Conjoint Measurement

The essay by the psychologist, Luce, and the statistician, Tukey (1964) can be viewed as the origin of conjoint analysis (Green and Srinivasan 1978; Carroll and Green 1995). Since its introduction into marketing literature by Green and Rao (1971) as well as by Johnson (1974) in the beginning of the 1970s, conjoint analysis has developed into a method of preference studies that receives much attention from both theoreticians and those who carry out field studies. For example, Cattin and Wittink (1982) report 698 conjoint projects that were carried out by 17 companies included in their survey in the period from 1971 to 1980. For the period from 1981 to 1985, Wittink and Cattin (1989) found 66 companies in the United States that were in charge of a total of 1062 conjoint projects. As regards Europe, Wittink, Vriens and Burhenne counted a total of 956 projects carried out by 59 companies in the period from 1986 to 1991 (Wittink, Vriens and Burhenne 1994 and Baier and Gaul 1999). A survey initiated in Germany in 1998 in which a total of 519 companies and chairs at universities of the country were to provide information on their activities in the field of conjoint analysis for the period from 1993 to 1998 shows that 52 institutions interested in the study design an average of 6 conjoint analyses per year (Melles and Holling 1999; for an earlier study Schubert 1991). If we project the number of analyses for the total period of five years, we get approx. 1531 projects.

One of the most frequent purpose of conjoint analysis is the measurement of price effects (Wittink and Cattin 1989; Wittink, Vriens, and Burhenne 1994). Usually this is be done by describing a number of product alternatives on a small number of attributes, including price, and collecting some kind of preference data for these product alternatives. From the estimated part-worth function for price one can infer price effects (Srinivasan 1979).

Many companies claim to be consumer-driven or focused. They often support this claim with evidence from extensive customer research programmes. They run focus groups, send out questionnaires, monitor customer satisfaction scores and analyse sales data.

In commercial applications of conjoint analysis to product design and product pricing it has become quite popular to further evaluate the estimated individual part-worth functions by predicting shares of choices for alternatives in hypothetical market scenarios (Wittink, Vriens and Burhenne 1994 and Baier 1999 for surveys on commercial applications). Wide-spread software packages for conjoint analysis (Sawtooth Software’s 1994 ACA system) already include specific modules to handle this so-called market simulation situation for which, typically, a threefold input is required: (I) The (estimated) individual part-worth functions have to be provided. (II) A definition of a hypothetical market scenario is needed that allows to calculate individual utility values for each available alternative. (III) A so-called choice rule has to be selected, which relates individual utility values to expected individual choice probabilities and, consequently, to market shares for the alternatives. In this context, the determination of an adequate choice rule seems to be the most cumbersome task. Well-known traditional choice rules are, e.g., the 1ST CHOICE rule (where the individuals are assumed to always select the choice alternative with the highest utility value), the BTL (Bradley,Terry, Luce) rule (where individual choice probabilities are related to corresponding shares of utility values), and the LOGIT rule (where exponentiated utility values are used). Furthermore, in newer choice rules implemented by various software developers, the similarity of an alternative to other alternatives is taken into account as a corrective when choice probabilities are calculated (Sawtooth Software 1994).

Over the past two decades conjoint measurement has been a popular method for measuring customers’ preference structures. Wittink and Cattin (1989) estimate that about 400 commercial applications were carried out per year during the early 1980s. In the 1990s this number probably exceeds 1000. The popularity of conjoint measurement appears to derive, at least in part, from its presumed superiority in validity over simpler, less expensive techniques such as self-explication approaches (Leigh, MacKay and Summers 1984). However, when considered in empirical studies, this superiority frequently has not been found (e.g. Green and Srinivasan 1990; Srinivasan and Park 1997). This issue is of major practical relevance. If, at least in certain situations, conjoint measurement is not clearly superior in validity to self-explicated approaches, it becomes highly questionable whether future applications for measuring customers’ preferences should be done by conjoint measurement, as self-explicated approaches are clear advantageous in terms of time and money effort.

The process of developing, introducing and disseminating a new product is one of the most frequently described and highly formalised subjects in marketing literature (Urban and Hauser 1993; Rogers 1995). The development of a software package, while not an exception to the more general models, presents a number of peculiarities which it is worth investigating (Urban and von Hippel 1988; Carmel 1995; Carmel and Becker 1995). The rate of technological progress in the IT industry is extremely rapid: this structural factor, combined with the low level of entry barriers, leads to intense competition within the sector. In this structural context one key performance factor is the ability to generate innovative ideas and put them into practice in the shortest possible time. It is in fact this time variable which plays a crucial role in securing a sustainable competitive advantage. Consequently, the reduction of time-to-market, i.e. the time lapse between the devising of the product and its marketing, acquires great importance.

