Eliciting a Counterfactual Sensitivity Parameter

Sensitivity analyses, wherein estimation is performed for each of a range of values of a sensitivity parameter, are of particular use in causal inference, where estimands often are identified because of untestable assumptions. Sensitivity parameters may have counterfactual interpretations, making their elicitation especially challenging. This article describes our experience eliciting a counterfactual sensitivity parameter to be used in the analysis of an ongoing HIV vaccine trial. We include instructions given to 10 subject-matter experts, the chosen ranges of our eight responders and some of their comments, their ranges applied to data from an earlier trial, and a brief discussion of some general issues regarding counterfactual elicitation.