Thirty Years of Conjoint Analysis: Reflectionsand Prospects

PAUL E. GREEN Suite 1400, Steinberg Halt-Dietrich Hallgreene@wharton.upenn.edu Tlie Wharton School

University of PennsylvaniaPhiladelphia, Pennsylvania 19104--6371

ABBA M . KRIEGER Suite 3000, Steinberg Hall-Dietrich Hallkrieger@wharton.upenn.edu The Wharton School

YORAM (JERRY) W I N D Suite UOO, Steinberg Hall-Dietrich Hallwindj@wharton.upenn.edu The Wharton School

Conjoint analysis is marketers' favorite methodology for find-ing out how buyers make trade-offs among competing prod-ucts and suppliers. Conjoint analysts develop and present de-scriptions of alternative products or services that are preparedfrom fractional factorial, experimental designs. They use vari-ous models to infer buyers' part-worths for attribute levels,and enter the part-worths into buyer-choice simulators to pre-dict how buyers will choose among products and services.Easy-to-use software has been important for applying thesemodels. Thousands of applications of conjoint analysis havebeen carried out over the past three decades.

Readers of Interfaces are no strangers methods extensively to important prob-

to multiattribute utility models lems in management and government de-[Keeney and Raiffa 1976; Starr and Zeleny cision making.1977; Bell, Raiffa, and Tversky 1988]. OR/MS researchers may be less familiarThomas Saaty [1980] introduced a differ- with another method—conjoint analysis—ent approach to multiattribute utility mea- that has been applied for over 30 years,surement: the analytic hierarchy process primarily by researchers in marketing and[AHP]. Both approaches emphasized small business. Conjoint analysis evolved fromnumbers of decision makers facing high- the seminal research of Luce and Tukeylevel decisions. Operations researchers and [1964]. Their theoretical contributions weremanagement scientists have applied both put to use by a number of psychometri-Ciipyrighl €• 2001 INFORMS MARKETING—CHOICE MODELS(XW2-2102/01 /31l)3/S056/Sn5.00 MARKEriNG-NEW PRODUCTS1526~551X electnmic ISSNThis paper was refereed.

INTERFACES 31: 3, Part 2 of 2, May-June 2001 (pp. S56-S73)

CONJOINT ANALYSIS

cians, including Carroll [1969], Kruskal119651, and Young [1969]. These research-ers developed a variety of nonmetric niod-els for computing part-worths (attribute-level values) from respondents' preferenceorderings across multiattributed stimuli,such as descriptions of products orservices.

Conjoint analysis is, by far, the mostused marketing research method for ana-lyzing consumer trade-offs. Surveys con-ducted by Wittink and Cattin [1989] andWittink, Vriens, and Burhenne 11994] at-test to its woddwide popularity.

It is not difficult to see why researchersdeveloped and applied conjoint analysisso rapidly. Conjoint analysis deals with acentral management question: Why doconsumers choose one brand or one sup-plier over another? Also, marketing re-search practitioners want to be part ofsomething new, and computer softwarefor implementing the methodology be-came readily available.The Basic Ideas of Conjoint Analysis

Conjoint analysis is one of many tech-niques for handling situations in which adecision maker has to deal with optionsthat simultaneously vary across two ormore attributes. The problem the decisionmaker faces is how to trade off the possi-bility that option X is better than option Yon attribute A while Y is better than X onattribute B, and various extensions ofthese conflicts.

Conjoint analysis concerns the day-to-day decisions of consumers—what brandof toothpaste, automobile, or photocopy-ing machine to buy [or lease]? Marketingresearchers may collect trade-off informa-tion for hundreds or even thousands of re-

spondents. Data collection and processingtechniques must be fairly simple and rou-tinized to handle problems of this scope.

Conjoint analysis is a technique for mea-suring trade-offs for analyzing survey re-sponses concerning preferences and inten-tions to buy, and it is a method forsimulating how consumers might react tochanges in current products or to newproducts introduced into an existing com-petitive array. Researchers have used con-joint analysis for consumer and industrialproducts and services and for not-for-profit offerings.

To understand the basic concepts ofconjoint analysis, assume that a marketer

Why do consumers chooseone brand or one supplierover another?

of credit cards wishes to examine the pos-sibility of modifying its current line of ser-vices. One of the first steps in designing aconjoint study is to develop a set of attri-butes and corresponding attribute levels tocharacterize the competitive domain. Fo-cus groups, in-depth consumer interview's,and internal corporate expertise are someof the sources researchers use to structurethe sets of attributes and levels that guidethe rest of the study.

In an actual study of credit-card sup-pliers, researchers used a set of 12 attri-butes with two to six levels, for a total of35 levels (Table 1). However, the totalnumber of possible combinations of levelsis 186,624. Conjoint analysts make exten-sive use of orthogonal arrays [Addelman1962] and other types of fractional factorialdesigns to reduce the number of stimulus

May-June 2001 S57

GREEN, KRIEGER, WIND

Annual price ($)0 10 20 50 80 100

Cash rebate (end-of-year, on total purchases)None {% 1%

800 number for message forwardingNone 9-5 weekdays 24 hours per day

Retail purchase insuranceNone 90 days' coverage

Common carrier insurance (death, injury)None $50,000 $200,000

Rental car insurance (fire, theft, collision,vandalism)

None $30,000

Baggage insuranceNone $2,500 depreciated cost

$2,500 replacement cost

Airport club admission (based on smallentrance fee)

No admission $5 per visit $2 per visit

Card acceptanceAir, hotel, rental cars (AHC)AHC and most restaurants (AHCR)AHCR and most general retailers (AHCRG)AHCR and department stores only (AHCRD)

24-hour medical/legal referral networkNo Yes ;

Airport limousine to city destinationNot offered Available at 20% discount

800 number for emergency car serviceNot offered Available at 20% discount

Table 1: These attributes (and levels withinattributes) describe the set of potential ser-vices that would be offered to credit-card sub-scribers.

descriptions that a respondent sees to asmall fraction of the total number of com-binations. In this problem, an array of 64profiles (less than 0.04 percent of the total)is sufficient to estimate all attribute-levelmain effects on an uncorrelated basis. The

study designers used a hybrid conjoint de-sign IGreen and Krieger 1996], and eachrespondent was asked to consider onlyeight (balanced) profile descriptionsdrawn from the 64 profiles.

For such studies, researchers may pre-pare prop cards (Figure 1). After the re-spondent sorts the prop cards in terms ofpreference, each card is rated on a 0 to 100likelihood-of-acquisition scale. In smallconjoint studies (for example, six or sevenattributes, each at two or three levels), re-spondents are given all of the fuH pro-files—16 to 32 prop cards. In these cases,respondents typically sort the prop cardsinto four to eight ordered categories beforethey give likelihood-of-purchase ratingsfor each separate profile within eachcategory.Types of Conjoint Data Collection

Four major types of data collection pro-cedures are currently used for conjointanalysis:(1) In full profile techniques, each respon-dent sees a complete set of the full-profileprop cards. After sorting the cards into or-dered categories, the respondent rateseach card on a 0 to 100 likelihood-of-purchase scale.

