ABSTRACT
Most narrowly defined, reliability refers to the extent to which the variance of the WTP amounts given by respondents in a contingent valuation survey is due to random sources, or "noise." If we have the results of a CV survey, the variance in the WTP amounts given for the good is the result of three principal factors. The first is the "true" underlying variation in willingness to pay for the good in the population of integer. Even if the values for the good were measured by a perfect instrument applied to a perfect sample, some people will be willing to pay more than others for the good. The second factor is the instrument: its concepts, its wording, its method of presentation. Imagine that the CV survey is administered at two different times, a few weeks apart, to exactly the same sample under identical conditions; and assume that when the survey is administered the second time the respondents have no memory of the WTP amounts they gave the first time (however unlikely), and that their preferences for the good in question have not changed in the interval. In this classic test-retest situation, the difference between the responses at time one and time two will reflect the true variance plus some noise introduced by the instrument's imperfections. The third source of variance in the mean WTP amount concerns the fact that we have only interviewed a sample of the population. Our sample of respondents is surrogate for a much larger population—perhaps 200,000, or even 240,000,000. In order to accurately represent the population, the sample was selected by rigorously following accepted random sampling procedures. If we had a perfect instrument and selected ten independent samples using identical random sampling procedures, the variance among the ten estimated willingness-to-pay amounts would reflect the variance introduced by our sampling procedures.
