CONSUMER RATIONALITY AND ECONOMIC EFFICIENCY: IS THE ASSUMED LINK

Consumer Rationality and Economic Efficiency: . .. .. Balakrishnan, Nataraajan and Desai

CONSUMER RATIONALITY AND ECONOMIC EFFICIENCY:IS THE ASSUMED LINK JUSTIFIED?

P. V. (SUNDAR) BALAKRISHNAN, University o/Washington at BothellRAJAN NA TARAAJAN, Auburn University

ANAND DESAI, Ohio State Universi~

In this paper, the authors investigate a basic assumption underlying most models based on social ormathematical psychology that consumers are rational in the sense of choosing the most economicallyefficient brand. Using Data Envelopment Analysis (DEA) for measuring the efficiency of individualconsumer choice, they report the results of two empirical studies, one involving the social psychology basedtheory of reasoned action and the other involving the mathematical psychology based traditional conjointanalysis, that were conducted to investigate the economic efficiency underlying the individual choicepredictions of the particular model. Results indicate that, in either case, the predictions are not independentof economic efficiency. Further, either model has a significantly higher first choice hit rate for the homoeconomicus group than for the other group. Implications of these results for market segmentation as wellas directions for future research are discussed.

.INTRODUCTION

It is common knowledge that the choice process hascaptured the attention of a diverse group ofresearchers spanning economics, social psychology,mathematical psychology, and of course, marketing.Although there are virtually countless contextscharacterizedby the choice process, choice behaviorenacted within the consumer-marketplace domain isthe most relevant for the practicing marketingmanager. By and large, the choice process in generaland consumer choice in particular have been wellresearched. Nevertheless,to our best knowledge,onecardinal assumption has remained untested in theliterature, and it has to do with the role of economicefficiency in the consumer choice process. Thispaper focuses on this aspect.

The theory of efficiency of consumer choice stemsfrom an axiom within the larger assumption ofrationality in decision making embedded in both

The Marketing Management JournalVolume 10, Issue 1, pages 1-11Copyright «:>2000, The Marketing Management AssociationAll rights of reproduction in any form reserved.

1

social psychology and mathematical psychology.This axiom states that consumers are rational in thesense of choosing the most economically efficientbrand. The primary purpose of this research is toinvestigate this basic assumptionunderlyingvariousmodels of consumer choice. Toward this end, twoseparate studies were conducted which are reportedhere. In both studies, we employed DataEnvelopment Analysis (DEA) for measuring theefficiency (or inefficiency) of individual consumerchoice. In view of this, before we describe thesestudies, it is imperative that we first elucidate theconcept of efficiency of individual consumer choiceand then furnish a brief description of DEA.

EFFICIENCY OF CONSUMER CHOICE

Lancaster (1966; 1971) modified the concept of agood, and consequently the analysis of consumerdemand, by suggestingthat consumershave demandfor attributes of a specific good rather than the gooditself. Ladd and Zober (1977) and Hauser andSimmie (1981) have extended the Lancasterianmodel from characteristics space to consequencemaps, thus providing a useful framework forthinking about consumer demand. In this paper, we

Marketing Management Journal, Spring/Summer 2000

f

Consumer Rationality and Economic Efficiency: . . . . .

adopt a perspective quite similar to that advanced byFomell, Robinson and Wemerfelt (1985) whereinthey liken consumer consumption of products to aproduction process. This framework views theproduct purchased by the consumer as an input whichgets converted to an output by means of thisconsumption process. We advance this input-outputorientation one step before the consumption of theproduct to the actual purchase of a product.Therefore, the product purchase itself, together withthe consumption, can now be viewed as a processinvolving both inputs and outputs. For example, theprice of a brand or a product could be an input.Another input could be the time spent in searchingfor a specific brand. Only after the consumer hasexpended these inputs into the purchase-consumptionprocess can he/she enjoy the associated outputs or thebenefits obtained from the brand.

The associated outputs or the product benefits canrange from such easily quantifiable aspects as "horsepower" and "miles per gallon" to intangibles as"gentle on hands" and "sparkling dishes". Theoutputs can be considered the dimensions usuallyused to represent products in traditionalmultidimensional maps. In such a frameworkinvolving inputs and outputs, the concept ofefficiency or inefficiency of products is a naturalconsequence. An efficient brand then is one which isnot dominated by another brand on all characteristicsor dimensions on which the product category isevaluated by the consumer. This definition, ofcourse, implies that an asymmetrically dominatedbrand (Huber, Payne and Puto 1982), for instance, isan example of an inefficient brand.

