Abstract
To explain persistent disparities in US electric vehicle adoption, this dissertation develops an innovative, scalable framework that models the mediating mechanisms of public perception in linking structural conditions to adoption outcomes by fusing large-scale social media data with state-level indicators. It introduces novel Large Language Model-assisted methods for high-accuracy sentiment analysis and fine-grained thematic identification, revealing that sentiment-adoption divergence must be examined through detailed perceptual channels. The final study integrates these text-derived perceptual measures with numerical predictors in a time-lagged panel Structural Equation Model to quantify the pathways. Bridging computational social science and econometrics, the research provides tailored and actionable insights for advancing more equitable policy design.