(2) In compositional techniques, such asthe CASEMAP procedure ISrinivasan1988], each respondent rates the desirabil-ity of each set of attribute levels on a 0 to100 scale and then rates the attributes onan importance scale. (This approach is typ-ically called self-explicated preference-datacollection.)

(3) In hybrid techniques, each respondentperforms a self-explicated evaluation taskand evaluates a subset of the full-profilecards [Green, Goldberg, and Montemayor

INTERFACES 31:3 S58

CONJOINT ANALYSIS

Annual price$20

Cash rebateNone

800 number for message forwardingNone

Retail purchase insuranceNone

Common carrier insurance$50,000

Rental car insurance$30,000

Baggage insurance$2,500 depreciated cost

Airport club admission$2 per visit

Card acceptanceAir, hotel, rental cars

Medical-LegalNo

Airport limousineNot offered

800 number for emergency carAvailable at 20% discount

Annual price$50

Cash rebate

800 number for message forwarding24 hours per day

Retail purchase insuranceNone

Common carrier insuranceNone

Rental car insurance$30,000

Baggage insuranceNone

Airport club admission$5 per visit

Card acceptanceAir, hotel, rental cars and mostrestaurants

Medical-LegalYes

Airport limousineAvailable at 20% discount

800 number for emergency carNot offered

Figure 1: These prop cards illustrate specific services that a credit card could offer. For eachcard Ihe respondent indicates how likely she would be to subscribe to the credit card on a0-100 point scale.

1981]. The resulting utility function is acomposite of data obtained from bothtasks.(4) In adaptive conjoint analysis, a hybridtechnique developed by Sawtooth Soft-ware [Johnson 1987], each respondent firstperforms a self-explication task and thenevaluates a set of partial-profile descrip-tions, two at a time. Partial profiles usu-ally consist of two or three attributes perstimulus card. Researchers vary the partialprofile descriptions depending upon re-sponses to earlier paired comparisons. Therespondent evaluates each pair of partialprofiles on a graded, paired comparisons

scale. Both tasks are administered by com-puter [Johnson 19871.Conjoint Models

Most conjoint analysts fit what is knownas the part-worth model to respondents'evaluative judgments, whether they obtainthese judgments using full-profile, self-explicated, or hybrid approaches. How-ever, they occasionally use vector andideal-point models. We assume that thereare P attributes and / stimuli used in thestudy design. For a given respondent, welet 1/;,, denote the desirability of the pih at-tribute for the jth stimulus; we first as-sume that i/,p is inherently continuous. The

May-June 2001 S59

GREEN, KRIEGER, WIND

Preference

Highestpreference

Idea! point x

Highestpreference

Lowest

Nexlhighest

I preferenceIJ ^

Amount of attribute p Amount of attribute p Chosen levels of attribute p

Figure 2: These diagrams illustrate what is meant by linear preferences, idea-point preferences,and discrete (part-worth) preferences. The third graph shows three part-worths. Source: Greenand Srinivasan [1978].

vector model assumes that the respondent'spreference Sy for the ;th stimulus is givenby

p

where w^ denotes the respondent's impor-tance weight for each of the P attributes(Figure 2).

In the ideal-point model, the analyst positsthat preference s, is inversely related to theweighted squared distance dj of the loca-tion yjf, of the /th stimulus from the indi-vidual's ideal point Xp, where dj is definedas • • :

dJ - f!P

In the part-worth model, the analyst as-sumes that

p

s, = 2i

where 1/, , is the category level and fp is afunction denoting the part-worth corre-sponding to level yip. In practice, the ana-lyst estimates fp(yjp) for a selected set ofdiscrete levels of i/, ,.

In Figure 3, we show illustrative (aver-aged) part-worths for the attribute levelsdescribed in Table 1. Part-worths are often

34 hours

fl-5pm

Sim.

Price Rebate Messaga Purctase Canier Car Baggage Airporl Acceptence Medicat/ Limougine Emergenc7ForvranJing Insurance Inautanca Inauiancs Insurance ClutB Legal Car Sarvics

Figure 3: This chart illustrates how part-worth scale values depend on price, message forward-ing, . . . emergency car service. For example, the most preferred price is $0 and the least pre-ferred price is $100.

INTERFACES 31:3 S60

CONJOINT ANALYSIS

scaled so that the lowest part-worth iszero, within each attribute. Strictly speak-ing, analysts evaluate part-worth functionsat discrete levels for each attribute. How-ever, in many applications, analysts inter-polate between levels of continuous attri-butes, such as price, when they enter thepart-worths in buyer-choice simulators.The scaling {vertical axis) is commonacross all attributes; this allows the analystto add up part-worths across each attri-bute level to obtain the overall (product orservice) utility of any profile composablefrom the basic attribute levels.Stimulus Presentation

In collecting conjoint data, analysts cur-rently emphasize the full-profile and hy-brid procedures, including Sawtooth'sadaptive conjoint analysis. While they stillemploy paragraph descriptions of attri-bute levels in some industry studies, theyusually use profile cards w ith terseattribute-level descriptions (Figure 1). An-alysts increasingly use pictorial material;these kinds of props make the respon-dent's task more interesting and conveyinformation easily with little ambiguity.Moreover, conjoint n:iethods are increas-ingly being applied to the design of physi-cal products (for example, foods and bev-erages, fragrances, and personal-careproducts). In these cases, researchers useactual, experimentally designedprototypes.Precursors to Conjoint Analysis

How did conjoint analysis come about?For marketing researchers working in dataanalysis and modeling, the 1970s provideda rich bounty of tools and techniques. Un-like the 1960s (a decade of borrowing toolsfrom operations researchers), the late '60s

and early '70s saw the strong influence ofdevelopments in the behavioral sciences(primarily psychometrics and mathemati-cal psychology) on marketing-researchmethods. In particular, three techniques,cluster analysis, multidimensional scaling[MDSl, often called perceptual and prefer-ence mapping, and conjoint analysis, wereintroduced to marketing at that time.

Cluster analytic methods almost imme-diately found application in market seg-mentation IGreen, Frank, and Robinson1967]. Unlike a priori market segmenta-tion, where researchers assume they knowsegment identifiers at the outset, clusteranalysis provided researchers and practi-tioners with tools for implementing posthoc, or cluster-based segmentation. Re-searchers could then base segmentation onthe needs or benefits sought, brand prefer-ences, problem-solving alternatives, psy-chographics, and a host of other variables.The idea of cluster-based segmentationwas to let the data speak for themselves interms of finding groups of consumers whoshare similar needs, attitudes, trade-offs,or benefits.