A typical assumption of most economic and strategicmodels is that all products in the market place areportrayed as efficient. However, the study byHjorth-Anderson (1984) indicates that at least at themicro level there do exist inefficient variants of acommodity. That is to say, it does not seemreasonable to assume that all of the products evokedby an individual consumer would lie on the efficientfrontier. While standard economic theory woulddictate that all inefficientbrands will be driven out of

the market place, this may not happen in the shortfUll. So at least in markets which are in a state offlux, perhaps due to factors such as new products


Balakrishnan, Nataraajan and Desai

being introduced or current products undergoingmodifications, and so forth, customers may evokebrands which do not lie on the efficient frontier.

From the supply side, consistent with the aboveobservation, Huber, Payne and Puto (1982) havestated that under certain conditions it may beactually to the benefit of the firm to introduce aninefficient brand (i.e., decoy). They argue that dueto the salience of an inefficientvariant on consumer

choice strategies, the decoy may actually provebeneficial by boosting the sales of its more efficientcounterpart. So, even if standard economic theorycalls for the elimination of inefficient products,sophisticated marketing strategy might actually callfor their preservation. Therefore, a consumer willquite likely encounter less than efficient brands inthe marketplace.

From the demand side perspective, however, arational consumer would always like to ensure thathe/she is making the "best" possible buy. After all,consumer rationality implies the notion of economicefficiency. In other words, of all the available orevoked brands, he/she will choose that whichprovides the highest utility per dollar (Hauser andGaskin 1984). Needless to say, in order to be ableto select the economically efficient brand, therational consumer needs to be aware of the differentproduct offerings (decoys and all) in themarketplace, their performance characteristics onsalient attributes as well as their related expenseswhich may include, among others, the amount ofeffort required to adapt and assimilate the product tohis/her individual needs. As stated earlier, this basicassumption of consumer rationality in the sense ofchoosing the most economically efficient brand nodoubt underlies most consumer models. In thispaper, we empirically investigate the validity of thiscardinal assumption.

DATA ENVELOPMENT ANALYSIS

Data Envelopment Analysis (DEA) has becomepopular since 1978 and interest in its potential forapplicability does not seem to have abated. This isevidenced by the spate of articles on DEA that haveappeared in a variety of outlets even recently (e.g.,Seiford and Zhu 1998; Soteriou and Zenios 1999).

2

--


For the context of this research, DEA has beenproposed as a technique for implementing theLancaster model (Chames et al. 1985). However,most applications of DEA have been in contextswhere the notion of a production technology isimmediately applicable and the unit of analysis, orthe Decision Making Unit (DMU), is generallya firmor some stylized form of production process (Seiford1990). The analogous DMU in the marketing contextwould be an aggregation of attributes in the form ofa brand or product. Such analyses which use a brandor a product as a DMU are useful in identifyingsuperior products and economically efficient brands.

We present here a simple graphical introduction toDEA. For a more formal mathematical treatment,werefer the reader to Banker et al. (1989). Consider anindividual comparing a number of cars on the basis oftwo criteria, engine condition and exterior condition.Further suppose that these criteria are measured on afive point scale at half point intervals. Then theresulting data for these cars can be represented by thescatter plot in Figure 1. The vehicles rated the bestare those which are farthest away from the origin.This set of undominated cars form what is termed tobe the efficient frontier and all the cars on thisfrontier are given efficiency scores of unity. Carswhich fall between the frontier and the origin aredeemed inefficient and their efficiency rating ismeasured in terms of their distance from the efficientfrontier. Hence, car models A,B,C and D are deemedefficient with an efficiency rating of 1.0. Theefficiency of the cars not on the frontier are measuredin distances along the ray from the origin through thepoint representing the car. For instance, theefficiency of the car at E is the ratio OE/OE'. Notethat this ratio is always less than or equal to one.

