Procedures
Research Workflow
Phase 1: Literature Review (4 weeks)
- Systematic search across Web of Science, Scopus, SSRN, NBER
- Three rounds of keyword-based searches (see
lr_search_incl_excl.qmd) - PRISMA-compliant screening and selection
- Literature classification (A/B/C grades)
- Conceptual model refinement based on literature gaps
Phase 2: Data Collection & Variable Construction (6 weeks)
- Week 1-2: Extract BoardEx education data
- Download director-level data with education fields
- Identify STEM-classified qualifications
- Extract AwardDate for vintage calculation
- Week 2-3: Classify STEM vs. non-STEM qualifications
- Create classification dictionary
- Manual validation of ambiguous cases
- Compute TKL for each STEM director
- Week 3-4: Download and parse 10-K filings
- Bulk download from SEC EDGAR (XBRL full-text)
- Extract relevant text sections (MD&A, Risk Factors, Business Description)
- Week 4-5: Run GenAI keyword extraction
- Apply validated keyword dictionary
- Compute TF-IDF weighted scores
- Calculate AI Adoption Speed (quarters to first mention)
- Week 5-6: Merge and validate
- Link BoardEx -> Compustat -> CRSP -> EDGAR via CIK/GVKEY
- Handle missing data (imputation for AwardDate)
- Construct panel dataset (firm-quarter)
- Winsorize outliers (1st/99th percentile)
Phase 3: Empirical Analysis (4 weeks)
- Week 1: Descriptive statistics and correlations
- Sample composition by vintage category
- Industry distribution
- Temporal trends in GenAI adoption
- Week 2: Main DiD estimation
- Triple DiD model (H1)
- Tech Committee interaction (H2)
- Vintage faultline analysis (H3)
- Reskilling moderation (H4)
- Week 3: Robustness and supplementary analyses
- Event study (parallel trends verification)
- IV estimation (birth year instrument)
- Placebo tests (non-GenAI technologies)
- Alternative vintage cutoffs (2010, 2012, 2015)
- Heterogeneity by industry and firm size
- Week 4: Results compilation
- Tables and figures
- Sensitivity analysis summary
- Effect size interpretation
Phase 4: Writing & Submission (4 weeks)
- Week 1: Draft Introduction, Theory, Hypotheses
- Week 2: Draft Method, Results
- Week 3: Draft Discussion, Conclusion
- Week 4: Internal review, revision, journal formatting
Quality Checkpoints
| Checkpoint | Criteria | Phase |
|---|---|---|
| Literature saturation | No new themes emerging in final round | Phase 1 |
| Data validation | <5% imputation rate for AwardDate | Phase 2 |
| Parallel trends | Non-significant pre-treatment coefficients | Phase 3 |
| Robustness | Results stable across alternative specifications | Phase 3 |
| Internal review | Co-author sign-off on all sections | Phase 4 |