Data structure
Define what one data row represents. For repeated observations, check links to the participant and assessment date. Use consistent group codes and clear variable names. Keep abbreviation definitions alongside the dataset.
Missing and extreme values
Distinguish a missing value from zero. Zero may be a genuine measurement, so automatic replacement distorts calculations. Check unusual values against source records. Deleting a record requires a substantive reason and a log entry.
Participant flow
Compare the numbers enrolled, assessed at each stage and included in analysis. For observational medical research, the STROBE checklist from EQUATOR Network can help assess reporting completeness in cohort, cross-sectional and case-control studies. Select it according to the design.
Final version
After checking, save the agreed dataset version and calculation date. Record corrections. If data changes, update all dependent tables and charts. Keep a short sample flow diagram and an explanation for every reduction in numbers for the defence.
Preserving the source material
Keep a separate, unchanged copy of the received data. Make working corrections in another version and maintain a log with date, variable, reason and processing rule. This allows you to return to the original value and explain the decision.
If data comes from several files, first check column names, units and category codes. Similar headings do not guarantee identical meaning: “duration” might mean days in one file and months in another. Find these differences before combining tables.
Variable dictionary
Describe every column: name, meaning, data type, allowed values and unit. Define categorical codes and missing-value notation. A blank cell, zero and “not applicable” may mean different things and must not automatically be treated as equivalent.
Add an example of a valid value. Specify one date format and the order of repeated measurements. For derived measures, retain the formula and source variables. A dictionary is especially useful when another specialist analyses the data or work resumes after a long break.
Checks before calculation
Count rows, unique observation IDs and records in each group. Check ranges and combinations of variables. Compare unusual numbers with source documents first. A value differing from the others is not necessarily an error.
Duplicates also need contextual checks. Two rows may represent different stages for one participant and be necessary. If a record was duplicated accidentally, document its removal rule. Recount groups after corrections and compare with the dissertation’s material description.
Sending the dataset for analysis
Prepare the working table, variable dictionary, research objectives and group explanations. Identify planned comparisons and later questions. Use anonymised IDs for external discussion and share only necessary information.
After receiving the report, check sample sizes and measure names against the dataset. Keep the version used for final calculations. Later changes may require recalculation. A filename with a date and version number prevents old and new results from being mixed.
