Labelled vs unlabelled designs in health DCEs
Labelled and unlabelled DCEs measure different things. The choice between them has direct consequences for what the estimates can and cannot say.
The difference between labelled and unlabelled DCE designs in health preference research, when each is appropriate, and what each produces.
Knowledge Base -> Experiment Design -> Health
Ben White, 07.07.2026
Labels vs generic alternatives
In an unlabelled DCE, alternatives are generic options described entirely by attributes. In a labelled DCE, alternatives carry real names - pembrolizumab, nivolumab, standard of care - importantly the label itself carries utility.
Labelled alternatives capture brand effects and label-specific preferences that unlabelled designs cannot measure. But label effects complicate interpretation - it may difficult to separate the utility of the label from the utility of the attributes unless an alternative specific design is used.
When labels matter and when they don't
Unlabelled designs produce pure attribute utility estimates uncontaminated by brand effects - appropriate for early-stage research, policy analysis requiring generic estimates, and regulatory submissions for treatments not yet approved.
Labelled designs are necessary when alternatives represent real competing products, when the status quo must be a specific named option, or when the research question involves preferences for specific named treatments.
TLDR Quick links
Setting up labelled and unlabelled designs in SurveyEngine
Step 1: Identify the research question. 'What attributes matter and by how much' - use unlabelled. 'How do patients choose between specific named treatments' - use labelled.
Step 2: Document the design choice rationale explicitly in the methods section. Reviewers expect justification for either choice.
Step 3: In labelled designs, the status quo or no-choice option utility must be estimated as a separate parameter.
Step 4: Test for IIA violations in labelled designs where alternatives may be correlated. Test with a nested logit specification.
Step 5: Report label parameters - alternative-specific constants - explicitly alongside attribute parameters.
Worked example - labelled vs unlabelled in oncology
A pharmaceutical company needs patient preference evidence for both an FDA submission and market access negotiations. Two separate studies are designed: an unlabelled DCE for the FDA covering efficacy, tolerability, and administration; a labelled DCE comparing the new treatment against two existing standard-of-care options for market access.
The unlabelled study produces pure attribute WTP estimates accepted in the FDA submission. The labelled study produces market share predictions for specific competitive scenarios used in pricing and contracting.
References
Hensher, D.A., Rose, J.M. and Greene, W.H. (2015). Applied Choice Analysis. Cambridge University Press.
Bridges, J.F.P. et al. (2011). Conjoint analysis applications in health. Value in Health, 14(4), 403–413.
Deciding between labelled and unlabelled designs? Contact SurveyEngine's health research team for guidance.
Or Contact us at support@surveyengine.com — we're glad to help.