Improving the quality of agricultural products is often considered an important step in the process of structural transformation, moving farmers from subsistence farming to market-driven, commercial agriculture. However, disorganized supply chains with many layers of intermediaries often prevent the transmission of quality premiums to upstream farmers, which discourages farmers from upgrading quality. This is especially true when quality is unobserved at the farm gate.
Certification as the Solution: Effective but Expensive
A growing body of literature finds that quality certification can reduce the market friction caused by asymmetric information and improve product quality. The cost to get certified, however, can be high and not financially feasible for smallholder farmers or even farmers' cooperatives. Therefore, in markets where the producers are mainly smallholders and products are aggregated by intermediaries, downstream buyers typically certify the quality in bulk after aggregation. Although studies show that producers respond to quality incentives and improve product quality, when quality is only certified further downstream in the value chain, whether and how the quality incentives are transmitted to upstream producers remain open questions.
In my job market paper, I examine a new potential solution to enhance accountability and incentivize quality improvement in value chains: establishing a traceability system that enables precise rewards for farmers who deliver high-quality products, particularly when quality is difficult and costly to observe at the farm gate.
I establish digital traceability systems for Kenyan dairy cooperatives and introduce an innovative quality monitoring method, using Bayesian statistical models to infer individual milk quality from pooled samples and reduce testing costs. The model-predicted quality shows a high correlation with random milk tests among 940 farmers from two different counties. I reveal randomly selected farmers' milk quality as determined either by the model or by random tests to both cooperatives and farmers. I find that cooperatives and farmers respond to both quality monitoring methods, and the milk quality improves at the endline. The digital traceability system outperforms conventional random testing in most metrics. Additionally, farmers in both treatment groups who consistently provide high-quality milk receive higher credit limits from the cooperative and use more credit on animal feed at the endline.
Kenyan Dairy (Formal) Value Chain
As Figure 1 shows, in the Kenyan dairy (formal) value chain, farmers sell their milk to dairy cooperatives. Dairy cooperatives hire milk transporters to collect milk from farmers. Milk is not tested at this point due to the prohibitively high testing cost (both time and monetary), and thus, farmers are paid based solely on quantity. Transporters usually aggregate the milk from multiple farmers to fill larger milk cans, which are then transported to collection centers. The collection centers pour the milk together into cooling plants (usually containing 10 to 100 of these cans). Cooperatives then sell the aggregated milk to processors who either accept it with a premium price, accept it without a premium, or reject it based on their comprehensive milk testing results.
Figure 1
The reason for the premiums is that high-quality milk can be used to produce high-value products like buttermilk and yogurt. Medium-quality milk is limited to use in ultra-pasteurized liquid packets, but low-quality milk cannot be traded by law. Therefore, quality incentives exist in the downstream markets but are not transmitted to the upstream farmers, which leads to a low-quality, low-price equilibrium.
Traceability and Bayesian Model of Quality Estimation
To address this market failure, I develop a traceability system to track the milk throughout the value chain so the system can provide a basis for continuous monitoring. This traceability system helps record who contributes to each aggregated milk can and their corresponding contribution percentages. In addition, I test these aggregated cans when they arrive at the cooling plant for several days. Using this testing data, I develop a novel Bayesian statistical model to infer whether individual farmers are producing high- or low-quality milk.
By testing aggregated milk cans with recorded information on which farmers contributed to each can, I infer individual milk quality. Since the composition of each container varies daily, the can-level testing outcomes reflect these differences, enabling individual-level assessments. As Figure 2 shows, the model-predicted quality shows a high correlation with the random individual milk tests (0.75 for added water and 0.62 for butter fat) among 940 dairy farmers.
Figure 2
Experimental Design and Results
To test both the random monitoring and model detection in this setting, I run an individual-level randomized controlled trial with 940 farmers from cooperatives in two counties in Kenya. I provide randomly selected farmers’ milk quality information either from the model or from random individual tests to dairy cooperatives, as well as to farmers themselves. After three rounds of quality monitoring and information sharing, I find that model group farmers reduce added water by a significant 21.9% compared to the control group, while random test group farmers show an insignificant 12.6% reduction. Farmers in both treatment groups who consistently deliver high-quality milk receive higher credit limits from the cooperatives and use more credit on animal feed. A back-of-the-envelope calculation indicates that this model achieved a more significant reduction in added water per testing dollar spent. Additionally, it saves time on individual milk samples taken in the field, and it provides aggregated can-level quality information to help differentiate good milk from bad milk before it is mixed at the cooling plant. These advantages suggest that this new quality monitoring approach could be a preferable method for continuously monitoring and regulating milk quality.
Policy Implication
The national long-term strategy, Kenya Vision 2030, recognizes the dairy sector as a key agricultural area and aims to boost exports of milk and dairy products. Currently, only a small fraction of Kenya’s milk production is exported, and a number of trade conflicts have arisen when regional importing countries rejected milk products processed in Kenya in recent years on the grounds that Kenya’s raw milk was of insufficient quality. There is a new law under development that will implement quality-based milk payment in Kenya. From October 2024, the Kenya Cabinet Secretary for Agriculture has collaborated with colleagues from the CGIAR ILRI Kenya to guide this policy. My results speak directly to these attempts to drive the structural transformation of the Kenyan dairy sector. Outside Kenya, the traceability system I developed and the innovative quality monitoring approach have the potential to scale not only to dairy cooperatives in other countries but also to a wide range of agricultural commodities and manufactured products.
Guanghong Xu is a Ph.D. candidate in Economics at the University of California, Santa Cruz.
Join the Conversation