Minimum Wages and Firms' Capital-Labour Decisions

Independent Replication and Methodological Extension

Author: Zechuan Li

Status: Phase 8 complete; publication-ready ECO1400 Econometrics project Original course context: ECO1400 Econometrics

Overview

This repository contains an independently developed replication and methodological extension of an earlier ECO1400 team coursework project. The research examines whether British firms changed their capital-labour ratios following the introduction of the UK National Minimum Wage in 1999.

The improved version rebuilds the analytical workflow from the data-validation stage onward. It corrects known treatment-coding issues, reproduces the original estimates, strengthens the difference-in-differences design, adds pre-trend and robustness analysis, improves the panel quantile regressions, and generates all tables, figures, and reports reproducibly.

Baseline replication, the firm-level low-wage DID/event-study, robustness checks, placebo tests, exploratory industry-proxy and quantile models, and final report artifacts are validated. Industry-derived part-time and retail/hospitality estimates are not presented as reliable minimum-wage causal effects.

Two reproducibility paths

Public path without FAME or Codex. Any user can run the wholly synthetic smoke test with python code/15_generate_synthetic_demo.py and Rscript run_public_demo.R. The committed aggregate outputs can rebuild the Markdown and PDF report with python code/14_generate_final_report.py. These steps validate the public workflow but cannot recreate the empirical estimates without licensed FAME data.

Licensed empirical path. An authorized FAME user can supply the checksum-matched source outside the repository and run Rscript run_all.R. The committed report/paper.html is a prebuilt portable snapshot of report/artifact.json; rebuilding that HTML wrapper requires the optional Data Analytics plugin and is not part of ordinary public analytical reproducibility.

Authorship and provenance

Zechuan Li is the sole author of this improved version. All new code, validation checks, methodological changes, tables, figures, documentation, and written analysis in this directory were independently developed for this version.

The project is based on an earlier ECO1400 team coursework study completed by Qikai Zong, Shuaiyu Li, and Zechuan Li. The original project files remain in the parent directory and are treated as historical source material. Additional provenance details are recorded in docs/project_provenance.md.

Research questions

1. Did firms exposed to the UK National Minimum Wage increase their capital intensity after the policy?
2. Did responses differ across low-wage firms, high-part-time industries, and retail or hospitality firms?
3. Were effects heterogeneous across the distribution of firms' capital-labour ratios?
4. Are the results robust to stricter pre-policy treatment definitions, alternative samples, placebo tests, and explicit pre-trend checks?

Implemented improvements

• Correct missing-value handling in the high-part-time treatment definition.
• Separate original-replication specifications from preferred improved specifications.
• Use strictly pre-policy information to define treatment exposure where feasible.
• Treat 1999 explicitly as a policy-transition year.
• Add event-study and parallel-trends analysis.
• Add sample-flow, missingness, outlier, and panel-integrity checks.
• Re-estimate panel quantile models with explicit year effects and reproducible settings, while retaining them as exploratory because contemporaneous controls may be post-treatment.
• Add alternative thresholds, placebo dates, sample restrictions, and outlier sensitivity checks.
• Generate every table and figure directly from code.
• Produce a reproducible English report in PDF and HTML.

Reproducibility target

The fast baseline pipeline is regenerated with:

Rscript run_all.R

run_all.R verifies the frozen source, rebuilds corrected and improved analysis data, runs data tests, reproduces Tables 1-5, constructs strictly pre-policy treatments, estimates improved DID and event-study models, creates validated figures, and records the R environment. The legacy panel-quantile bootstrap is separated because it takes roughly nine minutes on the verified machine:

Rscript environment/install_quantile_dependencies.R
Rscript run_quantile_replication.R
Rscript run_improved_quantile.R

Current quick start

1. Place the licensed frozen source in the documented sibling private-data directory, or set ECO1400_PRIVATE_DATA_FILE.
2. Confirm that its SHA-256 matches the value in docs/source_freeze.md.
3. Run Rscript run_all.R with R 4.3.0 or newer.
4. Review outputs/validation/data_quality_summary.md, outputs/validation/event_study_pretrend_tests.csv, and docs/phase5_did_event_study_results.md.
5. If quantile figures are required, install the isolated legacy dependency and run the two slow quantile entry points shown above.
6. Install environment/requirements-report.txt and run python code/14_generate_final_report.py to rebuild final tables, Markdown, artifact.json, and PDF.
7. Run environment/build_html_report.ps1 in an installed Data Analytics plugin environment to rebuild the self-contained HTML; otherwise use the committed, canonical-artifact-verified snapshot. Open report/output/pdf/minimum_wage_capital_labour_report.pdf for print.

The fast Phase 6 pipeline has been verified locally with R 4.5.1 and requires no mandatory third-party packages. Panel-quantile replication uses checksum-verified rqpd 0.6 plus the versions listed in environment/quantile_package_versions.csv.

Without licensed data, run python code/15_generate_synthetic_demo.py followed by Rscript run_public_demo.R. This validates the public schema and model path but does not reproduce the empirical estimates.

The frozen source manifest, initial data-quality findings, and original-project replication targets are available in the docs/ directory.

Repository structure

data/         Public documentation and a wholly synthetic sample only
code/         Ordered data-processing and analysis scripts
outputs/      Generated tables, figures, model objects, and validation outputs
report/       Generated Markdown, canonical artifact, bibliography, PDF, and portable HTML report
docs/         Provenance, data dictionary, methodology changes, and replication notes
environment/  Dependency and execution-environment documentation
tests/        Data-integrity and output-regression tests

Data availability

The analysis uses firm-level data derived from the FAME database. FAME data may be subject to institutional or contractual access restrictions. Real firm-level data will remain outside version control unless redistribution permission is verified. The public repository will provide code, a data dictionary, access instructions, and a synthetic or disclosure-safe sample where appropriate.

See data/README.md for the data-handling policy.

Licences

Independently authored code is released under the MIT License (LICENSE-CODE). Independently authored report text, documentation, tables, and figures are released under CC BY 4.0 (LICENSE-CONTENT.md). Neither licence covers FAME source data, firm-level derived data, historical team materials, or other third-party content.

Citation

Citation metadata and the GitHub repository URL are maintained in CITATION.cff.