# Discrete Choice and Conjoint Analysis: Measuring What People Actually Prefer Source: PaperSurvey.io Blog URL: https://www.papersurvey.io/blog/discrete-choice-conjoint-analysis-surveys --- Ask people to rate features on a scale of 1 to 5 and almost everything comes back important. Price matters, quality matters, speed matters, support matters. Rating questions are easy to answer and easy to inflate, because nothing forces the respondent to give anything up. Real decisions are different. When someone chooses a phone, a health plan, or a job offer, they cannot have the cheapest price and the best quality and the longest warranty at once. They trade off. Discrete choice experiments (DCEs) and conjoint analysis are built to measure exactly that trade-off: instead of rating attributes one at a time, respondents choose between complete alternatives, and the choices reveal how much each attribute is really worth. This is one of the best-validated methods in applied survey research, and it works on paper as well as online. Here is how it works, when to use it, and how to design a choice question that produces usable data. ### Rating vs Choosing A rating question measures stated importance. A choice question measures revealed preference within the survey. The difference is not cosmetic. Green and Rao (1971) introduced conjoint measurement to marketing precisely because direct importance ratings were unreliable: respondents rate everything as important, and the ratings do not predict behavior well. By presenting whole product profiles and asking people to evaluate or choose among them, you recover the implicit weight each attribute carries. The theoretical foundation is random utility theory, formalized by McFadden (1974), whose work on discrete choice modeling earned the Nobel Prize in Economics in 2000. The core idea is simple: each alternative has a utility made up of the utilities of its attribute levels, plus an error term, and people choose the alternative with the highest utility. From a set of observed choices you can estimate the part-worth utility of every attribute level, which tells you how much each one contributes to the decision. The practical payoff is that DCE results are expressed in the units decision-makers care about. You can rank attributes by importance, estimate willingness to pay for a feature, and simulate the market share of product concepts that do not exist yet. ### The Anatomy of a Choice Question Every discrete choice question is built from three ingredients: **Attributes** are the features being compared. For a phone: price, battery life, screen size, brand. These become the rows of the choice table. **Levels** are the possible values of each attribute. Price might be $299, $499, or $699. Battery life might be 12, 20, or 30 hours. The combination of one level per attribute defines an alternative. **Alternatives** are the complete profiles the respondent chooses between, shown as the columns. A typical task presents two or three alternatives, sometimes with a "None of these" opt-out so respondents are not forced to pick something they would never buy. A single choice task asks the respondent to pick one alternative. A full study repeats this across several tasks, each showing a different combination of levels, so that across the whole design every attribute level appears against the others enough times to estimate its effect. ### How Many Attributes, Levels, and Tasks The most common design mistake is asking respondents to compare too much at once. Discrete choice works because it mirrors a real decision, and real decisions become noisy when the cognitive load is too high. The ISPOR Good Research Practices task forces on conjoint analysis (Bridges et al., 2011; Johnson et al., 2013), which set widely used standards in health economics, recommend keeping the number of attributes manageable and the levels realistic and clearly ordered. In practice most well-run studies use around four to six attributes. Beyond that, respondents start simplifying, often by focusing on one or two attributes and ignoring the rest. For the number of alternatives per task, two or three is standard. More columns fit on a screen or page but increase the effort of each comparison. For the number of tasks, published DCEs commonly use somewhere between eight and sixteen choice tasks per respondent. Fewer tasks give you less information per person; more tasks increase fatigue and dropout. Orme (2010), whose practical guidance underpins much commercial conjoint work, emphasizes balancing statistical efficiency against respondent burden rather than maximizing tasks. **The practical recommendation**: Start with four to five attributes, three to four levels each, two to three alternatives per task, and eight to twelve tasks. Pilot it, and cut anything that makes respondents hesitate. ### Where Discrete Choice Beats a Rating Grid Discrete choice is not always the right tool. It is more work to design and analyze than a Likert grid, and for simple satisfaction measurement a rating scale is perfectly adequate. Reach for a choice experiment when the decision involves genuine trade-offs and you need to quantify them: - **Pricing research.** How much are customers willing to pay for a longer warranty or faster delivery? A choice design lets you put a monetary value on non-price features. - **Product and concept testing.** Which combination of features would win the most share, and which features are dispensable? - **Health and policy preferences.** Ryan and Farrar (2000) showed that conjoint analysis could elicit patient preferences for aspects of care that standard satisfaction surveys missed entirely. DCEs are now a standard method in health technology assessment. - **Employment and benefits.