Towards better reproducible practice

STAT1003 – Statistical Techniques

Dr. Emi Tanaka

Australian National University

These slides are best viewed on a modern browser like Google Chrome on a desktop or laptop. Some interactive components may require some time to fully load.

Table of contents

• Non-robust workflow
• A robust, reproducible workflow
• Analysis framework
• Statistical value chain
• Folder structure
• Sharing your documents

Non-robust workflow

What should have been submitted:

A robust, reproducible workflow

• Using a robust, reproducible workflow means:
  • you avoid manual, repetitive tasks
  • your results are computationally reproducible
• Using a robust, reproducible workflow doesn’t mean you won’t make mistakes, but it will help you minimise mistakes.

Literate programming is a programming paradigm introduced by Donald Knuth where it emphasises writing code for humans (i.e. intertwine code with natural language explanations).

• Literate programming includes documentation (detailed explanations, comments and annotations to provide context, rationale and insight into the program’s design and functionality).

Analysis framework

Tidy data

1. Each column is a variable.
2. Each row is an observation.
3. Each cell is a single value.

Tools

• Git/GitHub for version control and collaboration
• Open-source programming languages (e.g. R and Python) for coding
• Quarto with markdown syntax for interoperable reproducible reports

• Huebner et al (2016) A systematic approach to initial data analysis is good research practice.
  The Journal of Thoracic and Cardiovascular Surgery
• Huebner et al (2018) A Contemporary Conceptual Framework for Initial Data Analysis. Observational Studies
• Steegen et al (2016) Increasing Transparency Through a Multiverse Analysis. Perspect Psychol Sci
• van der Loo & de Jonge (2018) Statistical Data Cleaning with Applications in R.
• Wickham (2014) Tidy Data. Journal of Statistical Software

Statistical value chain

… a statistical value chain is constructed by defining a number of meaningful intermediate data products, for which a chosen set of quality attributes are well described …

– van der Loo & de Jonge (2018)

Schema from Mark van der Loo and Edwin de Jonge. 2018. Statistical Data Cleaning with Applications in R. John Wiley and Sons Ltd.

Folder structure

A suggested folder structure for data projects:

    project-root-folder/  # Root of the project folder
    │
    ├── README.md         # README file
    │
    ├── data/             # Raw and derived data
    │   ├── data-raw/     # Read-only files
    │   ├── data-input/   # Extracted and coerced from raw data
    │   ├── data-valid/   # Edit and imputed from input data
    │   └── data-stats/   # Analysed results (R objects, .csv, etc.)
    │
    ├── analysis/         # Scripts (not functions) to run analysis
    │
    ├── figures/          # Figures (.png, .pdf, etc.)
    │
    ├── misc/             # Misc
    │
    ├── report.qmd        # Report, paper, or thesis output

Sharing your documents

via Quarto Pubs

• Make sure you are logged in to your Quarto Pub account.
• Then run the following command in the Terminal:

quarto publish quarto-pub /path/to/your/quarto-document.qmd

Self-contained HTML document

format:
  html:
    embed-resources: true

then you can share your output HTML file with no external dependencies

Happy writing and sharing 😊