Review: Purchased agricultural input quality and small farms

Adoption of non-labor agricultural inputs remains low among small-scale farmers in many low-income countries. Accurate measurement of the quality of these inputs and quantities used is essential for assessing economic returns, understanding the drivers of agricultural productivity, and proposing and evaluating policies for increasing agricultural production. We review evidence regarding the quality of planting material, fertilizer and pesticides used by small farmers in low-income countries with a focus on Sub-Saharan Africa where the literature is most extensive. We distill four key findings. First, empirical evidence to date has centered on a limited set of agricultural inputs and locations. Second, some of this evidence is difficult to evaluate or may be misleading because best testing practices either were not followed or were not sufficiently documented. Third, while farmers are generally suspicious about input quality and therefore may hesitate to invest, these beliefs may exaggerate the severity of the problem. Farmers may attribute too much blame to poor quality inputs for bad crop yield outcomes. Fourth, most evidence comes from on-farm or in-shop samples; where input quality issues emerge at these downstream stages it is typically unclear where and how problems enter the upstream supply chain. We argue that while accurate documentation of measured and perceived agricultural input quality is important, the marginal productivity effects of input use hinge on the timing and method of application and on a host of complementary inputs (e.g., soils, moisture, labor). We conclude with specific priorities for future research that are linked to these key findings.


Introduction
Increasing agricultural productivity in Sub-Saharan Africa (SSA) remains essential to raising regional incomes and improving food security, as well as achieving structural transformation of the region's economies, a sustained transition of labor from low-productivity agriculture into higher productivity sectors (Gollin et al., 2002;McMillan et al., 2014;Timmer et al., 2015).While the SSA region has considerable agricultural and economic potential, it faces chronic challenges.Estimated yield gaps from primary staple cereal crops suggest that considerable productivity gains are indeed possible, but growing populations and the threats and uncertainties of climate change lend urgency and complexity to achieving needed agricultural growth (Leitner et al., 2020;Sileshi et al., 2010;Tittonell and Giller, 2013).Agricultural productivity gains in this region will require increased use of high quality agricultural inputs, including improved planting material, fertilizer, and agri-chemicals.Use of these inputs remains on average well below recommended levels across SSA, though with important variations across nations and farm types (Sheahan and Barrett, 2017).
Numerous studies have estimated potential returns from the use of improved (e.g., hybrid) seeds and the application of adequate fertilizer and herbicide by small farmers.Findings suggest that such inputs can be profitable, that they can raise yields and increase farm profits, but with important heterogeneity across farmers (Suri, 2011;Duflo et al., 2008;Marenya and Barrett, 2009;Harou et al., 2022;Beaman et al., 2013). 1  Bird et al. (2022) show that maize hybrids specifically-adapted to midaltitude areas in western Kenya increased yields by 26% on average and increased both yields and net revenue by nearly three times this amount for farmers who sowed hybrid maize previously and invested in (complementary) fertilizer.McCullough et al. (2022) use geocoded data from maize experimental trials in SSA to demonstrate the spatial variation in fertilizer use profitability due to differences in rainfall, soil quality, and market access.
Research on the returns to herbicide application has focused on quantifying the degree to which application reduces farmer labor costs (Haggblade et al, 2017).For example, Tamru et al. (2017) show that herbicide use increases labor productivity by nearly 100%, decreasing labor hours from 7.9 to 4.6 days per plot on average in Ethiopia, and Ashour et al. (2017) show that glyphosate application reduced labor time spent on plot weeding by 65%.Even so, nearly all of the existing studies estimating the returns to use of purchased agricultural inputs implicitly assume that the quality of the input is uniformly good and does not therefore introduce an additional source of variability in production outcomes related to agronomic response and profitability.
The prevalence and persistence of low agricultural input use by small farmers has been a focus of extensive study by researchers and a major concern of policy makers.Under-use is attributed to a range of related factors, including missing financial markets, uninsured risk, high transport and transactions costs, and information problems (Karlan et al., 2014;Emerick et al., 2016;BenYishay and Mobarak, 2019).Recently, research has begun to focus on the relationship of poor or uncertain agricultural input quality and input use (Michelson et al., 2021;Ashour et al., 2019;Bold et al., 2017;Bulte et al., forthcoming).Substandard quality, whether real or just perceived by farmers, could partially explain limited input use and low yields and profitability.Accurate measurement of the quality of agricultural inputs applied by farmers is therefore essential to raise farmer confidence about investing in such inputs and to improve yields and profitability through these input investments.
This paper reviews current evidence from peer reviewed academic studies and published expert reports regarding the quality of planting material, mineral fertilizer and herbicides as purchased and used by small farmers.We discuss best practices for definition and measurement of quality.We also summarize research eliciting and evaluating farmer assessment of seed, fertilizer and herbicide quality.Finally, we identify important evidentiary gaps and areas for future research.
Our review of available evidence produces four primary findings and a concluding reflection.First, we show that current research on the quality of purchased agricultural inputs has focused primarily on a limited set of inputs and geographies: primarily urea fertilizer, glyphosate herbicide, and hybrid seed and mostly in SSA.These are widely used agricultural inputsespecially by smallholder farmersand are critical for agricultural production.However, insights related to these inputs are not necessarily relevant to other purchased agricultural inputs. 2 In particular, we found no evaluation of other pesticides, including fungicides, insecticides, and inoculants and no assessment of agricultural lime.Nor did we find evidence regarding the quality of fertilizer with micronutrients including boron, iron, and zinc (all identified as relevant in SSA contextssee Bevis and Hestrin (2021), and Barrett and Bevis (2015)).
We restricted our review to papers and reports that discussed the methods used for sampling and testing; we found little such evidence regarding input quality for key agricultural regions of the world, including India, Latin America, and South Asia.A larger body of work with more geographic scope reports on farmer reports of quality problems without any independent assessment of the veracity of those claims.While such concerns are important on their own given farmer beliefs about quality (or even uncertainty about quality) affect use (Bulte et al. forthcoming;Maertens et al., 2023), it is striking that the preponderance of objective quality testing of agricultural inputs purchased and used by small farmers emanates from SSA.
Second, we find that some evidence about input quality is difficult to evaluate due to incomplete documentation of testing practices and protocols.At worst, claims based on non-transparent testing and sampling methods may mislead.Understanding testing techniques and procedures is critical as a pre-requisite to interpreting results but also challenging given how specialized these tools can be.Care and nuance are key to documenting test protocols because the interpretation of results and the expected magnitude of quality problems can vary widely by input.Genotyping of planting material, for example, can be relatively straightforward or extremely complicated depending on the genetics of the variety testedand getting these genetics accurately represented in a reference library is often a major undertaking unto itself.Similarly, though nutrient quality problems are exceedingly rare in urea fertilizer, nutrient shortages are more common and more likely in fertilizer blends including nitrogen, phosphorous, and potassium (NPK), calcium ammonium nitrate (CAN), and diammonium phosphate (DAP) and in glyphosate herbicide.
Third, evidence across a range of studies suggests that farmers are broadly suspicious about input qualitybeliefs that may be fueled by frequent local media claims (generally unverified) about fake fertilizer, seed, and agri-chemicals.With regard to urea fertilizer, for example, no credible evidence on measured quality supports farmers' widespread suspicions about this product in local markets.In other contexts, farmer suspicions seem consistent with available evidence regarding quality problems.Even so, their beliefs may exaggerate the degree of the problem; farmers may place too much blame on poor quality inputs for poor yield outcomes when other factors may be to blame including poor application timing, insufficient quantity applied, weather and other shocks, or incorrect fertilizers for soil nutrient limitations (see Hoel et al., 2022).
Fourth, evidence on purchased agricultural input quality is nearly exclusively based on samples taken from farmers or from agri-dealers. 1 Duflo et al. (2008) find in experimental work in Kenya that returns to fertilizer application are high on average -36% over a season or 69.5% annualized but that the full application package recommended by the government is not on average profitable for farmers in their sample.Barrett and Marenya (2009) show that fertilizer is profitable on average but that plots that are sufficiently degraded exhibit limited response to fertilizer, rendering it unprofitable for about a third of farmers in their Western Kenya sample.Harou et al. (2022) estimate that returns to fertilizer are significant in Tanzania but only for farmers who address a widespread sulfur limitation in the soils.Beaman et al. (2013) use an experiment in which they provided free fertilizer to female rice growers in Mali to show that women who received the full recommended quantity of fertilizer increased the value of their output by 31% but also increased labor and herbicide application on their plots, making it difficult to isolate the effect of the fertilizer alone.They conclude, "fertilizer's impact on profits is small compared with other sources of variation" (p.386).Other recent work assessing the effects of fertilizer taking into account heterogenous returns due to variability in chemical and physical properties includes Carter et al. (2021) in Mozambique, Kossoube and Nauges in Burkina Faso, Laajaj et al. (2020) in Western Kenya, Harou et al. (2017) and Burke et al. (2022) in Malawi, Chamberlin et al (2021) in Tanzania, Liverpool-Tasie et al. (2017) in Nigeria, and Burke et al. (2017) and Xu et al. (2009) in Zambia.Also see Jama et al. (2017) for a cross-country analysis of the returns to fertilizer use in Southern Africa.
2 For example, data from Malawi's Fifth Integrated Household Survey (2019/ 20) suggests that ~18% of cultivating households applied urea to at least one plot, while ~53% applied a different type of mineral fertilizer to at least one plot (rainy season; authors' calculation).Similarly, data from Nigeria's General Household Survey Panel (2018/19) suggests 24% of households applied urea on at least one plot, and 34% applied an alternative mineral fertilizer (authors' calculation).
H. Michelson et al.For inputs with documented quality problems including glyphosate herbicide, fertilizer blends, and hybrid seed, this lack of upstream testing means that there is little evidence as yet regarding the true source of the problem.Though farmer concern does seem to concentrate on input dealers, quality leakage could happen at the point of manufacturing, import, transport, packing, or storage.In the case of seeds, for example, viability (e.g., germination rate) steadily erodes over time even in good storage conditions; when subject to high humidity or heat in storage along the supply chain, seed viability can plummet (even in the absence of fraudulent actions by intermediaries).
Finally, we argue for the importance of understanding agricultural input quality and farmer beliefs about input quality, but also stress that these are among numerous factors determining the low average marginal productivity of agricultural inputs in SSA.These other factors of production, including soil quality, spatial heterogeneity in growing conditions, and other applied inputs, can interact in potent ways with agricultural input quality.Even more specifically, how and when purchased inputs are applied by farmers can themselves be critical dimensions of realized input quality.Objective and rigorous measures of input quality are an important innovation in empirical research methods, but the ultimate implications of input quality reflect a vast set of complexities inherent to agricultural production.
We begin with a review of the properties of planting material, mineral fertilizer, and pesticides3 and evidence on the use of these inputs among small farmers in SSA.Section III defines and discusses quality for planting material seeds, for mineral fertilizer, and for pesticides.In Section IV we focus on planting material quality measurement and evidence.In Sections V and VI we do the same for fertilizer and pesticides, respectively.Section VII reviews evidence regarding farmer beliefs about planting material, fertilizer and pesticide quality.We conclude with discussion of key evidentiary gaps related to measuring purchased agricultural input quality and use and offer recommendations regarding areas for future research.

