Protocol
Preserve the pair in a before-and-after comparison
Two columns may contain a relationship that their separate averages conceal.
When it fits
- The same units are measured twice or observations were deliberately matched.
When to avoid it
- A before-and-after change alone does not isolate the intervention from time trends or other causes.
Why it matters
Keep the correspondence between observations and examine the within-pair changes. A paired design is not the same as two unrelated samples. Verify identities and missing pairs before selecting an analysis appropriate to the design and data.
Steps
- Confirm what makes each pair meaningful.
- Link observations using reliable identifiers and inspect unmatched cases.
- Analyze changes with a method that respects the pairing and relevant assumptions.
An example
A person's later score should be compared with that person's earlier score, not an arbitrary row from a separately sorted list.
Check your result
Every claimed change uses the correct pair, and missing pairs are not silently replaced.
Keep this limit in mind
- A before-and-after change alone does not isolate the intervention from time trends or other causes.
Connected ideas
Useful withDo not mistake frequent readings for independent evidence
Evidence and sources
Supports
Paired observations have a meaningful one-to-one correspondence and can be analyzed through within-pair differences.
Pairing must reflect the study structure, not an arbitrary arrangement after seeing the results.
Two-Sample t-Test for Equal Means · Paired versus unpaired samples