Extensions to multiple attribute maps follownaturally. One can also develop "per dollar" maps bytaking the attribute rating and dividing that by theprice the car to obtain a per dollar rating on a givenattribute. The per dollar map and the resultingfrontier is analogous to the production function fromeconomic theory where the input resource ismeasured in terms of dollars and the outputs are theattributes of the car. Thus, Figure 1 would representdata shown in the output space. Efficiency scores formultiple attributes and multiple input resources (for

3


Figure 1Hypothetical Data on Car Attributes

5

4

c0

~ 3c0

()

CD.52C)cW

.1

0

1 3 52 4

Exterior Condition

instance, search effort, cost of other information),can be readily obtained by repeatedly solving, foreach car, a mathematical program.

In order to present the linear programming modelunderlying the measurement of efficiency, we firstintroduce some notation. Let Yijdenote the benefitaccruing along the ilbdimension (attribute) of the jibbrand. Let xkj denote the input required to beexpended along the klbdimension of the jibbrand inorder to obtain the associated benefits. Assumingthat there are M brands with N outputs and K inputattributes which characterize the product evaluationprocess, we denote by Y the matrix of outputattributes, of order N by M. Similarly, we let X bethe matrix of input attributes, of order K by M.

Assume that in attribute-space, these configurationscan be represented by combinations of these x andy vectors, then the set of all feasible configurations,such that a proportionality is maintained between xand y is

C={(x,y):y~Y.z, X.z ~ x, ZER.M} (1)

where, z = (Zl ,~ "",ZM) denotes the vector ofparameters which determine the combination of theattributes of the brands. Let (XO,l) be a givenobservation of attributes of some brand. Then therelative efficiency (RE) of that brand is:


r-


RE(xO ,l) = max{1I8: (XO,8yO)E C}

which is the optimal solution of the following linearprogram:

max 8

subject to:zY ~ 8yOzX ~ XOz~O

Thus, in this instance, the relative efficiencymeasure, 118,denotes the proportional expansion inthe output attributes that can be achieved whilemaintaining the current level of inputs (costs). Whilea variety of modifications and variations of this basicefficiency measure are possible, we restrict ourselvesto this simple measure to determine how it fares ininvestigating individual choice.

It is vital to mention that our point of departure fromthe standard Lancasterian approach is that themapping of products is not done in the characteristicsspace. Instead the products are mapped, as in Hauserand Gaskin (1984) and Hauser and Simmie (1981), inthe perceptual space. The reasons are that consumersrarely have the objective information on the variouscharacteristics of products and in many instances,such objective performance measures are hard toquantify. Instead, in keeping with the currentmarketing focus, we measure the consumers' utilityin terms of the satisfaction accruing to them for thevarious attribute levels. This is more consistent andin line with the marketing concept of maximizingcustomer satisfaction. This approach also has theadded desirability of allowing for heterogeneity inconsumer perceptions. That is, different levels of anattribute would provide differing perceived levels ofsatisfaction to different customers.

Consequently, employing the above linear program,we are now able to scale the product space in termsof utility of satisfaction with the variouscharacteristics on a per input basis. Unliketraditional approaches (Hauser and Shugan 1983),however, we estimate a satisfactionper dollar map atthe individual level as we do not assumehomogeneity in perceptions. Nor does this approach



(2) require the assumption that all alternative brands inthe market place are efficient. This thereforeenables us to determine the fraction of individualconsumerswho make economicallyefficientchoicesfrom a given set of alternatives.

STUDY 1: EFFICIENCY OF CONSUMERCHOICE IN SOCIAL PSYCHOLOGY

To our best knowledge, no one has yet investigatedthe choice predictions of social psychology basedformulations in terms of the efficiency of individualconsumer choice. In this section, we report theresults of an empirical investigation in the context ofefficiency of individual consumer choice employinga widely used attitude formulation, the theory ofreasoned action (Ajzen and Fishbein 1980). Thetheory of reasoned action (TRA) is possibly themost well known among the attitude formulations insocial psychology, and hence it was chosen as thepredictive model in this study. It has been welldocumented many times in the literature, and onemay refer to the work of Ajzen and Fishbein (1980)for a comprehensive review of TRA. A briefdescription of TRA is provided here.

