E

F

FF

G

HH

H

HH

HH

Witt, U., 2011. The dynamics of consumer behavior and the transition to sustainable


alassa, B., 1964. The purchasing-power parity doctrine: a reappraisal. Journal ofPolitical Economy 72 (6), 584–596.

andura, A., 1986. Social Foundations of Thought and Action: A Social CognitiveTheory. Prentice-Hall, Englewood Cliffs, NJ.

anks, J., Blundell, R., Lewbel, A., 1997. Quadratic Engel curves and consumerdemand? Review of Economics and Statistics 79 (4), 527–539.

arigozzi, M., Moneta, A., 2012. Identifying the independent sources of consumption.LEM Working Paper Series 2012/16. Scuola Superiore Sant’Anna.

lundell, R., Pashardes, P., Weber, G., 1993. What do we learn about consumerdemand patterns from micro data? American Economic Review 83 (3), 570–597.

owman, A.W., Azzalini, A., 1997. Applied Smoothing Techniques for Data Analysis.Oxford University Press, Oxford, UK.

aplan, A., 1978. The Sociobiology Debate. Harper & Row, New York.hai, A., 2011. Consumer specialization and the romantic transformation of the

British Grand Tour of Europe. Journal of Bioeconomics 13, 181–203.hai, A., Moneta, A., 2010. Retrospectives: Engel curves. Journal of Economic Per-

spectives 24 (1), 225–240.hai, A., Moneta, A., 2012. Back to Engel? Some evidence for the hierarchy of needs.

Journal of Evolutionary Economics 22, 649–676.harles, K.K., Hurst, E., Roussanov, N., 2009 May. Conspicuous consumption and race.

The Quarterly Journal of Economics 124 (2), 425–467.irera, X., Masset, E., 2010. Income distribution trends and future food demand.

Philosophical Transactions of the Royal Society B 365, 2821–2834.leveland, W.S., 1979. Robust locally weighted regression and smoothing scatter-

plots. Journal of the American Statistical Association 74 (368), 829–836.eaton, A., Muellbauer, J., 1980. Economics and Consumer Behavior. Cambridge

University Press.uesenberry, J.S., 1949. Income, Saving and the Theory of Consumer Behavior. Har-

vard University Press, Cambridge, MA.ngel, E., 1857. Die Produktions- und Consumptions Verhältnisse des Königreichs

Sachsen. Zeitschrift des Statistischen Bureaus des Königlich Sächsischen M inis-teriums des Inneren 8 and 9, reprinted in the Appendix of Engel (1895).

ngel, E., 1895. Die Lebenskosten belgischer Arbeiterfamilien früher und jetzt. Bul-letin de l’Institut International de Statistique 9, 1–124.

ngel, J., Kneip, A., 1996. Recent approaches to estimating Engel curves. Journal ofEconomics 63, 187–212.

rank, R.H., 1999. Luxury Fever. Money and Happiness in an Era of Excess. PrincetonUniversity Press.

rank, R.H., 2011. The Darwin Economy. Princeton University Press, Princeton, NJ.renzel Baudisch, A., 2006. Continuous market growth beyond functional satiation.

Time-series analyses of U.S. footwear consumption 1955–2002. In: Papers onEconomics and Evolution 2006-03. Max Planck Institute of Economics, Evolu-tionary Economics Group.

eorgescu-Roegen, N., 1954. Choice, expectations and measurability. The QuarterlyJournal of Economics 68 (4), 503–534.

arrod, R.F., 1933. International Economics. Cambridge University Press, London.art, J.D., 1997. Nonparametric Smoothing and Lack-of-Fit Tests. Springer Verlag,

New York.errnstein, R.J., 1990. Behavior, reinforcement and utility. Psychological Science 1

(4), 217–224.errnstein, R.J., 1997. The Matching Law. Harvard University Press, Cambridge, MA.eston, A., Summers, R., Aten, B., 2011. Penn World Table 7.0, Income and Prices at

the University of Pennsylvania. Center for International Comparisons of Produc-
tion.
irsch, F., 1978. Social limits to growth. Harvard University Press, Cambridge, MA.outhakker, H.S., 1957. An international comparison of household expenditure

patterns, commemorating the centenary of Engel’s law. Econometrica 25 (4),532–551.

