The best-financed, highest-profile CV study to date was conducted after the Exxon Valdez oil spill. Its central findings, reproduced in Table 1, illustrate CV's biases. After 4 describing the accident and the policy to prevent another, surveyors randomly assigned respondents to one of four versions of the questionnaire. In version A, respondents were first asked if they would support the policy if it cost them $10. If they said yes, the follow-up asked if they would support the policy if it cost $30. If instead they said no, the follow-up asked if they would support the policy at $5. Version B started at $30 and asked follow-ups at $10 and $60, and so on. Here's the problem. Suppose we want to estimate the fraction of the population willing to pay between $30 and $60 for the policy. There are multiple equally valid ways to calculate that number. Those who answered yes to the first question of version B but no to the follow-up have expressed willingness to pay at least $30 but not $60. They represent 26 percent of the randomly chosen respondents given version B of the survey. People who answered no to the first question of version C but yes to the follow-up are not willing to pay $60 and are willing to pay $30. They represent 10 percent of version C recipients. Or here's a third way. Of those given version B, 52 percent answered yes to the first question, and are willing to pay at least $30. Of those given version C, 50 percent answered no to the first question, and are unwilling to pay $60. The difference suggests only 2 percent were willing to pay between $30 and $60. So which is it, 26 percent, 10 percent, or 2 percent? The answer clearly depends on the sequence with which questions are asked. Something distinctly behavioral is biasing the respondents' answers.