Each colony sits on a field with a smooth, hidden forage gradient (a stand-in for any nuisance variable — sun exposure, wind, predator pressure — that varies in space). If the treatment assignment happens to track that gradient, the measured "treatment effect" is really a mix of the treatment and the environment: a confound.
- Completely randomized — every colony's assignment is drawn independently of position.
- Randomized block — the field is split into row blocks; treatment is randomized separately inside each block, cancelling out most of the gradient's influence.
- Haphazard — treatment is placed by "convenience" (the richest side of the field), which systematically tracks the gradient.
- Confounding index — the Pearson correlation between treatment assignment and the local gradient value; near 0 is a fair comparison, far from 0 means treatment and environment are entangled.