A Sample Selection Model for Protest Non-Response Votes in Contingent Valuation Analyses
A Sample Selection Model for Protest Non-Response Votes in Contingent Valuation Analyses|
Year of Publication|
Calia, P, Strazzera, E|
pay some money for a public good, for reasons that differ from a genuine indifference to the good. For example, some people may dislike the idea of placing monetary values to public goods like the environment or a historical monument. Some may protest against the inefficiency of the public administration in managing public funds, and refuse to pay a tax. Others may behave strategically, if they think that their answer could influence the actual level of taxation. A good survey design can effectively reduce them, but protest votes can hardly be completely removed from the dataset. The question is how to deal with them. Sometimes they are considered as true zero values, or, if the dichotomous choice method is used, as if they were below the minimum bid. Obviously, if the unwillingness to pay reflects only a protest and not a low or null valuation of the good, this procedure results in downward biased estimates of the wtp measure. It is of paramount importance that the questionnaire contains a follow up question for individuals that refuse payment, to investigate about their motivations, and interpret the responses. Alternatively, observations with protest votes are simply cut off the sample, and only the subsample with positive reservation prices is considered in the analysis. This procedure will not have any effect for the validity of the estimates only if there is no sample selection bias. Otherwise, it leads to incorrect estimates for the willingness to pay. In this paper we present a sample selection model that allows to take into account, and correct, the possible bias due to protest votes. It is shown that selection bias can sensibly affect the estimates for the willingness to pay for the public good. It will be seen that the model may present estimation problems because of flatness of the likelihood function. In some cases confidence intervals around the sample selection coefficient are too wide to give evidence of presence or absence of sample selection bias. It is argued that even in these circumstances the sample selection model with the protest votes should be preferred to the model without protest votes, since it takes into account the uncertainty about the estimates of the willingness to pay.