8
Model and Measurement Methodology for the Analysis of Consumer Choice of Food Products 1. Introduction This paper considers the problem of a consumer purchasing a food product within a certain product class (e. g. meat, bread, vegetables, soft drinks, cheese) and making a choice from the different alternatives that are available. These alternatives may be for ex- ample: different brands, different types (e. g. pork or beef or lamb for meat products, cauliflower or lettuce or French beans for vegetables), dif- ferent methods of preservation (e. g. canned or frozen or fresh), different packages or different nutritional contents (e. g. whole milk versus skim milk). It is important to know how these choices are made and which factors affect the outcome of the choice process, i. e. the alterna- tive actually chosen. This insight is important to the suppliers - produ- cers and retailers who attempt to of- fer products for which the consumers have a preference - and also to nu- trition scientists and people involved in nutrition education. If it is desir- able to change people's habits of buying and eating food, the factors that influence this behaviour must be known. 2. Views of consumer behaviour In the past, different approaches have been used to explain consumer behaviour. In the economist's view, the consumer is a rational consumer; knowing all the different products, their prices, qualities and other attri- butes, the consumer compares all possible combinations of products that he can buy with his income and Ernahrunqs-Urnschau 30 (1983). Supplement by Berend Wierenga, Wageningen ultimately chooses the combination that maximizes his total utility. For the axioms of economic theory of consumer behaviour, see, for exam- ple, GREEN (1971). A quite different school of thought - the motivations research school - believes that the consumer's motiva- tions are of psycho-analytical origin; this view was very dominant in the 1950s and 1960s. In this approach, what might be called the Freudian consumer, is considered as a passive being, irrational and driven by un- controllable forces from his subcon- scious. Experiences early in life, and sexual connotations play an impor- tant role in consumer behaviour, ac- cording to this theory (DICHTER, 1960). The consumer is very vulner- able to manipulations by his envi- ronment. The advertising industry in particular exploits the hidden forces in consumers to make them buy cer- tain products (PACKARD, 1957). The theoretical starting points of economics are charming in their clarity and simplicity, and economic theory has a good deal to say about how consumer expenditure is allo- cated among product categories such as food, clothing, leisure, etc. How- ever, at the more detailed level of choices between alternatives within a product class, the theoretical as- sumptions are not tenable. Usually the consumer is unaware of all the alternatives available and knows even less about all the specifics of each alternative. Also, the goals of a consumer are more than economic: in his consumption behaviour he may also strive to satisfy other needs, e. g. being accepted by other people, feelings of safety, status and prestige. It is now generally agreed upon that the claims of followers of the motiva- tion research school to explain con- sumer behaviour by very simple and straigthforward factors such as asso- ciations with sex, obedience to par- ents, etc. have not been substantia- ted. Often their explanations have been no more than speculative. Nevertheless, this approach has drawn attention to the role symbolic, non-economic aspects of products may play in consumer choice behav- iour and to the need of studying con- sumer motivation, perceptions, be- liefs, attitudes, etc. to explain con- sumer behaviour. In this paper a third view of con- sumer behaviour is adopted, which is emerging after a period of ever-in- creasing activity in consumer re- search: the consumer as an imperfect problem solver or decision-maker. This view can be found in a number of recent textbooks on consumer be- haviour, such as: ENGEL et al. (1978), SCHIFFMAN and KANUK (1978) and ZALTMAN and WAL- LENDORF (1979). Consumer behaviour is not irra- tional but purposeful and goal- oriented. But - as already mentioned - these goals are more than econom- ic. Moreover, the consumer has li- mited information about alternative choices and often his perceptions are biased. His memory and informa- tion-processing capabilities are limit- ed. Furthermore, a consumer can on- ly spend a very small amount of time on each individual decision. Within these limitations of information, cognitive skills, memory capacity and time, the consumer has to solve his problems of choice. 21

Model and Measurement Methodology for the Analysis of

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Model and Measurement Methodology for the Analysis of

Model and Measurement Methodology for theAnalysis of Consumer Choice of Food Products

1. Introduction

This paper considers the problemof a consumer purchasing a foodproduct within a certain productclass (e. g. meat, bread, vegetables,soft drinks, cheese) and making achoice from the different alternativesthat are available.

These alternatives may be for ex-ample: different brands, differenttypes (e. g. pork or beef or lamb formeat products, cauliflower or lettuceor French beans for vegetables), dif-ferent methods of preservation (e. g.canned or frozen or fresh), differentpackages or different nutritionalcontents (e. g. whole milk versusskim milk). It is important to knowhow these choices are made andwhich factors affect the outcome ofthe choice process, i. e. the alterna-tive actually chosen. This insight isimportant to the suppliers - produ-cers and retailers who attempt to of-fer products for which the consumershave a preference - and also to nu-trition scientists and people involvedin nutrition education. If it is desir-able to change people's habits ofbuying and eating food, the factorsthat influence this behaviour must beknown.

