Begin with the research question
A statistical test should answer a defined comparison, association, prediction, or estimation question. Starting with software menus often produces technically valid output that does not answer the study.
Identify the variable structure
Document outcome type, predictors, grouping, repeated measurements, clustering, censoring, and missingness before choosing a method.
Check assumptions
Assess independence, distribution, variance, linearity, sample size, influential observations, and model-specific diagnostics. Assumptions should be reported, not silently presumed.
Report uncertainty and meaning
Include effect estimates, confidence intervals, exact p-values where appropriate, diagnostics, sensitivity analyses, and practical interpretation.
Research quality reminder: The correct workflow depends on your question, design, data, ethics, and institutional requirements.