Why this matters
Organisations in Ghana constantly try changes: a new SMS reminder for loan repayments, a different fee message in a mobile money app, a radio programme for farmers, a home visit schedule for community health volunteers. The question is always whether the change caused the improvement. A fair experiment is the most reliable way to know, and adverts for stronger roles name experiment design and A/B testing directly.
Why random assignment works
If you simply compare people who chose the new thing with people who did not, the groups differ in other ways. Farmers who attend a training may already be more motivated. Randomly assigning who gets each version, by a coin flip or a computer, makes the groups alike on average in everything, including things you never measured. Then a difference in outcomes can fairly be put down to the change.
Designing an experiment
- Write the question and the decision. "Does reminder B raise on time repayment compared with reminder A? If yes by at least 3 points, we switch."
- Choose one main outcome before you start, such as repaid within 7 days of the due date.
- Randomise at the right level. If borrowers in the same savings group talk to each other, randomise whole groups, not individuals, or the versions leak.
- Work out the sample size you need to detect the smallest difference worth acting on.