This chapter discusses the fundamental elements, logic, and key principles in designing DCE studies. Accordingly, this chapter will be rather technical as we cover the nuts and bolts of the DCE methodology in detail. As we will demonstrate in this chapter, a key strength of DCE is that it involves low cognitive complexity (i.e., easy to understand and respond to) for respondents when performing choice tasks. In a DCE, respondents are offered two or more alternatives in each choice task in which each alternative contains different combinations of attributes and attribute levels. They will then be asked to indicate which alternative they prefer in each choice task (as a dependent variable) in categorical format (choice A or B, or C). In the experiment, attributes and their levels are constructed so that respondents explicitly indicate their preference (thus the “stated preference” method). The last part of the DCE technique is the examination of the importance of each attribute for all respondents. In other words, DCE allows the estimation of the marginal rates of substitutions of the attributes (i.e., the monetary value of forgoing one attribute in favor of another attribute).

3.1 A Brief Overview of DCE

As we have discussed previously in Chap. 2, a discrete choice experiment (DCE) is a quantitative research methodology that assesses the relative importance of a set of products, services, ideas, or policy attributes that influence the decision-making of individuals such as consumers, entrepreneurs, investors, employees, parents, citizens, or voters. The DCE is designed to mimic the real-life decision-making process, where individuals are presented with different scenarios with varying attribute levels and asked to make a choice based on their preferences (Louviere et al., 2000). Each decision we make always involves multiple attributes (e.g., making a holiday decision involves making choices involving hotel brands, locations, service quality, prices, etc.). DCE is an attribute-based measure of preference. DCE is built on the assumptions that a situation (i.e., a product, a service, or an intervention) can be described by its attributes and that a decision-maker’s valuation would depend upon different levels of these attributes. Attribute levels describe the variance over which attributes differ across alternatives. For example, when choosing a hotel for a holiday, a key attribute might be hotel brands, with three levels, such as Shangri-la vs. Hilton vs. Holiday Inn.

DCE, a form of quasi-experiment, is typically implemented in a survey format where respondents or participants are asked to make choices between a set of alternatives as defined by attributes and their corresponding levels. That is, respondents state their choice over different alternatives without requiring them to perform the alternatives (i.e., experiments without the actual intervention or quasi-experiment). As such, respondents provide quantitative information on their preference for each alternative, including trade-offs between attributes (Ryan & Gerard, 2003). DCE is based on the economic theory that humans or decision-makers will seek to optimize their utility or welfare (e.g., choosing Hilton over an unknown hotel brand for the same price per night) (Louviere et al., 2010). The data from the choice decisions made by respondents or participants allows researchers to uncover how respondents attach value to selected products, services, or intervention attributes.

Specifically, DCE provides insights into decision-makers’ preferences and quantified information on the trade-offs they are willing to make between different product or service attributes and levels. The quantified information in DCE can be monetary or non-monetary valuation. DCE, among other stated preference methods, proves to be a valuable tool in situations where it is not feasible to use revealed preference data to assess the choices made by decision-makers, such as when valuing a new product that is still in the development phase and not yet accessible in the market, or when there is a lack of variation in available choices (Mangham & Hanson, 2008).

3.2 Theoretical Foundations of DCE

DCE draws upon two theories as its theoretical foundations. First, DCE follows Lancaster’s economic theory of value (Lancaster, 1966), which argues that decision-makers have a preference for and derive utility from underlying attributes of products or services rather than the products or services per se. By presenting decision-makers with alternatives that vary in attributes and levels, DCEs provide researchers with insight into the preference structure of the respondents. This information is not readily available through other research methods. DCE helps elicit the hidden preference structure into an observable preference structure.

For example, it is difficult to know what people prefer from a box of chocolate for a Christmas gift. But we could scientifically describe the chocolate by flavor, calorie, packaging, and percentage of cocoa (using just three attributes at one level each as an example). Then, DCE can unravel what attributes of chocolate people value the most with high precision, transforming hidden preference structure into observable preference structure.

