Guidelines for the Creation of Analysis Ready Data

Harriette Phillips, Aiden Price, Owen Forbes, Claire Boulange, Kerrie L. Mengersen, Marketa Reeves, Rebecca Glauert

Introduction

In the field of data science, it is widely acknowledged that approximately 80% of a researcher's time is dedicated to data preparation rather than analysis. While the concept of "Analysis Ready Data" (ARD) has emerged to address this inefficiency, existing guidelines are often fragmented and specific to narrow domains like satellite imagery or social media analytics. This paper proposes a holistic, end-to-end framework covering the entire data lifecycle to mitigate "data debt" and enhance research reproducibility.

The 10-Step ARD Framework

The proposed framework consists of a 10-step workflow designed to be both sequential and concurrent, ensuring that data is not only technically clean but also ethically and legally sound. By synthesizing current literature, the authors established a standardized path from raw information to high-quality, immediately usable assets.

The 10-step workflow for generating Analysis Ready Data (ARD).

The workflow is categorized into ten distinct principles:

  1. Ethics: Establishing legal and ethical boundaries at the project's onset.
  2. Project Documentation: Maintaining a comprehensive "paper trail" for long-term project viability.
  3. Data Governance: Ensuring compliance with regional regulations (e.g., GDPR or the Australian Code).
  4. Data Management: Developing a formal Data Management Plan (DMP).
  5. Data Storage: Implementing secure, cloud-based infrastructure for backup and retention.
  6. Data Discovery and Collection: Sourcing and justifying data selection in alignment with research aims.
  7. Data Cleaning: Applying reproducible code (R/Python) to remove errors and inconsistencies.
  8. Quality Assurance: Detecting anomalies and quantifying "uncertainty" metrics.
  9. Metadata: Creating dataset-level descriptors following FAIR principles.
  10. Data Dictionary: Providing granular variable-level context for end-users.

Case Study: The Australian Child and Youth Wellbeing Atlas

The framework's efficacy was validated through the development of the Australian Child and Youth Wellbeing Atlas (ACYWA). This digital spatial platform harmonizes over 15 years of public health data (2006–2022) concerning individuals aged 0–24. The project integrated approximately 430 data items from over 30 different custodians, including the Australian Bureau of Statistics (ABS) and the Australian Institute of Health and Welfare (AIHW).

A critical technical challenge addressed was spatial and temporal alignment. As Australian geographical boundaries (ASGS) evolve every five years, the authors employed "Forward" and "Backward" correspondence techniques to map historical data onto a standardized 2016 boundary, enabling robust time-series analysis.

Key Insights and Discussion

ARD as a Lifecycle

A central thesis of this work is that ARD is a comprehensive lifecycle rather than a singular technical task. "Analysis ready" status encompasses the ethical approvals and governance structures that render data legally and safely usable.

Quantifying Uncertainty

Mapping data across shifting boundaries introduces inherent precision loss. To manage this, the authors implemented a standardized numerical scale (0 = low, 1 = medium, 2 = high) to track uncertainty. This allows for transparent pruning of high-uncertainty data while documenting lower levels in the data dictionary to guide researchers.

Metadata vs. Data Dictionaries

The framework clarifies a common point of confusion: metadata describes the dataset (authorship, licensing, scope), while the data dictionary describes the variables (definitions and calculation methods). This distinction is vital for ensuring data is truly Findable, Accessible, Interoperable, and Reusable (FAIR).

Conclusion

The successful deployment of the ACYWA as a public-facing asset demonstrates the practical value of the 10-step framework. While the process is resource-intensive—often involving long wait times for data custodians and significant costs—the resulting trust in the data is essential for informed policy decisions. Future work will explore the adaptability of this framework across different scientific domains where the weight of specific steps, such as ethics or cleaning, may vary.

Summary updated

BibTeX
@inproceedings{phillips2024guidelines,
  author = {Harriette Phillips and Aiden Price and Owen Forbes and Claire Boulange and Kerrie L. Mengersen and Marketa Reeves and Rebecca Glauert},
  title = {{Guidelines for the Creation of Analysis Ready Data}},
  booktitle = {Data},
  year = {2024},
  doi = {10.3724/2096-7004.di.2024.0007},
}
Figure