Introduction

Revealed preference (RP) or market data are commonly used by economic, marketing and transport analysts to estimate models that explain discrete choice behaviour, as discussed in chapter 2. Such data may have substantial amounts of noise that are the result of many influences, e.g., measurement error. In contrast, stated preference (SP) or choice (SC) data are generated by some systematic and planned design process in which the attributes and their levels are pre-defined without measurement error and varied to create preference or choice alternatives. Similarly, RP choices can be measured with relatively little (if any) error when direct observation is possible (e.g., one can record brands chosen by consumers in supermarkets, or modes chosen by travellers in the act of making trips). However, an individual's self-report of a choice ‘actually’ made is likely to be uncertain, and the uncertainty or noise probably increases as the time between the actual choice and the report of that choice increases. Additionally, SP and SC responses are ‘stated’ and not actual, and hence are uncertain because individuals may not actually choose the alternatives that they say they will/would.

In later chapters we will discuss the benefits of combining RP and SC data to take advantage of their strengths and (we hope) minimise their individual weaknesses. Before doing so, we have to introduce a set of analytical tools that provide the building blocks for the design of choice experiments.