Reproducible Research with Jupyter Notebooks for Public Health

A discussion paper advocating for reproducible research practices in epidemiology, authored by Arindam Basu of the University of Canterbury. It analyzes the state of reproducibility in observational epidemiological studies, citing a 2005 review where none of 90 studies provided statistical code. The paper argues for sharing data, code, and documentation as a minimum criterion for public health research.

Use Cases

• Studying reproducibility standards based on the framework proposed by Peng and Zeger (2006)
• Analyzing barriers to reproducible research based on the cited review of 90 epidemiological studies
• Designing reproducible analysis pipelines for public health data using Jupyter notebooks

Strengths

• Cites specific research findings, such as 0/90 studies providing code and 93% not reporting data processing methods
• Provides a concrete framework for reproducibility with three components: Data Available, Documentation, and Distribution

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download
• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment

Provenance

Source

Arindam Basu, University of Canterbury

Collection Method

Likely a review and discussion paper on research practices.