Multidimensional scaling methods alsoreceived considerable attention in the1970s. Marketing researchers learned thevalue of constructing attractive "maps" forpictorially representing large two-way (ormultiway) numerical tables as point orpoint-and-vector geometric representa-tions in two (or possibly three) dimen-sions. They could also profitably combineMDS with hierarchical clustering methodsto augment two-dimensional configura-tions with cluster-based representationsbased on solutions in higher dimensionali-ties (Carroll and Green [1997] summarize

May-June 2001 S6I

GREEN, KRIEGER, WIND

the history of the MDS field).As early appliers of MDS found (often

to their dismay), the brand-positioningmaps generated using MDS techniques ap-peared to be largely of diagnostic ratherthan predictive value. Researchers whowanted to use MDS to develop new orrestaged products faced a two-part prob-lem: translating brand scores on perceiveddimensions to manipulable dimensionsand relating manipulable attribute levelsto their counterparts in perceptual or pref-erence space. These problems illustrate thegeneral problem of reverse engineering, inwhich back-translation from perceived at-tributes to physical and chemical productcharacteristics is typically not one to one[Kaul and Rao 1994]. It is little wonderthat conjoint analysis, with its emphasison researcher-specified attribute levels,provided a basis for relating preference orchoice responses to explicit arguments, in-cluding physical attributes as well as ver-bal descriptions of product and service at-tribute levels.The Era of Conjoint Analysis

Conjoint analysis has been blessed withseveral sets of parents. The seminal pre-cursory paper to conjoint analysis. Luceand Tukey's paper on conjoint measure-ment, appeared in the journal of Mathemati-cal Psychology in 1964. The authors focusedon axiomatic approaches to fundamentalmeasurement. The idea was to obtain or-dered metric-scale data from rank-orderresponse data and a set of factorially de-signed stimuli. Not surprisingly, the ini-tial conjoint algorithm, called Monanova,designed by Kruskal [1965] and pro-grammed by Kruska! and Carmone, usedranked response data.

In the late '60s, Green and his colleagues(Vithala Rao, Frank Carmone, and ArunJain) started running numerous experi-ments with the Monanova program. Aworking paper by Green and Rao ap-peared in 1969, followed by the first mar-keting journal article on conjoint analysis[Green and Rao 1971]. Following this,Johnson [19741 and Westwood, Lunn, andBeazley [19741 pubhshed articles onJohnson's two~attributes-at-a-tlme trade-off model.

Users of Kruskal's Monanova andShocker and Srinivasan's [1977] Linmapprograms (both nonmetric models) soonlearned, however, that ratings-based de-pendent variables analyzed with dummy-variable regression techniques provided a

Choice-based conjoint studiescan be a mixed blessing.

robust alternative to ordinally based dataprocedures [Carmone, Green, and Jain1978; Cattin and Wittink 19761. Orthogo-nal main-effects plans, based on Addel-man's 119621 fractional factorial designs,significantly reduced respondents' cogni-tive burdens in responding to full-profiledescriptions.

Hence, in the mid-'70s conditions wereripe for the quick diffusion of metricmethods of conjoint analysis (usingdummy-variable regression). In the mid-'80s, Johnson [1987] introduced his adap-tive conjoint analysis program that usedgraded paired comparisons as one set ofinputs to the model. About the same time,Herman [19881 introduced a PC-basedpackage that used full-profile stimulibased on orthogonal designs. Both pack-

INTERFACES 31:3 S62

CONJOINT ANALYSIS

ages contained conjoint simulators. Bothsoftware packages were also easy to useand moderately priced for commercial re-search firms. The advent of PC-based con-joint packages opened a large and eagermarket for applying the methodology.

By the late '70s, conjoint analysis hadtruly come of age. Wittink and his coau-thors [Wittink and Cattin 1989; Wittink,Vriens, and Burhenne 1994] provided ex-tensive surveys attesting to the rapiditywith which conjoint analysis was beingadopted by researchers. Myers, Massy,and Greyser 119801 used conjoint analysisas a canonical case of how new ideas dif-fuse throughout the research community.Green and Srinivasan 11978] published areview of conjoint's progress, which theyfollowed with a further review 12 yearslater IGreen and Srinivasan 19901. The im-petus behind conjoint's diffusion reflectedthe joint influence of academic and practi-tioners' contributions to the methodology,the availability of easy-to-use software,and the early credibility of results frombusiness applications.

Conjoint analysis, unlike MDS and clus-tering, dealt with central problems—mea-suring buyer trade-offs for developingnew or reformulated products and esti-mating price-demand functions. In con-trast, cluster analysis and MDS tools areoften used as ancillary techniques for dataanalysis and presentation. Both, however,have been gainfully employed in conjointstudies involving buyer segmentation[Green and Krieger 1991] and in present-ing perceptual maps of conjoint results[Green, Krieger, and Carroll 19871.

Distant relatives to conjoint analysis in-clude the decompositional approaches of

Hoffman, Slovic, and Rorer [1968] and thefunctional measurement approach ofAnderson [1970]. However, the seminalpaper by Luce and Tukey [1964] and theexperimental-design papers by Addelman[1962] and Plackett and Burman [1946],among others, provided key theoreticalunderpinnings and motivation for devel-oping conjoint analysis. In particular, or-thogonal main-effects designs and theirmore sophisticated extensions still consti-tute a major activity in the continuingdevelopment of efficient experimentaldesigns. Considerable new research[Anderson and Wiley 1992; Huber andZwerina 1996; Kuhfeld, Tobias, andGarratt 1994; Lazari and Anderson 1994]on this topic occurred during the '90s.

The most publicized early paper on con-joint analysis appeared in Harvard BusinessReviezv [Green and Wind 19751. IronicallyGreen and Wind [1973] had earlier pub-lished a research monograph on conjointanalysis and multiattribute decision mak-ing that met with much less readerinterest.Conjoint Development in the '80s

Technical developments in conjointanalysis have proceeded swiftly over thepast two decades, accompanied by thou-sands of applications (Table 2). Duringthe '80s, two developments stand out withregard to model development andapplication:

—Choice-based conjoint models, and—-Hybrid conjoint models, includingJohnson's adaptive conjoint analysismodel.

In traditional conjoint analysis, respon-dents typically rate various product orsupplier profiles presented one at a time.

