Choice-Based Conjoint and MaxDiff (i.e., Maximum Difference, also known as Best-Worst scaling) are advanced methods for measuring and estimating consumer preferences by means of choice experiments and discrete choice models. Conjoint Analysis is a veteran methodology, initiated in the early 1970s, for measuring multi-attribute consumer preferences, based on ranking order or rating evaluations of product alternatives. However, a significant development took place in the 1990s that has transformed the family of conjoint methods for measuring consumer preferences, turning choice-based conjoint within a decade into the dominant methodology of conjoint studies. The MaxDiff approach may be seen as a complementary method that presents less information and is sufficient for situations or purposes where the conjoint experiments could be too demanding or superfluous.
Instead of going right into technical details of the methods and differences between them, let us consider an example on cars for illustration. In the purchase of a new car (e.g., compact, family or SUV), consumers will likely have to study and review many attributes and features. Attributes at a higher level of generality include, for example, safety, convenience, performance, design, driving experience, and entertainment. Such general-level attributes are sometimes called ‘dimensions’. A more specific attribute (e.g., number of gear drives, exterior colour, capacity of baggage cabin), is described by detailing several optional levels (e.g., colours red, blue, mustard or grey; brand names). Quantitative attributes are especially suitable for representing selected value-points over a range of values (e.g., fuel and/or electricity consumption, engine volume; price is a special case to be addressed later). Features, on the other hand, are often represented as binary options, such as whether the feature is included or absent (e.g., adaptive cruise control, lane keeping, but also digital vs. analogue-classic display on instrument panel, regular vs. metallic colour).
The technical profiles of cars may include quite a lot of information to ask respondents about in a survey. Even the larger conjoint experiments will usually focus on a subset of attributes of interest or concern to product developers and marketers. Yet, a conjoint study may combine multi-level attributes and features; researchers may describe attributes at a more abstract or concrete level, and the study may concentrate on aspects of a particular ‘dimension’ (e.g., safety, design). In each choice question, a respondent may be presented with three to five product profiles and he or she will be asked to choose a single most preferred option from them (or none of the options offered); an experiment usually includes 8 to 12 choice questions (Note: adaptive multi-stage methods are available for studies that aim to cover a larger span of product information).
The findings from conjoint studies can provide very helpful information in processes of new product development (NPD), for the configuration and design of products. However, the choice-based conjoint study may also be used for planning product upgrade and improvement, and in creating additional model variants. Especially, choice models can be constructed to reveal differences in consumer preferences, so as to design product versions or attribute add-on offerings that would appeal to different consumer segments (e.g., by using statistical methodologies of Latent Class or Hierarchical-Bayes estimation). The findings derived from the discrete choice (Logit) conjoint models normally include ‘utility’ values of attribute levels, relative importance weights of attributes (summed to 100%), and furthermore predictions of preference shares of product alternatives generated by simulations of hypothetical competitive market scenarios (e.g., for answering what-if questions and product or price optimisation).
However, the purpose of a research may not be concerned with product configuration and the details of attributes. For example, in early stages of planning and testing product concepts, the focal interest could be identifying key attributes of importance or value to consumers, without considering attribute specifics. The MaxDiff methodology is primarily suited for studying and setting priorities. A MaxDiff study may be conducted, for instance, to enquire about the priorities of drivers among comfort, convenience or ease of driving, safety, and performance of a car. In this case the relative importance of the attributes would be of main interest (parallel to the importance weights in a conjoint model but not similar!). There is another situation where MaxDiff could also be appropriate and efficient: Suppose the researchers concentrate on a particular dimension or system in a car and they can describe the attributes of interest as binary features — for examples the features included in the safety apparatus of the car or information and display features of the multimedia system. Hence, respondents are presented with sets of feature options that can be included (e.g., alerting on road signs, too-short distance alert, warning and correction of deviation from lane, speed alert), and they are asked to state which features are most and least important or desirable to them to have in their car.
The MaxDiff methodology has a simple and elegant concept: In each set of items (e.g., attributes, features, statements), a respondent chooses two items: one that is regarded the Best and one that is regarded the Worst (e.g., most and least important, preferable, desirable, valued, appealing items). A set would contain 4 or 5 items out of a complete list. An item must be included in at least one of the sets, but it would preferably appear two or three times in different sets (Note: it is desirable for a design to be balanced in the sense that any item is included equal times as others — a similar principle applies in Conjoint experiments for attribute levels). Across these sets of items (usually 8 to 12 sets) in different combinations of items, it is possible to derive from the Best-Worst choices an ordered list of all items (e.g., 20 items) by their priority to the consumers-respondents (based on the weights or scores estimated). The different combinations of items in the partial sets are essential for constructing the complete ordered list. MaxDiff studies may provide consumer researchers, marketers or product developers, for some research purposes or goals, adequate and revealing information with less burden to respondents.
- In the derived (descending) ordered list, items that were chosen as ‘Best’ at least once are assigned the higher scores and appear at top of the list; items that were chosen as ‘Worst’ at least once receive the lower scores and appear at the bottom of the list; the remaining items that were never chosen either way are assigned intermediary scores and appear in the centre of the list (importance scores or weights, summed to 100%, are allocated relative to the frequency of items being chosen as ‘Best’ or ‘Worst’).
Price has a different implication than most other attributes because it usually represents ‘disutility’, as monetary cost, against other attributes implying benefits (although in some conditions price may signal a positive value of quality or image to consumers, thus price could play a dual role). By including price as an attribute in a conjoint experiment with a few price-points, different forms of functions of response to price may be estimated (linear or non-linear). The quoting of price values in product profiles allows researchers to estimate consumer sensitivity to price.
Conversely, including price as an attribute in a MaxDiff study could face difficulties. Since this methodology does not present actual price information, it is not capable of providing a reliable measure of price sensitivity. The impact of price is susceptible to get ignored or discounted vis-à-vis most other ‘positive’ attributes — conjoint models are not immune to this problem, but not to the same extent, because in conjoint choices respondents can conceive the cost of products. Yet, in a particular segment of cost-driven consumers, respondents are more likely to instantly detect price and choose it as ‘worst’ item. The results and their interpretation in MaxDiff could get distorted when items-attributes have reverse meanings (i.e., refer to ‘desirability’ rather than ‘importance’), and when there are salient differences between consumer segments (e.g., with respect to price).
Identifying distinct groups or segments of respondents based on their choice responses can help considerably in distinguishing between sets of priorities that need to be addressed. It can resolve at least part of the problem described above by separating those who are more aversive to price (as cost) from the rest of the sample. However, segmentation should be beneficial in resolving issues of differences in attitude or preference with respect to various attributes or features. (Note: As in Choice-Based Conjoint studies, Sawtooth Software, an expert research firm in this field, proposes that Latent Class and HB analyses can also be used with the choices in MaxDiff models).
Choice-Based Conjoint and MaxDiff choice experiments and models are part of a powerful suite of methodologies for measuring consumer preferences, and assessing their implications, offered to researchers, marketers and product developers. These methods have strengths and weaknesses that make them fit for accomplishing different purposes or goals. Hence, they should be regarded as complementary methods. Furthermore, their content, structure and design can be adapted to satisfy the requirements and purposes of different stages in an NPD process, where they can be applied together with other methods of measuring consumer preferences. They open up a range of possibilities and opportunities for applied research on consumer preferences.
Recommended: Readers interested in more background information and learning in greater depth about these methodologies may turn to the website of Sawtooth Software, active in the field for forty years (look in the sections of Solutions and Resources).