[Paper Review] Tools and Recommendations for Reproducible Teaching

Mine Dogucu, Mine Çetinkaya-Rundel|arXiv (Cornell University)|Feb 19, 2022

Scientific Computing and Data Management19 references4 citations

TL;DR

This paper proposes a reproducible teaching framework centered on computational reproducibility, thorough documentation, and open sharing of teaching materials. It advocates for data science and statistics educators to adopt consistent, reproducible workflows—using tools like R, RMarkdown, Git, and GitHub—for creating, managing, and sharing course materials, thereby improving course sustainability, collaboration, and student learning through role modeling and open access.

ABSTRACT

It is recommended that teacher-scholars of data science adopt reproducible workflows in their research as scholars and teach reproducible workflows to their students. In this paper, we propose a third dimension to reproducibility practices and recommend that regardless of whether they teach reproducibility in their courses or not, data science instructors adopt reproducible workflows for their own teaching. We consider computational reproducibility, documentation, and openness as three pillars of reproducible teaching framework. We share tools, examples, and recommendations for the three pillars.

Motivation & Objective

• To address the gap in reproducibility practices specific to teaching materials in data science and statistics education.
• To propose a three-pillar framework—computational reproducibility, documentation, and openness—for consistent, sustainable teaching workflows.
• To provide educators with practical tools and recommendations to implement reproducible teaching, regardless of whether they explicitly teach reproducibility.
• To support course management at scale by reducing chaos in handling multiple files, student assignments, and collaborative teaching teams.
• To extend the benefits of reproducibility beyond research and student learning to include broader educational collaboration and visibility for instructors.

Proposed method

• Adopt a three-pillar framework: computational reproducibility, comprehensive documentation, and open sharing of teaching materials.
• Use R and RMarkdown for authoring course content, ensuring code, data, and output are version-controlled and re-runnable.
• Implement Git and GitHub for version control of course repositories, enabling tracking of changes and collaboration.
• Apply consistent file and folder naming conventions (e.g., lowercase, hyphen-separated names) aligned with the Tidyverse style guide.
• Use README files to document data provenance, structure, and dictionary information within data folders.
• Host course websites via GitHub Pages or RStudio Connect, and use platforms like Canvas, Gradescope, and Piazza for course delivery and grading, all integrated into the reproducible workflow.

Experimental results

Research questions

• RQ1 How can data science and statistics educators ensure that their teaching materials are computationally reproducible across time and different environments?
• RQ2 What documentation practices can improve the usability and maintainability of teaching materials for instructors and teaching teams?
• RQ3 In what ways does adopting open sharing of teaching materials enhance collaboration and accessibility in statistics and data science education?
• RQ4 How can reproducible teaching workflows improve course management in large-scale or remote teaching environments?
• RQ5 What role does modeling reproducible practices play in shaping student behavior and long-term adoption of best practices?

Key findings

• Adopting a reproducible teaching framework significantly improves course sustainability and manageability, especially in large courses with hundreds of students and multiple teaching assistants.
• Using version control with Git and GitHub enables instructors to track changes, collaborate effectively, and reproduce teaching materials across semesters or institutions.
• Consistent file and folder naming, along with README files for data documentation, enhances the clarity and reusability of course materials.
• Openly shared teaching materials—hosted on platforms like GitHub, Open Math Notes, or RStudio Education’s repository—can be reused, improved, and extended by other educators, fostering community collaboration.
• Instructors who adopt reproducible teaching workflows gain increased visibility and potential for collaboration, particularly beneficial for early-career academics.
• Even in non-reproducible-focused courses (e.g., theoretical probability), adopting reproducible workflows for teaching materials improves instructor efficiency and supports long-term course maintenance.