May-June 2001 S63

GREEN, KRIEGER, WIND

Choice-based conjoint [McFadden 1974; Gensch and Recker 1979; Batsell and Lodish 1981;Mahajan, Green, and Goldberg 1982; Louviere and Woodworth 1983]

Three-way multivariate conjoint analysis [DeSarbo et al. 1982]Number of attribute levels effects on derived conjoint importance [Wittink, Krishnamurthi, and

Nutter 19821Constrained parameter estimation in conjoint analysis ISrinivasan, Jain, and Malhotra 1983]Hybrid models for conjoint analysis [Green, Goldberg, and Montemayor 1981; Green 19841Introduction in 1985 of Bretton-C[ark's full-profile conjoint techniques [Herman 1988][ntroduction in 1985 of Sawtooth Software's adaptive conjoint analysis [Johnson 1987]Factor analytic approaches to individualized conjoint analysis [Hagerty 1985]Conjoint analysis and MDS in tandem [Green, Krieger, and Carroll 1987]Reliability and validity testing [Bateson, Reibstein, and Boulding 19871Simultaneous conjoint parameter estimation and segmentation [Kamakura 1988]Bretton-Clark's second generation, full-profile suite of programs [Herman 19881Conjunctive-compensatory self-explicated models [Srinivasan 1988|

Componential segmentation with optimization features [Green, Krieger, and Zelnio 1989]Compensatory model problems in negatively correlated environments [Johnson, Meyer, and

Ghose 1989]New experimental designs for conjoint [Steckel, DeSarbo, and Mahajan 1991]A reservation price model for optimal pricing [Kohli and Mahajan 19911A review of experimental choice analysis [Batsell and Louviere 1991]Latent class conjoint analysis [DeSarbo et al. 1992] ''Constrained part-worth estimation [van der Lans and Helsen 19921Modeling hierarchical conjoint processes [Oppewal, Louviere, and Timmermans 1994]Concomitant variable latent class modeling IKamakura, Wedel, and Agrawal 1994]Hierarchical Bayes models for conjoint analysis lAllenby, Arora, and Ginter 1995; Allenby and

Ginter 1995; Lenk et al. 1996]A comparison of metric conjoint models IVriens, Wedel, and Wilms 1996]Utility balanced experimental designs [Huber and Zwerina 1996]Competitive interaction simulators [Choi, DeSarbo, and Harker 1990; Green and Krieger 1997]Mixture models for segmentation [Wedel and Kamakura 1998]Commercial windows-based, choice-based conjoint [Sawtooth Software 1999]Krieger and Green's hybrid choice-based conjoint model [Vavra, Green, and Krieger 1999]Latent class conjoint analysis [Ramaswamy and Cohen 2000]Hierarchical Bayes applied to Internet recommendation systems [Ansari, Essegaier, and Kohli

2000]Response latencies and conjoint analysis [Haaijer, Kamakura, and Wedel 2000]

Table 2: This is a partial list of contributions to conjoint analysis over the 1974-2000 period.

on a likelihood-of-purchase sca]e. In 'choice-based conjoint, respondents typi-ca]]y see profile descriptions of two ormore explicit competitors, which vary onone or more attributes, hi this case, thetask is either to pick one's most preferredprofile from the set or alternatively to allo-

cate 100 points across the set of profiles,reflecting one's relative strength ofpreferences.

In choice-based conjoint analysis, ana-lysts typically employ multinomial logitmodels, although occasionally they useprobit-based models. They usually employ

INTERFACES 31:3 S64

CONJOINT ANALYSIS

traditional conjoint analysis when a newproduct is entering a new or stable marketin which competitors are either nonexis-tent or treated as passive in terms of re-sponding competitively to the new entry.In choice-based conjoint analysis, how-ever, the analyst assumes active competi-tors who can modify their profiles (includ-ing price and nonprice attributes) as well.Sawtooth Software's Adaptive ConjointAnalysis is a traditional (hybrid) conjointmodel. Its choice-based conjoint modeldeals with the case of explicit, active com-petitors. In choice-based conjoint analysis,analysts originally estimated parameters atthe total-sample level. Newer methods (hi-erarchical Bayes) now permit measure-ment of individual differences as well.

The seminal precursory paper to choice-based conjoint [using the multinomiallogit model] was written by an econome-trician [McFadden 1974]. ]VIcFadden's

Despite its maturity, conjointanalysis is far from stagnant.

work was soon recognized and adoptedby a number of marketing researchers, in-c[uding Punj and Stae[in [1978]. Genschand Recker [1979] a[so used this mode] indeveloping an alternative to regression foranalyzing cross-sectional choice data.BatseU and Lodish [19811 illustrated themultinomial logit's use in modeling indi-vidual choices over replicated choice sets.Their model yields a share-of-choice pre-diction for alternatives in competitivechoice sets. They indicate that the modelcan be profitably applied to market seg-ments, as well. Mahajan, Green, andGoldberg [1982] applied Theil's logit

model to a choice-based conjoint problem.Louviere and Woodworth 11983] ex-

tended the preceding research. They dis-cuss experimental designs that lend them-selves to choice-based conjoint problems.They also provide an extensive and richset of empirical examples based on vari-ous sets of fractional factorial designs.They focus on aggregate (pooled-over in-dividuals) consumer-choice studies. Theirpaper spawned subsequent research fordealing with the more complex experi-mental designs needed for choice-basedconjoint analysis.

Choice-based conjoint studies can be amixed blessii\g. The respondent's tasks areextensive, since respondents may have toevaluate 10 (or more) scenarios. Each sce-nario could contain eight or more brands,each with several attributes and with sev-eral levels within attributes. Nonetheless,choice-based conjoint analysis has mark-edly increased in popularity because it candeal with the complexity of choosingamong two or more competitive profiies,each of which can vary idiosyncraticallyacross attributes and levels.

In the early '80s, hybrid models IGreen1984; Green, Goldberg, and Montemayor1981] appeared in direct response to theincreasing popularity of conjoint analysis.Along with conjoint's increased applica-tion came the desire to expand the numberof attributes and the levels that could beaccommodated. Hybrid models employself-explicated data collected on both thedesirability and the importance of attri-butes and levels. Respondents then con-sider small subsets of the full profiles toevaluate.

The early hybrid models initially used

May-June 2001 S65

GREEN, KRIEGER, WIND

stagewise regression to fit simplified com-positional models to the self-explicateddata, later augmented by decompositionalmodels (fitted at the segment level). Thevalue of these models lies in the greateraccuracy they achieve, compared to non-hybrid models, in within-attribute estima-tion. The full-profile responses mainlyserve to refine self-explicated attributeimportances.

More recently, Green and Krieger [1996]extended hybrid models to allow parame-ter estimation at the individual level. Theydescribe four separate models of increas-ing generality. Again, the objective is touse the self-explicated data primarily forwithin-attribute part-worth estimationwMle using the full-profile analysis to pro-duce improved estimates of attributeimportances.

