What makes a good experimental design?
D-efficiency measures how precisely a design estimates preference parameters. Most researchers use it without understanding what drives it.
What D-efficiency is, how it is calculated, and how to generate D-efficient designs using Ngene in SurveyEngine.
Knowledge Base -> Experiment Design -> Methods & Academic
Ben White, 07.07.2026
What is D-efficiency and why does it matter?
The experimental design determines which combinations of attribute levels appear together in each choice set. The choice of design determines how precisely the model parameters can be estimated from a given number of responses.
D-efficiency is a summary measure derived from the determinant of the variance-covariance matrix of parameter estimates. A D-efficient design minimises the volume of the confidence ellipsoid around the parameter estimates - equivalently, it minimises the sample size required for a target level of precision.
D-efficient vs orthogonal designs
Orthogonal designs ensure attribute levels are statistically independent across choice sets. While this ensures balanced data, it does not guarantee efficient data. An orthogonal design with 20 choice sets may be less efficient than a D-efficient design with 12.
D-efficient designs minimise parameter uncertainty subject to the design constraints. For a given number of choice sets and respondents, a D-efficient design produces narrower confidence intervals than an orthogonal design. The gain is typically 20-40% reduction in required sample size for the same precision.
Generating a D-efficient design in SurveyEngine
Step 1: Enter attributes and levels in the SurveyEngine's Ngene Design Generator and specify the number of choice sets and alternatives per set and whether to include a no-choice option.
Step 2: Set priors if available from a pilot study. Use zero priors if no prior information exists.
Step 3: Generate the design. A generalised swapping algorithm or the modified Fedorov algorithm returns the D-efficient design. Check the D-error – finite values, often below 1, indicate that there is no multicollinearity or issue with model identifiability.
Step 4: Embed consistency checks and practice tasks manually after generation - the design generator does not include these automatically.
Worked example - comparing design options
A researcher designing a DCE with 5 attributes at 3 levels compares an orthogonal design against an efficient design. The orthogonal design requires 18 choice sets to satisfy orthogonality conditions. The efficient design makes smarter trade-offs in the choice tasks and can achieve the same D-error with 12 choice sets.
At 200 respondents, the orthogonal design collects 3,600 choices; the D-efficient design collects 2,400. Both data result in similar standard errors of the model parameter estimates, but the efficient design reduces respondent fatigue effects in later choice sets. If each respondent is given a subset of 6 choice sets, then the efficient design would require a 50% smaller sample size to obtain the same precision
References
Rose, J.M. and Bliemer, M.C.J. (2009). Constructing efficient stated choice experimental designs. Transport Reviews, 29(5), 587–617.
Huber, J. and Zwerina, K. (1996). The importance of utility balance in efficient choice designs. Journal of Marketing Research, 33(3), 307–317.
Ready to generate a D-efficient design for your DCE? Log in to SurveyEngine and use the built-in design generator.
Or Contact us at support@surveyengine.com — we're glad to help.