An alternative to using an experimental design is to instead ask people to describe their recent choices and build a model based on the resulting data. Such revealed preference methodologies are intuitively much more appealing than stated preference methodologies, as they lead to less concern about external validity (i.e., that the results of the analysis will align with what happens in the real world). Their estimation requires the analyst to have access to historical data documenting both consumer choice behavior and changes in the attribute levels of existing products. In many situations, however, analyses of historical data are not sufficient for the development of plausible choice models, due to:
- The technology frontier. Often managers are interested in assessing the likely impact of new innovations, meaning that historical data is unlikely to be of much value.
- The absence of historical data.
- Insufficient variation in the data. If there has never been any price competition, for example, price sensitivity cannot be measured using historical data. While there are vast quantities of data documenting changes in price and market share in many markets, more often than not firms change price simultaneously (either through collusion or reaction), making it impossible to determine the role of price in determining market share.
- The existing data may be misleading. In the carbonated soft drink market, for example, the higher priced brands, such as Coke and Pepsi, have a higher market share than the lower-priced generic brands. Consequently, analysis of historical data may lead to the erroneous conclusion that raising prices will lead to increased share. It is the norm rather than the analysis of historic data that can lead to unlikely conclusions. For example, one large bank received a consulting report advising that their share will increase if one of their competitors drops its price. Similarly, a brewer was told by another consultant that a 1% increase in price would lead to a 90.2% drop in sales.
For these reasons, revealed preference modeling is largely the domain of academic work where issues of external validity are less pressing (i.e., academics do not get sued when their work is not replicable, but consultants can). Commercially-oriented work tends to use choice-based conjoint.