Planting material
The foundation of all crop agriculture is the genetic content of planting material.For many staple crops (e.g., maize, rice, wheat, beans), this genetic material comes in the convenient form of seeds.Openly pollinated varieties (OPVs) produce seeds that can be saved and replanted.In contrast, hybrids are more responsive to inputs, especially fertilizer, and can generate higher yields in the right conditions, but also produce seeds that lose this hybrid vigor with each new generation.Hybrid seeds must therefore be replaced (repurchased) each season to maintain their initial yield advantage.Other familiar crops (e.g., cassava, banana, potato, sugarcane) are vegetatively propagated, meaning that farmers use cuttings from a mature plant to establish new plants rather than cultivating seeds (see Spielman et al., 2021).These crops rely on vegetative propagation because in their domesticated form they do not produce seeds or produce seeds that cannot easily be propagated directly as seedlings.The use of cuttings in agriculture ensures that genetic content remains identical from one generation to the next since each new plant is a clone of the plant from which it is derived.For many vegetatively (i.e., clonally) propagated crops, this conveys a clear advantage as it guarantees the preservation of desired genetic traits (e.g., resistance to a specific disease), but at the cost of limiting genetic variation in the population and thereby the genetic resilience and response to new pests or diseases.
These different modes of propagation have important implications for adoption and diffusion, particularly for small farmers who rely more heavily on informal sources of planting material.When stored under cool and dry storage conditions, seeds can be saved for months or years and retain their viability.Cuttings typically lose their viability more quickly.Generally, seeds are also easier to transport than cuttings.Whereas cuttings make it easy to preserve desired genetics and to share germplasm with neighbors, susceptibility to new diseases can make an entire production system vulnerable due to poor genetic diversification.As a result of these factors, seed supply chains tend to be much more developed and span vast geographies, while cuttings are generally sourced locally, distributed via informal transactions, and tend to attract much less private sector attention than seeds.Small farmers may acquire both seeds and cuttings via informal exchange (e.g., among neighbors or within farmer groups), but they are increasingly likely to buy packaged and branded seeds from an established retailer (i.e., agrodealer)something that rarely happens with cuttings.Unless otherwise specified, we focus primarily on planting material as seeds throughout this paper.
In contrast to fertilizer and pesticide, planting material is a lifeform with the potential to evolve over time, which introduces a host of complexities.Whereas a synthetic chemical fertilizer is categorically distinct from organic alternatives such as manure, innovation and improvement in planting material is often more subtle and less categorical.Since the dawn of agriculture, farmers have functioned as de facto breeders as they selected for desired traits and saved seeds accordingly.The advent of modern breeding brought more dramatic and discrete genetic improvements.These identifiable strains and varieties are collectively referred to as "improved" when compared to the backdrop of landraces adapted to a local agroecology that emerged as generations of farmers selected and saved planting material with specific production goals in mind.Given the pronounced spatial heterogeneity in growing conditions and the sensitivity of germplasm to these conditions, breeders typically improve planting material with a specific agroecological context in mind (see Bird et al., 2022), which further distinguishes the quality of planting material from other inputs.
In SSA, a minority of farmers report planting improved seed, as seen in Fig. 1 which uses data from FAO's Rural Livelihoods Information System (RuLIS) database. 4Statistics are presented separately for small farmers, also referred to as small-scale food producers, and larger broad category of agrichemicals that are used to address or remediate pressures (fungi, weeds, etc.) that inhibit crop growth.In this section we introduce pesticides as a class of purchased inputs that includes herbicide.The rest of the paper will focus on evidence and testing of herbicides as there is little evidence on other pesticides currently in the literature.household farms.Among small-holder farmers, fewer than one-in-five report using improved seed, and the proportion is only slightly higher among other farmers.Self-reported improved seed adoption data like these raise one important question: How does a farmer know whether he planted an improved seed or not?Similar questions do not naturally arise in the context of fertilizer and pesticide usage because these products are manufactured externally and distributed via retailers.But for seed, it is possible to save improved seed from a previous season or to acquire improved seed from a neighbor because seed can be produced locally and on-farm. 5In many surveys, it seems likely that farmers who report planting improved seed base this belief on the fact that they purchased the seed from a retailer rather than saving their own seed or acquiring it from a neighbor.Finally, it is worth noting that Fig. 1 only pertains to improved seed and not to cuttings: Farmers are even less likely to know whether they are growing improved planting material from cuttings than from seed because, as mentioned, they are much less likely to purchase cuttings in a formal transaction.

