Observations with protest votes in contingent valuation surveys could result in biased estimates of the willingness to pay measure, if treated as expressions of genuine indifference to public good or simply removed from the analysis. We propose an elicitation design and a sample selection model that allow correction of the possible bias due to protest votes. An application shows that selection bias can sensibly affect the estimate of the value of the public good. Inference in sample selection models may sometimes be difficult due the flatness of the likelihood function that causes uncertainty about the presence of selection bias. Even so, we recommend use of the sample selection model, since it takes into account this uncertainty.

Statistica; Vol 61, No 3 (2001)