Abstract
Data on willingness-to-pay (WTP) collected from contingent valuation surveys are usuallycensoredat zero. Insuch cases, ordinary least squares estimationof the WTP equation produces inconsistent parameter estimates. The maximum likelihood estimation of the Tobit model, which is widely used in this case, is not robust to heteroscedasticity and non-normal error structure. A least absolute deviations estimator allows for the censored data structure and is robust to these problems. In addition, the technique is useful in the case of small amounts of data. It is applied to household survey data in which the contingent valuation question involved the benefit of greenhouse gases reduction policy in Korea.