Skip to content

Empirical Methods and AI

The intersection of AI and empirical methods raises important methodological questions about how AI tools affect the rigor and validity of empirical research in economics.

Context & Background

AI makes it dramatically easier to run many specifications, test numerous hypotheses, and explore data in ways that raise concerns about:

  • P-hacking acceleration: AI can quickly generate hundreds of regression specifications, making it tempting to search for significant results
  • Specification searching: Automated exploration of control variables, sample restrictions, and functional forms
  • HARKing: Hypothesizing After Results are Known becomes easier when AI can rapidly generate post-hoc explanations
  • Data mining: AI-assisted pattern finding without theoretical grounding

At the same time, AI can improve empirical methods through better data cleaning, more comprehensive robustness checks, and automated replication verification.

Practical Implications

  • Pre-register analyses: Commit to specifications before running them, especially when using AI
  • Document all specifications: Record every analysis the AI ran, not just the ones you report
  • Use AI for robustness: After finding your main result, use AI to thoroughly stress-test it
  • Maintain theoretical discipline: Let theory guide hypotheses, not AI-discovered correlations
  • Transparency about AI involvement: Report how AI was used in the empirical analysis