nomics 47 (2013) 118– 134

Hull, C.L., 1943. Principles of Behavior. Appleton-Century-Crofts, New York.Ironmonger, D.S., 1972. New Commodities and Consumer Behavior. Cambridge Uni-

versity Press, Cambridge, MA.Kahneman, D., Wakker, P.P., Sarin, R., 1997. Back to Bentham? Explorations of expe-

rienced utility. The Quarterly Journal of Economics 112 (2), 375–405.Lades, L.K., 2013. Explaining shapes of Engel curves: the impact of differential

satiation dynamics on consumer behavior. Journal of Evolutionary Economics,DOI:10.1007/s00191-013-0324-6 (forthcoming).

Lanjouw, O.J., Lanjouw, P., 2001. How to compare apples and oranges: poverty mea-surement based on different definitions of consumption. Review of Income andWealth 47 (1), 25–42.

Lewbel, A., 2008. Engel curve. In: Durlauf, S.N., Blume, L.E. (Eds.), The New PalgraveDictionary of Economics. Palgrave Macmillan, Basingstoke.

Menger, C., 1871. The Principles of Economics. Free Press, Glenco, IL (1950).Moneta, A., Chai, A., 2010. The evolution of Engel curves and its implications for struc-

tural change. In: Discussion Papers in Economics 2010/09. Griffith University,Department of Accounting, Finance and Economics.

Musgrove, P., 1985. Food needs and absolute poverty in urban South America.Review of Income and Wealth 31 (1), 63–83.

Nadaraya, E.A., 1964. On estimating regression. Theory of Probability and its Appli-cations 10, 186–190.

Pavlov, I.P., 1927. Conditioned Reflexes: An Investigation of the Physiological Activityof the Cerebral Cortex. Oxford University Press, London.

Samuelson, P.A., 1964. Theoretical notes on trade problems. The Review of Eco-nomics and Statistics 46 (2), 145–154.

Scitovsky, T., 1981. The desire for excitement in modern society. Kyklos 34 (1),3–13.

Seale, J.L., Regmi, A., 2006. Modelling international consumption patterns. Reviewof Income and Wealth 52 (4), 603–624.

Skinner, B.F., 1966. Operant behavior. In: Honig, W.K. (Ed.), Operant Behavior – Areasof Research and Application. Meredith, New York, pp. 12–32.

Staddon, J.E.R., Cerutti, D.T., 2003. Operant conditioning. Annual Review of Psychol-ogy 54, 115–144.

Theil, H., Chung, C.-F., Seale Jr., J.L., 1989. International Evidence on ConsumptionPatterns. JAI Press Inc., Greenwich, CT.

Volland, B., 2012. Some economic consequences of human biology, unpublisheddissertation.

Watson, G.S., 1964. Smooth regression analysis. Sankhya: The Indian Journal ofStatistics, Series A 26 (4), 359–372.

Wilson, E.O., 1975. Sociobiology: The New Synthesis. Harvard University Press,Cambridge, MA.

Witt, U., 2001. Learning to consume – a theory of wants and the growth of demand.Journal of Evolutionary Economics 11 (1), 23–36.

Witt, U., 2008. What is specific about evolutionary economics? Journal of Evolution-ary Economics 18 (5), 547–575.

Witt, U., 2010a. Product characteristics, innovations and the evolution of consump-tion. A behavioral approach, mimeo. In: Paper prepared for the conference on“Technical Change: History, Economics and Policy” in honor of G.N. Tunzelmann,SPRU, March 2010.

Witt, U., 2010b. Symbolic consumption and the social construction of prod-uct characteristics. Structural Change and Economic Dynamics 21 (1),17–25.

consumption patterns. Environmental Innovation and Societal Transitions 1 (1),109–114.

Working, H., 1943. Statistical laws of family expenditure. Journal of the AmericanStatistical Association 38, 43–56.


































































































































































































































































































































































































































































http://dx.doi.org/10.1007/s00191-013-0324-6

















































































































































































































































































































































































Documents

Beyond Engel's law - A cross-country analysis