Handling Protest Responses in Contingent Valuation Surveys.

OBJECTIVES Protest responses, whereby respondents refuse to state the value they place on the health gain, are commonly encountered in contingent valuation (CV) studies, and they tend to be excluded from analyses. Such an approach will be biased if protesters differ from non-protesters on characteristics that predict their responses. The Heckman selection model has been commonly used to adjust for protesters, but its underlying assumptions may be implausible in this context. We present a multiple imputation (MI) approach to appropriately address protest responses in CV studies, and compare it with the Heckman selection model. METHODS This study exploits data from the multinational EuroVaQ study, which surveyed respondents' willingness-to-pay (WTP) for a Quality Adjusted Life Year (QALY). Here, our simulation study assesses the relative performance of MI and Heckman selection models across different realistic settings grounded in the EuroVaQ study, including scenarios with different proportions of missing data and non-response mechanisms. We then illustrate the methods in the EuroVaQ study for estimating mean WTP for a QALY gain. RESULTS We find that MI provides lower bias and mean squared error compared with the Heckman approach across all considered scenarios. The simulations suggest that the Heckman approach can lead to considerable underestimation or overestimation of mean WTP due to violations in the normality assumption, even after log-transforming the WTP responses. The case study illustrates that protesters are associated with a lower mean WTP for a QALY gain compared with non-protesters, but that the results differ according to method for handling protesters. CONCLUSIONS MI is an appropriate method for addressing protest responses in CV studies.