A NOTE ON NONLINEARITY BIAS AND DICHOTOMOUS CHOICE CVM: IMPLICATIONS FOR AGGREGATE BENEFITS ESTIMATION

It is a generally known statistical fact that the mean of a nonlinear function of a set of random variables is not equivalent to the function evaluated at the means of the variables. However, in dichotomous choice contingent valuation studies a common practice is to calculate an overall mean (or median) by integrating over offer space (numerically or analytically) an estimated logit or probit function in which sample mean values for the concomitant variables are used. We demonstrate this procedure to be incorrect and we statically test the procedure against the correct method for nonlinear models. Using data resulting in a well-behaved logit model, we reject the hypothesis of congruence between the two means. Such a finding should be considered in future single response dichotomous choice CVM studies, particularly when aggregation is of interest.