Not Another Politics Podcast The Future of Empirical Research in the Age of AI
20 snips
Feb 6, 2026 Andy Hall, a Stanford political scientist who used Claude and Claude Code to replicate and extend a vote-by-mail study. He recounts prompting strategies, how Claude mostly reproduced results, where automated code stumbled, the audit process that caught data and legal-timing errors, and wider implications for research speed, oversight, and how AI might reshape empirical social science.
AI Snips
Chapters
Transcript
Episode notes
AI Generated The Work Plan
- Andy asked Claude to plan and generate instructions.md to update the paper through to the present.
- He then launched Claude Code to run the data collection and replication with minimal intervention.
AI Can Faithfully Replicate Analyses
- Claude Code replicated the original estimates correctly and updated regressions with most new data.
- The extended results matched earlier conclusions: similar turnout effect and near-zero partisan impact.
AI Suggested Unguided Robustness Checks
- Claude proposed extra analyses like California-specific estimates and event studies without clear justification.
- Some suggested robustness checks were vague and led Claude Code to implement odd choices.