The Theory of Reasoned Action

TRA views intention as the immediate antecedent ofbehavior. Intention is determined by an attitudinal(personal) component and a normative (social)component. The model can be symbolically statedas

B - BI = w(.(AB)+ wz.(SN) where,

B is overt behavior, BI is behavioral intention(subjective probability of intending to performbehavior B), AB is attitude toward performingbehavior B (e.g., attitude toward buying a brand),SN is subjective norm (normative influence; thecollective perceived influence of "importantothers"), and WIand Wzare empirically determinedweights denoting the relative influence of the twocomponents. ABis determined as Lbiej(i=1...n)where,bi is the subjective probability that performing thebehavior will result in outcome I, ei is theindividual's evaluation of outcome I, andn is the

4


number of salient outcomes. SN is determined as

LNBjMCj G=L.N)where, NBj is the belief that referentj thinks the individual should/should not perform thebehavior, MCj is the individual's motivation tocomply with referent j, and N is the number of salientreferents.

It is pointed out that the extent of agreementbetweenBI and B depends upon the degree of measurementcorrespondence on the elements of action (refers tothe behavior of interest), target (theobject/issue/person figuring in the behavior ofinterest), context (the context in which the behaviorof interest is enacted), and time (i.e., the time atwhich the behavior of interest occurs). Thehypothesis underlying the application of TRA to aconsumer choice situation is that a consumer has acertain level of BI (the predictor of behavior) foreach of the available alternatives, and would end upselecting that alternative for which he/she has thehighest level ofB!.

TRA is a person based, compensatory andcompositional technique, and relies on direct inputfrom the consumer. Further, it has a normative(influence of important others) construct through thesubjective norm construct. Although it assumeshuman rationality to prevail in decision making,constraints are not endogenous to the model.However, it implicitly assumes that the individualtakes contextual constraints into consideration whileresponding to measurement items. For more detailson these characteristics, seeNataraajan and Warshaw(1991).

The Study

In this study, we investigate the economic rationalityunderlying the predictions of TRA. After all, TRAassumes rationality in behavior; the qualification"reasoned" substantiates this. It appears to us that ifin fact TRA is based on rationality, it should then atleast imply the notion of economic efficiency.Following from this, is the primary question, "DoesTRA predict choices that are economicallyefficient"? Specific questions of interest are:

5


1. What is the proportion of economically efficient(the homo economicus group) consumers amongthose whose first choices are correctly predictedby TRA ?

2. What is the proportion of economicallyinefficient consumers among those whose firstchoices are correctly predicted by TRA?

3. What is the proportion of economicallyefficientconsumers(the homo economicus group) amongthose whose first choices are not correctlypredicted by TRA?

4. What is the proportion of economicallyinefficient consumers among those whose firstchoices are not correctly predicted by TRA?

For the sake of completeness, the average number ofefficient options available for the homo economicusand for the other group (comprising those that makeinefficient choices) would also have to bedetermined to see if there was any unreasonabledisparity.

Method

The predictive ability of TRA was tested in ahypothetical setting involving forced choice amonga set of ten real apartments in a southern universitytown. Exhibit 1 depicts the attributes and levels onwhich these ten apartmentsdiffered from each other.Undergraduatebusiness students enrolled in severalmarketing courses at a large southeastern universitywere recruited as subjects for the study. Althoughparticipation was voluntary, the subjects receivedpoints as part of their grade for "class participation."In fairness to the non-participants, those who did notwish to be subjects were given another assignmentfor the same number of points as part of their "classparticipation." In all, 117 subjects comprised thetest sample.

The Prediction Phase

Subjects were provided with profile descriptions ofthe ten apartments along with a questionnaire. Acover letter informed the subjects that they were toassume that they needed an apartment for their own


,.#

-

Consumer Rationality and Economic Efficiency: ... .. Balakrishnan, Nataraajan and Desai

Exhibit I

Apartment Attributes and Levels in Study I

A. Monthly Rent1. $1502. $2503. $3504. $4505. $5506. $650

C. Size

1. One room plus kitchen2. Two bedrooms plus kitchen3. Two bedrooms, living room plus kitchen4. Three bedrooms, living room plus kitchen

E. Recreational Facilities

1. Tennis courts, sauna, club house & pool2. Tennis courts, sauna & pool3. Tennis courts & pool4. Only pool5. None

B. Distance from Campus1. Less than one mile2. Between 1and 2 miles3. Between 2 and 3 miles4. Between 3 and 4 miles

D. Renovation

1. Throughout2. Only kitchen and bathroom3. Only bathroom4. None

F. Washer & Dryer1. No washer & dryer2. On-location (not free)3. On-location (free)4. In-house

Used Car Attributes and Levels in Stud.YlI

A. Year of Make1. 19862. 19873. 19884. 1989

C. Odometer Mileage1. Less than 30,000 miles2. More than 30,000 but less than 50,000 miles3. More than 50,000 but less than 80,000 miles

