UC Berkeley

The increasing availability of big data and recent developments in causal inference both open new opportunities in natural resource economics, but not without appropriate tools. This dissertation consists of three papers, each of which make progress in econometric techniques for the use of big data in natural resource economics. In chapter one, I show that predicted datasets such as those commonly used for crop-specific land use data are commonly afflicted by misclassification, which can threaten empirical results derived from this data. I then introduce a Bayesian nonparametric multiple imputation technique that enables making use of imperfect predicted data when a small amount of validation data is available, achieving consistency with narrower confidence intervals than would be possible with validation data alone. Chapter two estimates the recreational value of a rare species at a forest site, using a change in taxonomic status as a natural experiment. Though this paper uses the long-established logic of travel cost demand modelling for this valuation exercise, the specific implementation of the travel cost demand model is novel, using a difference-in-differences approach rather than a discrete choice random utility model. This avoids the need to collect data for or even characterize the entire choice set and enables use of the natural experiment. In chapter three, I theoretically extend this development, and argue for the adoption of a Poisson ratio-in-ratios model (the multiplicative equivalent of the additive difference-in-differences model) for estimating welfare changes resulting from random shocks affecting recreational sites generally.