2. Views of consumer behaviour

In the past, different approacheshave been used to explain consumerbehaviour. In the economist's view,the consumer is a rational consumer;knowing all the different products,their prices, qualities and other attri-butes, the consumer compares allpossible combinations of productsthat he can buy with his income and

Ernahrunqs-Urnschau 30 (1983). Supplement

by Berend Wierenga, Wageningen

ultimately chooses the combinationthat maximizes his total utility. Forthe axioms of economic theory ofconsumer behaviour, see, for exam-ple, GREEN (1971).

A quite different school of thought- the motivations research school -believes that the consumer's motiva-tions are of psycho-analytical origin;this view was very dominant in the1950s and 1960s. In this approach,what might be called the Freudianconsumer, is considered as a passivebeing, irrational and driven by un-controllable forces from his subcon-scious. Experiences early in life, andsexual connotations play an impor-tant role in consumer behaviour, ac-cording to this theory (DICHTER,1960). The consumer is very vulner-able to manipulations by his envi-ronment. The advertising industry inparticular exploits the hidden forcesin consumers to make them buy cer-tain products (PACKARD, 1957).

The theoretical starting points ofeconomics are charming in theirclarity and simplicity, and economictheory has a good deal to say abouthow consumer expenditure is allo-cated among product categories suchas food, clothing, leisure, etc. How-ever, at the more detailed level ofchoices between alternatives within aproduct class, the theoretical as-sumptions are not tenable. Usuallythe consumer is unaware of all thealternatives available and knowseven less about all the specifics ofeach alternative. Also, the goals of aconsumer are more than economic:in his consumption behaviour hemay also strive to satisfy other needs,e. g. being accepted by other people,feelings of safety, status and prestige.

It is now generally agreed upon thatthe claims of followers of the motiva-tion research school to explain con-sumer behaviour by very simple andstraigthforward factors such as asso-ciations with sex, obedience to par-ents, etc. have not been substantia-ted. Often their explanations havebeen no more than speculative.Nevertheless, this approach hasdrawn attention to the role symbolic,non-economic aspects of productsmay play in consumer choice behav-iour and to the need of studying con-sumer motivation, perceptions, be-liefs, attitudes, etc. to explain con-sumer behaviour.

In this paper a third view of con-sumer behaviour is adopted, which isemerging after a period of ever-in-creasing activity in consumer re-search: the consumer as an imperfectproblem solver or decision-maker.This view can be found in a numberof recent textbooks on consumer be-haviour, such as: ENGEL et al.(1978), SCHIFFMAN and KANUK(1978) and ZALTMAN and WAL-LENDORF (1979).

Consumer behaviour is not irra-tional but purposeful and goal-oriented. But - as already mentioned- these goals are more than econom-ic. Moreover, the consumer has li-mited information about alternativechoices and often his perceptions arebiased. His memory and informa-tion-processing capabilities are limit-ed. Furthermore, a consumer can on-ly spend a very small amount of timeon each individual decision. Withinthese limitations of information,cognitive skills, memory capacityand time, the consumer has to solvehis problems of choice.

21

Page 2: Model and Measurement Methodology for the Analysis of

SIMON uses the expression"bounded rationality": an individualcan only have rational behaviour"that is compatible with the access toinformation and the computationalcapacities that are actually processedby organisms" (SIMON, 1979, p. 7).For an information-processing ap-proach to consumer behaviour, seealso HEITMAN (1979).

In this paper, keeping the con-sumer as an imperfect problem-sol-ver in mind, I shall consider the vari-ous factors that affect consumer be-haviour in choosing food products:economic factors such as prices andincome, psychological variables suchas perceptions, beliefs, motives andattitudes, and sociological factorssuch as culture, social class and ref-erence groups.

Figure 1: A model jar the choice oj a jood product

To integrate all these factors andtheir interactions, a comprehensivemodel of consumer choice behaviouris required. Such a model for thechoice of food products is presentedin the next section. This model,which draws upon more generalmodels from the literature, e. g.HOWARD and SHETH (1969) andENGEL et. al. (1978) offers an ana-lytical framework for the analysis ofconsumer food choice. The modelhas no pretentions to being completeor the only possible model. Its func-tion is merely to offer guidance tothinking about and analysing con-sumer behaviour in choosing food.The model is given in two stages(figs. I and 2 respectively): first, acomprehensive general model, andsecondly, a more detailed model de-scribing the formation of perceptions

+

commercial. ",-

Information <,non-commercial

Experience

1...-------"---=---,IIIIIIIIL

(a)PerceptionsandBeliefs

(c)Attitudeor Utilityor Preference

Choice

Outcome

Motives/needsPersonalityLife-styleCultureSocial classReference groups

(b)Preference weights,importances ofevaluative criteria

AlternativeI Evaluation

III

//

//

/--_.../

Income;Other anticipatedor unanticipatedcircumstances

22

and preferences with respect to foodproducts.

The subsequent section deals withmethodology. Methods are discussedfor the measurement and analysis ofimportant components of the model:perceptions and preferences. To givean idea of the possible results thatmay be obtained, some applicationsare also briefly described and dis-cussed.

