Methodology
Research Paradigm
This study employs a positivist research paradigm with a quantitative, archival research design. The goal is to establish causal relationships between education vintage and GenAI adoption using large-scale panel data and quasi-experimental methods.
Research Design: Difference-in-Differences (DiD)
The study exploits the sudden and largely unanticipated release of ChatGPT in November 2022 as an exogenous shock, creating a natural experiment that demarcates “pre-GenAI” and “post-GenAI” periods.
Why DiD?
- ChatGPT’s release was a discrete, identifiable event
- It was largely unanticipated in its timing and impact
- It allows comparison of firms with different board vintage compositions before and after the shock
- Firm and time fixed effects control for unobserved heterogeneity
Key Assumptions
- Parallel Trends: Absent the ChatGPT shock, firms with different board vintages would have followed similar AI adoption trajectories
- No Anticipation: Firms did not systematically alter board composition in anticipation of ChatGPT
- SUTVA: One firm’s treatment does not affect another firm’s outcome
Validation Strategy
- Event Study: Plot quarterly coefficients around the ChatGPT launch to visually verify parallel trends
- Placebo Test: Test whether vintage effects appear for non-GenAI technologies (cloud, blockchain)
- Robustness: Vary vintage cutoff points (2010, 2012, 2015)
Unit of Analysis
Firm-quarter panel, covering U.S. publicly listed firms with available BoardEx and Compustat data.
Time Period
- Pre-treatment: 2019 Q1 – 2022 Q3
- Treatment event: 2022 Q4 (ChatGPT release, November 30, 2022)
- Post-treatment: 2022 Q4 – 2025 Q4
Population and Sample
- Population: All U.S. publicly listed firms in BoardEx universe
- Sample criteria: Firms with at least one STEM-degreed director, available financial data in Compustat, and 10-K filings in SEC EDGAR
- Expected sample: ~3,000-5,000 firms, ~50,000+ firm-quarter observations