add

Author: julio-garbers

README.md

@@ -0,0 +1,74 @@
+# AI Adoption Share by Industry and Country
+
+Industry-level AI adoption shares derived from the **MAP-AI** (Mapping Artificial Intelligence in Firms) dataset, as described in:
+
+> Garbers, Julio and Terry Gregory. "Mapping Artificial Intelligence: Evidence from Firm-Level Web Data in Europe."
+
+The MAP-AI data underlying this study -- including firm-level indicators of AI adoption, AI ecosystem roles, and AI technology types for European firms between 2016 and 2024 -- are publicly available via GitHub: <https://github.com/MAP-AI-data/data>.
+
+## File
+
+`ai_adoption_share_by_industry_country.parquet` (also available as `.csv`)
+
+A single long-format table with **60,584 rows**, covering two industry classification systems (NACE Rev. 2 and NAICS 2022) at four digit levels, for four European countries plus a pooled aggregate, annually from 2016 to 2024.
+
+## Schema
+
+| Column | Type | Description |
+|---|---|---|
+| `classification` | string | Industry classification system: `"nace"` or `"naics"` |
+| `code` | string | Industry code (e.g., `"C"`, `"62"`, `"620"`, `"6201"`) |
+| `digit_level` | int | Aggregation level: 1, 2, 3, or 4 |
+| `country` | string | `"Germany"`, `"France"`, `"Luxembourg"`, `"Belgium"`, or `"Pooled"` |
+| `year` | int | Year of observation (2016--2024) |
+| `ai_adoption_share` | float | Share of firms classified as using AI (null if `n_firms` < 30) |
+| `n_firms` | int | Number of unique firms in the industry-country-year cell |
+
+## Coverage
+
+**Countries:** Belgium, France, Germany, Luxembourg, and a pooled aggregate combining all four.
+
+**Years:** 2016--2024 (9 years).
+
+**Industry codes:**
+
+| Classification | 1-digit | 2-digit | 3-digit | 4-digit |
+|---|---|---|---|---|
+| NACE Rev. 2 | 21 sections (A--U) | 88 divisions | 272 groups | 615 classes |
+| NAICS 2022 | 9 sectors | 24 subsectors | 96 industry groups | 304 industries |
+
+NACE codes follow the Eurostat NACE Rev. 2 classification. NAICS codes follow the 2022 NAICS structure published by the U.S. Census Bureau. NACE 1-digit codes correspond to section letters (A--U); higher levels are numeric. NAICS codes are numeric at all levels.
+
+## Methodology
+
+### AI adoption measure
+
+The `ai_adoption_share` column reports the share of firms in a given industry-country-year cell whose websites indicate AI usage, as identified by the MAP-AI indicator. MAP-AI uses a Large Language Model (LLM) to classify firm website content and determine whether a firm adopts AI. The indicator captures *realized* AI adoption -- firms that actively use, develop, or deploy AI technologies -- rather than potential AI exposure. See the paper for full details on the classification methodology and validation.
+
+### Construction of this dataset
+
+For each industry-country-year cell, `ai_adoption_share` is computed as the mean of a binary AI adoption indicator across all firms in the cell. The number of unique firms (`n_firms`) is reported alongside each share.
+
+### Minimum firm threshold
+
+Cells with fewer than 30 firms have `ai_adoption_share` set to null to avoid reporting unreliable shares based on small samples. The `n_firms` count is always reported regardless of the threshold.
+
+### Code validation
+
+Industry codes are validated against official classification files (Eurostat NACE Rev. 2 RDF and 2022 NAICS Structure) before aggregation. Higher-level codes (1-digit through 3-digit) are derived from valid 4-digit codes.
+
+## Citation
+
+If you use this data, please cite:
+
+```bibtex
+@article{garbers_mapping_ai,
+  title   = {Mapping Artificial Intelligence: Evidence from Firm-Level Web Data in Europe},
+  author  = {Garbers, Julio and Gregory, Terry},
+  year    = {2025}
+}
+```
+
+## License
+
+This dataset is licensed under the [Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/) license. You are free to share and adapt this data for any purpose, provided you give appropriate credit by citing the paper above.