Experimental Design
Our sample included 1,098 farmers from Mchinji district in central Malawi, the major groundnut-producing area in the country . Farmers were randomly selected from a list of members of the National Smallholder Farmers’ Association of Malawi (NASFAM), a farmer-based organization that has over 43 associations across the country. Each NASFAM association had sub-units at the community level, called Group Action Centers (GACs). GACs were typically about 10 to 35 kilometers apart. A single NASFAM association counted 21 GACs (or communities) on average, with each GAC had an average of about 15 farmer clubs. A club was made of 10 farmers who reside within the same village; village were typically 1-5 kilometers apart from each other. We targeted two associations for the study, and we randomly selected 16 GACs from each association to form the study sample. Within each of the 32 GACs, we randomly selected 25 farmers, subject to the condition that at least 2 (and at most 5) farmers were selected in each club. The resulting initial sample included 830 farmers, who participated in auctions twice: in the harvest and lean seasons. To these we added 268 randomly-sampled new farmers in the lean season auction, to control for and test possible learning effects arising from the bulk of our sample bidding in the same auction twice; first in the harvest season and then again in the lean season.
For our intervention, we provided aflatoxin training to the treatment group and the random assignment for our RCT into treatment (received information through training about aflatoxins) or control (did not receive training about aflatoxins) groups was done at the GAC level. We assigned treatment at the GAC level to avoid potential information spillover across clubs (or villages) within the same GAC. This arrangement also ensured cost-effective administration of the study activities (aflatoxins training and auction). Although GACs are far enough apart to limit possible information contamination, GACs that fall within the same association are generally similar in terms of member demographics.
The information provided to treatment group participants included facts about aflatoxins, the crops they affect and the way they affect crops (in the field and during harvest, drying and storage), the health and economic effects of aflatoxins, and how to avoid or reduce aflatoxins contamination (practices available and appropriate for smallholder farmers). In the second round of the auction during the lean season, participants in the treatment group were not given the aflatoxin training again. However, we added some new participants assigned to the treatment group in the second round to control for possible learning effects. These new participants were given the same training as the original treatment group received at harvest. For compliance with IRB requirements, participants in the control group were provided with the training at the end of the study, after completing the auctions in the lean season.
For our outcome variable which is willingness to pay (WTP), we elicit farmers’ WTP for grain quality with incentive-compatible, revealed preference auctions using the Becker-Degroot-Marschak (BDM) mechanism (Becker et al., 1964). The BDM is commonly applied in field experiments in developing countries (Channa et al., 2019; De Groote et al., 2011; De Groote et al., 2016; Prieto et al., 2021) . BDM auctions provide revealed preferences estimates based on bidding real money and actually purchasing the item at the bid price. In our setting, because participants bid on three quality grades of groundnut, one of their three bids was randomly selected as a binding bid.
Participants were first oriented about the BDM goals and procedures, then went through two practice rounds with sweets to ensure they understood the process as well as understood that strategic behavior was not beneficial. Once this was done, participants completed the real auction. All the three groundnut grades were auctioned in one-kilogram units, and participants were allowed to inspect the groundnut before bidding. They then bid on the three grades of groundnut that were presented in random order. Once they bid for all the grades, the enumerator rolled a die in the presence of the participant to determine which of the three grades of groundnut was the binding bid. The participants then drew a paper from a bag that had uniformly distributed numbers around the median market price in each village, as reported by NASFAM farmers. These were used as “offer prices” at which the binding bid was determined. The participant bought the kilogram of groundnut of the selected grade if their bid was higher than the randomly drawn “offer price” from the bag, and they paid the “offer price” rather than the price they bid. Conversely, they did not buy the groundnut if their bid was below the “offer price.” In all analyses, we use the amount that participants bid as our measure of WTP. Participants were given a fixed participation fee at the start of the survey to eliminate liquidity constraints that would limit participation and bias their WTP.
The auction was implemented twice, first during the harvest season (June 2019) when farmers had abundant stocks of grains, and then again targeting the same participants during the lean season (January 2020). In the lean season we recruited an additional sample of 268 farmers (155 in the control group and 113 in the treatment group) during the second auction to tease out possible learning effects among the farmers in the original sample from the repeated auctions.
We purchased all groundnut from a single trader during the 2019 harvest in order to reduce heterogeneity in other grain attributes. The grain was then used to simulate the different grain quality grades prevalent in local markets (i.e., sorted and unsorted grain) for both auctions. Appendix B shows pictures of the three quality grades. For the auction implemented in the lean season, we used the same grain that was purchased during the harvest season and stored in hermetic (airtight) bags to ensure minimal variation in grain quality (Baributsa et al., 2017). Aflatoxins testing of groundnut was done by a laboratory in Malawi’s capital, Lilongwe (Appendix C). The aflatoxins-safe certificate was shown to participants when they were presented with the 1-kg sample of aflatoxins-safe groundnut on which they bid. All groundnut used in the auctions came from the same sample, in which the aflatoxins level was 2.1 ppb (below the 15 ppb limit in Malawi and the 4 ppb limit in the European Union); the aflatoxins level was not mentioned when presenting participants with the samples of unsorted and sorted grades