The most-used commercial conjointmodel is Johnson's adaptive-conjoint-analysis program. AGA is a hybrid modelthat incorporates self-explicated desirabili-ties and importances, followed by thepresentation of pairs of partial profiles(typically consisting of levels on two orperhaps three attributes) drawn from thefull set of attributes. A respondent isasked to choose between the members ofeach pair and to include his or her prefer-ence intensity as well via graded pairedcomparison; AGA developers [SawtoothSoftware 1999] have continued to intro-duce useful refinements to the originalversion [Johnson 1987].Conjoint Developments in the '90s

Probably the most far-reaching develop-ments in the '90s use hierarchical Bayesianmodeling of individual differences inchoice-based models. Before this, choice-

based models were either estimated fromdata pooled across al] individuals or by la-tent class methods (partial dissaggrega-tion) as applied by DeSarbo et al., [1992]and Ramaswamy and Cohen [2000].

The work of A]lenby, Arora, and Ginter[19951, Ailenby and Ginter [19951, andLenk, DeSarbo, Green, and Young [1996]has enabled choice-based-conjoint users toobtain individua[-level, part-worth esti-mates based on hierarchical Bayesianmethods. (Sawtooth Software has recentlyadded this type of module to its choice-based conjoint software.) To the extentthat an individual's parameters are bothself-consistent and different from the ag-gregated data, the individual's data willreceive more weight in the estimation ofhis or her part-worths. Individuals whosepart-worths are estimated poorly (that is,with large error from his or her own data)will receive more weight from the aggre-gate data.

Recently, Vavra, Green, and Krieger[1999] proposed another approach to thechoice-based-conjoint probiem. They de-veloped a hybrid choice-based model thatcombines self-explicated data with full-profile responses. This model requires noweighted averaging of an individual'sdata with that from the entire group. Onthe downside, however, it requires self-explicated (in addition to full-profile) data.Illustrative Applications

Of the extensive list (Table 3) of conjointapplicafions, two are of particular interest:the customer-driven design of Marriott'sCourtyard Hotels [Wind et al. 1989], andthe design and evaluation of the New Jer-sey and New York EZ-Pass electronic tollcollection system [Vavra, Green, and

INTERFAGES 31:3 S66

CONJOINT ANALYSIS

AT&T's first cellular telephone—Chicago-based study of 1,000 drivers' reactions to cell phonefeatures of the new "honey-comb" relay system

Ford Fairlane—conjoint analysis used in redesign of Ford to reflect general automotivedownsizing objectives

IBM R[SC 6000 workstation—conjoint used to measure potential buyer reactions to variations inperformance and reliability features of a new workstation

Squibb's captopril antihypertensive—six-country study of physicians' evaluations of Capoten'sefficacy and safety features

Tagamet (SKF) and Zantac (Giaxo) ulcer drugs—competitive pricing and analysis of demandelasticities

Healt[i maintenance plans—study conducted by the American Association of Retired Persons;results submitted to Congress

US Navy benefit packages for reenlistment—conjoint used to develop menu of new reenlistmentplans based on individual differences in types of duties, health needs, and sign-over bonuses

Antidumping litigation-^AT&T vs. Pacific-rim manufacturers legal dispute regarding smallbusiness telephone equipment; case adjudicated in AT&T's favor

Continental vs. American Airlines litigation—conjoint study of travel agents' trade-offs amongairline flight selections

FedEx new services study—trade-off study of customer reactions to new methods for trackingdelayed and lost letters and packages

Intermittent windshield wipers litigation—study to determine consumer evaluations of thederived "willingness to pay" for the intermittent wiper feature

MasterCard and Diner's Club—new travel and entertainment features evaluated for each caseMonsanto's herbicide packaging study—consumer reactions to advanced packaging devices for

liquids, solids, and aerosolsPolaroid's instant camera design—study of consumer reactions to new features and camera design

aestheticsMarriott time-share units—development of optimal interior and exterior decors, services, and

priceRitz Carlton—conjoint used to develop hotel decor and servicesJapanese cable TV—conjoint survey of Japanese consumers' trade-offs among services and prices

of satellite TVShell and Texaco's brand equity—conjoint study of price, brand, and alternative company

positions for the Shell and Texaco CorporationsUPS services study—conjoint used to examine customers' evaluations of four major suppliers of

overnight letter and package deliveryTable 3: These illustrative conjoint applications were conducted by the authors and their col-leagues.

Krieger 1999]. marily to business travelers who had noTo the best of our knowledge. Court- need for many of the features provided by

yard by Marriott is the largest project in up-scale hotels, such as Marriott andterms of attributes and levels ever under- Hyatt.taken using conjoint analysis. In the early In addition to price, the study design in-'80s, Marriott management wished to de- eluded seven hotel facets—external decor,sign an "optimal" hotel chain catering pri- room decor, food service, lounge facilities.

May-June 2001 S67

GREEN, KRIEGER, WIND

general services, leisure activities (for ex-ample, a fitness club), and security fea-tures. Analysts developed some 50 attrib-utes with a total of 160 attribute levels.The models included hybrid conjoint andan early type of choice-based conjoint. Theanalysts also used computer simulators.The study made extensive use of visualprops {pictures and three-dimensionalmodels) as well as experimental rooms inwhich the furnishings and decor were sys-tematically varied according to experimen-tal designs.

By al] counts, the Courtyard study wasa success. Marriott implemented almost allof the design recommendations and laterextended the approach to other new prod-ucts (for example, Marriott Suites) andused the findings in the design of Court-yard advertisements and brochures. By1990, Marriott had over 300 Courtyards,employing over 15,000 people. Today,there are 450 Courtyards worldwide, withannual sales in the billions of dollars.Marriott has since used conjoint analysisin such related endeavors as designingtime-share vacation units and in room andamenities pricing.

In the EZ-Pass tol] collection project.New Jersey and New York developed areally new product aimed at speeding upand simplifying vehicle passage on theirtoll highways, bridges, and tunnels. Com-muters use an electronic tag (transponder)attached to the vehicle's inside wind-shield. As the vehicle approaches a tolllane, an antenna in the lane reads the cus-tomer's vehicle and account informationembedded in the tag. The information iselectronically forwarded to an in-line com-puter. The computer, in turn, deducts the

toll from the customer's account.The project began in 1992. There were

two main questions:—How should EZ-Pass be configured?—What level of resources should be allo-cated to its implementation?The two states conducted a large conjointstudy (over 3,000 respondents) by atelephone-mail-telephone procedure inwhich qualified respondents received apacket of questionnaire materials and avideo cassette. The video contained an 11-minute infomercial that described theproblems associated with the current toll-road system and the benefits provided byEZ-Pass.

Analysts used seven conjoint attributesdealing with such issues as number oflanes available, tag acquisition, cost, tollprices, invoicing, and other uses of the tag.The study designers analyzed the individ-ual respondents' data at both the overaUsample level and by region and facility.The overall (at equilibrium) forecast madein 1992 of "take rate" was 49-percent us-age. The actual take rate (seven yearslater) was 44 percent; future usage is ex-pected to be higher than 49 percent. (EvenDavid Letterman liked the system!)