E. Interior Condition1. A few stains and tears2. A few stains but no tears3. Virtually no stains or tears

use, and were to choose to rent one of the tenapartments. It also stated that while the apartmentswere all livable, unfurnished, located in generallyclean complexes and allowed pets, they all varied onother important attributes as shown in Exhibit 1.The questionnaireconsisted of two parts, one part forTRA items and the other for DEA items. To avoid


B. Price1. $7,000-$7,9992. $8,000-$8,9993. $9,000-$9,9994. $10,000-$10,9995. $11,000-$11,9996. $12,000-$12,9997. $15,000-$15,999

D. Exterior Condition1. Slight rust and few dents2. Slight rust, but no dents3. Virtually no rust or dents

order effects/position bias, there were two versionsof this questionnaire. In one version, the TRA partappeared first, and in the other version the DEA partappeared first.

For TRA, the items were ABy (attitude towardbehavior --purchasing a particular apartment -- for

6

Consumer Rationality and Economic Efficiency: ... ..

each subject and for each apartment; y=apartment)and SNy(subjectivenorm -- influencing the purchaseof a particular apartment -- for each subject and foreach apartment). These were patterned after thestandard formats furnished by Ajzen and Fishbein(1980) and measured directly. Since prediction andnot explanation was the main interest, the belief andevaluation stage of TRA (the Ebjej and ENBjMCjterms) was deemed quite unnecessary, and

measurement took place only in the subsequentstages. This is in line with what Ajzen and Fishbein(1980) have suggested, and others (e.g., Warshaw1980) have followed.

ABywas computed as a summated score of foursemantic differential scales (coded from +3 to -3 orreversed depending on the polarity) with thefollowing bipolar adjectives: 1. Pleasurable-Painful2. Wise-Foolish 3. Unpleasant- Pleasant 4.Punishing-Rewarding. Similarly, SNywas measuredusing a scale ranging from extremely likely toextremely unlikely (coded from +3 to -3).

For efficiency,evaluations through DEA, satisfactionscores were obtained. The subjects were asked toindicate their satisfaction on a nine point scaleranging from unsatisfactory to satisfactory with eachof the levels of the various attributes as shown inExhibit 1.

The Actual Choice Phase

After the subjects had completed the abovequestionnaire, they were given another questionnairethat required them to rate each of the ten apartmentson a 0 to 100 likelihood of renting scale and alsoindicate their most preferred apartment if they had topick one.

Analysis

The satisfaction data enabled the computation ofefficiency scores for the subjects. The efficiencyscores for each of the ten apartments used in theholdout were estimated at the individual level. Forthis, the satisfaction with the different levels ofapartment characteristics was scaled per dollar rent(input) of an apartment to obtain the efficiencyscores. Then, for each individual, we examined

7

- -- --


whether his/her choice was from the set of

apartments on his/her computed Pareto-frontier.This led to the determination of the fraction ofsubjects who chose from those apartments estimatedto be on their efficiency frontiers.

Next, first choice hit rates for TRA were

determined. For this, the dependent variable, BIy,was computed as B~ = Wly.Aay+ w2y.SNy,where W1yand W2ywere the apartment specific beta weights.As suggested by Ajzen and Fishbein (1980), theseweights were estimated from a separate and smaller

sample of subjects. Thus, for each subject, 10 BIyscores were computed. In each case, the apartmentwith the highest score was deemed the first choiceof a subject as predicted by TRA. The predictedfirst choices were compared to the actual firstchoices, and hit rates were computed. Whether ornot TRA predictions were independent of economicefficiency was tested using the Chi square test. Thesignificance of the difference in the hit rates for thehomo economicus group and the group that maderelatively inefficient choices was conducted using aZ test for proportions.