Secondly, DCE is rooted in the random utility theory (McFadden, 1974; Thurstone, 1927). Random utility theory is a long-standing theory of choice behavior that sees a decision-maker’s behavior as an interlinked set of factors. The theory proposes that every decision-maker is a rational decision-maker who seeks to maximize utility relative to their choices and considers utility as a latent structure that cannot be observed directly. That is to say, random utility theory considers that a decision-maker has a “utility” for each choice alternative. Still, these utilities cannot be directly seen by researchers nor known precisely by the decision-maker.

Random utility theory originated from Thurstone’s theory of paired comparisons (1927), in which Thurstone proposed that the modeling of individual choice is the outcome of a process of associating the random variable with each alternative and the alternative with the greatest realization of value is the one selected. Many years later, building on Thurstone’s work, McFadden introduced the random utility framework as a tool in econometric analysis and assumed that each decision-maker faces a finite choice set and selects an alternative that maximizes their utility. In this framework, a choice is defined by the differences in attributes, marginal utilities for attributes, and a stochastic term (McFadden, 1973). The utility generated by an alternative is assumed to depend on the utility associated with its composing attributes and attribute levels.

Specifically, based on the random utility theory framework, the utility for individual i conditional on choice j can be separated into a systematic or explainable component Vij and a stochastic or non-explainable component εij (Mazzanti, 2003). The individual’s utility Uij for a particular alternative is represented as (Eq. 3.1):

$${\mathrm{U}}_{ij}={V}_{ij}+{\varepsilon }_{ij}, j=1, \cdots , J$$
(3.1)

The non-explainable component εij comprises all unidentified factors that impact individual choices from unobservable attributes, specification errors, and/or measurement errors.

The explainable component Vij is a function of attributes of the product or service or idea or policy and characteristics of individual decision-makers, comprising attributes explaining differences in choice alternatives and covariates explaining differences in differences in individual choices, often represented as Eq. 3.2:

$${V}_{ij}={X{^\prime}}_{ij}\beta +{Z^{\prime}}_{i}\delta$$
(3.2)

where Vij represents the utility or value associated with alternative j for individual i, \({X^{\prime}}_{ij}\) is a vector of attributes of the alternatives such as the hotel brand (e.g., Shangri-la), of good j as viewed by individual i, and \({Z}^{\prime}\) is a vector of personal characteristics for individual i, and β and δ are vectors of parameters representing vectors of coefficients to be estimated.

The utility Uij depends on attributes of the alternative j and attributes of the individual decision-maker i. Random utility theory assumes that individuals will choose alternative j if, and only if, the utility derived from this alternative is higher than the utility of any other option in the set of J alternatives (Ben-Akiva et al., 1997; Louviere et al., 2010). Therefore, the probability P that individual i chooses option k from a set of competing alternatives Ci is represented as (Eq. 3.3):

$$\mathrm{P}\left(k|{C}_{i}\right)=\mathrm{P}[\left({V}_{ik}+{\varepsilon }_{ik}\right)>\mathrm{Max}({V}_{ij}+{\varepsilon }_{ij})],$$
(3.3)

for all options j ≠ k in the choice set Ci

3.3 The Key Elements of DCE Design

Before we explain the process of designing and implementing a DCE study, we first need to understand the key elements of DCE. As mentioned earlier, DCE is usually presented in a survey format where respondents are given a set of alternatives and asked to choose among different alternatives in a choice set and repeat the process for other choice sets. For example, suppose we want to study consumers’ preference for bottled orange juice. In that case, we can present them with different choices of bottled orange juice described by a set of attributes and attribute levels. Tables 3.1 and 3.2 are examples of the potential choice sets (also known as a choice question in DCEs) that are presented to respondents in a typical DCE study. In Table 3.1, only two alternatives (Alternative A or B) are prompted for respondents to choose as the most preferred answer; while in Table 3.2, three alternatives (Alternative A or B or Neither) are used to allow respondents to choose “Neither” if both A or B are not considered attractive. The number of choices in each choice set is one of the key decisions researchers must make in the design stage.