Both projects. Courtyard by Marriottand EZ-Pass, illustrate the ability of con-joint analyses to lead to actionable find-ings that provide customer-driven designfeatures and consumer-usage or salesforecasts.Future Prospects

After 30 years of development and ap-plication, conjoint analysis seems to havesurvived the test of time. While newbreakthroughs may well be less frequent,the method continues to grow in depth

INTERFACES 31:3

CONJOINT ANALYSIS

and breadth of usage. We expect to see thefollowing further developments:—New simulator-optimizers that can max-imize either financial return or marketshare I Vavra, Green, and Krieger 1999];—New classes of problems, includingmenu selection and bundling models fortelecommunications and banking services[Ben-Akiva and Gershenfeld 19981;—More realistic imagery for describing at-tribute levels, for example, using virtualreality displays;

—Continued extensions of conjoint appli-cations to such fields as tourism, entertain-ment, health maintenance, gambling, andlegal disputes;—New application venues, such as con-joint's recent implementation on the Inter-net, including activebuyersguide.com,personalogic.com, and conjointonline.com,sites that typically use hybrid conjointmodels to elicit buyer preferences forWeb-based merchandise [Ansari,Essegaier, and Kohl 2000];—Additional studies of conjoint reliabilityand validity [Haaijer, Kamakura, andWedel 2000; Vriens, Wedel, and Wilms1996];

—Extension of consumer-based applica-tions to other stakeholder groups, such asemployees, suppliers, stockholders, andmunicipalities;—New "dynamic" conjoint simulators thatconsider competitive action-reaction se-quences IChoi, DeSarbo, and Harker 1990;Green and Krieger 19971; and—Prototype simulators (for example, testcars) that permit analysts to measure re-spondents' preferences for type of ride, ac-celeration, cornering, and so forth in real-istic surroundings.

In short, despite its maturity, conjointanalysis is still far from stagnant. Becausethe methods deal with the pervasive prob-lem of buyer preferences and choices, con-joint's future promises continued develop-ment and application.The Role of Software Development inthe Diffusion of Conjoint Analysis

Throughout the development of conjointanalysis and the precursory techniques ofcluster analysis and MDS, the avai]abilityof inexpensive and easy-to-use softwarehas been crucial to their dissemination.Early on. Bell Laboratories and the Mar-keting Science Institute played importantroles in making mainframe software avail-able to both academic and industry users.With the growth of the personal computer,Johnson's Sawtooth Software and Her-man's Bretton-Clark companies providedaffordable PC software to business usersand academic versions for teaching pur-poses. Sawtooth Software has maintainedcontact with business and academiathrough its newsletters, annual meetings,and continued development of software[Sawtooth Software 19991 to implementnew research ideas, such as choice-basedmodels. The American Marketing Associa-tion's annual Advanced Research Tech-niques Forum provides a meeting placefor the fruitful exchange of ideas betweenacademics and practitioners.A Science vs. Engineering Postscript

From its inception, conjoint analysis hasdrawn on ideas from mathematical psy-chologists, psychometricians, statisticians,econometricians, and operations research-ers. These ideas concern experimental de-sign, parameter estimation, descriptivemodel building, normative model build-

]VIay-June 2001 S69

GREEN, KRIEGER, WIND

ing, and the comparative evaluation of

various models' reliability and validity.

The method's practical consequences

attest to its value and staying power

IGustaffsson, Hermann, and Huber 2000].

Through the interplay of theoretical con-

tributions and practical applications, con-

joint methodology continues to grow, as

academics and practitioners learn useful

things from each other.

Acknowledgments

We thank Eric Bradlow and Wes

Hutchinson of the Wharton Schoo], an

anonymous reviewer, and the editors for

helpftjl comments on the paper's initial

draft.ReferencesAdde[man, Sidney 1962, "Orthogona] main-

effect plans for asymmetrical factorial experi-ments," Techiwmetrics, Vol. 4, No. 1, pp. 21-46.

Allenby, Greg M.; Arora, Neeraj; and Ginter,James L. 1995, "Incorporating prior knowl-edge into the analysis of conjoint studies,"journal of Marketing Research, Vol. 32, No. 2(May), pp. 152-162.

Allenby, Greg M. and Ginter, James L. 1995,"Using extremes to design products and seg-ment markets," journal of Marketing Research,Vol. 32, No. 4 (May), pp. 152-162.

Anderson, D. A. and Wiley, James B. 1992, "Ef-ficient choice set designs for estimating cross-effects models," Marketing Letters, Vol. 3, No.4 (October), pp. 357-70.

Anderson, Norman H. 1970, "Functional mea-surement and psychophysical judgment,"Psychological Review, Vol. 77, No. 3, pp. 153-70.

Ansari, Asim; Essegaier, Skander; and Kohli,Rajeev 2000, "Internet recommendation sys-tems," Journal of Marketing Research, Vol. 37,No. 3 (August), pp. 363-375.

Bateson, John E. G.; Reibstein, David J; andBoulding, William 1987, "Conjoint analysisreliability and validity: A framework for fu-ture research," Review of Marketing, ed. M. J.Houston, American Marketing Association,

Chicago, Ilinois, pp. 451-481.Batsell, Richard R. and Lodish, Leonard M.

1981, "A model and measurement methodol-ogy for predicting individual consumerchoice," journal of Marketing Research, Vol. 18,No. 1 (February), pp. 1-12.

Batsell, Richard R. and Louviere, Jordan J. 1991,"Experimental analysis of choice," MarketingLetters, Vol. 2, No. 3 (August), pp. 199-214.

Bell, D. F.; Raiffa, Howard; and Tversky, Amos,eds. 1988, Decision Making: Descriptive, Nor-mative, and Prescriptive Interactions, Cam-bridge University Press, New York.

Ben-Akiva, Moshe and Gershenfeld, Shari 1998,"Multi-featured products and services; Ana-lyzing pricing and bundling strategies," jour-nal of Forecasting, Vol. 17, No. 3, pp. 175-196.

Carmone, Frank J.; Green, Paul E.; and Jain,Arun K. 1978, "The robustness of conjointanalysis: Some Monte Carlo results," journalof Marketing Research, Vol. 15, No. 2, pp. 300-303.

Carroll, J. Douglas 1969, "Categorical conjointmeasurement," paper presented at the An-nual Meeting of the Association of Mathe-matical Psychology, Ann Arbor, Michigan.

Carroll, J. Douglas and Green, Paul E. 1997,"Psychometric methods in marketing re-search: Part 2, multidimensional scaling,"journal of Marketing Research, Vol. 34, No. 2(May), pp. 193-204.

