14.1 Introduction

The purpose of this chapter is to illustrate how to use Case 3 BW data to understand preference differences between groups and between individuals in a similar way to what we would do with data from a traditional DCE that collected only first-choice (best) responses. Traditionally, collecting best and worst choices has been thought of as a data augmentation process (for example, see Louviere, Street et al., 2008). This view is consistent when best and worst choices are inversely related because, by observing both best and worst choices for a particular choice set, one gets more information about order, which is what matters in choices.

Consider the potential advantages of collecting Case 3 BW data:

1. (1)more data per respondent from each choice task, thus providing greater precision for model estimates; and

2. (2)using simple BW measures instead of (or in place of) more sophisticated choice models as a way to estimate preferences.

When the inverse relationship between best and worst choices does not hold, choice models that combine data from best and worst choices generally make poorer predictions of best or worst choices than models that rely, respectively, on either best or worst choices alone. When and if this occurs, the advantage of case 3 BW data is different:

1. (3)one can gain additional insights into individual decision processes by comparing differences between best and worst choices.

This chapter focuses on illustrating how to use a balanced incomplete block design and simple BW measures introduced in earlier chapters (and formally in Chapter 5) to (1) test assumptions about the substitutability of best and worst data, (2) show how to cluster the data based on BW count data (BW scores) and (3) estimate aggregate- and individual-level models. We illustrate how to apply these ideas about case 3 BW data at each level of aggregation, using two data sets from online choice experiments involving delivered pizzas and toothpaste.

To anticipate our results, we find considerable evidence for differences between best and worst choices in both data sets.