Agricultural intensification Program

Forest restoration took land out of production in Ghana and Sierra Leone. The programme had to put the same output back outside the restored area — plantain in Ghana, rice, cassava and groundnut in Sierra Leone — and be able to show it had done so before carbon credits could be issued.

That claim only holds if you can say what each farmer used to produce, what they produce now, and that the difference came from the programme rather than from the weather or a good year. At the start none of that existed on paper. Farmers were known by name in a community, plots were known by sight, and past yields were whatever someone remembered.

I built one management system covering both countries, holding the whole chain in a single record: the farmer profile captured at enrollment, the plots that farmer was allocated, the boundary of each plot, and every input handed out against them. One farmer, one record, in Ghana and in Sierra Leone.

Baseline grower records were collected in ODK Collect at enrollment — a socio-economic survey plus self-reported yields going back three years for each piece of land. A farmer moving onto new land is recorded twice: the history of the new plot, and their own history on the land they farmed before. Without both, an increase in yield could just be a better field.

Every plot was walked and mapped. The survey captures the crop and the boundary, and plot maps are generated automatically with satellite layers and coordinates, so any plot in the programme can be pulled up and matched to the participant it was assigned to.

For the end-of-year impact assessment I wrote the design as a matched comparison rather than a before-and-after. Treatment and control farmers are paired using coarsened exact matching on geography, farm size, distance from home to farm, land and soil type, tenure, crop mix, gender, age band, input use and time to the nearest market — so like is compared with like. The comparison is set at 95% confidence, a 5% margin of error and 80% power.

Field teams screen farmers against the matching criteria as they go, and only farmers who match take the evaluation survey. Where a community cannot supply enough matched controls, it stays in the study with the shortfall written down and the bias it introduces stated, rather than being quietly dropped.

The programme can answer the question it exists to answer: for any enrolled farmer, what was grown before, on which piece of land, and what has changed since. Enrollment, plot allocation, mapping and input distribution stopped being four separate exercises and became one record.

The yield claim rests on a matched control group built from stated criteria rather than on a before-and-after that any auditor could pick apart, and the places where matching fell short are written into the record instead of hidden.