Website Data, Replication, and Research Design
Project status: Archived as a completed teaching project. This repository preserves the final lecture materials; no further analysis is planned. See STATUS.md.
This repository includes Lecture Explorer, a project-specific agent skill for navigating the materials, configuring a fork, checking data access, and planning your own extensions. With pi, trust the repository and try:
/skill:lecture-explorer Where should I start?
Or ask for a complete local workflow:
/skill:lecture-explorer Outline the steps to allow me to extend this project on my own machine.
Lecture Explorer is a repository guide, not a tutor or grader. See the student setup page for installation, data access, and replication options.
This lecture explores how modern data sources can revisit a classic study on earnings management during import relief investigations.
- Part 1 – Three papers: Internet/Wayback data and CompuCrawl (
@haans2024internet), website-based disclosure measures (@boulland2025company), and the original earnings management setting (@jones1991earnings). - Part 2 – Research in action: A transparent walk-through of a Jones (1991) replication, a quick website text exploration, and a structured critique/brainstorming session.
Themes: replication, extending classic designs with new data, and practicing research design thinking.
- Slides: lecture-slides
- Lecture website/index: index page
- Student setup and Lecture Explorer: software, data access, and replication levels
- Jones replication: Jones replication
- Website analysis: Website analysis
- Paper summaries: Jones (1991), Boulland et al. (2025), Haans and Mertens (2024), and Godsell et al. (2017)
-
@haans2024internet– The Internet Never Forgets Wayback Machine + web scraping methodology; CompuCrawl database; focus on longitudinal website data and coverage/quality issues. Takeaway: Website data offers rich history, but coverage gaps and bias matter. -
@boulland2025company– Company Websites: A New Measure of Disclosure Website-based disclosure measures (size/content) and validation against traditional metrics; applications to information asymmetry and private firms. Takeaway: Website measures complement classic disclosure metrics. -
@jones1991earnings– Earnings Management During Import Relief Investigations Clean incentive setting at the ITC, the Jones model for discretionary accruals, and income-decreasing earnings management in the investigation year. Bridge to Part 2: What happens if we revisit this setting with modern data and tools?
-
Phase 1 – Replication Revisit Jones (1991) with a modern sample and cross-sectional specification; handle timing (fiscal year vs. investigation dates) and sample construction; compare core results to the original.
-
Phase 2 – Website exploration Quick-and-dirty keyword analysis on website text around investigations using archival data; check whether there is a signal; list major measurement and design concerns.
-
Phase 3 – Structured critique Small-group prompts to triage concerns, propose design improvements, think of alternative website outcomes, or pivot to other data sources.
-
Wrap-up Emphasis on replication as a starting point, quick checks before big investments, and making research decisions (go/no-go) explicit.
Details for each phase (timing, prompts, extra exhibits) live in the slides and HTML pages, not here.
- During the lecture: Open the slides and the HTML pages (replication, website analysis, paper summaries) when prompted.
- After the lecture: Use the lecture website and HTML reports for full details on data sources, variable construction, and empirical results.
- As a template: Reuse the structure for your own projects (separate data prep, scripts, and rendered reports).
index.qmd– Landing page for the lecture website.lecture-materials/– Slides and handouts.data-prep/– Data cleaning, construction, and supporting scripts.docs/– Rendered HTML outputs (e.g. replication, website analysis, site pages).assets/– Figures, images, bibliographic files, styles.
More fine-grained documentation is available on the website/index page and inside the HTML reports.
Start with the student setup guide. It distinguishes viewing the archived materials, exploring the code in a fork, reproducing the analyses, and extending the project.
USITC inputs can be regenerated from the public website. The Jones analysis requires WRDS credentials with Compustat access. The combined cleaned website text is available from the open-access CompuCrawl database. Generated CSV and DuckDB files remain local and are intentionally excluded from Git.
By the end of this lecture, you should be able to:
- Understand the potential and limitations of archival website data.
- Evaluate new disclosure measures critically.
- See how a classic setting (Jones 1991) can be revisited with modern data.
- Practice structured critique and research design thinking.
- Use this repo structure as a starting point for your own research projects.
Caspar David Peter Rotterdam School of Management peter@rsm.nl
For questions, ideas, or potential extensions, feel free to reach out.
This repository is made available under the MIT License. You are free to use, adapt, and extend the code and materials, provided that you include appropriate attribution.
If you use this repository or lecture materials in your own work or teaching, you can cite it as:
@misc{peter2025phdlecture,
author = {Caspar David Peter},
title = {Earnings Management Then and Now: Revisiting Jones (1991) -- Website Data, Replication, and Research Design},
year = {2025},
howpublished = {GitHub repository},
url = {https://github.com/CasparDP/phd-lecture-dec}
}AI tools (local and hosted LLMs) helped with brainstorming, structuring the lecture, and polishing some wording. The research ideas, empirical designs, and code are mine—any remaining errors are proudly human.