The last section of the paper dis-cusses the potential insights that canbe obtained using the approach pre-sented here and discusses the desira-bility of and the possibilities for ex-tending the analysis.

3. The model

Figure I gives a schematic pictureof the model. When the consumerhas to make a choice between thealternatives in a product class, he hascertain perceptions and beliefs aboutthese alternatives. When evaluatingalternatives he weighs these percep-tions with his preference weights, in-dicating the importance he attachesto the various aspects. This results inan overall utility or an attitude withrespect to the various alternativechoices. In principle, he will choosethe alternative with the greatest utili-ty and this choice leads to a certainoutcome: the satisfaction obtainedfrom the product chosen. In a gene-ral model of consumer decision-making (ENGEL et al. 1978, Chap-ter 2), it is customary to distinguish 5stages in the decision process: Prob-lem recognition (I), Search (2), Al-ternative evaluation (3), Choice (4)and Outcome (5).

Here we assume that stage I hasbeen passed: through triggering me-chanisms such as physiological needs(hunger, thirst) or observation ofproducts in shops, advertisements orelsewhere, the consumer has some-how become involved in a problemof choice. Stage 2 refers to the infor-mation-seeking activity of the con-sumer before he makes his choice.Since most food products are low-cost items, we assume that the con-sumer does not become involved inan extensive information-search pro-cess at the moment of the purchase.(This is not the case when a more ex-

Publications of the AGEV, vol. 2

Page 3: Model and Measurement Methodology for the Analysis of

pensive product, for example an au-tomobile, is being purchased.) Usual-ly the consumer will base his deci-sion on the information about pro-ducts that is currently stored in hismemory. As figure I indicates, thisinformation is acquired from experi-ences with the product in the past,advertisements seen, communicationwith other consumers, etc. For acomprehensive study of consumerinformation-processing with respectto food products, see RUDELL(1979). For an important part, con-sumer choice of food products is ofthe routinized response behaviourtype (HOWARD, 1977). Only occa-sionally, e. g. with the introductionof a new product or brand, or newpackage for an established brand,will the consumer become involvedin some information-acquiring activ-ity at the moment of purchase. Sohere we assume that the product per-ceptions have already been formed.

The treatment in this paper con-centrates on the Alternative Evalua-tion stage (3) (the area within thedotted lines of fig. I). Here we havethree important elements.

(a) Perceptions of products/be-liefs about products

This refers to the way products are"seen" by consumers, e. g. whetheror not a product is considered as nu-tritions, high in fat-contents, tasty,safe, modern, exclusive, status-en-hancing, etc. Since consumers oftenhave imperfect information, percep-tions of products may well differfrom objective reality. (For example,Dutch consumers wrongly perceivebutter as containing more fat thanmargarine.)

As already mentioned, percep-tions are formed by information andexperience. However, as the secondarrow to box (a) in figure I indicates,the formation of perceptions may al-so be affected by psychological fac-tors, such as needs and motives (SI-MON, 1967) and by socio-culturalelements. Certain food products maybe perceived as inferior in one cul-ture, but as a luxury in another cul-ture.

(b) Preference weights indicatingthe importance of the different eval-uative criteria in the determinationof preferences

Erniihrungs-Umschau 30 (1983), Supplement

The evaluative criteria or attri-butes of choice are the dimensions ofthe product that play a role in theformation of preferences. Evaluativecriteria may include: taste, nutrition-al value, price, and ease of prepara-tion. However a certain product isperceived in terms of these attri-butes, a consumer may assess theseattributes differently when forminghis preferences. For example, a con-sumer may attach great importanceto taste and nutritional value, butfind price and ease of preparationless important. This is reflected bydifferent weighting factors in hispreference or utility function. Usual-ly, different persons have differentpreference weights. These weights, asfigure I indicates, are related to mo-tives, needs, personality factors, cul-ture, social class, reference groups(family), etc.

(c) Attitude or utilityThe way the product is perceived

in terms of the various attributes onthe one hand and the preference orutility weights on the other hand to-gether determine the consumer'soverall attitude to or utility for thatproduct. For example, the consum-er's overall attitude to a product orchoice alternatives can be deter-mined with the FISHBEIN-ROSEN-BERG model (ENGEL et aI., 1978,p. 396):

Ab = L w, Bibi =1

where Ab = attitude toward alterna-tive b

Wi = weight or importance ofevaluative criterion i

Bib = belief with respect tothe ability of alternativeb to satisfy evaluativecriterion I, i.e. score ofalternative b in terms ofattribute i

n = number of salient attri-butes

Subsequent stages in the decision pro-cess

The next stage in the choice pro-cess after alternative evaluation isthe Choice (4). Of course, normallyone would expect a consumer to buythe alternatives he prefers the most

(most favourable attitude). However,there may be a discrepancy betweenpreference and choice. For example,he simply may not have enough in-come to purchase the most preferredalternative, he may not have theequipment necessary to prepare theproduct, the shop in which the pur-chase is made may not have the mostpreferred product in stock, etc. Afteran alternative has been chosen, it isconsumed, which results in a certainOutcome (5). The product mayormay not live up to expectations. Theexperience constitutes a feedback-loop in figure I. According to theoutcome of the process, the informa-tion and experience is updated,which may change the perceptions ofthe alternative choices and may ef-fect the preferences and actualchoices at the next purchasing occa-sion.

