Abstract |
There is no established protocol for the analysis of data collected in contingent valuation surveys. This has led to seemingly arbitrary removal of some outlier observations, which leads to questions about the representativeness of the results reported in the literature. The report critiques the methods researchers have used to analyze the data from contingent valuation surveys and suggests procedures for future analyses. Three procedures regress the responses on a set of economic and demographic variables (e.g., income, age, sex, prices). Then if the unweighted residuals appear to be normally distributed, analysis should proceed. The first process would drop outliers (iteratively) that are more than (say) six standard deviations from the mean response. The second process would weight observations close to the regression line more heavily than those further away, so that those with true extreme preferences are not eliminated. The third procedure would incorporate known independent sources of bias in the regression. A forth approach would examine the median response rather the mean; this would be more like a voting outcome but ignores extreme preferences. Testing of competing approaches is recommended. |