Modeling Discrete Choice with Uncertain Data: an Augmented Mnl Estimator

When estimating the value of natural resources, the applied analyst must often work with noisy or otherwise imprecise measures of both dependent and independent variables. To help control for this uncertainty, this article introduces a multinomial logit model (MNL) that uses ancillary information to control for uncertainty in observed choices and in the attributes of a respondent’s choice set. For example, surveys of rural recreationists may encounter difficulties when identifying exactly where people visited—since many sites may be “informal” (such as the “the park down by the river”).1 Second, even when sites can be identified, good information on site attributes may be lacking; a problem that is more likely to be true for “informal” sites (such as sites that are not intensively managed by government agencies). When faced with such a problem, one can construct “regions,” and assume that a respondent chooses to visit a region, rather than a particular site (Feather, Hellerstein, and Hansen). Environmental attributes, such as water quality or land use, are often available on a