After this more or less general de-scription of the choice process, Ishall now elaborate on the nature ofperceptions and preferences in thecase of food products. The attributesof a food product that playa role inthe determination of preferences andchoice can be categorized into threeclasses:

(I)

(i) Hedonic attributes

These attributes are related to feel-ings of joy, pleasure and delight inconsuming the product. Hedonic at-tributes refer to aspects such as taste,smell, flavour, i. e. the sensoryaspects of a product. Delicious foodscores very favourable on hedonicattributes.

(ii) Instrumental or functional attri-butes

Apart from the joy of consump-tion, food products have to performcertain very important functions forthe consumer. Food products are theinputs to the physiological processes,produce energy and help to buildand maintain the various cells andorgans of the body. Attributes suchas ingredients, contents of specificnutritional components (e. g. carbo-hydrates, proteins, vitamins), typesof components, presence/absence ofadditives can be classified as instru-mental or functional. Use-relatedaspects of products, e. g. package

23

Page 4: Model and Measurement Methodology for the Analysis of

size, ease of preparation, method ofpreservation, baking quality of flour,spreading quality of butter, can alsobe classified as instrumental attri-butes.

(iii) Expressive or symbolic attributes

The consumption of a food pro-duct may also have symbolic conno-tations, for example, a consumermay express status, exclusiveness,distinction, "savoir vivre", progres-siveness, thrift, sobriety or modestyin consuming specific food items.These symbolic aspects may be im-portant to the consumer himself (try-ing to buy products that are in agree-ment with his "self-concept") or maybe meant to convey something aboutthe consumer to persons in his socialenvironment (conspicuous consump-tion). These symbolic functions of aproduct may be rather removed fromthe physical product itself. For thesesymbolic functions the meaning of aproduct in a specific culture, sub-cul-ture or social class is important; theway consumers talk about the pro-duct and the way the product is com-municated in advertisements andpackage designs. Sometimes also, the

price of the product and the outletswhere the product is available contri-bute to the expressive values of aproduct.

Figure 2 shows how the differenttypes of attributes are related to pro-duct characteristics. For this purposethe product is depicted in three lay-ers.

The hedonic/sensory characteris-tics are a function of the innermostlayer: the physical and chemical pro-perties of the product.

The instrumental/functional attri-butes refer to an extended productconcept: the physical/chemical pro-perties and aspects such as type ofprocessing, package size and typeand other aspects that determine thein-use properties of the product.

The expressive/symbolic attri-butes refer to the most extended pro-duct concept, including brand name,advertising, package design, price,etc. Consumer choice is not only de-termined by the objective product at-tributes as such, but also by the waythese,attributes are perceived by con-sumers. Therefore in figure 2 a "per-ceptual filter" is shown, indicatingthat in the perception process stimuli

Figure 2: Formation of perceptions and preferences with respect to food products

may be lost or stored in the consum-er memory in a biased way. Falseproperties may be attributed to pro-ducts.

The attribute perceptions are, soto speak, the inputs to the process ofpreference formation. Through thepreference or utility function of theconsumer (of which equation (I) is avery simple version) the attributeperceptions are converted into atti-tude or utility ratings. In figure 2,separate preference functions havebeen introduced for sensory attri-butes and for instrumental and ex-pressive attributes.

4. Methodology for measurement andanalysis and examples of application

When starting from the frame-work of figure 2 we wish to explainhow a consumer makes a choicewithin a specific product class, thefollowing questions are relevant:

(i) What are the dimensions or at-tributes: hedonic, instrumental andexpressive that are important in con-sumers' perceptual judgements about

PRODUCT CONSUMER'S CONSUMER'S PREFERENCEPERCEPTIONVV hedonic/

sensory - sensory --- sensoryVV attributes: L-- evaluation preferencetaste, smell, function

cultural meaning of the product colour, texture ~-" VV~package size, type is. V overt::

all.S t>t:: :;; VO/J 1: physical! instrumental! preference.~-0 ~ chemical V functional functions" properties attributes: ingre- L-- I- for ->O/J 0.

~

'" '0 of the dients, nutritional instrumental-'"u" product ./ components, addi- and'"0. -0

" tives, package, expressive pre-0 0.fer-E ::- ease of prepara- attributes

3i t> tion encepreprocessing, built-in services ::;0 :/r.brand, advertising theme -V expressive/sym-

bolic attributes:~V exclusiveness, di- -

stinction, progres-

Vsiveness, sobriety

per!ptualfilter

24 Publications of the AGEV, vol. 2

Page 5: Model and Measurement Methodology for the Analysis of

the product? In other words: whichattributes do consumers use to distin-guish between the alternative choicesin a specific product class.

(ii) How do the various alternativechoices load (score) on these per-ceptual dimensions? How do theconsumers "see" these alternatives?