Results

Table la gives the break down on efficiency and hitrates. A significant Chi square (5.56, Idf, p<0.02)established the dependence between TRApredictions and the notion of economic efficiency.Overall, 58 of the 117subjects (49.57percent) madeeconomically efficient choices. Also, TRA had anoverall hit rate of 58.17 percent. For the homoeconomicus group, the hit rate was 68.97 percent(40 out of a maximum possible of 58), and for theother group, it was 47.46percent (28 out of 59).The differencebetween these hit rates is statisticallysignificant (Z = 2.358, p=O.O18) at the 0.02 level.Table 1b furnishes the further break down on theaverage number of efficient available options foreach group.

Table laTRA: First Choice Hit Rates

EfficiencyEfficient Inefficient

Correct

Incorrect

40

1858

28

3159

68

49117


Consumer Rationality and Economic Efficiency: . . . . .Table Ib

TRA: Average # of Efficient Options AvailableEfficiency

Efficient Inefficient

Correct

Incorrect

5.875

7.167

5.607

4.970

STUDY 2: EFFICIENCY OF CONSUMERCHOICE IN MATHEMATICAL

PSYCHOLOGY

As mentioned in the introduction to this paper, to thebest knowledge of the authors, no one has yetinvestigated the choice predictions of mathematicalpsychology based formulations in terms of theefficiency of individual consumer choice. In thissection, we report the results of an empiricalinvestigation in the context of efficiency ofindividual consumer choice employing traditionalconjoint analysis, possibly the most widely used toolfor measuring consumers' preferences amongmultiattribute options (Green 1984). Traditionalconjoint analysis is well documented in the literature,and one may refer to the work of Green and Wind(1975) for a brief review.

The Study

In terms of administration, this study was conductedsimilarly to Study 1 except for three differences.First, a traditional conjoint analysis task replacedTRA. Second, the product category employed wasused cars (see Exhibit 1 for the attributes and theirlevels). Third, data were collected fromundergraduate business students enrolled in severalmarketing courses at a large mid-western university.A fractional factorial design yielded a reducedstimulus set of 32 used car profiles, which were thenprovided to the subjects for the estimation of the partworths. Twelve realistic used car profiles culledfrom the classifieds were employed as part of thevalidation study, and the subjects were asked toindicate their most preferred profile given that theywere in the market to purchase a used car.

Akin to Study 1, the individual level efficiencyscores for each of the twelve used cars used in the



holdout were computed, and the fraction of subjectswho chose from those used cars estimated to be ontheir efficiency frontiers, was determined. Next, foreach subject, the used car with the highestcalculated total utility score was deemed theconjoint's predicted first choice. These werecompared with the subjects' actual first choice, andhit rates were computed. As in Study 1, thesignificance of the association between the conjointpredictions from the notion of economic efficiencyas well as the statistical significance of thedifference in the hit rates for the homo economicusgroup and the group that made relatively inefficientchoices were tested.

A total of 106subjects participated in the study. Ofthese, we had 102 usable responses as four of thesubjects did not indicate their first choice from anyof the available options. Table 2a gives the breakdown on efficiency and hit rates. Here again, asignificant Chi square (6.08, Idf, p<0.02)established the dependence between the conjointpredictions and the notion of economic efficiency.Overall, 65 of the 102 subjects (63.7 percent) madeeconomically efficient decisions. The total numberof correct conjoint predictions was 41 (out of amaximum of possible 102) for a predictive accuracyof 40.2 percent. The homo economicus group had ahit rate of 49 percent (32 out of a maximum possible65). The other group had a hit rate of 24.3 percent(9 out of a possible 37). The difference in these hitrates is statistically significant (Z=2.467, p=0.0136)at the 0.02 level. Table 2b furnishes the break downon the average number of efficient brands for theconsumers in the four cells.

Table 2aConjoint: First Choice Hit Rates


Correct

Incorrect

9

28

37

41

61

102

32

33

65

8

--


Table 2b

Conjoint: Average # of Efficient Options Available


Correct

Incorrect

4.780

5.760

4.110

4.500

DISCUSSION AND CONCLUSIONS

The empirical studies do seem to indicate, perhapsnot surprisingly, that not all consumers makeeconomically efficient choices. It is, however,surprising that such a significantly large fraction ofthe consumers do seem to choose brands primarilyfrom those that lie on the Pareto-frontier.

Consequently, the recommendation of models basedon the axiom of individual consumer rationality maybe more generalizable than previously conjectured.

