Choosing attributes and levels for your DCE
The attributes and levels in a DCE determine the quality of the preference estimates. Poor attribute selection cannot be corrected in analysis.
How to select attributes and levels for a DCE, what makes a good attribute, and the common mistakes that compromise study validity.
Knowledge Base -> Foundations -> Methods & Academic
Ben White, 07.07.2026
The problem with attributes and levels
Attribute selection is the most consequential decision in DCE design. The attributes define what the study measures - if the wrong attributes are included, or the right attributes are described poorly, the preference estimates will not answer the research question regardless of how well the rest of the study is executed.
Levels define the range and granularity of variation in each attribute. Levels that are too close together produce insufficient variation to identify preferences; levels that are implausible reduce respondent engagement and produce biased estimates.
Why attribute selection determines study validity
Attribute selection errors are not recoverable in analysis. A study can have a perfectly efficient experimental design, rigorous fieldwork, and correct model specification, and still produce invalid estimates if the attributes do not reflect what actually drives choices.
This is why qualitative precursor research is not optional. Interviews or focus groups with members of the target population identify the attributes that matter, the language used to describe them, and the ranges of levels that respondents find credible.
Selecting attributes and levels in SurveyEngine
Step 1: Conduct qualitative research. Use interviews or focus groups with the target population to identify relevant attributes. Ask directly what factors influence their decisions, and probe for attributes that might be implicit or taken for granted.
Step 2: Apply attribute inclusion criteria. Each attribute must be: relevant to the decision, understandable to respondents, independently variable from other attributes, and policy-actionable. Exclude attributes that fail any of these criteria.
Step 3: Set level ranges. Levels should span the range of values respondents might realistically encounter, from the current or worst-case value to the best achievable improvement. Use quantitative values where possible rather than qualitative descriptors.
Step 4: Check for dominance. If one alternative is unambiguously better than all others across all attributes, respondents will always choose it regardless of the other alternatives. Dominated alternatives produce no useful preference data.
Worked example - treatment preference in oncology
A pharmaceutical company designing a patient preference study for a treatment with three dosing options - daily oral, weekly injection, monthly infusion - conducts eight cognitive interviews before the main study. Interviews reveal that administration frequency is less important to patients than injection site reactions and the requirement to attend a clinic.
The attribute list is revised from the clinical team's initial specification to include clinical attendance requirements and injection site reaction severity alongside the originally planned attributes. The revised study produces preference estimates that are directly relevant to the prescribing context and are accepted in the regulatory submission.
References
Lancsar, E. and Louviere, J. (2008). Conducting discrete choice experiments to inform healthcare decision making. PharmacoEconomics, 26(8), 661–677.
Mangham, L.J., Hanson, K. and McPake, B. (2009). How to do (or not to do)... Designing a discrete choice experiment for application in a low-income country. Health Policy and Planning, 24(2), 151–158.
Ready to design your attribute set? Log in to SurveyEngine and use the design generator to check D-efficiency as you add attributes and levels.
Or Contact us at support@surveyengine.com — we're glad to help.