Which variety of conjoint should I use?
Conjoint analysis, one of the most effective tools for understanding buyer decision making, was first introduced in the early 1980’s in the form of a card sorting task. The typical task consisted of 18 or more physical cards, each bearing a product description that was constructed by combining features according to experimental design principles, and respondents sorted the cards in rank order from most preferred to least preferred. Monotonic analysis of variance (“MONANOVA”) was used to estimate “part-worths” for each of the features that made up the product concepts. Once the researcher had these part-worths in hand it was possible to simulate the choices consumers would make under different competitive market scenarios.
This fundamental idea of conjoint analysis–building a model of decision making by simulating the purchase process in some fashion–has carried through a host of variations on the original process, including adaptive conjoint analysis (ACA), choice-based conjoint analysis (CBC), menu-based choice (MBC) and adaptive choice-based conjoint (ACBC). The three methods that directly observe choices in a simulated purchase situation, CBC, ACBC and MBC, have, for the most part, replaced methods that infer choice probabilities from other preference measures, like traditional rankings- (or ratings-) based conjoint and ACA methods.
Considering just the three choice-based methods, here are some guidelines on when to use each technique.
1. Basic Choice-based Conjoint
In basic choice-based conjoint, respondents are presented with a series of choice tasks (usually about twelve “modeling” tasks plus two for holdout validation) that include between 3 and 6 choice alternatives. These alternatives are defined by combining different levels or manifestations of various product features or attributes, which can include brand and price.
Basic choice-based conjoint is best suited for problems with about 3 to 7 attributes with 6 or fewer levels per attribute. We have designed CBC exercises with more attributes (up to about 12) which is feasible if the attributes are not technically complex (requiring lots of words to describe) and have only 2 or 3 levels, on average.
Responses to basic choice-based conjoint questions can be “single select” or constant sum allocations. The determination of whether to use single select or constant sum is based on whether the in-market choice being modeled requires selecting one alternative and rejecting all others, or the in-market choice reflects demand across multiple choice settings or contexts (such as prescribing for a heterogeneous set of patients or usage occasions). Allocation should be used only when the choices represent an actual competitive market context (i.e. “branded” alternatives that may share all the same features, some features, or no features). The competitive set can include “outside goods” that are not in the same product class or category. For example, a conjoint task asking people to allocate trips across various transportation modes might include these alternatives: private car, bus, light rail, and bicycle. A variable like “cost per mile” will not apply all to traveling by bicycle.
In most cases a basic choice-based conjoint task will include a “none” option. This helps to make the choice set collectively exhaustive when there are some alternatives that are not included in the design (such as low market share brands) and makes it possible to estimate potential demand without having to correct for “overstatement.”
In general, basic choice-based conjoint will be the preferred approach (either single select or constant sum allocation) to most insight problems that require conjoint or trade-off modeling (e.g., pricing research, feature optimization, and new product research). This approach is robust and offers considerable flexibility in the design of the choice tasks.
2. Adaptive Choice-based Conjoint
Adaptive choice-based conjoint (ACBC) was developed by Sawtooth Software in response to research showing that in many cases respondents do not use a simple additive process in making their choices. Under a simple additive process, consumers assess the features of each choice and then mentally add or sum the utilities of all the choices and select the one alternative with the highest utility. In reality, consumers often employ non-compensatory processes, such as screening out all alternatives that are priced above a certain level or that lack a particular feature. This is particularly true for products or services that have many (i.e. more than 5) defining attributes.
ACBC is designed to improve the respondent experience by allowing individuals to identify their screening rules before going into a choice-based-conjoint exercise. This is achieved by first asking respondents to identify their “ideal” product from the available attributes, then presenting them with some choice alternatives that vary slightly from that ideal in order to determine if they are using any screening rules. Finally, respondents complete a “standard” single-select choice-based conjoint exercise. These tasks are adaptive in that the respondent only has to consider those attributes that the previous two steps have identified as important (the attributes are visible but do not vary and have a different background shading in the tasks).
ACBC has a number of features that make it desirable for specific problems and research has shown that the estimation of utilities is as good as for basic CBC studies. In our view, ACBC is most suited for problems where there are 5 to 15 attributes, where the overall price of the product is strongly determined by the features that are included, and where the competitive alternatives (e.g. other brands) can be defined in terms of the same attributes, although not necessarily the same levels of those attributes. All other things being equal (e.g., number of attributes and levels and competitive similarity), ACBC may be better for smaller sample sizes because there is less noise in the individual-level data. ACBC does not support constant sum allocation responses.
3. Menu-based Choice
I developed our approach to menu-based choice (which has been incorporated into Sawtooth Software’s MBC product) to deal with a specific situation known as “mixed bundling.” When a company employs mixed bundling (which is essentially a pricing strategy), they offer pre-configured bundles at a specific price and they offer the option of buying the components of the bundles separately. Thus, a consumer can choose either a bundle or can “mix and match,” choosing different components of the bundle from different makers if they wish.
Menu-based choice can also be used when consumers are presented with a “basic” package or product and have the option of adding additional features. An example might be choosing a particular brand of refrigerator and then choosing from among three different extended service plans (or none). Menu-based choice is implemented by linking two or more basic choice-based conjoint models and is appropriate only in the specific situations described above.
Share and Enjoy
aca adaptive choice-based conjoint acbc adaptive conjoint analysis additive process buyer decision making card sorting task cbc choice probabilities choice-based conjoint analysis choice-based methods competitive market competitive market scnarios conjoint conjoint analysis constant sum allocations experimental design priciples feature optimization in-market choice insight menu-based choice mbc mixed bundling monanova monotonic analysis of variance new product research non-compensatory processes part-worths pricing research pricing strategy rank order sawtooth software simulate single select trade-off modeling