Identifying inconsistent responses in dichotomous choice contingent valuation with follow-up questions

We develop a new but simple non-parametric method to diagnose inconsistency in double-bounded contingent valuation questions in the presence of both perfect and imperfect correlation between initial and follow-up response distributions. The proposed method can identify inconsistency in iterative responses at each bid interval. We apply this method to data from five well-known double-bounded contingent valuation surveys. The predictions of our model match closely with parametric outcomes. Further, we find that the inconsistency patterns generally vary for different data sets and different bid intervals within data sets. Therefore no single behavioral model can explain all latent inconsistency patterns either within or across data sets. In addition, we examine the impact of inconsistency in responses on bias and efficiency of the double-bounded format in the absence of correction for inconsistency. We conclude that the commonly cited benefits of the double-bounded format hold only in the special case of nearly perfect consistency between initial and follow-up response distributions. Our method provides a simple tool researchers can use to determine the similarity in response distributions between the initial and follow-up responses and whether incorporating collected follow-up responses are likely to actually increase efficiency without introducing bias.