Start with the question you need to answer
A useful analysis brief begins with the question, the people or records it concerns, and the decision it should inform. You do not need to arrive knowing the name of a statistical test. Explain what you want to compare, estimate or understand, and bring any approved research objectives or hypotheses.
For example, a hypothetical training study might ask whether participants' scores changed between two assessments. Useful context includes when each assessment happened, whether the same people completed both, and how participants were recruited. A spreadsheet of scores alone cannot explain those details. Note any changes between the original study plan and what actually happened.
Explain what each row and column represents
Write a short data dictionary: the column name, its meaning, units, allowed values and any special codes. Include the questionnaire or measurement instructions when relevant. The UK Data Service's documentation guidance highlights variable labels, value labels and reasons for missing data as information that helps others interpret a dataset.
State whether a row represents a person, a visit, an organisation or a transaction. If a person can appear several times, explain how those records connect. In the training example, two rows with the same participant code may represent two assessment dates. Treating them as accidental duplicates would remove part of the study.
Make uncertainty visible in the data
Document what blank cells mean. An unanswered question, a question that did not apply and an unavailable measurement can have different explanations. Keep missing values distinguishable from genuine zeros. If codes such as 99 or -9 indicate missingness, flag them clearly so they are not mistaken for measured values.
Add a short list of known issues: inconsistent dates, suspicious values, incomplete records or categories that changed during collection. You do not need to solve every issue before asking for support. A note saying that a coding decision is unresolved is more useful than an unexplained replacement. Avoid guessing values merely to make the sheet look complete.
Keep a traceable version of the dataset
Keep the original export and work on a separate copy. Record corrections and transformations, including the affected field, the reason and the date. Oxford's research data guidance recommends reversible preparation steps and retaining raw files so changes can be checked later.
Use a descriptive filename and identify which version is ready for review. Include any existing analysis files and explain how they relate to the data. If you have already calculated a score, provide its formula and explain how incomplete answers were handled. This allows the calculation to be reviewed without reconstructing your decisions from the final numbers.
Share the minimum necessary and agree the support
Begin an enquiry with a project summary. Before transferring data, agree which files are needed, who can access them and how they will be shared. Remove unnecessary names, contact details and identifiers. Participant codes can support record matching, but replacing names with codes does not automatically make data anonymous. Keep any identifying key separate and restricted.
For academic work, clarify permitted support with your institution or supervisor. Agree how contributions will be acknowledged and retain responsibility for understanding and writing your submission. Oxford's research integrity guidance recommends discussing authorship and attribution during the project. A useful handover should help you explain the methods, decisions and limitations in your own words.