Experimental design and the estimation of willingness to pay in choice experiments for health policy evaluation

This chapter focuses on stated preferences obtained from discrete choice experiments (DCEs also known as SPDCEs), as opposed to data that reflect real market choices (revealed preferences (RP) as discussed in Chapter 9). DCEs try to simulate the essential elements of real market options that consumers might face in the future. Unlike real market choice data, DCEs rely on constructed markets in which key factors that are hypothesized to drive choices are systematically varied. To the extent that the consumers in a DCE make choices in a manner consistent with the way in which they would actually choose in a real market, one can derive standard welfare estimates for policy changes. The remainder of this chapter is devoted to discussing and illustrating how this can be accomplished with DCEs. More details on DCEs can be found in Louviere, Hensher, and Swait (1). In order to collect DCE data information from consumers, one must identify factors that drive the choices of interest. These factors are called ‘attributes’ of the choice options. Once these attributes are identified, one must assign them values, known in experimental design parlance as ‘levels’. Taken together, the attributes and levels define and determine possible choice options that can be offered to consumers in a DCE survey. That is, a factorial combination of attribute levels completely defines the possible choice options. So, for example, if there are three attributes, say A(4), B(3), and C(2), with the associated number of levels in parentheses, the factorial combinations, or all possible options are 4 × 3 × 2 = 24. We refer to that factorial combination as a ‘full’ or ‘complete’ factorial design. Typically, the number of combinations in a full factorial design is too many to use in practical field applications of DCE surveys, so one has to sample from the full factorial to reduce the size of the problem. There are many ways to sample from full factorials, but a common approach in DCE surveys is to sample based on what is known as a ‘fractional factorial design’. We return to these ideas in more detail later in the chapter. The type of experimental design used to construct a DCE survey is important for three reasons: 1. it determines the economic quantities of potential interest that