Cattin, Philippe and Wittink, Dick R. 1976, "AMonte Carlo study of metric and nonmetricestimation techniques," Paper 341, GraduateSchool of Business, Stanford University.

Choi, S. C; DeSarbo, Wayne S.; and Harker,Patrick T. 1990, "Product positioning underprice competition," Management Science, Vol.30, No. 3, pp. 175-199.

DeSarbo, Wayne S.; Carroll, ]. Douglas;Lehmann, Donald R.; and O'Shaugnhessy,John 1982, "Three-way multivariate conjointanalysis," Marketing Science, Vol. 1, No. 4(Fail), pp. 323-350.

DeSarbo, Wayne S.; Wedel, Michel; Vriens,Marco; and Ramaswamy, Venkatram 1992,"Latent class metric conjoint analysis," Mar-keting Letters, Vol. 3, No. 4 (July), pp. 273-288.

Gensch, Dennis H. and Recker, Wilfred W.1979, "The multinominal muUiattribute logitchoice model," journal of Marketing Research,

INTERFACES 31:3 S70

CONJOINT ANALYSIS

Vol. 16, No. 1 (February), pp. 124-132.Green, Paul E. 1984, "Hybrid models for con-

joint analysis: An expository review," journalof Marketing Research, Vol. 21, No. 3 (May),pp. 155-159.

Green, Paul E.; Frank, Ronald E.; andRobinson, Patrick J. 1967, "Cluster analysis intest market selection," Management Science,Vol. 13, No. B, pp. B387-^00.

Green, Paul E.; Goldberg, Stephen M.; andMontemayor, Mila 1981, "A hybrid utility es-timation model for conjoint analysis," journalof Marketing, Vol. 45, No. 1 (Winter), pp. 33-41.

Green, Paul E. and Krieger, Abba M. 1991,"Segmenting markets with conjoint analy-sis," journal of Marketing, Vol. 55, No. 4 (Oc-tober), pp. 20-31.

Green, Paul E. and Krieger, Abba M. 1996, "In-dividualized hybrid models for conjoint anal-ysis," Management Science, Vol. 42, No. 6(June), pp. 850-867.

Green, Paul E. and Krieger, Abba M. 1997, "Us-ing conjoint analysis to view comp>etitive in-teraction through the customer's eyes," inWharton on Dynamic Competitive Strategi/,eds., George S. Day, D. J. Reibstein, and R. E.Gunther, John Wiley and Sons, New York,pp. 343-367.

Green, Paul E.; Krieger, Abba M.; and Carroll,J. Douglas 1987, "Conjoint analysis and mul-tidimensional scaling: A complementary ap-proach," Journal of Advertising Research, Vol.27, No. 5 (October/November), pp. 21-27.

Green, Paul E.; Krieger, Abba M.; and Zelnio,Robert N. 1989, "A componential segmenta-tion model with optimal design features,"Decision Sciences, Vol. 20, No. 2 (Spring), pp.221-238.

Green, Paul E. and Rao, Vithala R. 1971, "Con-joint measurement for quantifying judgmen-tal data," journal of Marketing Research, Vol. 8,No. 3 (August), pp. 355-363.

Green, Paul E. and Srinivasan, V. 1978, "Con-joint analysis in consumer research: issuesand outlook, " Journal of Consumer Research,Vol. 5, No. 2 (September), pp. 103-123.

Green, Paul E. and Srinivasan, V. 1990, "Con-joint analysis in marketing: New develop-ments with implications for research andpractice," journal of Marketing, Vol. 54, No. 1(October), pp. 3-19.

Green, Paul E. and Wind, Yoram 1973, Multiat-

tribute Decisions in Marketing: A MeasurementApproach, Dryden Press, Hinsdale, lUinos.

Green, Paul E. and Wind, Yoram 1975, "Newway to measure consumers' judgments,"Harvard Business Revieiv, Vol. 53, No. 4 (July-August), pp. 107-117.

Gustaffsson, Anders; Hermann, Andress; andHuber, Frank, eds. 2000, Conjoint Measure-ment: Methods and Applications, Springer-Verlag, Berlin, Germany.

Haaijer, Rinus; Kamakura, Wagner; and Wedel,Michel 2000, "Response latencies in the anal-ysis of conjoint choice experiments," Journalof Marketing Research, Vol. 37, No. 3 (August),pp. 376-382.

Hagerty, Michael R. 1985, "Improving the pre-dictive power of conjoint analysis: The use offactor analysis and cluster analysis," Journalof Marketing Research, Vol. 22, No. 2 (May),168-184.

Herman, Steve 1988, "Software for full-profileconjoint analysis," in Proceedings of the Saw-tooth Conference on Perceptual Mapping, Con-joint Analysis, ami Computer Intervieiving, ed.M. Metegrano, Sawtooth Software, Ketchum,Idaho, pp. 117-130.

Hoffman, Paul J.; Slovic, Paul; and Rorer, L. G.1968, "An analysis of variance model for theassessment of configural cue utilization inclinical judgment," Psychological Bulletin, Vol.69, No. 5, pp. 338-349.

Huber, Joel and Zwerina, Klaus 1996, "The im-portance of utility balance in efficient choicedesigns," journal of Marketing Research, Vol.33, No. 3 (August), pp. 307-317.

Johnson, Eric; Meyer, Robert J.; and Ghose,Sanjay 1989, "When choice models fail: Com-pensatory models in negatively correlatedenvironments," journal of Marketing Research,Vol. 26, No. 3 (August), pp. 255-270.

Johnson, Richard M. 1974, "Trade-off analysisof consumer values," Journal of Marketing Re-search, Vol. 11, No. 2 (May), pp. 121-127!

Johnson, Richard M. 1987, "Adaptive conjointanalysis," in Proceedings of the Sawtooth Soft-loare Conference on Perceptual Mapping, Con-joint Analysis, and Computer Intervieuiing, ed.M. Metegrano, Sawtooth Software, Ketchum,Idaho, pp. 253-265.

Kamakura, Wagner 1988, "A least squares pro-cedure for benefit segmentation with conjointexperiments," journal of Marketing Research,

May-June 2001 S71

GREEN, KRIEGER, WIND

Vol. 25, No. 2 (May), pp. 157-167.Kamakura, Wagner; Wedel, Michel; and

Agrawal, Jagdish 1994, "Concomitant vari-able latent class models for conjoint analy-sis," International journal of Research in Mar-keting, Vol. 11, No. 5, pp. 451-464.

Kaul, Anil and Rao, Vithala R. 1994, "Researchfor product position and design decisions:An integrative review," International journalof Research on Marketing, Vol. 12, No. 4, pp.293-320.

Keeney, Ralph L. and Raiffa, Howard 1976, De-cisions with Multiple Objectiz<es: Preferences andValue Tradoffs, John Wiley and Sons, NewYork.

