Topics
Data-driven analysis on food security, transit equity, and public policy. All research includes methodology documentation and replication materials.
AI Methodology
How we do research: data validation, transit routing, API collection, and AI-integrated workflows with code examples.
Logistic regression beats LLM readouts on survey prediction
On a real public-health prediction task, a plain logistic regression on seven demographic facts beat a language-model activation pipeline, AUC 0.769 vs 0.747; steering the model's internals changed nothing about the ranking.
Steering vectors estimate an average regression gradient
Activation steering approximates an average regression gradient: directional alignment at cosine 0.63, and a covariance adjustment lifts the steering effect by 14 percent.
Prediction-powered inference corrects AI-imputed survey estimates
Treating AI-imputed survey responses as real observations can understate a prevalence estimate by a factor of three; a regression adjustment lets predictions sharpen the estimate without ever making it worse.
Food Security
Food access, SNAP enrollment, grocery store distribution, and the complex relationship between geography and food insecurity in California.
The Food Desert Myth
Zero food deserts by federal definition, yet SNAP participation varies 4.7x across similar communities.
Hidden Mobility Deserts
When nearby stores are unreachable: 13 census tracts with grocery stores under 0.5mi but 30+ minutes by transit.
The Vulnerability Paradox
Why high-vulnerability communities sometimes have better geographic food access than low-vulnerability areas.
Transit Equity
How transit access shapes economic opportunity. Mobility measurement, accessibility indices, and transportation barriers.
Hidden Mobility Deserts
2.7M transit routes reveal 13 census tracts where nearby destinations are unreachable by public transit.
Transit Equity and Race
Analyzing racial disparities in transit access across California communities.
Housing Tenure and Transit
How housing stability affects transit access patterns in urban California.
Education Policy
School funding, special education budgets, and how districts respond to fiscal constraints, stimulus funding, and federal mandates.
How California Protected Special Ed, Then Shifted Cuts
California's 2008 funding floor prevented $1.2B in special ed cuts, but districts shifted cuts to general education.
Stimulus Saved Schools, Then Made Things Worse
How ARRA funding created a fiscal cliff for school districts after the Great Recession.
The Unfunded Mandate
IDEA promised 40% federal funding for special education. Four decades later, it's still at 15%.
Health Policy
Medicaid fraud detection, public-coverage data, and the gap between administrative records and the populations they describe.
The Data Landscape
What 227 million rows of Medicaid claims data can and can't tell us about fraud.
The Label Problem
Why training a fraud classifier is harder than it looks: positive-only labels and selection bias.
The Classifier
SHAP-interpreted XGBoost for Medicaid fraud detection, with positive-unlabeled learning.
Methods & tools
- AI research workflows — Claude Code guides for academic research.
- Open-source methods — reproducible code and validation studies.
- Grocery tier validation — a scalable food-environment quality method.
- Glossary — methods and terms used across the research.
- Tools & data — public analysis repositories.
Get Research Updates
New analysis delivered to your inbox. Policy-relevant findings with methodology. No spam, unsubscribe anytime.
Subscribe for Free