Technological improvements alone are unlikely to modernise agriculture if farmers are uncertain about the quality of new innovations

The modernisation of African farming has by-and-large failed to take off. Adoption of modern inputs by smallholder farmers remains disappointingly low, which depresses crop quality and productivity, and perpetuates rural poverty. Adoption of modern inputs is constrained by many well-known constraints, including high transaction costs and lack of information or liquidity. Another constraint is that the quality of agricultural inputs traded on local markets tends to be highly variable. Active ingredients may be missing in fertiliser or pesticide samples, germination rates of modern seed are sometimes surprisingly low, and mislabeling of inputs seems to occur regularly. Maybe traders purposefully fake or adulterate inputs (Bold et al. 2017), or maybe quality deterioration occurs along the supply chain due to poor handling (Barriga and Fiala 2020). Regardless of the cause, the result is the same. Farmers are unsure what they are buying and, as a result, are less likely to purchase modern inputs. But this is only part of the story.

The experiment

We organised an experiment in rural Tanzania to study the impact of modern seed when seed type is uncertain (Bulte et al. 2023). More than 600 smallholder maize farmers participated in this study, and each received a trial package containing maize seed for planting on their farm. The package contained 1.2 kg of seed, enough for approximately 0.1 hectares. Farmers were free to plant the seed wherever they wished and could manage the crop as they deemed most appropriate. In our experimental design we varied whether farmers received modern or traditional seed, and we varied the information that farmers received about their seed package. The result was a so-called factorial design with four different groups:

Half of the farmers received an open-pollinated modern maize variety that, according to station trials, should produce high yields and be resistant to droughts and low-nitrogen soils. These were farmers in Groups 1 and 3. The farmers from Groups 2 and 4 received a traditional seed variety.

Half of the farmers were informed about which seed they received (Groups 1 and 2). These farmers participated in an experiment that resembles a conventional randomised controlled trial (RCT) in economics. The farmers from Groups 3 and 4 were not told what seed they received. Instead, they only learned there was a 50% probability that their seed was of the modern type, and a 50% probability that their seed was of the traditional type. The aim of this “blinded experiment” was to mimic the situation farmers face when buying modern inputs on local input markets in Tanzania––we estimate there is only a 50/50 probability that purchased inputs are of the advertised type and perform as intended. To enhance the credibility of our treatments we selected modern and traditional seed types that were similar in size and appearance, and dusted all seed with a pink fungicide to further conceal its type.

We asked farmers to keep track of their labour time and provided them with special bags to store their harvests. This allowed us to monitor how many hours they worked on their “experimental plots” and to accurately measure their yields. We focused on labour because labour is a key input, and because labour and seed are complements in agricultural production––it makes sense for farmers to work harder when using high-quality seed. 


By comparing outcomes across different experimental groups we can learn a few things about the impact of modern seed and uncertainty, and their interaction, on farm production. Comparing Groups 1 and 2 enables us to study the impact of using modern seed under full information. Farmers from Group 1 harvest, on average, some 40% more maize than farmers from Group 2. This is a large treatment effect. Part of the yield increase is due to the fact that farmers spend more time tending their experimental plot when they know they received modern seed. The total yield effect of improved seed rests on the improved genetics of the modern seed as well as on the behavioural response of the farmer.

If we compare farmers in Groups 3 and 4 (the blinded experiment) we obtain a much smaller yield gain due to modern seed. The treatment effect is about half of what we obtained in the conventional RCT (Groups 1 and 2). The explanation is simply that farmers in Group 3 did not work harder on their modern seed plots than farmers in Group 4 worked on their traditional seed plots, because they were uncertain about what seed they were growing. The absence of a labour response implies that the yield effect of modern seed is muted. 

This is also evident from comparing outcomes for farmers in Groups 1 and 3, which picks up the effect of uncertainty on harvests (and labour) conditional on growing modern seed. Uncertainty lowers yields by some 15% because farmers, on average, allocate 15% less effort to cultivating their experimental plot. This suggests how poorly working input markets and uncertainty about seed type can undermine agricultural performance—even if input quality turns out to be “high”.

A final comparison is between Groups 2 and 4, which captures the effect of uncertainty about seed type conditional on growing the traditional variety. We expected that the probability of receiving modern seed would “crowd in” farmer labour (compared to the sure case of growing traditional seed), but this expectation turned out to be wrong. We do not document any changes in labour or yields between these groups. This means the behavioural response to uncertainty is highly non-linear. Increasing the probability of receiving “good seed” from 0% to 50% has no effect on behaviour. Further increasing this probability from 50% to 100% produces a strong behavioural response.

These results suggest an important insight. Uncertainty about seed type mutes the behavioural response to using modern seed, which in turn reduces harvests and profits. As farmers fail to experience the potential benefits of modern seed, they may be less likely to continue adopting modern seed. Instead, they are more likely to abandon the innovation. 

Sustained adoption of modern inputs

To study the issue of sustained adoption in more detail we revisited the same farmers who received modern seed during the trial (Groups 1 and 3) some six years after completing the experiment. During the follow-up visit we asked them whether they purchased any improved seed during the past six year interval. Uncertainty about seed type discouraged the subsequent adoption of modern seed––farmers from Group 1 were 25% more likely to purchase modern seed than farmers from Group 3. This suggests that fully-informed farmers were much better able to learn about the potential profitability of using modern seed than uninformed farmers.

An important take-home message of the study is therefore that technological improvements, such as the introduction of modern seed, are unlikely to invite a process of agricultural modernisation on their own. Their introduction should be accompanied by an institutional innovation. Farmers need access to modern seed, but they also need access to a low-cost and trusted seed certification system for sustained adoption and agricultural development to take off.


Barriga, A and N Fiala (2020), “The supply chain for seed in Uganda: Where does it go wrong?” World Development, 130: 104928

Bold, T, K Kaizzi, J Svensson and D Yanagizawa-Drott (2017), “Lemon technologies and adoption: Measurement, theory and evidence from agricultural markets in Uganda.” Quarterly Journal of Economics, 132(3): 1055–1100.

Bulte, E, S di Falco, M Kassie and X Vollenweider (2023), “Low quality seeds, labor supply and economic returns: Experimental evidence from Tanzania.” The Review of Economics & Statistics, In Press.

Tanzania Seed quality Agricultural development Modernisation Technology adoption