(iii) How do the consumers weigh(trade-off) the product dimensionsagainst each other when arriving attheir overall preferences? For exam-ple, one alternative may be very tastybut not so good for the health, an-other alternative may have a lesspleasant taste but better health con-notations. How does the consumertrade-off these dimensions? Thisquestions refers to the preference ofutility function. This section dealswith measurement and analyticmethods that can be used to answerthese questions. Let us first considerperceptions: questions (i) and (ii)above.

In principle, a great many attri-butes can playa role in the forma-tion of perceptions. It is not difficultto list 20 attributes that are potential-ly important for the choice of a givenfood product. However, as men-tioned earlier, human capacity forstoring and processing informationis limited and in practice much fewerattributes really playa role. It is im-portant then to be able to determinethe nature of these attributes and theloadings of the different alternativechoices on these attributes. Generallyspeaking, we need methods directedat finding the underlaying factors be-hind a possibly complex responsepattern. This immediately leads todimension-reducing techniques suchas principal component analysis,factor analysis, discriminant analysisand multidimensional scaling. Let ussee how these techniques can be ap-plied -in the context of this percep-tion analysis.

Sensory perception

Here the purpose is to determinethe "sensory profile" of the choicealternatives: the underlying dimen-sions of the sensation with respect toappearance, smell and taste of theproduct. A possible procedure is tostart with a relatively large numberof attributes or sensory parameters

Ernahrunqs-Urnschau 30 (1983), Supplement

and have members of a trained panelrate a number of different formula-tions of the food product on these at-tributes. Afterwards the correlationsbetween these attribute ratings arecomputed and the underlying dimen-sions are found by means of princi-pal component or factor analysis.Here it is important for the test per-sons to react to the physical productas such and for the results to be unaf-fected by brand name, package de-sign, price, etc. Therefore all thesecues should be removed from theproducts to be tested.

Using this procedure, HORS-FIELD and TAYLOR (1976) foundthat there are 3 more or less inde-pendent dimensions on which meator meaty products are judged byconsumers: toughness, succulenceand flavour. Together these threefactors explained 96% of the var-iance in an original set of 11 attri-butes. FRIJTERS (1976) applying asimilar approach, found that the ba-sic sensory dimensions in cookedchicken breast meat are: cohesive-ness, dryness and fatness.

Three aspects of these sensorymeasurements are important. In thefirst place, the test persons should, ifat all possible, really see, smell andtaste the products.

HUGHES (1976), also in a studyabout meat, asked consumers to ratespecific meat cuts, such as lambchops, stewing steak, pork chops ona number of scales, such as tastiness,tenderness, flavour, juiciness andleanness. Here the respondents didnot actually taste products, but hadto base their answers on the informa-tion about the different productsstored in their memories. The prob-lem is then that the respondents mayuse different reference products intheir answers. Some may think offirst quality lamb chops, others ofchops of a very low quality. Also inthis way sensory judgements mayeasily be contaminated with aspectssuch as price, quality perceptions ofoutlets, etc.

A second aspect of this type ofmeasurement is the desirability ofhaving a trained panel of test per-sons. It is important that the panelistattach the same meaning to the dif-ferent points of a specific attribute

scale. For this purpose, training ses-sions are held where the test personsgive ratings to a number of test pro-ducts and discuss the ratings givenamong each other, paying special at-tention to large discrepancies. In thisway it is possible to develop a panelthat is able to give scores to new pro-ducts (not used during the training)with some degree of unanimity.

Finally, care should be taken thatonly perceptual and not evaluativequestions are asked. The respondentsshould say how much or how little ofa certain attribute a specific productcontains, not how he appreciates thatlevel (e. g. how near this is to hisideal point). If perceptual and evalu-ative aspects are mixed, the resultsare difficult to interpret. For exam-ple HUGHES, in the study men-tioned earlier, uses attributes such as"good flavour", "tasty" (evaluative)as well as "juicy" and "tender" (per-ceptual/ descriptive).

After the determination of the ba-sic perceptual dimensions, the pref-erence functions of these dimensions(the sensory evaluation functions)can be estimated. But here we reachthe stage of preference analysis. Nowthe respondents should be a repre-sentative sample of real consumers.It should be known where the prefer-ences of the real consumers lie. Thepreferences of a possibly heavily bi-ased group of trained test personsare not so interesting. HORSFIELDand TAYLOR (1976) followed thisapproach. After they had determinedthe basic sensory dimensions using atrained panel of testers, they recruted390 housewives and asked them foracceptability (or preference) scoresfor products whose sensory profileswere known. This enabled them toestimate sensory evaluation func-tions and obtain indications of pro-mising new products.

Perception with respect to instrumen-tal and expressive attributes

Here again the purpose is to re-duce the large number of potentiallyimportant attributes to a limitednumber of underlying perceptual di-mensions. Now we want to have therespondents' reaction to the com-plete product, including packagetype and design, way of processing,

25

Page 6: Model and Measurement Methodology for the Analysis of

brand, price, etc. Broadly speakingthere are two approaches to findthese underlying perceptual dimen-sions.