Mineral fertilizer
Fertilizers are critical agricultural inputs, providing essential nutrients to crop growth and development and to the preservation and enhancement of soil fertility (Henao and Baanante, 2006).Nutrients can be delivered via organic (manure, compost) or mineral additions to the soil.Primary macro-nutrients delivered by mineral fertilizers, also referred to as inorganic fertilizers, include nitrogen, phosphorous, and potassium.Secondary nutrients include calcium, magnesium, and sulfur.Mineral fertilizer blends can also include micronutrients such as copper, manganese, and zinc.Globally, the most widely-deployed plant nutrients in agriculture are nitrogen, phosphorous, and potassium while the most widely used mineral fertilizer in the world is urea, accounting for more than 50% of global nitrogen fertilizer use (Heffer and Prud'homme, 2016). 6nder-use of fertilizer in crop cultivation is associated with low crop yields in the near term and soil nutrient depletion in the long term if nutrients are not added back into the soils through some other mechanism.Over-use or misapplication of mineral fertilizer contributes to environmental problems including nitrate and phosphate water contamination (Keeney and Olson, 1986;Sebilo et al., 2013) and increased greenhouse gas emission (Snyder et al., 2009).Use of an inappropriate fertilizer for a given soil type can reduce yield response and profitability of use (Kihara et al., 2016;Tittonell et al., 2008) and further deplete already deficient nutrients, contributing to degraded soils, lower yields, and, perhaps, perceptions by farmers of poor fertilizer quality (this sort of potential misattribution is discussed in Hoel et al., 2022).
Most fertilizer is applied in solid granule form but liquid forms of ammonia-based fertilizers are also available.Fertilizers are sold either as single nutrient, "straight" fertilizers or multi-nutrient fertilizers in the form of compounds or blends.Urea fertilizer, for example, is a single nutrient fertilizer, 46% nitrogen by weight.Compound fertilizers contain multiple nutrients in each granule with all granules manufactured to have the same nutrient composition.Fertilizer blends are made by mixing granules of different straight fertilizers to achieve a desired nutrient composition.Blends are becoming more common than compound fertilizers as a means of delivering multiple nutrients to crops; blends provide more flexibility for tailoring formulations to meet soil and crop requirements.Blending also offers fertilizer producers some flexibility in sourcing components from a range of sources, reducing costs.Sheahan and Barrett (2017) review Living Standards Measurement Study-Integrated Surveys on Agriculture (LSMS-ISA) data to characterize input use among households cultivating at least one agricultural plot in the primary growing season in Ethiopia, Malawi, Niger, Nigeria, Tanzania, and Uganda.7They find that approximately 35% of farm households use some amount of mineral fertilizer in the primary growing season.Variation in the use of fertilizers both across and within countries, however, is evident.This can be seen in Fig. 2, which presents the incidence of mineral fertilizer use among crop-farming households for 12 countries in SSA (left panel) and 5 countries from other regions (right panel), drawn from the RuLIS database.Small farmers have a lower incidence of fertilizer use in nearly all represented countries, with the gap between the rate of use among small and larger household farms as large as 24 percentage points (Benin).
Variation in use of mineral fertilizer across and within countries is compounded by variation in quantity of fertilizer applied.Sheahan and Barrett (2017) in their review of LSMS-ISA countries also document that the average application of mineral fertilizer per unit of cultivated area across the six countries is 57 kg per hectare (26 kg per hectare in nutrients) but emphasize considerable variation across countries.Average application rates range from 1.2 kg per hectare (Uganda) to 146 kg per hectare (Malawi), across all households irrespective of fertilizer application.Average application rates among households using mineral fertilizers varies substantially by country, with 81 kg/ha on average in Ethiopia, 189 kg/ha in Malawi, 26 kg/ha in Niger, 310 kg/ha in Nigeria, 96 kg/ha in Tanzania, and 38 kg/ha in Uganda (Sheahan and Barrett, 2017).Average use rates and even some application rates conditional on use are generally below government recommendations based on experimental trials.8

Pesticides
Pesticides are agri-chemicals whose application protects crops from pressures that impede plant growth and development.Pesticides include insecticides, which protect against insect infestation and damage; herbicides, which kill weeds that compete with crops for nutrients, soil, and sun; and fungicides, which protect crops from fungi including rusts, mildews, and blights.These inputs are labor saving relative to the manual work of pulling weeds and dealing with insects and fungi (Tamru et al., 2017).Chemical pesticides can have human health consequences for agricultural workers and for individuals who are exposed to them through water pollution or through direct consumption in food (Jepson et al., 2014) and poor quality and improperly used pesticides can contribute to the emergence of resistance to active chemical ingredients among common weeds and insects (Cerdeira and Duke, 2006).These agro-chemicals have environmental effects as well, contaminating surface water and impacting aquatic life (Annett et al., 2014).Sheahan et al. (2017) use LSMS-ISA data from Ethiopia, Nigeria, Tanzania, and Uganda to characterize the relationship between pesticide use and farmer reported health outcomes as well as pesticide use and farm household health expenditures.Households with higher pesticide use report more health problems and higher health-related expenditures.They find that while pesticide use is associated with higher crop productivity and harvest value, farm households may bear the cost of the negative externalities from use, especially given evidence that farmers may not apply pesticides using correct dosage nor use correct safety equipment or precautions. 9 Sheahan and Barrett (2017) also documents pesticide use in Ethiopia, Malawi, Niger, Nigeria, Tanzania, and Uganda using the LSMS-ISA data: 16% of farming households apply at least one agri-chemical to their crop during the primary season but again Sheahan and Barrett find considerable heterogeneity in use both across and within countries.The incidence of agri-chemical use among crop farming households in 18 SSA countries (left panel) and 12 countries from other regions (right panel) is illustrated in Fig. 3, for small-scale food producers and larger household farms separately.Similar to the observed patterns of fertilizer use, chemical input use is considerably lower among small-scale food producers than among larger farms, particularly in SSA.In Benin, for example, only 43% of small-holder crop-farming households apply any chemicals, while 73% of households operating larger farms report using chemical inputs.Among the SSA countries included here, Cote d'Ivoire exhibits the highest incidence among small-scale food producers, with 57% of households applying chemical inputs.Haggblade et al. (2022) document that pesticide use has tripled in West Africa in the last ten years, arguing that growth in use is driven by a combination of increasing labor costs, increasing pest pressures, and falling prices of generic pesticides in markets.To the contrary, rates of pesticide use are very high and of similar magnitude in small-scale food producers and larger farms in several countries in other regions, such as India, Iraq, and Pakistan.
The most commonly used pesticide in the world is glyphosate herbicide (Benbrook, 2016).Glyphosate is a non-selective herbicide, often used by farmers to kill weeds before planting.It is sold in concentrated form and diluted with water before application with a sprayer.Use of agri-chemicals remains relatively low in SSA but sales of glyphosate have increased in recent years; prices have declined by as much as 50% in some countries since 2000 as global patent protections on production have expired and new manufacturers have entered markets.Pesticides tend to be sold at the retail level in concentrated form as a liquid in bottles, often of one liter or half liter, sealed and wrapped in plastic.Farmers sometimes purchase glyphosate from agri-dealers in a prediluted form from open receptacles such as jerry cans.Haggblade et al. (2022) argue that the recent rapid growth of pesticide use in SSA has outpaced regulatory capacity, creating scope for quality problems.Unregistered brands and in some cases locally banned ingredients proliferate in these local and regional markets (Murphy et al., 2012;Haggblade et al., 2021 and2018), with further potential implications for human health and environmental problems.

Non-labor agricultural input quality: definition
Agricultural inputs are generally experience goods as actual quality is observable by most customers only after purchase and use. 10 Especially in locations where regulation and enforcement of product standards is weak or nonexistent, farmers are largely on their own with regard to quality inference and often lack information about product quality at the time of purchase.
Sowing unviable or genetically impure seeds can severely limit productivity and profitability.Application of mineral fertilizer with inadequate nutrient content will impact yields, reducing the economic benefits of application accordingly.Using pesticides with less active ingredient than advertised will similarly decrease efficacy, requiring additional sprayings to deal with infestations or the use of labor for weeding or field preparation; these costs further reduce the profitability of use.Substandard quality of these inputs could therefore partially explain limited uptake.But beliefs about quality problems could have similar effects over time (Hoel et al., 2022).Accurate measurement of the quality and quantity of agricultural inputs applied by farmers as well as their beliefs about quality and efficacy are therefore essential to understand the degree to which quality is a problem in these markets and the degree to which quality problems or beliefs about quality problems may explain or help explain limited adoption in some areas.

Planting material quality
Because planting material is a lifeform, its quality is complex and multi-dimensional.Three primary quality dimensions are especially important. 11First, analytical purity indicates the proportion of a given sample that is seed for the correct species, which is reduced by the presence of debris or seeds of a different crop.
Second, the viability of the planting material indicates whether it germinates and grows with sufficient vigor to produce a harvestable crop.From farmers' perspective, viability is a particularly salient dimension of planting material quality as it is both an important and observable outcome.While it can be difficult to predict prior to planting, one key element of viabilitygood or bad germinationis readily apparent to farmers within weeks of planting.Typically, when farmers recount an experience with bad seed, they often have in mind seed that germinated at much lower rates than they expected.Third, varietal purity indicates that the sample is genetically what it claims to be.This is the most complex dimension of quality since it describes the genetic composition of a given seed.Given that seed is the product of parental genes, the varietal identity of a given seed can be quite mixed.Consequently, varietal purity is fundamentally a probabilistic concept that hinges on how genetically similar it is to a pre-defined reference (e.g., the variety indicated on the label).It is possible for a given seed to be most similar to the stated reference material as compared to other genetic material, but to have low levels of genetic purity.Such a circumstance would indicate the presence of foreign genetic material, but at a level low enough that the seed still matched the reference.
To varying degrees and in different ways, these three dimensions of planting material quality are relevant for both seeds and clonal cuttings.While viability is critical for both since seeds and cuttings are both living organisms that can die and become inviable as planting material, cuttings may be more susceptible to disease.The fact that cuttings are typically not as storable as seed also changes the quality dimensions of greatest consequence.Whereas commercially produced seeds are typically treated with fungicides to prevent disease prior to germination (i.e., in storage and during transportation), cuttings are more difficult to treat topically.Instead, the primary determinant of the health of cuttings is the prevalence of disease in the production environment in which they were produced (e.g., Diiro et al. forthcoming).In contrast, varietal purity is generally more straightforward for cuttings, which are genetic clones of the parent material, than for seeds.Since varietal purity is still not evident by visual inspection, however, it generally requires genetic testing in the case of both seeds and cuttings.

