2026 Theses Master's

Predicting Formal Protest Petitions Against Housing Development: Evidence from Austin, Texas

Lewis, Daniel Hardesty

This thesis investigates whether formal protest petitions against proposed housing developments in Austin, Texas can be anticipated before public hearings, using only the administrative and neighborhood records available at the time of application filing. The central question is whether parcel characteristics, neighborhood demographics, and application features observable by planning staff at filing can meaningfully identify which cases are likely to attract organized opposition.

Using discretionary zoning cases from Austin's pre-HB 24 petition regime (2007-2024), predictive models are trained and evaluated on a temporal holdout that mimics forward administrative use. The results are qualified: filing-date features can rank petition risk above chance -- the best model achieves moderate out-of-distribution discrimination with a wide uncertainty interval (PR-AUC = 0.718, 95% CI: [0.48, 0.91]) -- but the rarity of petitioned cases and the model's limited external validity mean predictions are best used as relative triage rankings, not as case-level probabilities for high-stakes decisions.

The core finding is that administrative data at filing carries real predictive signal for petition risk, yet that signal is too uncertain and context-dependent to support uncritical case-level decisions. The broader contribution is twofold: first, a methodology -- time-stamped features, temporal holdout evaluation, and explainability audits -- that is portable to other discretionary review systems; and second, the

identification of neighborhood agency as a predictive bound. The study finds that the volatility of performance across petition thresholds serves as empirical evidence that mobilization is a strategic social choice that cannot be reduced to mechanistic administrative records.