Mother-Baby Trial Design

The mother trial is a replicated experiment, run at a few representative sites under good management, that carries the complete set of treatments and provides precise, analyzable estimates of their effects. The baby trials are many simple trials on individual farmers' fields, each managed by the farmer and each testing only a small subset of the treatments against the farmer's usual practice, with one replicate apiece. The mother gives statistical precision and the full comparison; the babies give realism, breadth, and farmer evaluation. They are analyzed together so each compensates for the other's weakness.

Replication in the mother-baby design comes from the large number of farms rather than from repeats within a farm. Any single baby trial is noisy, but hundreds of them together sample the real distribution of soils, management, and seasons and let treatment effects be estimated across environments using mixed models that treat farms as random sites. The replicated mother trials supply the controlled error structure and full treatment comparison needed to interpret this dispersed signal, so the babies' value is collective coverage and farmer judgment, not individual precision.

Because the design's purpose is to learn what will actually be adopted, not only what yields most under measurement. The dispersed baby trials make it feasible to collect structured farmer rankings and reasons at scale, covering quality, labor, risk, and other traits that determine real-world acceptance. Integrating these farmer evaluations with the agronomic estimates — following the participatory-selection logic of Witcombe and colleagues — is what lets the mother-baby design speak to adoptability, distinguishing it from a conventional multi-environment yield trial.