The first is to develop a largenumber of attribute scales and havethe respondents (drawn from theconsumer population in question)rate the various product alternativeson these scales. The semantic differ-ential scale (OSGOOD) is very ap-propriate here. From these ratings,the correlation coefficients betweenscales can be computed and thecorrelation matrix in turn can serveas the input to factor or principalcomponent analysis. It is then hopedthat these techniques will produce asmall number of dimensions thatsummarize most of the original rat-ings and which can be readily inter-preted. Discriminant analysis can al-so be used to find the basic dimen-sions that distinguish between thedifferent products. This type of ap-proach, in which the researcher hasto define the various scales before-hand is called compositional: fromthe ratings on the different scales aperceptual picture of the product iscomposed (WIERENGA, 1980 (b),pp. 279-280).

The second approach which iscalled decompositional, takes the op-posite starting point. Here we startwith the total perception of the pro-duct and try to decompose it into itsbasic components. This is the proce-dure followed in multi-dimensionalscaling. An advantage of this ap-proach is that no perceptual attri-butes have to be specified before-hand (with the risk of overlookingimportant attributes). On the otherhand it is not always easy to interpretthe dimensions resulting from mul-tidimensional scaling. Often this in-terpretation can be made easier byusing ratings on specific attributes.So a mixture of a decompositionaland a compositional approach offersthe best prospects. This approachwill be illustrated here with some re-sults about consumers' perceptionsof vegetables in the Netherlands,based on a sample of 150 house-wives.

According to the requirements ofmultidimensional scaling, first simi-lar data were collected. A similarity

26

Figure 3: Perceptual configuration of vegetables; projections in 1-2 and 1-3 plane respectively

ASP

RHUB

MUSH

CUC

dim2

CAR

FRB CAUL

SWPdim I

LET

ON

LET

RHUB

SPIN WHCAB

RCABEND

LE EK

dim3

END

SPIN

FRB

CAR

CUC

ASP

SWP

MUSH ON

Key:ASPCARCAULCUCEND

asparaguscarrotscauliflowercucumberendive

FR BLEEKLETMUSHON

.) In the Netherlands, rhubarb is used as a vegetable.

number for a specific pair of vege-tables indicates how similar thesetwo vegetables are in the eyes of therespondents. In multidimensionalscaling these similarity data are thenused to construct a perceptual confi-guration: a spatial structure of points

dim I

CAUL

LE EK WHCAB

RCAB

French beansleeklettucemushroomonions

RCABRHUBSPINSW PWHC

red cabbagerhubarb'spinachsweet pepperwhite cabbage

(each point denoting a specific vege-table), which reflects the similarityinformation as well as possible. Thismeans that in this configuration, si-milar vegetables will be close to eachother, and vegetables that are notseen as very similar will lie further

Publications of the AGEV, vot. 2

Page 7: Model and Measurement Methodology for the Analysis of

apart from each other. More specificinformatiorl about multidimensionalscaling can be found in GREEN andCARMONE (1970), KRUSKAL andWISH (1978), WIERENGA (1980(a) and (b)). In the vegetable study,15 different vegetables were used. Itturned out that a good solution couldbe obtained in three dimensions. Fi-gure 3 gives the resulting perceptualconfiguration, projected into the 1-2and 1-3 planes respectively. For theinterpretation of the dimensions letus consider the correlations of thecoordinate scores with ratings on at-tribute scales. Dimension I is highlypositively correlated with "real veg-etable" (r = 0.73), "for people withheavy work" (r = 0.80), "carbohy-drates" (r = 0.93) and negativelycorrelated with "for a side-dish" (r =

0.85). So evidently the scores on di-mension I indicate how much (in theeyes of consumers) a vegetable is aproducer of energy. Dimension 2 hasstrong correlations with "expensive"(r = 0.62), "festive meal" (r = 0.67)and "for people with high incomes"(r = 0.62). Therefore we call dimen-sion 2 the distinction dimension. Di-mension 3 has high correlations with"iron" (r = 0.74), "vitamin A" (r =0.76) and "vitamin C" (r = 0.76).This dimension can be called micro-component dimension.

Thus it turns out that the percep-tion of vegetables is composed ofthree major dimensions: the extent towhich a vegetable is a producer ofenergy, the extent to which a vege-table delivers micro-components andthe extent to which a vegetable hasconnotations of distinction. The firsttwo dimensions can be classified asinstrumental; distinction clearly isan expressive dimension. Further re-search showed, that the energy di-mension was related to the extent towhich a vegetable is considered as amain vegetable (not intended toserve only as a salad or garnish). Thedistinction dimension is related tothe day of the week the vegetable isconsumed. Vegetables with a highrating on the distinction dimensionare predominantly eaten at the week-end, especially on Sunday.

It has already been mentionedthat product perceptions may well beat variance with objective reality.

