Methods someone can follow

The methods section has one job: to let a competent reader repeat what you did, and judge whether it was sound. Not "roughly what you did". What you did.

The test

Hand your methods to a classmate. Could they redo the study without asking you a single question? Every question they'd need to ask is a missing detail.

What to include

• Data: source, size, how it was collected, inclusion and exclusion rules, how it was split.
• Procedure or model: every step, in order; architectures and settings.
• Measures: exactly how each outcome was measured.
• Analysis: which tests or comparisons, the software and versions, how many runs or seeds.
• Ethics: approval and consent, where people are involved.
• Availability: where the code and data are.

Use a reporting guideline

Many fields have checklists of what methods must report:

• CONSORT for randomised trials.
• ARRIVE for animal research.
• ML reproducibility checklists at NeurIPS and other venues.

The EQUATOR Network keeps a library of these guidelines.

Style

Methods are the one place where short, plain, even repetitive writing is fine. Past tense, clear order, subheadings. Clarity beats elegance here.

RememberWrite methods so a competent stranger could repeat the work without asking you anything.