Abstract
Using contingent valuation methods (CVM) is increasingly common in project analysis. Referendum-type questions are thought to be easier to answer than the open-ended variety, but there is a downside: econometric techniques must be applied to the referendum data to infer the mean or median willingness to pay (WTP) of the sample and, thus, of the population of potential beneficiaries. This is not just a technical point as is demonstrated with data from a referendum CV study of a sewage and wastewater treatment project in Brazil. A factor of 4 separates lowest from highest central tendency estimates, ignoring the outlier that is 14 times larger than the largest of the other figures. This is ample variation to make a difference in the cost-benefit analysis. Analysts using referendum CV must be sensitive to the problems they buy into, and must decide how to deal with the uncertainty in the results.