Applications of Forest-Based Machine Learning Methods in Environmental Economics

Abstract

This dissertation sits within applied environmental economics and relies on advanced forest‑based machine‑learning methods. The first two papers use a double‑residualized Causal Forest framework to analyze a contingent‑valuation survey on water quality, estimating how self‑selected respondent behaviors—interacting with interactive maps or issuing protest responses—affect the probability of voting for a hypothetical cleanup program. The third paper turns to a Local Linear Forest for high‑accuracy prediction, linking Residential Energy Consumption Survey data to high‑resolution climate grids to identify U.S. "dual‑vulnerability" hotspots where households face both extreme temperatures and high energy burdens. Together, the papers show how flexible forest algorithms can either isolate causal effects in stated‑preference studies or deliver precise forecasts in energy‑expenditure modeling, thereby deepening our understanding of how environmental factors influence economic decisions.