A more interesting result arising from theexperimental investigation is the potential foridentifying consumers who make economicallyefficient decisions. A segmentationof the consumersalong such lines seems to hold promise especially inthe arena of optimal product design. It is of someinterest to note that even when the subjects, as in thiscase, are walked through the process of estimatingthe most economical purchase (without beingactually told how), we find that a large proportion ofthe subjects across different populations and productcategories are unable to make such estimations. This,backed up by the evidence of an older study byFriedman (1972) on unit pricing, seems to suggestthat perhaps the only way to avoid such largeconsumer welfare losses may be through expandedpolicy initiatives such as public education programs.

Additionally, the consumer welfare losses may belarger than their pure numerical estimate of thepopulation. That is to say, these losses may befurther compounded by the magnitude for the"inefficient" consumers. To put it differently, asboth the social psychological (TRA) andmathematical psychology (conjoint) models predictconsumer choices better for the efficient decisionmakers, it is only natural that products and serviceswill continue to be better designed and targeted to

9


this group. However, it may now be possiblethrough the use of DEA based models to identifyconsumers who may be unable to engage in therequisite cognitiveprocessing of information. Thisin turn may lead to the better targeting of publicpolicy programs. In any case, it is clear that theconcept of efficiency (or inefficiency) of consumerchoice deserves further investigation, especiallygiven the result that predictive validity of both thetheory of reasoned action and traditional conjointanalysis for the individually rational consumersegment is significantly higher than that forconsumers who make relatively inefficient choices.

Efficacy of DEA: An Assessment

DEA can aid decision makers by summarizing thewealth of information about the inputs expendedand the outputs obtained for each brand in a productcategory. It allows us to represent concisely, for aconsumer, the relative efficiency of the differentoptions. Thus, consumers, having identified thesalient attributes, can then easily determine how thedifferent brands stack up relative to one another. Inaddition, DEA does not assume, unlike moststrategic models (Hauser and Shugan 1983), that allchoice alternatives for a consumer are efficient.Further, it does not require homogeneous consumerperceptions as each consumer could have differentsalient attributes and differing performanceperceptions for the same brand. This is especiallyimportant when one considers that DEA isemployed at the individual level and its output cantherefore be used as benchmarks or basis for furtherdiscussions. Importantly,DEA is adaptive in naturein that it allows for a combination of choicestrategies on the part of the purchaser.

Furthermore, DEA is able to explicitly take intoaccount that consumers might, and in fact do,consider a number of different aspects along whichthey may have to "invest" in order to obtain themultiple benefits a product has to offer. Forinstance, the advertisement campaign by AppleComputers (before or after their resurgence!),whichexplicitly suggests that their products are superior tothose of their competitors is a case in point. In thiscampaign, the case for the superiority of the productis made by emphasizing that Apple computers are


""

-~ ~--


more efficient since the users require much lesstraining before the products can be fruitfully used. Inother words, the lower training and support costs ofthis particular product translates into higher overallefficiency (even with performance on all otherattributes being the same). However, this aspect ofDEA, which is both a strength as well as a weakness,is that the axes, unlike, say, in multidimensionalscaling, are explicitly defined. In any case, theresults presented here, although preliminary, stronglysuggest that DEA certainly has the potential for useas a tool for operationalizingthe Lancasterian modelat the individual level.

Future Directions

A number of avenues for future research are nowopen, and deserve attention. First, various issuesrelating to data collection need to be examined. Forinstance, the approach taken here requires subjects toindicate their utility of satisfaction with all levels ofeach attribute. Thus, this approach requires as manyquestions as the total number of attribute levelsemployed in the study. Additional research toinvestigate the possibility for reducing data collectionfrom subjects needs to be undertaken.

Another critical issue, namely that of multipleefficient points, deserves to be addressed. Note thatthe efficiency score obtained for E (Figure 1) is withrespect to the frontier defined by the cars A and B.Other inefficient cars would be compared to otherfacets of the frontier. These different facets

essentially represent different trade-off rates orweights for the attributes. Thus, informationon theirrelative weights for the attributes can be used todetermine which facet of the frontier is the mostdesirable and thereby reduce the size of the choiceset. This means that it could be possible in the futureto predict the alternative model that would be mostpreferred. In other words, extending this approachhas the added potential for identifying the brand thatwill be chosen by each consumer.



REFERENCES

Ajzen, leek and Martin Fishbein (1980),UnderstandingAttitudes and Predicting SocialBehavior, Englewood Cliffs, NJ: Prentice-Hall.

