Benjamin Norton is a PhD student in Applied Economics at Cornell University. His Co-authors include Jessica Hoel (Colorado College), Hope Michelson (University of Illinois) and Victor Manyong (IITA).
On the home page of this site, you’ll find the Theodore Schultz quote “Most of the people in the world are poor, so if we knew the economics of being poor, we would know much of the economics that really matters. Most of the world’s poor people earn their living from agriculture, so if we knew the economics of agriculture, we would know much of the economics of being poor”. This sentiment may hold its greatest weight in Sub-Saharan Africa where most of the regions estimated 433 million people living in extreme poverty work in agriculture. Much of this agriculture operates at a subsistence level, being characterized by low productivity. While this low productivity contributes to the persistence of extreme poverty, it does not ex-ante have to exist. There is good evidence that yields can be increased using modern inputs like pesticides, herbicides, and mineral fertilizers, possibly leading to poverty alleviation.
However, we know adoption of these inputs is low, likely because there is widespread suspicion of the quality of these inputs. In one piece of evidence farmers in Uganda on average believed that fertilizer in their local markets contained 38% less nitrogen than the advertised amount [1]. In our own work in Tanzania and Uganda, we discovered that 70% and 84% of farmers respectively believe that some fertilizers in their local markets are counterfeit or adulterated. In addition, these beliefs seem to be ambiguous, as many farmers report uncertainty about the rates of counterfeiting or adulteration.
What’s weird is that these beliefs disagree with empirical evidence from fertilizer testing. Multiple rounds of randomly sampled fertilizer testing in Tanzania, Uganda, Malawi, Kenya, Cote d`Ivoire, Ghana, Nigeria, Senegal, and Togo have found that fertilizer counterfeiting and adulteration are extremely rare. Even though mineral fertilizer has been available for over half a century, there is still persistent doubt about its quality.
In our paper, we posit an answer to how these beliefs about mineral fertilizer quality persist in equilibrium. Guided by surveys and focus groups of smallholder farmers and interviews with extension agents, agro-input dealers, fertilizer regulators, and fertilizer manufacturers, we home in on two key mechanisms: farmers have multiple priors about fertilizer quality, and farmers learn from yield signals with misattribution. A farmer might think it possible that there are different rates of fertilizer adulteration or counterfeiting and not know which one to believe, and farmers might attribute low crop yields to counterfeit fertilized when the true limiting factor is unobserved.
Multiple priors are one way to model ambiguous beliefs, and there is much ambiguity surrounding mineral fertilizer quality. An extension agent might say that fertilizer is a boon for crop productivity, while a neighboring farmer could relate a story about how they used fertilizer and it didn’t help there crops that much. An agro-input dealer could say that their fertilizer was the top quality, while media prints stories about how nearly 40% of fertilizer in Tanzania is fake. With all these different signals, a farmer could hold multiple priors about fertilizer quality: It could be that 50% of the fertilizer in the market is fake, or 20%, or 10%, or 0%. But all are possible! When there are multiple different sources of information, it is hard to judge the authenticity of or know what weight to give to any one source.
Misattribution occurs when an agent mistakenly infers the quality of specific input from the outcome of a process in which the input plays a role – they “miss”-attribute the outcome to that specific input. For example, this could be attributing a sunken soufflé to the age of the eggs used when instead it was because you didn’t preheat the baking sheet in the oven. In the case of fertilizer, our focus groups and interviews all indicated that farmers complain of bad quality fertilizer after using fertilizer and receiving a bad yield. Farmers are told that fertilizer will increase their yields, so when they get a poor yield they are apt to blame the fertilizer when the poor yield could be due to other things like bad weather, pests, and random variation in general.
We combine misattribution and multiple priors into a single learning model of a farmer learning about the quality of fertilizer by observing crop yields over multiple seasons. Our baseline simulation shows that when misattribution and multiple priors are present, beliefs do not converge to the empirical measures of fertilizer tampering nor to a single prior. Figure 1 compares the evolution of the range of a farmer’s beliefs about the proportion of good quality fertilizer in their marketplace; the four simulations are with misattribution and multiple priors, with just one or the other, and without either. Our model predicts that beliefs will be more biased when outcomes are more variable; more lower tail events are likely to lead to more misattribution. Our model also predicts that beliefs will be more uncertain when outcomes are more variable, because it is harder to dismiss a wider range of priors. We test these predictions with precipitation and farmer beliefs data in Uganda. As predicted, we find that farmers who live in regions with higher historic precipitation variation have more biased and more uncertain beliefs about fertilizer quality.
What’s this all mean, though? For one, it points to the value of a trusted regulatory and verification mechanism. So long as a regulatory authority remains underfunded without the reach to test and verify the quality of fertilizer, for farmers that fertilizer can be poor quality remains a possible state of the world. In addition, one should attempt to understand learning mechanisms when introducing new products. For example, new fertilizer is introduced and sold with test plots. If in addition to the new fertilizer these test plots also have the best quality seeds, best quality herbicides and pesticides, and best quality management, then they are likely to have decent, above average yields which will be attributed to the new fertilizer. A farmer might buy the new fertilizer and have an unrealistically high expectation of its performance given the production conditions on their plots – this would increase the likelihood of misattribution and could actually lead to lower believed quality of the new fertilizer. Most importantly, however, is to understand that if we observe a puzzle – farmers are not using fertilizer – we should not say “they are being irrational” but to explore the causes of this puzzle. The problem the farmer is facing is not the one we observe, and it is important to get into this problem and understand what it truly is. It is only from there we can work on improving conditions faced by these farmers.
[1] Bold, T., K. C. Kaizzi, J. Svensson, and D. Yanagizawa-Drott (2017). Lemon technologies and adoption: Measurement, theory and evidence from agricultural markets in Uganda. The Quarterly Journal of Economics 132(3), 1055–1100.
