Impacts of pre-harvest and post-harvest treatments on reducing aflatoxin contamination in smallholder farmers’ maize

The study will include 3 interventions: pre-harvest treatment, post-harvest treatment, and pre- and post-harvest treatment. The pre-harvest treatment group will receive Aflasafe, a new bicontrol agent that outcompetes aflatoxin-producing fungi in the field during maize growth. The post-harvest treatment group will receive a tarp. Tarp is a plastic sheet that meets food safety standards and is used during maize drying as a barrier between the grain and soil where the aflatoxin-producing fungi live. The pre- and post-harvest treatment group will receive both Aflasafe and a tarp. All treatment groups will also be trained on the existence, dangers, and methods of preventing aflatoxin. All inputs will be distributed at baseline, before maize planting, and participants will be trained on their use at that time. After baseline survey and distribution of inputs, we will follow up with the farmers who received Aflasafe to remind them to apply Aflasafe on time. We will randomly select a subset of farmers to receive a follow up phone call reminding them about the use of Aflasafe. Orthogonal to the assignment to the input treatment, another randomization at the household level will determine whether the input (or inputs) is provided to female or male household member. Households in the control group will not receive inputs nor training.

2021-10-11

2021-11-20

Aflatoxin levels in maize.

Samples of maize will be tested for aflatoxin contamination using a test that shows whether the total aflatoxin level in a crop sample is below or above 10 ppb. The maximum allowable level of aflatoxin in maize in Kenya and in the European Union is 10 ppb. Maize samples will be collected at harvest and during storage (between 1-4 months after harvest).

Knowledge scores.
WIllingness to pay for 2 kg of Aflasafe.

In each household, respondent and one more adult member (e.g., spouse) will answer a set of questions about aflatoxin, Aflasafe, and/or tarp, depending on random assignment to an experimental group. The number of correct answers on the knowledge test will be normalized to generate an index that ranges between 0 and 1, i.e., a knowledge score. Knowledge scores will be created for each topic separately (aflatoxin, Aflasafe, tarp). Knowledge scores will be used as a measure of: i) knowledge retention of the person that received input(s), training on its (their) use, and training on aflatoxin; ii) diffusion of knowledge between the person that received training and input(s) and other adult member in the household who is of different gender (e.g., spouse).

A sample of rural smallholder households will be randomly drawn from the population of maize farmers in two aflatoxin-prone counties in upper eastern Kenya, Meru and Tharaka-Nithi. Working closely with local extension, we will identify maize-producing sub-counties, wards, location, and sub-locations in the two counties. Then, we will randomly select 60 sub-locations from within these. Lastly, we will randomly select 4 villages from complete village lists of each sub-location, that will generate 240 villages in total. Sub-locations will serve as randomization blocks (i.e., one village within each sub-location will be randomly assigned to each of the three treatments and one village will be assigned to the control group in each block). Aflatoxin levels in maize depend on climate conditions and the contamination levels can vary significantly between different geographical areas. Sub-location is the lowest administrative unit in Kenya, after village. Therefore, using sub-locations as blocking factor reduces the impact of geographical characteristics as a source of variability in our dependent variable.

The assignment to input treatment groups will be randomized at the village level. The 4 villages from each sub-location will be randomly assigned to one of four groups (1 village to each group): a control (C) group and three treatment groups, including (T1) pre-harvest input only; (T2) post-harvest input only; (T3) pre- and post-harvest inputs. In this design, sub-locations (lowest administrative unit after village) serve as blocking factors.

After villages have been randomly assigned to control or treatment groups, we will work with village elders to obtain up to date household lists for each village. We will then randomly select 8 dual-adult households from each village to participate in the study. All eight households in each treatment village will receive the relevant input(s), a training on its (their) use, and aflatoxin training. In control villages, eight households will be surveyed at the same interval as in the treatment villages, but they will not receive inputs nor training.

In addition to the main research question on the (cost) effectiveness of pre- and post-harvest inputs, we will also study the effect of gendered input and training allocation on aflatoxin levels in maize, knowledge retention, and diffusion of information within a household. Therefore, orthogonal to random assignment to treatment at the village level, we will randomly determine whether a female or a male adult member from each sampled household (all of which will be dual-adult households) will receive the input(s) and training.

Before the second follow-up survey, we decided that we wanted to understand more about the demand for Aflasafe. Therefore, we will add a Becker-Degroot-Marschak (BDM) auction at the end of the second follow-up survey. The BDM auction gives farmers the chance to purchase a 2-kilogram bag of Aflasafe, which is how the product is currently packaged for sale in Kenya. Participants are also given evidence from the first round of the follow-up survey, that based on data from fellow farmers in the study, those households who received Aflasafe at harvest were on average 10 percentage points less likely to have an unsafe level of aflatoxin in their maize at time of harvest than were those who did not receive Aflasafe.

The potential bids for Aflasafe in the BDM auction range from Kenya Shillings (Ksh) 250 to Ksh 550 with a uniform distribution around the market price of Ksh 400. The bids increase in increments of Ksh 50. The participants will play two practice rounds of the BDM auction where they bid for cookies and pens before they bid for Aflasafe. The participants are given Ksh 600 for their maize samples at the beginning of the survey. This will potentially alleviate any credit constraints around purchasing the Aflasafe that participants may experience. However, the payment is not tied to the auction and participants are under no obligation to buy the Aflasafe.

The objective of this BDM auction is to answer the following questions.
1. What percentage of farmers in the sample are willing to pay at or above the market price of Ksh 400 for 2 kilograms of Aflasafe?
2. Is there a difference in willingness to pay for Aflasafe between participants who were randomly chosen to receive Aflasafe for free in the previous season and those who were not selected to receive it for free in the previous season?

In office by a computer using Excel.

Village (assignment to control or treatment) and household (assignment of input to male or female adult in household).

Yes

240 villages (level of randomization and clustering).

8 farmers per village, totaling 1,920 farmers for baseline and follow-up surveys.

1. T1 (Pre-harvest treatment): 60 villages (480 farmers)
2. T2 (Post-harvest treatment): 60 villages (480 farmers)
3. T3 (Pre- and post-harvest treatment): 60 villages (480 farmers)
4. C (Control): 60 villages (480 farmers).

Aflatoxin levels will be analyzed at the household level. Our power calculations made use of published information on aflatoxin levels in maize in eastern Kenya, our primary outcome. We were able to obtain percentage of samples that tested below the Kenyan regulatory limit of 10 ppb (61%) as well as intra-cluster correlation (0.07) from Hoffmann et al. (2018). Our power calculations were done for the intent to treat (ITT) effect and two-way comparisons between control and one treatment group, or between two treatment groups. Based on data from Hoffmann et al. (2018) described above, a 0.05 significance level, and 0.8 statistical power, a sample of 60 clusters (villages) and a cluster size of 8 farmers (per village) provides a minimum detectable effect size of a 10.5 percentage point difference in the percentage of samples testing below 10 ppb. This represents a 17% increase in the share of samples testing below 10 ppb, yielding Cohen’s h of 0.22 which is considered a small effect size (Cohen, 1988). Our final sample with all four groups includes 1,920 farmers in 240 villages. References: Cohen, J. (1988) Statistical Power Analysis for the Behavioral Sciences (2nd ed.). Hillsdale, NJ Lawrence Erlbaum Associates. Hoffmann, V. et al. (2018) ‘Can markets support smallholder adoption of a food safety technology? Aflasafe in Kenya’, Project Notes. International Food Policy Research Institute (IFPRI).