Data Collection

Primary Data Sources

Database Provider Variables Access
BoardEx WRDS Director education (AwardDate, Qualification, Subject), Committee membership, Employment history University subscription
Compustat S&P Global via WRDS Total assets, Revenue, R&D expenditure, SIC/NAICS codes University subscription
CRSP WRDS Stock returns, market capitalization University subscription
SEC EDGAR SEC 10-K annual filings (full text for NLP analysis) Public access

Variable Construction

Independent Variables

Tech Knowledge Latency (TKL)

  • Formula: Current Year - STEM Degree Award Year
  • Source: BoardEx AwardDate field for STEM-classified qualifications
  • Board-level aggregation: Mean TKL across all STEM-degreed directors

Education Vintage Category

Category Graduation Period Technological Era
Legacy Vintage Pre-2000 Mainframe, early internet
Transitional Vintage 2000-2012 Web 2.0, mobile, early ML
Modern Vintage Post-2012 Deep learning, transformers

Cognitive Vintage Faultline

  • Formula: Variance of TKL within the board
  • Measures: Intra-board heterogeneity in technical education recency

Reskilling Index

  • Formula: Count of technical certifications or degrees obtained after age 45
  • Source: BoardEx additional qualifications with AwardDate

Dependent Variables

GenAI Adoption Score

  • Method: NLP-based keyword extraction from 10-K filings
  • Keywords: “Large Language Model,” “Transformer,” “Generative AI,” “GPT,” “Foundation Model,” “Prompt Engineering,” “AI-generated,” “LLM”
  • Scoring: TF-IDF weighted frequency per 10-K filing
  • Exclusion: Generic tech terms (“Big Data,” “Cloud,” “Machine Learning”) to isolate GenAI-specific adoption

AI Adoption Speed

  • Formula: Number of quarters from ChatGPT launch (2022 Q4) to first GenAI keyword appearance in 10-K
  • Lower values = faster adoption

Control Variables

Variable Source Measurement
Firm Size Compustat Log(Total Assets)
R&D Intensity Compustat R&D Expenditure / Revenue
Industry Compustat SIC 2-digit / NAICS codes
Board Size BoardEx Number of directors
Board Independence BoardEx % independent directors
CEO Tenure BoardEx Years as CEO
Firm Age Compustat Years since IPO
Prior IT Investment Compustat IT capital expenditure

Missing Data Strategy

BoardEx AwardDate Imputation

When AwardDate is missing but QualificationType is available:

Degree Imputation Rule
Bachelor’s Birth Year + 22
Master’s Birth Year + 24
Doctoral Birth Year + 28

Validation

  • Compare imputed vs. actual dates where both are available
  • Report imputation rate and sensitivity analysis excluding imputed observations

Data Collection Timeline

Step Activity Duration
1 Extract BoardEx education data Week 1-2
2 Classify STEM vs. non-STEM qualifications Week 2-3
3 Download and parse 10-K filings from EDGAR Week 3-4
4 Run GenAI keyword extraction Week 4-5
5 Merge BoardEx + Compustat + CRSP + EDGAR Week 5-6
6 Data cleaning and validation Week 6