Kohli, Rajeev and Mahajan, Vijay 1991, "A res-ervation price model for optimal pricing ofmultiattribute products in conjoint analysis,"journal of Marketing Research, Vol. 28, No. 3(August), pp. 347-354.

Kruskal, Joseph B. 1965, "Analysis of factorialexperiments by estimating monotone trans-formations of the data," Journal of the RoyalStatistical Society, Series B, Vol. 27, No. 2, pp.251-263.

Kuhfeld, Warren F.; Tobias, Randall D.; andGarratt, Mark 1994, "Efficient experimentaldesigns with marketing research applica-tions," journal of Marketing Research, Vol. 31,No. 4 (November), pp. 545-557.

Lazari, Andreas G. and Anderson, Donald A.1994, "Designs of discrete choice set experi-ments for estimating both attribute and avail-ability cross effects," Journal of Marketing Re-search, Vol. 31, No. 3 (August), pp. 375-383.

Lenk, Peter J.; DeSarbo, Wayne S.; Green, PaulE.; and Young, Martin R. 1996, "HierarchicalBayes conjoint analysis: Recovery of part-worth heterogeneity from reduced experi-mental designs," Marketing Science, Vol. 15,No. 2, pp. 173-191.

Louviere, Jordan and Woodworth, George1983, "Design and analysis of simulated con-sumer choice or allocation experiments. Jour-nal of Marketing Research, Vol. 20, No. 4 (No-vember), pp. 350-367.

Luce, R. Duncan and Tukey, John W. 1964, "Si-multaneous conjoint measurement: A newtype of fundamental measurement," Journalof Mathematical Psychology, Vol. 1, pp. 1-27.

Mahajan, Vijay; Green, Paul E.; and Goldberg,Stephen M. 1982, "A conjoint model for mea-

suring self- and cross-price demand relation-ships," journal of Marketing Research, Vol. 19,No. 3 (August), pp. 334-342.

McFadden, Daniel 1974, "Conditional logitanalysis of qualitative choice behavior," inFrontiers on Econometrics, ed. P. Zarembka,Academic Press, New York, pp. 105-421.

Myers, John G.; Massy, William F.; andGreyser, Stephen A. 1980, Marketing Researchand Knoidedge Development, Prentice-Hall, En-glewood Cliffs, New Jersey.

Oppewal, Harmen; Louviere, Jordan; andTimmermans, Harry 1994, "Modeling hierar-chical conjoint processes with integratedchoice experiments," journal of Marketing Re-search, Vol. 31, No. 1 (February), pp. 92-105.

Plackett, R. L. and Burman, J. P. 1946, "The de-sign of optimum multifactorial experiments,"Biometrika, Vol 33, No. 4, pp. 305-325.

Punj, Girish N. and Staelin, Richard 1978, "Thechoice process for graduate businessschools," journal of Marketing Research, Vol.15, No. 4 (November), pp. 588-598.

Ramaswamy, Venkatram and Cohen, Steven H.2000, "Latent class models for conjoint analy-sis," in Conjoint Measurement: Methods andApplications, eds. A. Gustafsson, A. Hermann,and F. Huber, Springer-Verlag, Berlin,Germany.

Saaty, Thomas L. 1980, The Analytical HierarchyProcess, McGraw-Hill, New York.

Sawtooth Software 1999, CSC for Windows,Sequim, Washington.

Shocker, Allan D. and Srinivasan, V. 1977,"LINMAP (Version II): A FORTRAN IV com-puter program for analyzing ordinal prefer-ence (dominance) judgments via linear pro-gramming techniques and for conjointmeasurement," journal of Marketing Research,Vol. 14, No. l ,pp. 101-103.

Srinivasan, V. 1988, "A conjunctive-compensatory approach to the self-explication of multiattributed preferences,"Decision Sciences, Vol. 19, No. 2 (Spring), pp.295-305.

Srinivasan, V.; Jain, Arun K.; and Malhotra,Naresh K. 1983, "Improving the predictivepower of conjoint analysis by constrained pa-rameter estimation," journal of Marketing Re-search, Vol. 20, No. 4 (November), pp. 433-438.

Starr, Martin K. and Zeleny, M. 1977, MultipleCriteria Decisions Making, North-Holland,

INTERFACES 31:3 S72

CONJOINT ANALYSIS

Amsterdam, Holland.Steckel, Joel H.; DeSarbo, Wayne S.; and

Mahajan, Vijay 1991, "On the creation of fea-sible conjoint analysis experimental designs,"Decision Sciences, Vol. 22, No. 2, pp. 435-442.

van der Lans, Tvo A. and Helsen, WiUem H.1992, "Constrained part-worth estimation inconjoint analysis using the self-explicatedutility model," International journal of Researchin Marketing, Vol. 9, No. 3, pp. 325-344.

Vavra, Terry G.; Green, Paul E.; and Krieger,Abba M. 1999, "Evaluating EZ-Pass," Market-ing Research, Vol. 11, No. 3 (Summer), pp. 5-16.

Vriens, Marco; Wedel, Michel; and Wilms, Tom1996, "Metric conjoint segmentation meth-ods: A Monte Carlo comparison, journal ofMarketing Research, Vol. 33, No. 1 (February),pp. 73-85.

Wedel, Michel and Kamakura, Wagner A. 1998,Market Segmentation: Conceptual and Methodo-logical Foundations, Kluwer, Boston,Massachusetts.

Westwood, Dick; Lunn, Tony; and Beazley,David 1974, "The trade-off model and its ex-tensions," journal of the Market Research Soci-ety, Vol. 16, No. 3, pp. 227-241.

Wind, Jerry; Green, Paul E.; Shifflet, Douglas;and Scarbrough, Marsha 1989, "Courtyard byMarriott: Designing a hotel facility withconsumer-based marketing models," Inter-faces, Vol. 19, No. 1 (January-February), pp.25-47.

Wittink, Dick and Cattin, Philippe 1989, "Com-mercial use of conjoint analysis: An update,"Journal of Marketing, Vol. 53, No. 3 (July), pp.91-96

Wittink, Dick R.; Krishnamurthi, Lakshman;and Nutter, Julia B. 1982, "Comparing de-rived importance weights across attributes,"journal of Consumer Research, Vol. 8, No. 4,(March), pp. 471^74.

Wittink, Dick; Vriens, Marco; and Burhenne,Wim 1994, "Commercial u.se of conjoint inEurope: Results and critical reflections," In-ternational journal of Research in Marketing,Vol. n . No. 1, pp. 41-52.

Young, Forrest W. 1969, "Polynomial conjointanalysis of similarities: Definitions for a spe-cial algorithm," Research paper No. 76, Psy-chometric Laboratory, University of NorthCarolina.

May-June 2001 . S73