Mineral fertilizer quality
Fertilizer has two quality dimensions: agronomic quality that can be measured using lab-based methods and visually observable characteristics.The agronomic quality of all fertilizers is based on the degree to 10 Given the weather-driven stochasticity of agricultural production and spatially variable and sometimes unknown soil conditions (requiring application of different nutrient combinations) it may be difficult for farmers to detect whether mineral fertilizer is effective at all.For discussion of mineral fertilizer as a credence good, see Hoel et al. (2022).  1The International Seed Testing Association (ISTA) provides and regularly updates handbooks that provide more detail on these seed quality dimensions (https://www.seedtest.org/en/publications/handbooks-1170.html,Accessed 30 Jan 2023).
H. Michelson et al. which the measured nutrient content is consistent with its manufacturing standard.For example, urea fertilizer, which is 46% nitrogen by weight, should have at least 45.5% nitrogen by weight to be considered in compliance in most countries.Countries set tolerance limits that determine when a fertilizer is out of compliance.These tolerance limits vary based on whether the fertilizer is a single nutrient fertilizer or a blend, with larger tolerance limits for multi-nutrient fertilizers.Tolerance limits for nutrient content deviations in West Africa are set by the 15-member Economic Community of West African States (ECOWAS). 12Standards in East and Southern Africa are set by individual countries, though some work is being done to harmonize standards across the 21 countries that make up the Common Market for Eastern and Southern Africa (COMESA). 13 Low agronomic fertilizer quality (deficiencies in nitrogen or other nutrients compared to package specifications and compliance standards) can result from manufacturing problems, mismanagement along the supply chain, adulteration, or counterfeiting.Manufacturing problems are more common for blended fertilizers and relatively rare in single nutrient fertilizers such as urea.Adulterated fertilizer has been deliberately mixed with non-fertilizer material including sand, rocks, dirt, or salt.Counterfeit fertilizer is when an entire bag of non-fertilizer material is sold as fertilizer.
A second dimension of fertilizer quality is its visually-observable properties: the degree to which the fertilizer is too wet, expired, sold in short bags (bags that are underweight relative to the labeled size), powdered, clumped into hard aggregates, discolored or dirty.While observable quality issues like clumping are not associated with nutrient shortages in the fertilizer, powdered, wet, and clumped fertilizer can contribute to nutrient segregation in the bag.Clumping and powdering also potentially increase the costs and complexity of application for farmers.Michelson et al. (2021) find no relationship between the observed quality characteristics and the unobserved agronomic nitrogen content in Tanzania.The International Fertilizer Development Center (IFDC) (2013) analysis of fertilizer quality in West Africa finds that high measured moisture content was strongly associated with fertilizer caking.IFDC also found that moisture content and nutrient segregation were strongly associated in NPK samples.
In locations with capital-constrained mineral fertilizer supply chains, degradation of observable quality is likely to be a fundamental and recurring challenge, due to limited resources to support investment, transportation, and storage.Storage and handling conditions including humidity and temperature control, preservation of the integrity of the bag and the type of bag used (laminated or merely woven material) and use of pallets for stacking bags in transport and storage can result in caking, powdering, and discoloration.

Pesticide quality
As with fertilizer, pesticides can be adulterateddiluted with another substance like wateror counterfeitin which an entirely different product such as water is sold as herbicide or insecticide.Pesticides can also have quality problems due to errors in manufacturing.
Counterfeits may present in the market as sophisticated copies, with high-quality branding and packaging that can pass for the legitimate product.Application of adulterated or counterfeited pesticides can adversely affect crop growth.Pesticides are sold according to a labeled concentration of the active ingredient.
An issue raised by Haggblade et al. (2018) in their work on pesticides in Mali is the widespread presence of unregistered pesticides in markets, including pesticides that are registered in other neighboring countries but not in Mali and pesticides with no registration in the region.Haggblade et al. also document the presence of pesticides with banned ingredients in Mali including pesticides containing atrazine and paraquat.Murphy et al. (2012) evaluate the contents of 128 samples of pesticides purchased in The Gambia and find most products are unlabeled, sold in plastic bags or unlabeled plastic containers; they find a wide range of pesticides for sale, nearly half of which contained components that are banned by the World Health Organization or in the United States.

Measuring planting material quality
Because planting material quality is multi-dimensional, reliably measuring quality demands a variety of techniques and tests.Some of these quality assessment methods require very little expertise to deploy; othersgenotyping, in particulardemand a level of capacity currently beyond most countries in SSA.This capacity is, thankfully, expanding quickly and may soon be sufficient enable full domestic testing of all dimensions of planting material quality.

Analytical purity
Visual inspection is typically the basis for measures of analytical purity.While this process can be formalized and quantified with varying degrees of sophistication, it essentially entails sifting through a sample of seed and removing all debris or seeds of other crops.This inspection is part of the standard suite of tests performed by crop inspectors, but it is simple enough for farmers to do as part of their cultivation routine as well.

Viability
Measuring planting material viability requires a battery of tests.These rely on sampling to infer the viability of a population of seed (i.e., a seed lot) because the tests themselves destroy individual seeds and render them unviable as seeds.The most common test is to measure the moisture content of a seed sample.This is an indicator of seed viability as it can affect germination: Sustained exposure to high humidity can quickly degrade seed viability by making it more susceptible to damage due to pathogens, insects or handling.The moisture content of a sample of seeds reflects the prevailing humidity in the conditions in which the seeds were recently stored, whether in an airtight package or exposed to open air.Measuring moisture content of a seed sample involves a combination of measuring the ambient humidity and weighing the seeds (potentially after grinding them to flour) before and after heating them to infer water weight loss.Well-established and crop-specific moisture thresholds indicate whether the seed is at-risk of losing viability.For maize, for example, moisture content should be no more than 12% (Setimela and Kosina, 2006). 14While precise measurement of moisture content is beyond most farmers, it is standard practice for seed companies and crop inspectors.
Another standard assessment of seed viability is provided by germination tests, which indicate the proportion of a sample that germinates in response to a standard protocol that exposes the seed to 12 Tolerance limits for single nutrient fertilizers with up to 20% nutrient content have a tolerance limit of maximum 0.3 units and those with more than 20% nutrient content have a maximum tolerance of 0.5 units.Multi-nutrient fertilizers and blends have a tolerance limit of maximum 1.1 units for individual nutrients for primary nutrients and 2.5% for all primary nutrients combined.These are presented and discussed in Sanabria (2013).Tolerance limits are also set for secondary nutrients (Ca, S, Mg) and for maximum deviations in fertilizer weight for 50-kg bags. 13A range of standards exist across East African countries.The Kenya Standard 158 set in 2011 permits a maximum lower limit for solid compound fertilizers of 1.1%.The Ugandan government has no set tolerance limits for fertilizers.IFDC used Kenyan standards to evaluate Ugandan samples (Sanabria et al. 2018). 14Although not typically a concern in normal conditions, seed also becomes unviable at extremely low moisture content levels.
H. Michelson et al. optimal conditions for germination.This indicates the share of a sample considered viable at the time of testing.Since viability as measured by the germination rate is steadily decaying with time, the timing of these tests and the storage conditions during any delays in testing can directly affect these measures.Standard germination test protocols are relatively easy to follow, but are rarely, if ever, conducted by farmers.Instead, farmers conduct their own version of a germination test each time they plant seeds and wait and watch for seedlings to emerge.Such an on-farm germination test does not constitute a replication of standardized tests and will likely show lower germination rates for a given seed lot since conditions are typically less than optimal for seed germination.
The next level of testing for viability extends germination tests beyond optimal conditions and through seedling development (e.g., Kim et al., 1994).These vigor tests measure the proportion of a sample likely to produce normal seedlings under the kind of growing conditions that may occur on a reasonably well-managed plot in the field (AOSA, 2009).While correlated with germination tests, vigor tests reflect factors that affect seedling development after germination, which include storage conditions (Ferguson et al., 1991).Standard vigor test protocols entail phenotypical inspection of seedlings, including roots, stem and emerging leaves.Attentive farmers with reasonably good growing conditions effectively conduct on-farm vigor tests each season.For saved seed, vigor is strongly shaped by how, where, and how long the farmer stored this carryover seed.
Finally, seed health testing measures the presence of pathogens and diseases in or on the seed.Evaluating seed health is important both as an indicator of viability of the seed and as part of efforts to limit the spread of diseases as planting material is distributed across space (ISTA, 2022; Gitaitis and Walcott, 2007).A range of tests can detect seedborne pathogens, beginning with careful visual inspection of the seed and proceeding to grow-outs in controlled conditions and ultimately to DNAbased methods (for a review of these approaches, see Tsedaley, 2015).Commercially produced seed is generally treated with fungicide prior to packaging as a seed health measure, but similar treatments are rare for informally sold or saved seed.Vegetative cuttings are often more susceptible to pathogens prior to cultivation than seed and also less likely to be treated or tested as they are more likely to be distributed informally between neighbors or within groups.

