Write the purpose first
Are you describing responses, comparing groups, estimating an association or modelling an outcome? Record the sampling unit, repeated measurements, group structure and outcome type before choosing software or a test.
Inspect measurement and data quality
- Review missing responses, invalid codes, duplicate records and consent eligibility.
- Document category coding and the treatment of ordinal responses.
- Use a composite score only when the construct and measurement evidence justify it.
- Assess scale reliability where appropriate; reliability alone does not establish validity.
Consider candidate methods
A categorical association may suggest a contingency-table analysis, subject to count assumptions. A two-group comparison of a continuous outcome may suggest a mean-difference model, subject to design and distribution considerations. Ordinal, repeated, clustered or non-independent observations may need different approaches. These are starting points, not automatic test prescriptions.
Report what the evidence supports
- Explain method choice and checked assumptions.
- Report effect estimates and uncertainty alongside relevant significance tests.
- Distinguish planned comparisons from exploratory analysis and address multiple testing when necessary.
- Discuss sampling limitations, measurement issues and alternative explanations.