Table 3.1 An example of a choice set in DCE (e.g., choosing orange juice)
Full size table
Table 3.2 An example of a choice set in DCE (e.g., choosing jobs)
Full size table

The design stage in DCE is critical because it could influence how much utility information researchers can extract from respondents’ answers (Louviere et al., 2000). A typical DCE design usually involves four key elements, which are choice sets, alternatives, attributes, and attribute levels:

  • Choice sets: In a DCE study, respondents are presented with multiple pairs of alternatives for evaluation. A pair of alternatives is called a choice set. The examples in Tables 3.1 and 3.2 demonstrate a choice set in each study that examines respondents’ preference toward orange juice and job choice, respectively. A DCE study usually contains multiple choice sets that are presented and evaluated in multiple choice tasks by respondents.

  • Alternatives: Alternatives are choices that are presented to respondents in a DCE study. Each alternative or scenario refers to a product or service that is described by a set of attributes or characteristics. In DCE, respondents are asked to choose one alternative they prefer more (or the most) out of a given number of alternatives (two or more) in each choice task. Table 3.1 provides two alternatives (A or B) for respondents to evaluate, while three alternatives (A or B or Neither) are exemplified in Table 3.2.

  • Attributes: Attributes refer to characteristics of a product or service or idea or policy of interest to researchers in a DCE study. The attributes are used to describe different characteristics of the alternatives. From the choices made by respondents in a DCE study, researchers can elicit the relative importance that respondents place on the attributes of a product or service and how they are willing to make trade-offs of one attribute over another. In Table 3.1, which examines decision-makers’ preference for orange juice, four attributes are included to describe orange juice: Organic Production, Sugar, Country of Origin, and Price.

  • Attribute levels: Attributes levels refer to values assigned to each attribute. We can think of attribute level as the operationalization of each attribute. The attribute levels need to be constructed so that the respondents are willing to make trade-offs (i.e., sacrificing one alternative in favor of another) between combinations of different attributes. For example, the attribute Country of Origin in Table 3.1 has two attribute levels, including locally produced and imported. This will consciously ask respondents to state their choice of the country of origin of the orange juice.

3.4 Designing and Implementing DCE

The process of designing and implementing DCE can be divided into six steps, which are adapted from the model proposed by Ryan (1996). A simple process flowchart is illustrated below (see Fig. 3.1).

Fig. 3.1
A flow diagram of stages of a D C E study. It has 6 steps and they are as follows. Identification of attributes, identification of attribute levels, experimental design of D C E, questionnaire design and construction, data collection, and data analysis.

Stages of a DCE study

Full size image

In designing a DCE experiment, a researcher needs to first establish a clear understanding of the issue (e.g., research question or unresolved theoretical puzzle and the research objective) for examination. For instance, if your study aims to understand patients’ preference for flu vaccination, then the DCE should be designed and constructed around factors that may influence people’s decisions about whether to receive a vaccination. The researcher should justify the utilization of a DCE design over other approaches (i.e., limited vaccination alternatives available; lack of access to patients’ records; new vaccines available to be tested on humans).

Step 1: Identification of Attributes: The first step in designing a DCE study involves identifying the attributes of interest that will be used to describe the product or service or idea or policy under investigation. Selecting and defining attributes require a good understanding of the topic, literature, and the target population’s perspective (Coast & Horrocks, 2007). There are multiple ways of identifying attributes, such as conducting systematic reviews to identify potential attributes to be used in the study and collecting primary data via in-depth interviews or focus group discussions with potential respondents or participants. The selected attributes should be demand relevant (i.e., something consumers are willing to pay a price for) and measurable. Normally, a DCE study consists of no more than ten attributes (i.e., usually up to 7 attributes) for ease of operationalization and to avoid respondent fatigue in completing the study.

Step 2: Identification of Attribute Levels: After deciding which attributes to use to describe the product, service, idea, or policy for evaluation, the next step is to decide the values associated with each attribute or the attribute levels. To determine attribute levels, researchers could consider different situations that potential decision-makers may encounter. Typically, the attribute levels selected should reflect the most plausible range of situations that target populations are expected to experience. The levels could be either qualitative categories (e.g., locally produced vs. imported for Country of Origin attribute) or quantitative measures ($10, $15, or $20 for Price attribute) (see Table 3.1). The attribute levels should be actionable to the respondents in a way that they are willing to make trade-offs between combinations of the attributes. Researchers can do in-depth interviews and or focus group discussions as part of a due diligence process to ensure that the attributes and attribute levels are appropriate in the study.