Varietal purity
Measuring whether a given seed is genetically what it is purported to be is complicated.Breeders have long used an indirect, phenotypical approach to assessing varietal purity that uses the phenotype of a mature plant and the crop it produces to identify the germplasm of the planting material that produced it.These grow-out tests remain an important part of breeders' toolkit to ensure that the parent material they work with is genetically true-to-type and to eliminate genetic off-types before they infiltrate the seed system.Grow-out tests obviously require that the phenotypes of the desired varieties be carefully and clearly documented since this is the reference against which mature plants in the grow-out are compared.Conducting these tests requires training and expertise.Although it is typically performed manually, robots with high resolution sensors are increasingly used to phenotype plants in advanced research contexts (Atefi et al., 2021).Farmers and breeders alike often develop an intuition about distinctive phenotypical traits of familiar or favorite varieties and can often quickly tell whether a given plant looks like a given variety.
Dramatic reductions in the cost of DNA sequencing have opened the door for more direct genotyping of planting material as a way to more precisely and quantitatively assess varietal purity.This DNA fingerprinting approach, the details of which are beyond the scope of this survey (see Poets et al., 2020 for an accessible introduction to the topic), uses distinctive segments of a target genome (i.e., SNPs) as a reference library against which newly sequenced material is evaluated.If the DNA of the tested material matches the distinctive SNPs of the reference variety, it is considered to be the same variety.Because DNA is complex and can be the product of multiple genetic sources, however, this is more a probabilistic than a deterministic declaration (i.e., which reference variety does this sample match best).In addition to indicating which variety in the reference library best matches the tested DNA, DNA fingerprinting provides a measure of genetic purityessentially, the percentage of distinctive SNPs that match the declared variety (i.e., how good is the best match).The quality of any given DNA fingerprinting exercise hinges crucially on how complete and accurate the reference library is.In turn, this crucial input hinges on getting reliable and genetically pure reference material, which can itself be challenging given the often sensitive nature of this proprietary material.Producing a reference library can be relatively easy for some crops and wildly more complex for others.For clonal cuttings, DNA fingerprinting is often more straightforward than for seeds because the scope of genetic mixing is more limited.The rise of DNA sequencing techniques has led to a proliferation of DNA fingerprinting options that range widely by cost and resolution (see Poets et al., 2020).While labs equipped to conduct this analysis are spreading rapidly around the world, most low-and lowermiddle-income countries still lack the capacity to conduct DNA fingerprinting at scale.

Evidence on planting material quality
Establishing an effective system for regulating and certifying planting material is an essential element of a functional seed system (Bigirwa, 2020).Such a system requires the institutional capacity to test dimensions of seed quality using standard protocols at key points in the system. 15These standard tests provide measures of analytical purity, moisture content, and germination as the basis for certification for commercial distribution and sale.More advanced regulatory systems routinely conduct vigor tests, monitor seeds for pathogens and disease, and ensure that breeders in the public sector maintain clean parental lines using grow-out tests and other best practices.The use of genotyping to assess varietal purity in seed system regulation remains experimental in SSA, but with falling costs and expanding capacity these more precise genetic tests will almost surely become standard practice in the coming decade or two.While seed regulation and oversight has expanded steadily in recent decades throughout SSA, there appears to be considerable quality variation in commercialized seed from country to country.Since the informal seed systemincluding saved and shared planting material, which constitutes the majority of farmers' access to cuttings for many cropsfalls outside formal regulation and certification, seed from informal sources is of far less reliable quality.
Although measures of seed quality are routinely collected and analyzed as part of these regulatory systems, they are typically confidential records and not published publicly.As interest in input quality has swelled in recent years, economists have begun to use some of the tests reviewed above to evaluate seed quality. 16 Barriga and Fiala (2020) tested 112 samples of maize seed in Uganda, primarily purchased by mystery shoppers from retailers, which included 21 varieties of maize seed sold by 16 companies.They find uniform and very high levels of analytical purity (99.6% on average) and acceptable although more variable results from moisture tests (12.9% on average), germination tests (86.8% on average) and vigor tests (71.5% on average).One of the challenges of using these standard tests is that only analytical purity, 15 Regulating and certifying planting material quality for cuttings is quite different than for seeds and more challenging in a variety of ways (see Spielman et al. 2021). 16As an alternative use of these tests, they have also been used to evaluate the impact of interventions aimed at improving seed quality.For example, a number of studies have used germination rate as a key outcome variable in evaluations of PICS bags to enable farmers to save seed without suffering the normal losses due to pests, diseases and high humidity (Baoua et al., 2014, Williams et al., 2017).
H. Michelson et al. which is typically an uninteresting outcome because it is readily apparent by visual inspection, is not affected by storage time and conditions.Moisture, germination and vigor tests, in contrast, are more important and interesting, but can vary dramatically based on storage time and conditions, so they are not fixed quality attributes.
Varietal purity has been a more popular dimension of planting material quality among researchersparticularly as access to genotyping tools has expanded. 17The fact that it is difficult to observe for intermediaries and farmers only adds to the allure of variety purity and motivates researchers to explore the implications of this critical but often invisible quality dimension.The first wave of evidence from DNA fingerprinting targeted cassava (Maredia et al., 2016a;Floro et al., 2018;Wossen et al., 2019a), sweet potato (Kosmowski et al., 2016) and bean (Rabbi et al., 2015). 18Since it is difficult to discuss the results of these studies without reference to what farmers think they are growing, we delay this discussion until Section 7.More generally, having a reliable frame of reference is crucial for being able to interpret DNA-based varietal purity measures.Without a reference library, which itself can be a major investment, these measures are practically useless and can only provide a statistically sense of genetic variation in a given sample without shedding any light on whether this variation is of any real consequence. 19

Measuring agronomic quality
Lab-based measures are required for accurate assessment of fertilizer nutrient content.Nitrogen has measurement requirements that are distinct from other nutrients.The International Fertilizer Development Center (IFDC) uses Inductively Coupled Plasma-Optical Emission Spectrometry (ICP-OES) to measure potassium (K2O), Calcium (Ca), Magnesium (Mg), Zinc (Zn), Boron (B), and Cd (Cadmium) in fertilizer samples and has used both the Kjeldahl method and combustion analysis to measure nitrogen (N) and sulfur (S) (Sanabira, 2013;Sanabria et al., 2018a;Sanabria et al., 2018b).Note that determining adulteration once the nutrient content is found to be out of compliance requires a further analysis step to identify and document the presence of non-fertilizer material fillers; this step is essential in order to distinguish adulteration from errors in manufacturing.In addition, best practices for fertilizer testing include calculation and reporting of the analytical error, which is the error from the chemical analysis itself.Analytical error can be due to instrument malfunction (due to mis-calibration for example) or analyst error and is calculated by double-testing samples and comparing the results.

Nitrogen
The two primary lab-based measures to assess nitrogen content are the wet-chemistry based Kjeldahl method in which samples are ground and then diluted and distilled, and the Dumas method, which uses sample combustion at high heat.Mass spectroscopy (discussed below) is not used to measure nitrogen in samples because of strong background effects in the measurement caused by the presence of atmospheric nitrogen.
The Kjeldahl method is well-established and widely used in nutrient analysis but labor intensive and relatively slow, with a 100-minute analysis time per sample.Developed in 1883 by Johan Kjeldahl, the method is used in a range of applications including analysis of soils, feed, and wastewater.The Kjeldahl method involves several manual steps that can introduce human error.In contrast, the combustion-based Dumas method 20 is fast and automated, with a 4-minute analysis time, and relatively inexpensive.Tate (1994) compared the Kjeldahl and combustion methods and found the two analyses to produce statistically equivalent estimates of nitrogen content in analyzed fertilizer but concluded that combustion analyses were "more time efficient, more accurate, and less hazardous than Kjeldahl analyses" (p.829). 21

Non-nitrogen nutrients
Inductively coupled plasma mass spectrometry (ICP-MS) and inductively coupled plasma-optical emission spectroscopy (ICP-OES) are two forms of mass spectrometry used to measure the presence and quantities of non-nitrogen elements in fertilizer including Ca, Fe, K, Mg, Na, P, and S. The spectrometer converts elements to a gaseous state and then assesses the wavelengths of the light to detect the presence and quantity of elements present in the sample.

Agronomic quality: evidence
Assessment of bag characteristics including bag weight and expiration date and of observable fertilizer quality attributes is based on observation.Bag weight shortages within 1% of the weight reported on the fertilizer label are permitted in international regulatory systems though these can vary by region and country; a 50 kg bag missing less than 0.5 kg would be in compliance and those more than 0.5 kg short are out of compliance (OOC).Granular integrity, caking, and presence of impurities and discoloration is often evaluated based on visual assessment.In Michelson et al. (2021) these characteristics were independently coded by two enumerators from photographs of the sampled fertilizers with each characteristic assessed as either present or not.IFDC assesses bag condition and weight of the bag, granule segregation, granule integrity, caking, moisture content by observation and/or feel of the fertilizer and qualitatively and separately rates these characteristics for a given sample as none, low, medium, or high (Sanabria et al., 2013).
Nutrient content shortages in fertilizers tend to be quantified in terms of both the frequency with which they occur in sampling and their severity.Samples are classified as in compliance or out of compliance but also by how much they are out of compliance.

