Briefing, not a guideline

Statistical methods, with the assumptions left visible.

A method is a formula plus the situation where it misleads. The product implements some of these and not others. Where it does, the job should name the snapshot. Public plots, if you see one, are synthetic.

What this covers

Intervals

A percentage without an interval hides the sample size.

Diagnostic accuracy

Sensitivity and specificity need the reference standard and the population.

Survival

Kaplan–Meier and Cox models need a time origin and a censoring rule. They are not claimed as a finished module unless the statistics page says so.

Comparison of methods

Bland–Altman and regression methods answer different questions.

Missing data

A complete-case analysis is a choice. It is not neutral.

Multiplicity

Many looks need a plan. The software does not invent one.

How a team uses it

Name the question

Difference, accuracy, or agreement.

Write the population

Who was included.

Freeze

Then compute.

State the limit

In the report, not in a footnote nobody sees.

Related reading

Questions

Which of these run in the product?

The statistics capability page lists what is implemented. This page is the briefing.

Can I paste patient data here?

No. This is a public website.

Available with limitations means the described software behavior exists. A professional service is work a person does under a separate scope. Nothing here is a certification.

See the workflow on your protocol.

The conversation is about your study design and what the software can and cannot do today.