Good-practice guide

1. Organise the dataset

• Use stable variable names and clear labels.
• Define value labels and codes for missing, refused or not-applicable responses.
• Keep unique case identifiers and separate directly identifying information.
• Describe the unit of observation, case count and any filtering applied.

2. Retain instruments and context

Archive the questionnaire or interview guide as it was actually used, including instructions, routing and question order. Every variable should be traceable to its corresponding question or source.

• study purpose, universe and collection period;
• sampling, selection and recruitment procedure;
• collection mode and interviewer instructions;
• response rate, departures from the plan and known limitations.

3. Explain transformations

Document recoding, cleaning, imputation, weights and derived variables. Include formulas or code where possible; a computed variable without provenance quickly becomes impossible to verify.

4. Keep durable formats

Deposit the original files and include an open representation when it preserves the meaning of the data. Use UTF-8 for text, do not use formatting as a substitute for values, and include a README explaining the package structure.

Do not silently overwrite a published version. Record changes and retain links between versions.

5. Check rights and confidentiality

• Confirm who may authorise deposit and reuse.
• Remove unnecessary identifiers and assess indirect identification risk.
• Separate public files from material requiring controlled access.
• Assign an appropriate licence and citation conditions.

A good package lets someone who did not participate in the project understand what every file, variable and code represents.