Urea
Evidence on urea fertilizer quality suggests that nitrogen deficiencies in urea are extremely rare.As a single nutrient fertilizer, errors in manufacturing are unlikely; the urea molecule contains 46% nitrogen (N) and reducing that content during manufacturing is difficult and uncommon.Adulteration and counterfeiting in urea are also unlikely given that (1) the color, size, and textural uniformity of the urea prills which makes successful adulteration difficult and (2) it is only profitable to dilute fertilizer with something cheaper than the fertilizer itself, and urea is low-cost relative to plausible adulterants (table salt, kaolin). 17While most of the evidence surveyed here uses genotyping tools, Bold et al. (2017) instead use an experimental grow-out test to estimate the responsiveness of farmer saved maize seed, maize hybrid seed purchased from local retailer and maize hybrid seed purchased directly from the seed company.This provides a loose indirect test of the genetic composition of the seed and may therefore correlate with varietal purity. 18It is no coincidence that many of these early DNA-fingerprinting studies focused on vegetatively propagated crops (i.e., cassava, sweet potato) because these are genetically simpler crops to identify, with less genetic variation since the cuttings are clones of the parent material. 19 Barriga and Fiala (2021) take this approach.For a discussion of the limitations of DNA fingerprinting untethered from a reference library, see https:// blogs.worldbank.org/impactevaluations/devil-details-measuring-seeds(Accessed 20 November 2022). 20The Dumas method is also known as method number AOAC 993.13 by the Association of Official Agricultural Chemists. 21A research team at the University of Illinois Urbana-Champaign has developed a means of verifying urea fertilizer quality using a smart phone application.The machine-learning based image classifier can detect the presence of non-urea materials.The tool is designed to be used by farmers and fertilizer sellers to verify urea quality before purchase.
H. Michelson et al.Foreign substances like sand are visually detectable and hard to pass unnoticed.
Table 1 (based on a table in Hoel et al. ( 2022)) summarizes the results of recent IFDC reports and academic studies based on sampling and testing the nitrogen content of urea fertilizer.Nearly all studies have been conducted in SSA.Nearly all academic studies have focused on urea fertilizer.Studies include samples collected from retail shops, wholesalers, importing ships, and farmers.Quality issues prove rare with the exception of two studies: an analysis of fertilizer quality in Ghana conducted by IFDC in 2010 in which 9% of the samples (21 out of 222) had insufficient nitrogen and a study conducted by Bold et al. in 2014 in Uganda in which 100% of the 369 samples tested were found to be missing nitrogen, and missing on average 30% of their nitrogen, an outlier result in published studies and reports.
The International Fertilizer Development Center (Sanabria et al., 2018) Uganda fertilizer assessmentwhich finds no problems in urea testeddiscusses the Bold et al. (2017) findings of widespread and considerable nitrogen shortages in urea, arguing that Bold et al.: "does not identify or quantify the presence of materials that may be used to dilute nitrogen content in the urea samples.Dilution is the only possible way of reducing nitrogen content in urea.The nitrogen content in the samples used as evidence could be below 46% as a result of deficiencies in the use of the Kjeldahl method, especially when the method is applied manually and by personnel with limited experience analyzing fertilizers.A very common mistake is assuming that a lab with experience analyzing soils will perform well analyzing fertilizers.".
Several recent research studies have used two separate labs to test the same samples to assess differences in the analytical error across labs.Ashour et al. (2019a) for example sent 115 samples of urea fertilizer and 72 samples of NPK to both a Ugandan labwhich used the Kjeldahl method -and to a lab based in the United Stateswhich used combustion (also known as the Dumas method) -and the results were substantially different. 22 Michelson et al. (2021) shipped samples from Tanzania, where facilities for measurement of nitrogen content were limited, to both Kenya and to the United States for testing.Ten percent of the samples were tested in both labs.Initial results from the Kenyan lab suggested significant and widespread nutrient problems in the samples but these results were not consistent with the lab results for the same samples from the United States.All samples were subsequently reanalyzed in the Kenyan lab with re-calibration based on results from the US labs. 23This reanalysis found virtually no nutrient problems in the urea samples, with only 1% out of compliance and then only slightly so. 24 Asante et al. (2021) found discordance between results of NPK fertilizer samples double tested in a Canadian lab and a Ghanaian lab.Both labs had used the Kjeldahl method.The Ghanaian lab initially identified significant nitrogen deficiencies but retested the samples after the Canadian lab results indicated that the samples met nitrogen standards.Upon retesting, the Ghanaian lab identified a problem with the nitrogen digester and found that nitrogen levels were also good in the samples.Thus, limited findings based on tests of samples sent to multiple labs suggest that the differences in results across labs can substantially exceed the within-lab analytical error for some methods, especially methods like the Kjeldahl method which involve manual steps that can introduce human error.

Multi-nutrient fertilizers
Evidence regarding the measured agronomic quality of multinutrient fertilizers suggests the presence of far more quality issues than in urea.Michelson et al. (2021) report results of tests of DAP and CAN collected in Tanzania and find 15% of DAP samples out of compliance and 63% of CAN samples.While frequency is relatively high, severity is relatively low: mean nitrogen deviations are 7% for DAP and 6% for CAN.Asante et al. (2021) find that NPK sampled in Ghana (15 samples) meets standards for nitrogen, phosphorous, and potassium.Apart from Michelson et al. (2021) and Asante et al. (2021), nearly all published evidence regarding the quality of multi-nutrient fertilizers comes from IFDC reports on West Africa (2013), Uganda (2018), andKenya (2018).IFDC reports that quality problems are most frequent in NPK fertilizers manufactured through blending; 51% of the 106 samples of NPK 15:15:15 (the most commonly found NPK fertilizer in West Africa in the study) were out of compliance based on the ECOWAS standards as were 86% of the 20:10:10 blend samples and 96% of the 15:10:10 blend samples.More work is required to understand the degree to which these widespread but generally modest shortages are economically and agronomically significant.
The IFDC report on the quality of fertilizer in West Africa (Sanabria et al., 2013) argues that nutrient deficiencies in some blends is likely attributable to uneven distribution of granules in the bags due to improper blending (mixing).In such cases, the nutrient composition of the entire bag may be in compliance with the labeled nutrient content but non-uniform distribution within the bags means that samples taken from the bag are not in compliance.In addition, small farmers purchasing blends from open bags or in repackaged smaller bags are likely to receive products with nutrient contents that do not reflect the manufactured standard.
Granule segmentation among blends can also happen in bulk transport before bagging, with consequences for the nutrient composition of the bagged fertilizer.Fertilizer is generally imported in bulk and then bagged in country.Segmentation in bulk transport is studied experimentally in Hall et al. (2014), who show that blends of varying granular size exhibit significant nutrient segregation during bulk transport before bagging.
Nutrient deficiencies in blends can also be attributable to manufacturing problems, poor control of blending procedures, and poor-quality blending equipment.Blends often contain filler in addition to combinations of nutrients including N, P, and K, and low nutrient content can mean more filler or oversupply of another, depending on the nature of the production problem.Bulk blending often uses separate conveyer belts for separate nutrients and fillers; miscalibration in the timing and speed of these conveyor belts on the production line can contribute to blending errors.Srichaipanya et al. (2014) discuss the problem and present experimentally-elicited data about the magnitude of the errors that can obtain from miscalibration in the production process.Phosphorous in a 16-16-8 NPK blend ranged between 13% and 18.5%, potassium between 7.2% and 9.7%, and nitrogen between 14.6% and 20%.
Quality issues were much less prevalent in the compound fertilizers sampled in the IFDC studies. 25Results from IFDC studies in Uganda and Kenya are consistent with the findings of quality issues arising in blends, likely due to problems in manufacturing (Sanabria et al., 2018a and2018b).
Adulteration and counterfeiting in mineral fertilizer are rare.The 22 While the Ugandan lab found wide variation in the nitrogen content of both the urea and NPK samples; between 20% and 70% nitrogen in the urea and between 5% and 30% for the NPK, the US-base lab found that nearly all urea samples contained 46% nitrogen and that nearly all NPK contained 17% nitrogen. 23The Kenyan lab analysis of the fertilizer samples was based on spectrometry readings using a spectrometer that had been calibrated for soil nutrient assessment rather than fertilizer assessment.The initial spectra suggested widespread nitrogen shortages relative to the manufacturing standards but did not detect the presence of additional fillers or other nutrients to make up for these significant nitrogen and phosphorous shortfalls. 24Costs for fertilizer testing can vary considerably based on whether the lab is located in the United States or in Sub-Saharan Africa and whether the lab is a university, government, or private lab.
25 Crystal and liquid fertilizers, which comprise a small share of the market, were found to have significant quality issues with considerable nitrogen shortages documented in Kenya (Sanabria et al., 2018a).
H. Michelson et al.IFDC report argues, "the perception that fake or adulterated fertilizers in West African markets is a dominant quality concern is not supported by the findings of this study."Of the 2,037 samples IFDC collected in West Africa, only seven were found to contain no materials with fertilizer properties and were classified by the IFDC researchers as adulterated and misbranded. 26, 27These seven samples were all of the same type and location of origin: a superphosphate fertilizer collected from Nigeria.IFDC analyses conducted in Uganda (2018a) and Kenya (2018b) similarly found no evidence of fillers or foreign materials suggesting adulteration.

