Description
Recording where every dataset came from, when and how it was obtained, and every transformation applied, so findings can be traced back to source data.
Provenance answers the question 'where did this number come from?'. For UX research this means logging the origin of each raw file (survey platform export, analytics query, recording batch), its retrieval date, query or filter used, consent basis and checksum, then logging every transformation step (exclusions, recoding, imputation, model runs) with inputs, outputs and code version. Rigour comes from automatic rather than manual logging, from hashing raw inputs so silent changes are detected, and from keeping a machine-readable lineage that connects each reported figure to its inputs. This is particularly important when analytics data are re-queried later and return different results, or when participant data must be deleted on request.
Attributes
- Discipline
- Research Design & Rigor
- Loss type
- Unmigrated
- Stage
- Post-launch, Definition
- Data
- Analytical / expert
- Effort
- Days
Steps
- Create a data manifest recording source, retrieval date, query or export settings, row count, checksum and consent basis for each raw file.
- Log each processing step automatically (function, input files, output files, parameters, code commit, timestamp) to a provenance log.
- Record exclusion counts at each stage (e.g. 1,240 started, 1,102 completed, 1,031 after attention checks) as a flow table.
- Re-verify raw file checksums before each full rerun and stop if any have changed.
- Link each figure and table in reports to the provenance entries and code commit that produced it.
- Maintain a deletion register so data subject requests can be traced through all derived files.
Sources
- Python (pandas, numpy, pathlib)
- Jupyter Lab
- Git
- conda/venv
- cookiecutter project templates
- Buneman, P., Khanna, S. & Tan, W.-C. (2001) Why and where: A characterization of data provenance. ICDT
- W3C (2013) PROV-DM: The PROV Data Model
- The Inquiry Series, C1 — The Python Research Environment — Setup, Philosophy & Reproducibility
Also known as
- Data lineage
- Provenance tracking
- Data audit trail
Categories
- Research Operations
- Research Design & Rigor