Discrete Choice Experiment

Imagine you’re at a cafe, perusing the lunch menu. Their recommended items are a soup, a traditional chicken sandwich, or a chickpea salad sandwich. You’re too hungry for just soup, so that leaves you to choose between the sandwiches. You always try to go for the vegetarian options, so you opt for the chickpea salad sandwich. But as you’re about to order, you see they have a new falafel wrap that looks incredible. It’s a bit more expensive, but it seems worth it, so you order the falafel wrap. Now you just have to choose a drink...

Without realizing it, you’ve made a series of consecutive choices that ultimately revealed your lunch preference. By deciding between each option presented, an outsider may be able to infer what you wanted for lunch on that day, allowing them to infer how you would usually rank lunch menu items as a whole. For example, upon hearing your decision, a friend might be able to guess what attributes you’re basing your choices around for your lunch dish (for example, How filling will this food be? Does this food fit with my ethical framework?). They might also sense that specific menu item characteristics vary in importance (the food looking appetizing is very important, while the cost is a less critical component), even if how you value these things may change over time.

An everyday situation like this is reminiscent of the more complicated, formal, and sophisticated discrete choice experiment method scientists use to model individuals’ decision-making processes. Like your series of choices between lunch items, discrete choice experiments typically use distinct decisions (e.g., A vs. B vs. C) to estimate the "utility" or value that individuals assign to each alternative based on the attributes of the options. The relative utility values help infer a person's likelihood of choosing one option over another. Despite this setup, DCEs are not primarily focused on creating a strict ranking of the items presented. Instead, they reveal the trade-offs and preferences based on the attributes, leading to utility scores for the different options, which can indirectly inform a ranking.¹

In most standard consumption models seen in economics, consumption is typically modeled as a continuous variable (e.g., the quantity of a good can be any value along a continuum).¹ However, in a discrete choice experiment, the options that are given are discrete, meaning categorical or whole-number terms. For example, an economist may ask, “What is the optimal number of cereal boxes for a family to consume per year?” and use regression analysis or calculus methods to model demand based on average consumption.

Furthermore, discrete choice analysis examines decisions involving distinct alternatives. Rather than asking a continuous variable question like, “How much cereal,” an economist might also ask a discrete choice question like, “What kind of cereal?” to study families’ preferences for different brands. This allows DCEs to study both qualitative and quantitative questions. Although there are an astounding number of cereal options (some might say too many), a DCE could group together different categories or brands of cereal and still provide helpful insight. Most often, a DCE is used when there are only a few distinct quantities or qualities available from which to choose. This might be something like, “How many boxes does your family consume per week?” or, “Which Cheerios variety is your favorite?” With a narrow set of options available, preferences for these types of choices can be more definitively revealed through DCEs. 

Discrete choice experiments are used to examine choices at an individual level and by larger organizations like business firms or government agencies or to estimate consumers' willingness to pay for certain features. Particularly in these larger organizations, it’s essential to have more objective methods of analysis that are easy for extensive scale application. Techniques like logistic regression and probit regression can be used to analyze discrete choice experiments empirically.¹