Visually observed quality: evidence
Michelson et al. ( 2021) find a high incidence of fertilizer with degradation in observable physical characteristics including the presence of powdering (in which granules lose their structural integrity), caking (in which the fertilizer forms a hard aggregate), and impurities.More than 30% of fertilizer sampled in Morogoro region, Tanzania, had evidence of at least one of these issues.IFDC reports also document that these visually observed quality characteristics are commonly degraded: 15% of bags sampled in Uganda, 90% of urea samples in Senegal, and 59% of samples of urea from Togo and Côte d'Ivoire exhibited medium or high levels of caking.Fifty percent of the samples from West Africa were assessed by IFDC to have fine particles, with granular integrity moderately or significantly compromised.Asante et al. (2021) find that 13% of NPK samples from Ghana have some degree of modest granule segregation and 19% have some caking.
IFDC reports (Sanabria et al., 2013;2018a;2018b) document that fertilizer bags are often underweight. 28For example, 41% of bags in Nigeria, 28% in Côte d'Ivoire, 13% in Senegal, 12% in Ghana, 7% in Togo, and 10% in Uganda were underweight.Results in Kenya showed that the frequency of underweight bags increases as the bag size decreases, with 38% of sampled 10 kg bags underweight, 28% of 25-kg bags, and 19% of 50-kg bags.IFDC researchers caution that they are unable to ascertain whether they are underweight due to deliberate tampering or due to poor process control during bagging or re-bagging.
As noted above, powdering and granule segregation can lead to uneven distribution of nutrients in fertilizer bags.This can be important both for sampling to assess quality and for small farmers who may purchase fertilizer in small quantities of one or two kilograms scooped directly from open bags or repackaged for sale, or for famers who purchase larger quantities but with use spanning a longer time period, such as multiple agricultural seasons.IFDC notes a relationship between high moisture content of the samples and high caking.Moisture content and granular segregation were found to have a negative relationship with nutrient content in NPK blends in West African samples (Sanabria et al., 2013) but this mapping is not clear.Urea, for example, often exhibits caking and granular degradation without any deviations in nitrogen content.Samples with observed quality issues related to powdered granules and discoloration are often found to have good nutrient content and samples with no observable problems (especially blends) can be found to have deficiencies.Evidence in Michelson et al. (2020) suggests Notes: a not precisely discernable from the report.20 samples were out of compliance with nitrogen content between 44% and 45.5%. 26The classification of these samples as adulterated was based on careful assessment.The samples were initially tested in West Africa, then sent to the IFDC laboratory in Alabama, United States.After initial analyses indicated that the samples contained no phosphorous, the researchers used X-ray mineralogical methods to characterize the spectrum of each sample. 27The authors of the 2013 IFDC report write (p.xiv): "Trained inspectors reported evidence of adulteration in 31 of 134 (23 percent) samples collected in Coˆte d'Ivoire but only 14 of 414 (3.4 percent) samples from Nigeria.However, the only cases of completely proven adulteration are the seven samples of SSP from Nigeria that were found to have no P2O5 content or any of the minerals that carry P in phosphate rock.While high percentages of nutrient deficient samples in some NPK blends found in some countries could be interpreted as fraud during manufacturing or along the distribution chain, this is not substantiated by findings of this study; the lack of or poor control of blending procedures and use of inadequate blending equipment are also possible explanations.
28 As discussed, tolerance limits for weight shortages are 1% of the labeled weight based on international standards.A 50-kg bag with a shortage greater than 0.5kg is considered out of compliance, for example.
H. Michelson et al. that farmers use these observables as a signal of unobservable nutrient content (more below on farmer assessment).

Pesticide: quality measurement and evidence
For pesticides, we focus our discussion on the most widely used product.Glyphosate herbicide is generally sold in concentrations of 36 and 43.9 percent glyphosate by weight.Glyphosate formulation verification is performed in a lab using high-pressure liquid chromatography with ultraviolet detection.The method compares tested samples to a reference sample and is the standard procedure to measure glyphosate concentration (Morlier and Tomkins, 1997).Samples are tested in duplicate.Haggblade et al. (2021) use this method to test 100 samples of glyphosate acquired in Mali and Ashour et al. (2019b) use the method for their 483 Ugandan samples.Haggblade et al. (2021) note that Mali had no accredited lab to test for glyphosate concentration and so samples were tested both in a West African lab outside of Mali and also in the United States.Costs for glyphosate testing are high relative to fertilizer testing: $175 per test at a private Ugandan lab (2014), $195 per test in a private U.S. lab, and $50 per test at a Ugandan university laboratory.
Several recent studies and reports (Counterfeit Pesticides Across Europe, 2008;Fishel, 2009) argue that counterfeit pesticides are increasing globally.Only a handful of academic studies have tested the quality of pesticides for sale in markets in low-income countries.These studies have primarily focused on glyphosate.Haggblade et al. (2021) test 100 samples of glyphosate collected from four primary agricultural markets in Mali and find that on average the samples contained only 87% of the labeled concentration of the active ingredient.Samples varied substantially however: glyphosophate concentration varied between 59% and 103% of the labeled value.Haggblade et al. find a large experimental error in the tests conducted however, and stress that improvements in laboratory testing capability in West Africa are critical.Ashour et al. (2019b) find that sampled bottles (483 samples from 120 markets in 25 districts in Uganda) are missing on average 15% of the advertised amount of glyphosate, with 31% of the samples containing less than 75% of the advertised concentration.Ashour et al. are unable to distinguish between counterfeiting, adulteration, storage problems, or manufacturing errors. 29aggblade et al. ( 2022) find a strong relationship between bad formulation and a brand being unregistered in their study.They find no relationship between price and the accuracy of the stated concentration, nor do they find that older products (based on the labeled date) are more likely to have quality problems.

Farmer beliefs
It can be difficult for farmers to evaluate the quality of planting material, fertilizers, and herbicides.The quality signal may be somewhat easier for farmers to detect with herbicides as evaluating the effectiveness eliminating a weed is likely more direct and immediate than fertilizer's effect on plant growth and yields, however problems related to herbicide suitability and application technique can obfuscate and obstruct learning.Non-labor agricultural inputs are experience goods but the weather-driven stochasticity in agricultural production and lack of knowledge about proper use can make them effectively credence goods whose quality cannot be evaluated even after use.Farmers may not be able to assess for days or weeks after application whether the applied input "worked".They need to see if the plant grew, if the leaves developed discoloration characteristic of unaddressed nutrient shortages, whether the weeds or the insects died.But other factors can also contribute to the effectiveness of these inputs: whether the farmer uses the right input or formulation given particular pest pressures, soil quality, or growing conditions; the timing and amount of rainfall; the timing of application and whether the input was correctly applied.Bold et al. (2017) and Hoel et al. (2022) demonstrate through modeling and simulations the difficulties that these factors present for farmer learning about the quality and effectiveness of the inputs they purchase and apply.

