- Produce a dictionary, cleaning log and access/retention record.
Data dictionary
Define variable names, meaning, units, types, valid ranges and missing-value codes. Separate participant identities from research identifiers. Record transformations and preserve an unchanged raw-data copy.
Cleaning and quality control
Check ranges, duplicate records and inconsistencies. Excluding outliers or imputing missing data requires methodological justification. Report sample counts through cleaning stages so their consequences are visible.
Protection and transfer
Transfer only necessary information and restrict access. Align encryption, backup and retention with institutional requirements. A public messaging channel is not a substitute for an agreed secure process for sensitive data.
Reproducing results
Retain analysis code, software versions and execution order. Data sharing requires consent, permission and regulatory compatibility. If sharing is restricted, document methods and dictionaries or clearly labelled synthetic examples.
A data-dictionary example
For a teaching variable named score, record the construct, instrument, numeric type, permitted range, measurement time, missing code and derivation. Do not impose an invented range on real data. Record formulas and inputs for derived variables.
Versions and change control
Use meaningful filenames and versions. Separate raw and processed data and preserve transformations in code or a decision log. Test backup restoration rather than merely counting copies. Assign folder responsibilities and permissions in collaborative work.
Close the project carefully
An authorized colleague should understand the dataset using the README and dictionary. Retention or deletion follows institutional rules, consent and project agreements rather than a universal period. Match publication availability statements to actual releases and permissions.
Worked case and implementation decisions
The following is a fictional teaching case. Do not use its numbers or wording as actual study findings.
In the fictional dictionary, score is numeric, measured in points, with a teaching range of 0–100 and blank missingness. A value −99 used as a documented missing code must not enter the mean, but replacing it without consulting the dictionary is also wrong. Preserve raw data and encode transformations. Separate linkage IDs from personal identifiers and restrict access in real projects. Retention follows consent, contracts and institutional rules rather than one period for all research.
| Stage | Teaching example | Verification question |
|---|---|---|
| Variable | score, points and measurement time | Is the actual definition documented? |
| Missingness | Documented code, not analyst assumption | Is conversion correct before calculation? |
| Transform | Code from raw to processed | Are changes counted? |
| Access | Research ID separate from identity | Are permissions and controls adequate? |
| Preservation | Versions, backup and restoration test | Does retention match project policy? |
Exercise output: Produce a dictionary, cleaning log and access/retention record.
Sources and further reading
Official sources for verification and further reading