Banker, Rajeev D., Abraham Charnes., William W.Cooper, John Swarts, and D.A. Thomas (1989),"An Introduction to Data EnvelopmentAnalysiswith Some Models and Their Uses," Researchin Governmental Accounting, 5, 125-163.

Charnes, Abraham,William W. Cooper, andEdwardo Rhodes (1978),"Measuring Efficiencyof Decision Making Units," European Journalof Operational Research, 3(4),429-444.

Charnes, Abraham, William W. Cooper, David B.Leamer and Fred Y. Phillips (1985),"Management Science and MarketingManagement," Journal of Marketing, 49(Spring),93-105.

Charnes, Abraham, William W. Cooper, and RobertM. Thrall (1986), "Classifying andCharacterizing Efficienciesand InefficienciesinData Envelopement Analysis," OperationsResearch Letters, 5(3), 105-110.

Fornell, Claes, William T. Robinson, and BirgerWernerfelt (1985), "Consumption Experienceand Sales Promotion Expenditure,"Management Science, 31(9) (September), 1084-1105.

Friedman, M. (1972), "Consumer PriceComparisons of Retail Product: The Role ofPackaging and Pricing Practices and theImplications for Consumer Legislation,"Journal of Applied Psychology, 56, 439-446.

Green, Paul E. (1984), "Hybrid Models for ConjointAnalysis: An Expository Review," Journal ofMarketing Research, 21 (May), 155-169.

Green, Paul E. and Yoram Wind (1975), "New Wayto Measure Consumer's Judgements," HarvardBusiness Review, 53,107-117.

Hauser, John R. and Steven P. Gaskin (1984),"Application of the Defender ConsumerModel," Marketing Science, 3 (Fall), 327-351.

Hauser, John R. and Steven Shugan (1983),"Defensive Marketing Strategies," MarketingScience, 2 (Fall), 319-360.

10

Consumer Rationality and Economic Efficiency: . . . . . Balakrishnan, Nataraajan and Desai

Hauser, John R. and Patricia Simmie (1981), "Profit-Maximizing Perceptual Positions: An IntegratedTheory for the Selection of Product Features andPrice," Management Science, 27 (January) 33-56.

Hjorth-Anderson, Chr. (1984), "The Concept ofQuality and the Efficiency of Markets forConsumer Products," Journal of ConsumerResearch, 11 (September), 708-718.

Huber, Joel, John W. Payne, and Christopher Puto(1982), "Adding Asymmetrically Dominatedalternatives: Violation of Regularity andSimilarity Hypothesis, " Journal of ConsumerResearch, 9, 1 (June), 90-98.

Lancaster, Kelvin J. (1966), "A New Approach toConsumer Theory," Journal of PoliticalEconomy, 74, 132-157.

Lancaster, Kelvin J. (1971), Consumer Demand,New York: Columbia University Press.

Ladd, George W. and Martin Zober (1977), "Modelof Consumer Reaction to ProductCharacteristics," Journal of ConsumerResearch, 4,89-101.

Nataraajan, Rajan and Paul R. Warshaw (1991), "Arethe Models Compatible for EmpiricalComparison? An Illustration with An IntentionsModel, An Expectations Model, and TraditionalConjoint Analysis," in John Sherry and BrianSternthal (eds.), Advances in ConsumerResearch, (Vol. 19, p. 472-481). Provo, UT:Association for Consumer Research.

Seiford, Lawrence H. (1990), A Bibliography ofData Envelopment Analysis, Amherst, MA:University of Massachusetts.

Seiford, Lawrence H. and Joe Zhu (1998), "StabilityRegions for Maintaining Efficiency in DataEnvelopment Analysis," European Journal ofOperational Research, 108(1), 127-139.

Soteriou, Andreas C. and Stavros A. Zenios (1999),"Using Data Envelopment Analysis for CostingBank Products," European Journal ofOperational Research, 114(2),234-248.

Warshaw, Paul R. (1980), "Predicting Purchase andOther Behaviors from General and ContextuallySpecific Intentions," Journal of MarketingResearch, 17,26-33.

11 Marketing Management Journal, Spring/Summer 2000

Documents

CONSUMER RATIONALITY AND ECONOMIC EFFICIENCY: IS THE ASSUMED LINK