<STRONG>Discrete Choice<BR>Warren F. Kuhfeld<BR>Abstract</STRONG><BR>Discrete choice modeling is a popular technique in marketing research, transportation, and other areas,<BR>and is used for understanding people’s stated choice among alternatives. We will discuss designing<BR>a choice experiment, preparing the questionnaire, inputting and processing the data, performing the<BR>analysis, and interpreting the results.<BR>Introduction<BR>This chapter shows you how to use the multinomial logit model (McFadden, 1974; Manski and McFadden,<BR>1981; Louviere and Woodworth, 1983) to investigate consumer’s stated choices. The multinomial<BR>logit model is an alternative to full-profile conjoint analysis and is extremely popular in marketing<BR>research (Louviere, 1991; Carson et. al., 1994). Discrete choice, using the multinomial logit model,<BR>is sometimes referred to as “choice-based conjoint.” However, discrete choice uses a different model<BR>from full-profile conjoint analysis. Discrete choice applies a nonlinear model to aggregate choice data,<BR>whereas full-profile conjoint analysis applies a linear model to individual-level rating or ranking data.<BR>Several examples are discussed.† There is also a very basic introductory example starting on page 73<BR>in the introduction to experimental design chapter, which starts on page 47. Be sure to read the design<BR>chapter before proceeding to the examples in this chapter.<BR>• The candy example (page 144) is a first, very simple example that discusses the multinomial logit<BR>model, the input data, analysis, results, and computing probability of choice.<BR>• The fabric softener example (page 156) is a small, somewhat more realistic example that discusses<BR>designing the choice experiment, randomization, generating the questionnaire, entering<BR>and processing the data, analysis, results, probability of choice, and custom questionnaires.<BR>• The first vacation example (page 184) is a larger, symmetric example that discusses designing the<BR>choice experiment, blocks, randomization, generating the questionnaire, entering and processing<BR>the data, coding, and alternative-specific effects.
|