Design of experiments is an established technique for product and process improvement that has its origin in the 1920s and the work of Sir Ronald Fisher. Conjoint analysis shares the same theoretical basis as traditional design of experiments, but was originally used within the field of psychology and it was not until the early 1970s that the methodology was introduced into marketing research to form what is called conjoint analysis (Luce and Tukey 1964; Green and Rao 1971; Johnson 1974). Today, conjoint analysis is an established technique for investigating customer preferences.

It is the goal of conjoint analysis to explain and predict preferences of customers (Schweikl 1985). Variants of predefined manifestations of attributes of various product concepts (both real and hypothetical) are created, and these are presented to test persons for evaluation. The contributions (partial benefits) the various attributes make to overall preference (overall benefit) are estimated on the basis of overall preference judgments (Green and Srinivasan 1978).

It is clear that conjoint analysis has had a substantial impact upon research practice (Wittink and Cattin 1989; Wittink, Vriens and Burhenne 1994). Conjoint analysis has evolved, and along with that evolution has been a gradual shift in the types of responses collected, from rankings to ratings to choices.

Conjoint analysis has already been widely accepted by marketing researchers as a popular instrument for the measurement of consumer preferences. Typical applications of conjoint analysis include new product design based on the relationship between product features and predicted choice behavior, benefit segmentation based on attribute preferences, etc. The popularity of conjoint analysis among marketing researchers hinges on the belief that it produces valid measurements of consumer preferences for the features of a product or service, and that it provides accurate predictions of choice behavior.

It is widely known that preference is not an observable phenomenon per se but a construct and in so far ”a term specifically designed for a special scientific purpose generally to organize knowledge and direct research in attempt to describe or explain some aspect of nature“ (Peter 1981, 134) for which ”[...] measures can be developed which at least partially represent the constructs.“ (Peter and Churchill 1986, 1). Consequently, measuring a construct implies that “numbers are assigned to objects [...] in such a way that the relations between the numbers reflect the relations between the objects [...] with respect to the characteristic involved” (Green, Tull and Albaum 1988, 243).

The purpose of this chapter is to place conjoint analysis techniques within the broader framework of preference elicitation techniques that are consistent with the Random Utility Theory (RUT) paradigm. This allows us to accomplish the following objectives: explain how random utility theory provides a level playing field on which to compare preference elicitation methods, and why virtually all conjoint methods can be treated as a special case of a much broader theoretical framework. We achieve this by:

Conjoint choice experimentation involves the design of product profiles on the basis of product attributes specified at certain levels, and requires respondents to repeatedly choose one alternative from different sets of profiles offered to them, instead of ranking or rating all profiles, as is usually done in various forms of classic metric conjoint studies. The Multinomial Logit (MNL) model has been the most frequently used model to analyze the 0/1 choice data arising from such conjoint choice experiments (e.g., Louviere and Woodworth 1983; Elrod, Louviere and Davey 1992). One of the first articles describing the potential advantages of a choice approach for conjoint analysis was by Madanski (1980). His conclusion was that conjoint analysts could adopt the random utility model approach to explain gross trends or predilections in decisions instead of each person’s specific decision in each choice presented. The real breakthrough for conjoint choice came with the Louviere and Woodworth (1983) article in which they integrated the conjoint and discrete choice approaches.

Conjoint analysis was introduced to market researchers in the early 1970s as a means to understand the importance of product and service attributes and price as predictors of consumer preference (e.g., Green and Rao 1971; Green and Wind 1973). Since then it has received considerable attention in academic research (see Green and Srinivasan 1978, 1990 for exhaustive reviews; and Louviere 1994 for a review of the behavioral foundations of conjoint analysis). By systematically manipulating the product or service descriptions shown to a respondent with an experimental design, conjoint analysis allows decision-makers to understand consumer preferences in an enormous range of potential market situations (see Cattin and Wittink 1982; Wittink and Cattin 1989; and Wittink, Vriens, and Burhenne 1994 for surveys of industry usage of conjoint analysis).

One of the reasons conjoint analysis has been so popular as a management decision tool has been the availability of a choice simulator. These simulators often arrive in the form of a software or spreadsheet program accompanying the output of a conjoint study. These simulators enable managers to perform ‘what if’ questions about their market—estimating market shares under various assumptions about competition and their own offerings. As examples, simulators can predict the market share of a new offering; they can estimate the direct and cross elasticity of price changes within a market, or they can form the logical guide to strategic simulations that anticipate short- and long-term competitive responses (Green and Krieger 1988).

Conjoint analysis was introduced in the 1970’s to quantify consumer tradeoffs. The methodology has become very popular for market-based strategic decisions (see e.g. Green and Srinivasan 1990; Wittink, Vriens and Burhenne 1994). This popularity has both been influenced by and has led to the development of a substantial body of research on conjoint analysis. Surprisingly, however, conjoint applications are almost exclusively limited to one-shot, ad hoc surveys.