Paired-data tests

Learning objectives

• Identify legitimate pairing structures.
• Analyse within-pair differences.
• Avoid pseudo-pairing by row order.

Pairing structure

Pairs may be the same garment before/after, the same operator under two methods, or matched units. The test analyses the distribution of within-pair differences, not two independent group means.

Pairing controls unit-to-unit variation when the pair identity is legitimate and preserved.

Garment-factory example

Measure shrinkage on the same panels before and after wash and analyse the paired differences; the pair ID is the panel serial number.

Method

1. Define the pair and record the ID.
2. Plot the differences.
3. Test the mean/median of differences.
4. Report CI for the difference and practical effect.

Common mistakes

• Pairing unrelated observations by row order.
• Losing pair IDs during data entry.