Conference Agenda

Discrete choice experiments (DCEs) typically elicit a single “best” alternative per choice task, mirroring market settings in which only one chosen option is observed. Alternative elicitation formats—such as Best–Worst and Best–Best—collect additional preference information per task and are often motivated by gains in statistical efficiency. Yet efficiency gains require that multi-stage responses be behaviourally consistent and that the elicitation protocol not distort choice behaviour. We provide a systematic split-sample comparison of three elicitation formats—Best-only, Best–Worst, and Best–Best—in a DCE valuing peatland restoration outcomes in Scotland. Respondents were randomly assigned to treatments that differ only in the elicitation protocol, while the choice tasks and experimental design were held constant. Using mixed logit models estimated in preference space (with treatment-specific scale) and willingness-to-pay space, we test two questions. First, does the elicitation format affect first-best choices, even when the first question is identical across treatments? Second, are preferences recovered from follow-up responses equivalent when the follow-up elicits the “worst-of-two” versus the “best-of-two,” which are informationally equivalent in theory? We find that (i) first-best preferences are not invariant to the elicitation format—consistent with an observer effect—and (ii) follow-up responses reveal format-dependent preference information, including systematic differences between worst-of-two and best-of-two framing. The results caution against treating ranked-elicitation formats as an “insurance strategy” for small samples and have direct implications for welfare measurement and DCE design.