Step 3: Experimental Design of DCE: In this step, researchers design and construct the alternatives or choice sets based on the attributes and attribute levels identified earlier. Good experimental design is crucial for the success of a DCE experiment. Researchers will make a series of decisions in designing a DCE experimental design. Examples include deciding between using a full factorial design or a fractional factorial design, the number of alternatives in each choice set, the allocation of the alternatives, the number of choice sets to present in each choice task to respondents, and the number of respondents to study. We will provide step-by-step instructions with examples to illustrate how to design a DCE experiment using R in Chap. 7.

Step 4: Questionnaire Design and Construction: The next step is to develop a questionnaire that will be distributed to respondents or participants for primary data collection. Usually, a DCE questionnaire includes these sections: (1) information on the objectives of the study and an explanation of the DCE task, (2) screening questions based on certain (inclusion or exclusion) criteria, (3) the DCE choice sets, (4) demographic questions, and (5) follow-up questions to examine respondents’ preference further. In this step, researchers should consider whether the task instructions are appropriate and easy to understand and whether sufficient background and contextual information are provided to respondents. Doing a pre-test of the DCE questionnaire prior to the actual launch is advisable to ensure the wording and definition of attributes and attribute levels and that participants fully understand choice tasks. Researchers can take advantage of software such as Qualtrics, Google Survey, Survey Monkey, or Question Pro to assist in implementing the DCE survey. Notably, after completing the questionnaire design and before data collection, it is important for researchers to ensure that they obtain ethics approval as a DCE involves human participants.

Step 5: Data Collection: After the finalization of the DCE questionnaire, researchers need to decide on the methods to collect DCE data. Common methods used in DCE data collection include face-to-face interviews, telephone interviews, mailed paper questionnaires, online questionnaires, or combining multiple methods (e.g., qualitative interviews followed by an online DCE survey). The advantage of using an online questionnaire in DCE for data collection is that researchers can use the randomization process enabled by survey software for DCE data collection. In certain DCE designs, randomization is needed because the experimental design is too complex to do. Thus, the DCE study is divided into multiple blocks design, and each respondent is randomly assigned to only one of the block designs. R statistical programming offers a powerful suite of packages to assist in the optimal design of choice sets or fractional/full factorial design.

Step 6: Data Analysis: The final step in DCE is data analysis, reporting, and result interpretation. Generally, the analysis of DCE data involves logit-based regression models with a dichotomous or polychotomous categorical dependent variable (i.e., the dependent variable in DCE is usually a choice of alternative A, B, or C). When the choice presented to respondents is binary (a single yes or no vote on an alternative) or comprises two alternatives (i.e., alternative A or B), binary probit or logit (logistic regression) models are more appropriate. When “Neither” or “Opt-out” alternatives are included in the choice set, or there are more than two alternatives, McFadden’s multinomial logit (MNL) regression is more suitable (McFadden, 1974). We will discuss the data analysis process of DCE in more detail in Chap. 9 later.

Figure 3.1 only provides a simplified overview of the key steps involved in conducting a DCE. However, it should be noted that some of these tasks are not strictly sequential but rather inter-related and may influence each other. For example, the experimental design is also closely linked to the identification of attributes and attribute levels. The selection of attributes and their levels is a crucial step in a DCE, as these determine the factors that will be varied in the choice sets presented to participants. The experimental design aims to create a well-balanced and efficient set of choice sets that allow for accurate estimation of preferences and meaningful analysis of the data. The choice of attribute levels and the arrangement of choice sets should be guided by statistical principles, such as orthogonality and efficiency, to optimize the information obtained from participants.

3.5 Various Applications of DCE

DCE experiments have been widely applied in different disciplines. It has become one of the most applied choice modeling methods. In the tables below, we summarized several empirical studies that used the DCE method in transportation economics, healthcare, tourism management, and environmental economics. We highlighted the topics of investigation, selected attributes, number of respondents, and number of alternatives in each choice set in Table 3.3. These will offer benchmarks for any researchers wishing to use DCE.

Table 3.3 Sample DCE applications in different disciplines
Full size table