Experimental Design
Phase 1
We drew a random sample of 509 farmers in four sub-counties (two from each of two counties) in Western Kenya. We drew this sample by drawing a random sample of 50 Enumeration Areas or Census Tracts (corresponding to villages in Kenya) from Emuhaya and Luanda sub-counties in Vihiga county and Ugenya and Ugunja sub-counties in Siaya county. In each of these 50 villages, ten farmers were chosen at random to be part of the study (with replacements possible if the household could not be found). Half were allocated to the treatment group and half to the control group. A baseline survey was conducted over March and April 2024 which included collecting soil samples. An endline survey was conducted over October and November 2024. A willingness to pay exercise was conducted before planting in 2025 for both treatment and control groups.
Phase 2
This phase will test PROVEN40 with a sample of approximately 300 treatment farmers (and an accompanying 300 control farmers) in Eastern Kenya, in two sub-counties in each of two counties (Kathiani and Matungulu sub-counties in Machakos county and Makueni and Mbooni East sub-counties in Makueni county). Across these, we selected 60 Enumeration Areas at random to be in the study. In each villages, we conduct a short household listing exercise to help in drawing the sample. From the listing exercise, we will also collect each farmer’s acreage planted with maize. We need farmers to have at least 1 acre of maize, given the experimental design below so we will screen farmers for this requirement during the listing exercise. From those that meet this requirement, we will sample 10 at random to be part of the study, of which 5 will be allocated to the treatment described below.
This Phase 2 trial will also allow for a more robust experimental trial. In particular, the trial is designed as follows:
The treatment farmers were offered enough PROVEN40 at no cost to them for a quarter of an acre of maize. In addition, they were asked to to compare this to the use of the extender (the sugar water that is added to the bacteria) alone on a (different) quarter of an acre. The remaining half acre is planted as per usual practices. This will allow the famers to compare across their own plots, comparing the plot that has PROVEN40 to the plot that does not and also allow us to confirm whether the extender has impacts on its own.
The 300 control farmers will not receive anything. However, this sample lets us measure any farmer adaptation to the technology across plots in the treatment group (we can compare non-PROVEN40 plots to control plots).
For the treatment and control farmers, before we implement any of the experimental design explained above, we collect a detailed baseline survey as well as soil samples before the planting season (in September and October 2025). In mid January 2026, we will collect some plant material, one sample from each type of plot (PROVEN40, extender alone and control) and send this to a lab to test for nitrogen uptake by the plant.
The endline survey will be collected in February 2026. For the harvests, our survey team will visit the farmers at harvest and weighing their harvests to get more accurate harvest data. All other data on cropping will be collected by self-reports in the endline survey. During the endline survey, we will also collect some qualitative data on how treatment farmers perceived PROVEN40, whether they would use it in the future, and whether control farmers had heard of PROVEN40.
Finally, we will conduct a willingness to pay (WTP) experiment with all the treatment and control farmers in the following season (either March 2026 or September 2026). We will only conduct this part of the study if we do see improved yields from PROVEN40 in the endline survey (if we do not, of course, we should not try and sell PROVEN40 to the farmers).