Choice-Based Conjoint

Traditional conjoint asks respondents to rate or rank individual product profiles, whereas CBC asks them to choose one alternative from a competing set, often with a 'none' option. The choice task more closely resembles an actual purchase, avoids the scale-use idiosyncrasies of ratings, and connects directly to random utility theory and the multinomial logit model introduced by Louviere and Woodworth. The trade-off is statistical: a single choice carries less information than a full rating, so CBC needs more tasks or hierarchical Bayes pooling to estimate individual-level utilities, but the realism and the natural link to share simulation have made it the dominant approach.

Each respondent answers only a handful of choice tasks, which is far too few to estimate a stable set of part-worths in isolation. Hierarchical Bayes solves this by treating every respondent's utilities as a draw from a population distribution and combining each person's own choices with information borrowed across the sample. As Orme describes, this shrinkage produces reliable individual-level utilities that capture genuine preference heterogeneity, which aggregate logit cannot. Those individual utilities are what make realistic market simulation and segmentation possible, which is why HB has become the default estimation method in commercial CBC.

Yes, indirectly. If price is included as an attribute, the estimated part-worths put feature value and price on the same utility scale, so willingness to pay for a feature can be computed as the price change that offsets the feature's utility. This is one of CBC's most valued outputs, but it must be interpreted carefully: the estimate depends on the price range shown, assumes the linear or piecewise utility specification used, and reflects stated rather than revealed preference. Best practice is to read willingness-to-pay figures as directional guidance and to confirm pricing conclusions with holdout choices and, where available, real market data.