Experimental Design
Producers
First, we will randomize our interventions at the market cluster level. Treatment clusters will receive both the producer and consumer intervention, and control clusters will receive no intervention whatsoever.
Next, we will randomize recruitment into the producer intervention within clusters. In treated market clusters, we will randomly select half of the producers to invite to the training, testing, and certification program. The other half will be invited at the end of the study. This will allow us to test for effects on certified and non-certified producers, who may be affected through information spillovers, shifts in consumer preferences, or changing prices. Furthermore, offering these producers training and certification later will allow us to measure demand for certification not directly incited by the research team (see below) and reduce the potential of harming initially non-certified producers or causing conflict between producers. In control clusters, we will randomly select half of the producers to be in the study so that we have three equally sized treatment arms: 240 treated producers, 240 untreated producers in treated market clusters, and 240 control producers.
We will stratify the randomization across market clusters based on geography and the number of producers in the cluster. We believe geography will be the best determinant of where a producer sources their groundnuts, which is they factor we believe will be most highly correlated with aflatoxin levels. Stratifying based on the number of producers will help us get the same number of producers in the three treatment arms. Because we will only stratify on these variables, we can assign treatment before collecting baseline data, which has two advantages. First, it allows us to construct a sample with an average of 12 producers in treatment market clusters (half of whom will be treated) and an average of six producers in control clusters. Second, it makes it possible to invite (treatment) producers to training upon completion of the baseline survey.
Consumers
Coinciding with producer training, we will implement the consumer intervention in all treated market clusters. This means we will estimate the combined effect of a producer and consumer intervention, which we believe is the most likely way a program like this would be implemented in practice. This design does not allow us to estimate the effects of the producer intervention alone, or to estimate the \emph{additional} effect of the consumer intervention when the produce intervention is in place. However, it gives us maximum power to estimate the (combined) treatment effect we are most interested in.