Contingent Valuation and Revealed Preference Methodologies: Comparing the Estimates for Quasi-Public Goods
A literature search provides 83 studies from which 616 comparisons of contingent valuation (CV) to revealed preference (RP) estimates are made.Summary statistics of the CV/RP ratios are provided for the complete dataset, a 5% trimmed dataset, and a weighted dataset that gives equal weight to each study rather than each CV /RP comparison.For the complete dataset, the sample mean CV/RP ratio is 0.89 with a 95% confidence interval [0.81-0.96]and a median of 0.75.For the trimmed and weighted dat:asets, these sumrnarJ statistics are (0.77; [0.74~0.81];0.75) and (0.92; [0.81-1.03];0.94), respectively.Tne Spearman rank correlation coefficients between the CV and RP estimates for the three datasets are 0.78, 0.88, and 0.92, respectively, with the Pearson correlations a bit larger.Non-parametric density estimates are provided, as well as the results of regressions of the observed CV /RP ratios on the basic RP technique used and the broad class of goods valued. STUDY INCLUSION CRITERIATo help find studies that contain both CV and RP estimates, we systema!icttJly reviewed entries in the Carson et al. (1994) bibliography of over 1.600 contingent valuation papers.To be eligible for inclusion in our sample, a study must provide at least one contingent valuation estimate and one revealed preference estimate for essentially the same quasi-public good; thus, no studies of private goods (e.g., Neill et al., 1994) are included.The goods valued are various forms of recreation (mostly outdoor), changes in health risks, and changes in environmental amenities such as air pollution, noise pollution, water pollution, or parks.Consumers (individuals or households) had to have been interviewed to obtain contingent valuation estimates.Thus, we did not include studies where the respondents were not consumers such as Bohm's (1984) study of local governments' willingness to purchase statistical information.Furthermore, we considered only contingent valuation estimates of willingness to pay (WTP); we excluded estimates based on willingness to accept compensation or on contingent behavior responses. 5Otherwise, we have tried to include all available study estimates.The studies we examined span almost thirty years, 19661994.The earliest study is Knetsch and.Davis' (1966) well-known contingent valuation-travel cost comparison of outdoor recreation in Maine.The latest stud.yconsidered is Choe, Whittington, and Donald (1.994) who value the opening of a polluted urban beach in the Philippines.Due to well-known, potential biases in relying upon only the published literature to summarize research findings, we spent considerable effort searching the unpublished literature SWe do include CV estimates derived from willingness to drive questions if they were intended to be: .directlycompared to a travel cost estimate.CV questions phrased in terms of willingness to give up other gOQds .arenot included.No comparisons between CV willingness to pay ~estimates,and ~~tual willingness to . .ac:ceptcoiJlpellS8tion (e.g . BishQp and Heberlein, 1979) are used.However, our initial investigation ~ggested that CVIRP r~tios in,