Erniihrungs-Umschau 30 (1983), Supplement

This is demonstrated clearly in thisvegetable research. When we look atthe correlation coefficients betweenperceived and actual contents for anumber of nutritional componentswe find the following results:

ironvitamin Avitamin Ccarbohydratesproteins

0.820.83

-0.Q30.220.70

So for the components vitamin Cand carbohydrates there is virtuallyno relationship between perceivedand actual values. Probably this is aconsequence of the limited capacityof information storage by consum-ers. There is a tendency in human in-formation-processing to "chunk" in-formation, i. e. to simplify the pro-cessing and storing of informationby storing information about similarstimuli in the same chunk, MILLER(1956). This seems to have happenedhere with the micro-components ironand vitamins A and C. For examplethe correlation between perceivedcontents of vitamin A and vitamin Cis 0.88, whereas the correlation be-tween the actual contents of thesecomponents is -0.27. In the same waythe correlation between vitamin Cand iron contents is 0.82 (perceived)and -0.43 (actual). This explains thepoor results for vitamin C mentionedabove. Apparently the informationabout the different micro-compo-nents is stored together and consum-ers simply think that a vegetable witha high rating on one micro-compo-nent is also relatively good on theother ones.

Preference measurement and analysisThe parameters of consumer pref-

erence functions can be estimatedfrom preference judgements for dif-ferent choice alternatives, whose rat-ings on the perceptual dimensionsare known. Sometimes a sensorypreference function is estimated se-parately, as was the case in the re-search done by HORSFIELD andTA YLOR, referred to already. Theyregressed consumer acceptabilityscores on the ratings on the threesensory dimensions: toughness, sue-

culence and flavour. They did notonly try a linear specification (likeequation (I)), but also a quadraticone, which implies an ideal-pointpreference model. (For a taxonomyof preference models, see WIEREN-GA, 1980 (b), pp. 271-276.)

Although their estimated prefer-ence function had a reasonablepredictive power, there are two im-portant limitations in the approachtaken by HORSFIELD and TAY-LOR. In the first place they estimat-ed a preference function at theaggregate level. However, prefer-ences tend to vary between consum-ers and an aggregate preferencefunction has a very restricted mean-ing. A much more realistic approachis to estimate preference parametersat the individual level and then con-sider the distribution of preferenceparameters over the parameter space.(For example the locations of idealpoints or the directions of individualpreference vectors.) There are goodalgorithms available to estimatethese preference parameters for indi-vidual respondents, e. g. PREFMAP(CAROLL, 1972) and LINMAP(SRINIVASAN and SHOCKER,1973).

A second limitation in HORS-FIELD and TAYLOR's approach isthe restriction to sensory attributes.But, as figure 2 indicates, sensorypreference is only one (albeit veryimportant) component of overallpreference. Other product attributes:instrumental and expressive, relatedto package, price, brand, etc. mayhave a dominant influence on con-sumer choice behaviour. Maybe aconsumer will not buy the most tastyproduct, but a product from a trust-ed brand or with attractive package.Ultimately the preference functionshould incorporate all these differentdimensions that affect consumerchoice.

For the vegetable data, preferencefunctions were estimated at the indi-vidual level. These were based onpairwise preference judgments whichthe respondents had given for allpairs of vegetables in the study. Thedimensions included in the prefer-ence functions were the three per-ceptual dimensions discussed above:distinction, energy and micro-com-

27

Page 8: Model and Measurement Methodology for the Analysis of

ponents and a fourth variable: taste.The respondents had indicated foreach vegetable whether or not theyliked the taste of that vegetable. Thisvariable, taste, can be conceived ofas a direct measure of sensory prefer-ence according to figure 2. (We didnot determine the sensory percep-tions of vegetables, neither was asensory evaluation function estimat-ed.) The estimated preference func-tions are rather different for differ-ent respondents. As an illustration,for 3 different respondents the esti-mated preference functions are givenbelow. Here:p = preference valued = score of the vegetable on dis-tinctione = score of the vegetable on energym = score of the vegetable on mic-ro-componentst = score of the vegetable on tasteConsumer I: p = 0.55 d - 0.60 (e -1.18)2 + 0.51 m + 0.33 t

Consumer 2: p = -0.27 (d - 1.64)2 -0.24 (e - 1.13)2 + 0.94 m + 0.90 tConsumer 3: p = -0.17 (d - 1.59)2 -0.60 (e - 0.81)2 + 0.68 m + 0.55 t

We see that consumer I gives aweight of 0.55 to distinction, 0.51 tomicro-components and 0.33 to taste.For energy his preference function isof the ideal-point type. Here prefer-ence increases with increasing energyuntil e = 1.18, but at energy levelsbeyond this point the preference de-creases with further increasing ener-gy scores. Consumer 2 puts a highweight on taste (0.90) and in additionto energy also has an ideal-pointmodel for distinction. (A vegetablecan also be too festive or expensive).This is also true for consumer 3,whose preference function, althoughof the same mathematical form, hasquite different parameter valuescompared with consumer 2. All inall, we were quite succesful in esti-mating these preference functionsfor vegetables. Over all 150 respon-dents the correlation coefficient(root-mean-square) between originalpreference order and preference or-der reproduced by the model was0.82. Once the preference functionsare estimated, these can be used topredict the effects of changes in pro-duct perceptions (i. e. other scores ondistinction, energy, micro-cornpo-

28

nents) on preferences and to predictthe preferences for new products:new combinations of the attributes.