** What mix of salary, remote work, and vacation would attract the strongest candidates? In each case the value comes from forcing the trade-off. Green and Srinivasan (1990), reviewing two decades of conjoint practice, noted that its enduring advantage over self-reported importance is that it recovers the weights people act on rather than the ones they report. ### Discrete Choice on Paper Choice experiments are usually associated with online panels, but there is nothing about the method that requires a screen. A choice task is a small table: attributes down the side, alternatives across the top, and a single checkbox per alternative. That prints and scans cleanly. Paper administration is valuable when your population is not reliably online, such as patients in a clinic waiting room, attendees at an in-person event, employees on a factory floor, or residents in a rural community. It also avoids the device-dependent layout problems of web forms, where a choice table that looks balanced on a laptop can wrap awkwardly on a phone (Louviere, Hensher, and Swait, 2000, stress that the visual presentation of the choice set is part of the stimulus and should be held constant). The main paper-specific consideration is fatigue. Each choice task takes real effort, so a long paper booklet of twenty tasks will see quality decline toward the end. Keep the task count moderate, make the table visually clean, and give respondents a clear instruction to mark one alternative per task. ### Running a Discrete Choice Question in PaperSurvey PaperSurvey.io includes a dedicated discrete choice question type that prints as a clean choice table and reads back automatically through optical mark recognition, the same way a single choice question does. Setup is structured rather than free-form. You list the **attributes** you want to compare, then add each **alternative** with its value for every attribute. The editor keeps the values aligned to the attributes and shows a live preview of the table as you build it. [Image: Setting up a discrete choice question in PaperSurvey] Each choice question represents one choice task, with one checkbox per alternative and an optional opt-out alternative such as "None of these." To present several tasks, you add a separate discrete choice question for each one, varying the levels between them. Because each alternative is a real scannable checkbox tied to the question, a marked paper form comes back as a clean selection with no manual data entry. [Image: A printed discrete choice task] Responses are stored the same way as any other choice question, so the chosen alternative appears in your results, exports, and the responses table by its alternative name. And because PaperSurvey supports both paper and web collection for the same survey, you can run the identical choice design in the field on paper and online through a link, then analyze the combined data together. ### Analyzing the Results The output of a choice experiment is a set of choices, and the analysis turns those choices into part-worth utilities using a discrete choice model (most commonly a multinomial or mixed logit). From the utilities you derive the quantities that matter to decision-makers: - **Attribute importance**, the relative weight each attribute carries in the decision. - **Willingness to pay**, obtained by expressing the utility of a feature in the units of the price attribute. - **Preference share simulation**, predicting how a set of product concepts would split choices. For a first study you do not need advanced modeling to get value. Simple counts of how often each alternative and each level was chosen already reveal the dominant drivers. The formal logit estimation refines those insights and lets you simulate concepts, and it is well supported in standard statistics packages. Because PaperSurvey exports your data to CSV, Excel, SPSS, Stata, SAS, and R, you can take the collected choices straight into whichever tool you use for choice modeling. ### Common Mistakes - **Too many attributes.** Respondents cope by ignoring most of them. Keep it to what genuinely drives the decision. - **Unrealistic level combinations.** If your design shows the best phone at the lowest price, respondents learn to trust the exercise less. Some designs deliberately exclude implausible combinations. - **Dominated alternatives.** If one alternative is better on every attribute, the choice is not informative. Vary levels so each alternative wins on something. - **No opt-out when one is realistic.** Forcing a choice when "I would buy none of these" is a real option inflates demand for every alternative. - **Overlong task sequences.** Fatigue degrades the later tasks. Fewer, cleaner tasks beat a long grind. ### The Bottom Line Rating scales measure what people say is important. Discrete choice experiments measure what they would actually choose when they cannot have everything, which is almost always the question you really care about. The method has decades of validation behind it, from McFadden's random utility theory to its routine use in pricing, product design, and health policy, and it runs on paper as readily as online. If your research involves trade-offs, a choice question will tell you more than a page of importance ratings ever could. ### References - Bridges, J. F. P., Hauber, A. B., Marshall, D., Lloyd, A., Prosser, L. A., Regier, D. A., Johnson, F. R., & Mauskopf, J. (2011). Conjoint analysis applications in health, a checklist: A report of the ISPOR Good Research Practices for Conjoint Analysis Task Force. *Value in Health*, 14(4), 403-413. - Green, P. E., & Rao, V. R. (1971). Conjoint measurement for quantifying judgmental data. *Journal of Marketing Research*, 8(3), 355-363. - Green, P. E., & Srinivasan, V. (1990). Conjoint analysis in marketing: New developments with implications for research and practice. *Journal of Marketing*, 54(4), 3-19. - Johnson, F. R., Lancsar, E., Marshall, D., Kilambi, V., Muhlbacher, A., Regier, D. A., Bresnahan, B. W., Kanninen, B., & Bridges, J. F. P. (2013). Constructing experimental designs for discrete-choice experiments: Report of the ISPOR Conjoint Analysis Experimental Design Good Research Practices Task Force. *Value in Health*, 16(1), 3-13. - Louviere, J. J., Hensher, D. A., & Swait, J. D. (2000). *Stated Choice Methods: Analysis and Applications*. Cambridge University Press. - McFadden, D. (1974). Conditional logit analysis of qualitative choice behavior. In P. Zarembka (Ed.), *Frontiers in Econometrics* (pp. 105-142). Academic Press. - Orme, B. K. (2010). *Getting Started with Conjoint Analysis: Strategies for Product Design and Pricing Research* (2nd ed.). Research Publishers LLC. - Ryan, M., & Farrar, S. (2000). Using conjoint analysis to elicit preferences for health care. *BMJ*, 320(7248), 1530-1533. [Start your free trial](https://www.papersurvey.io/app/auth/register) and build your first discrete choice question for paper or web. --- Need more? The complete PaperSurvey.io help center and blog is available as a single document at https://www.papersurvey.io/llms-full.txt.