5. Concluding remarks

Looking back at figure 2, whathave we attained so far and whichfurther steps should be taken? Withthe approach described here we cananalyse how consumers perceiveproducts and how these product per-ceptions influence preferences. Thiscan provide important insights thatcan be used for decisions about whatand how to communicate about foodproducts when the purpose is tochange buying and eating habits. Al-so this can give cues about whichcombinations of attributes should bechosen for new products when theobjective is to offer products thatwill meet a high degree of accept-ance from consumers. However, thisanalysis of perceptions and prefer-ences clearly falls short of complete-ly explaining how consumers choosefood products. Interesting relation-ships not discussed here are:

(i) the link between the left part offigure 2 (the products at the differentlevels) and product perceptions.What is the relationship betweenphysical product properties and sen-sory attribute scores? How does thepresence of additives affect percep-tions? How can advertising and nu-trition education communicationchange the image of a product?

(ii) the relationship between socio-psychological variables and productperceptions and preferences (fig. 1);

(iii) the relationship between pref-erence (attitude) and actual choice.

For some of these relationships,interesting models and empirical re-sults can be found in the literature.These elements should be integratedwith the measurement and analysisprocedures discussed so far. Furtherelements to be considered in moreextended models for the explanationof food choice behaviour are thephenomenon of buying combina-tions of products (so far we haveconcentrated on the purchase of oneitem from a product class) and theeffect of group influences (otherfamily members) on consumer be-haviour.

References:[I) BEITMAN, J. R.: An Information Pro-

cessing Theory of Consumer Choice,Reading, Mass., 1979.

[2) CAROLL, J. D.: Individual differencesand multidimensional scaling. In: SHEP-ARD, R. N., A. K. ROMNEY and S. B.NERLOVE (eds.): Multidimensional Scal-ing, Theory and Applications in the Behav-ioral Sciences, Part I: Theory, New York(1972) pp. 108-155.

[3) DICHTER, E.: The Strategy of Desire,New York 1960.

[4) ENGEL, J. F., R. D. BLACKWELL andD. T. KOLLAT: Consumer Behavior, 3rded., Hinsdale, Ill. 1978.

[5) FRIJTERS, J. E. R.: Evaluation of a tex-ture profile for cooked chicken breast meatby principal component analysis. In: Poul-try Science, vol. 55 (1976) No. I, pp.229-234.

[6) GREEN, H. A. J.: Consumer Theory, Pen-guin books, Harmondsworth 197\.

[7) GREEN, P. E. artd F. J. CARMONE:Multidimensional Scaling and RelatedTechniques in Marketing Analysis, Boston1970.

[8] HORSFIELD, S. and L. J. TAYLOR: Ex-ploring the relationship between sensorydata and acceptability of meat. In: Journalof the Science of Food and Agriculture,vol. 27, (1976) pp. 1044-1056.

[9) HOWARD, J. A.: Consumer Behavior:Application ofa Theory, New York 1977.

[10) HOWARD, J. A. and J. N. SHETH: TheTheory of Buyer Behavior, New York1969.

[II) HUGHES, D. R.: Consumer attitudes tomeat cuts - a further study. In: Report 21Department of Agricultural Marketing,University of Newcastle upon Tyne, 1976.

[12] KRUSKAL, J. B. and M. WISH: Multidi-mensional Scaling. In: Sage Publications,Beverly Hills 1978.

[13) MILLER, G. A.: The magical number se-ven, plus or minus two: Some limits on ourcapacity for processing information. In:Psychological Review, vol. 63 (1956) pp.81-97.

(14) PACKARD, Y.: The Hidden Persuaders,Harmondsworth 1957.

[15) RUDELL, F.: Consumer Food Selectionand Nutrition Information, New York1979.

[16) SCHIFFMAN, L. G. and L. L. KANUK:Consumer Behavior, Englewood Cliffs, N.J.,1978.

[17) SIMON, H. A.: Motivational and emo-tional controls of cognition. In: Psycholo-gical Review, vol. 74 (1967) pp. 29-39.

(18) SIMON, H. A.: Models of Thought, NewHaven 1979.

[19) SRINIVASAN, V. and A. D. SHOCKER:Linear programming techniques for mul-tidimensional analysis of preferences. In:Psychometrika, vol. 38 (September 1973)pp.337-369.

[20) WIERENGA, B.: Multidimensional mo-dels for the analysis of consumers' percep-tions and preferences with respect to agri-cultural and food products. In: Journal ofAgricultural Economics, vol. 31 (1980 a)No. I, pp. 83-97.

[21) WIERENGA, B.: Multidimensional pref-erence analysis: a review of models andmeasurement methodology. In: Proceed-ings 33th ESOMAR-congress, Monte Car-lo 1980 (b) pp. 267-289.

[22) ZALTMAN, G. and M. WALLENDORF:Consumer Behavior, Basic Findings andManagement Implications, New York1979.

Publications of the AGEV, vol. 2