Farmer beliefs about planting material quality
Farmers naturally care deeply about the quality of the planting material they use.All other inputs they apply in the hopes of a good harvest are based on the belief that the seed will respond favorably to the right growing conditions.Forming accurate beliefs about seed quality can be challenging, particularly for smallholder farmers subject to so many shocks and factors outside their control.This degree of background production noise can make it hard to infer seed quality based on one's own or others' experience alone.In the context of branded seed purchased from retailers, packaging, certification and trust in the agrodealer can help to reassure farmers about seed quality, but even these beliefs are often subject to a barrage of accusations and anecdotes from other farmers and from local media that raise suspicions about fraudulent or counterfeit seeds (see Zahur, 2010or Han, 2009 as examples of unsubstantiated media reports).
While farmers' first concern with seed is whether it will even germinate, there is little research about farmers' germination expectations.For decades researchers have elicited farmers' subjective yield expectations for various applications and questions, but these techniques cannot isolate the independent contribution of planting material to these beliefs.Recent and ongoing work has, however, shed light on farmers' beliefs about the varieties they sow and how these beliefs change their input investments, agronomic practices and production outcomes (e.g., Wossen et al., 2019a, Wossen et al., 2022). 30The fact that farmers often adopt local names for varieties (often describing a distinctive phenotypical feature of the variety) that differ from the official name and may differ from one location to the next poses a core complication in this work.In response, researchers have generally opted for a higher-level classification of improved or non-improved seeds, which can be quite imprecise in practice.The criteria for improved seeds may be clear to breeders, as it suggests a role for purposeful formal breeding and approval of the variety by regulatory process, it is much less clear what farmers have in mind when they report that a variety they are growing is "improved." 31In some cases, farmers may see as improved any seed they purchased from an agrodealer, especially if it was packaged and branded.In other (arguably rare) cases, distinctive traits (e.g., orange fleshed sweet potato) may clearly convey the coming years.Much of this work is explicitly part of an effort to leverage improved measurement tools in data collection to better understand how nonclassical measurement error affects statistical inferencein ways that can either produce misleading conclusions and reveal new insights about farmer beliefs, decisions and behavior (see Abay et al. 2021, Abay et al. 2022, Abay et al. 2023). 31There is a large literature from other disciplinary perspectives related to this topic that emphasize indigenous knowledge and agroecological diversity (e.g., Briggs 2005), which is beyond the scope of this paper and the papers reviewed herein that focus on whether farmers can identify whether the specific varieties they grow have been improved by modern breeding efforts.
H. Michelson et al. improved status of the planting material. 32ncertainty about seeds has real consequences for farmers.In two double-blind studies in Tanzania, Bulte et al. (2014, forthcoming) demonstrate that farmers make input investments based on their beliefs about whether they are growing an improved variety or not.In the case of cowpea, a farmer who believes he might be growing an improved variety increases inputs that he perceives to be complementary to improved germplasm to the extent that his yields are indistinguishable from the improved seedseven when he actually planted unimproved seeds (Bulte et al., 2014).In the case of maize, uncertainty about whether the planted seeds are improved triggered a different response entirely, leading farmers to reduce labor inputs, which may reflect their baseline suspicions about maize seed quality available in the market (Bulte et al. forthcoming).Surrounded by such suspicions and by local media reports about counterfeit seeds in the market, farmers fret about seed quality but a lab-in-the-field experiment in Kenya suggests they do not rely on informative packaging signals to update their beliefs unless specifically trained to do so (Gharib et al., 2021).
Genotyping allows researchers to resolve uncertainty about the genetic content of farmers' planting material.The first wave of genotyping work aimed to assess whether farmers' beliefs about whether their crops are improved or not align with DNA fingerprinting-based identification of their crops.These studies find quite high levels of misclassification of cassava (Maredia et al., 2016;Floro et al., 2017;Wossen et al., 2019a, Wossen et al., 2019b, Kosmowski et al., 2016).Represented as a confusion matrix with matched beliefs about improved status (i.e., true positive, true negative) on the diagonal and misclassified beliefs (i.e., false positive, false negative) off-diagonal, Kosmowski et al. (2016) finds 20% of farmer beliefs as false positive and 19% as false negative for sweet potato in Ethiopia.Wossen et al. (2019a) find 25% and 10%, respectively, for cassava in Nigeria.
A few of these studies (Maredia et al., 2016;Kowsmowski et al., 2016) experiment with other elicitation techniques to test whether this degree of misclassification is a reflection of farmers' inaccurate beliefs or researchers' poor elicitation of these beliefs.They find that visual aids to help farmers identify improved varieties help reduce misclassification, but only slightly.It seems that a non-trivial share of farmers simply have mistaken beliefs about whether they are growing improved seeds.Subsequent research suggests that misclassification may lead to biased estimates of the productivity of improved germplasm.Wossen et al. (2019a) show that because beliefs are correlated with farmer characteristics, which are in turn correlated with productivity, misclassification "produces a bias of about 22 percentage points in the productivity impacts of adoption" (p.1).In a follow-up study, Wossen et al. (2022) show that most of this misclassification is due to misperception rather than misreporting (i.e., intentional misrepresentation) by farmers and that these false beliefs potentially lead to suboptimal input investments and production practices.

Farmer beliefs about mineral fertilizer quality
Evidence suggests that small farmers have concerns about fertilizer nutrient content and that those concerns may negatively affect their purchasing.Sanabria et al. (2013) indicate in a study of fertilizer in West Africa that farmers report beliefs that adulterated urea is widespread in their markets but that this suspicion lacks scientific support (p. 39).A possible contributing factor: many small farmers purchase fertilizer in quantities less than the 50 or 25 kg bags available from manufacturers.Small farmers purchase one or two kilograms at a time from open bags in agri-dealer shops or in small plastic bags that are repacked for sale from opened bags by agri-dealers.This practice of selling fertilizer from open or repacked bags is a source of considerable suspicion regarding fertilizer quality in many markets, both among farmers and among fertilizer regulators.In several countries the practice of purchasing from open bags is illegal but still a primary means by which small farmers acquire fertilizers.
Five studies (Bold et al., 2017;Ashour et al., 2019a;Hoel et al., 2022;Maertens et al., 2021;and Asante et al., 2021) have directly elicited farmer beliefs about mineral fertilizer quality.This work draws on an emerging literature measuring and analyzing subjective beliefs in lowincome country contexts (for good surveys of this work see Attanasio, 2009 andDelavande andGine, 2011).Consistent with other academic studies on fertilizer quality in low-income countries, these elicitations have focused on urea.These studies all find that farmers on average believe that there are quality problems in urea in their local markets, despite evidence to the contrary.Hoel et al. (2022) and Maertens et al. (2021) elicit farmer beliefs about fertilizer quality in Tanzania by asking farmers to think of their local market and "imagine that ten farmers from your village would visit agro-dealer shops in [this market] during the long rains season and each purchase 1 kg of fertilizer."Then they ask, "If 10 farmers in your village purchase 1 kg of fertilizer at [this market] during the long rains season, how many would get good quality bags?They ask farmers how certain they are about their response.Hoel et al. find that 70% of farmers believe that some fertilizer in their local market is bad; on average, farmers believe that 34% is bad quality.Farmers also report considerable uncertainty about these beliefs.Maertens et al. find that farmers believe that about 30% of fertilizer in their local market is bad.Bold et al. (2017) ask farmer respondents "to assess the quality of fertilizer on a scale of 1 to 10, where 0 means there is no nitrogen, 5 means that half of the official nitrogen is there, and 10 is the best possible quality".Bold et al. find that farmers expect urea in their local shop to contain 38% less nitrogen than the manufactured standard; this means that farmers on average expect urea to contain 28.4% nitrogen by weight rather than 46%.Ashour et al. (2019a) in Uganda also ask how many out of ten farmers who went to a local market to purchase fertilizer would buy bad fertilizer.They find that farmers believe that 35 percent of the fertilizer in their local market is of bad quality; but elicited distributions exhibit considerable uncertainty in this stated belief. 33 Asante et al. (2021) asked both input dealers and farmers in Ghana to assess the quality of fertilizer in their district by estimating how many bags out of every ten bought and sold were good quality versus substandard quality.Researchers asked the question separately for commercial fertilizer and for fertilizer available through Ghana's subsidy program.Input dealers proved more pessimistic than farmers.Based on their own experience, input dealers estimated that 45% percent of the commercial fertilizer was of bad quality and 31% of the subsidized fertilizer.Farmers reported that 28% of the commercial fertilizer in their district was likely substandard and 19% of the subsidized; 47% of agri-dealers reported that fertilizer quality issues are among the most frequent complaints they receive from their customers.
Fertilizer quality can be difficult for a farmer to assess based on observation or experience, especially given the stochasticity of production outcomes driven by weather variability (Bold et al., 2017).In particular, Hoel et al. (2022) argue that farmers are prone to misattribute low yields to bad fertilizer rather than to weather shocks, misapplication, or timing issues.In the presence of uncertainty about fertilizer quality beliefs, Hoel et al. show that this tendency to misattribute bad outcomes to bad fertilizer can make it impossible for farmers to learn about the true (good) quality of urea fertilizer over time.Some of the fertilizer was likely counterfeit/adulterated: all of it, most of it, some of it, none.
H. Michelson et al. error could stem from information problems related to appropriate fertilizer type or correct application rates.Other error could result from bias in farmer estimates of plot size as they transfer application quantity recommendations to their own plots (Abay et al., 2022;Bevis and Barrett, 2020;Abay et al., 2021;Gourlay et al., 2019).
Evidence suggests farmer beliefs about quality issues may affect fertilizer adoption.Hoel et al. (2022) and Michelson et al. (2021) provide some evidence of the relevance of beliefs to purchasing using willingness to pay assessments.Both studies find that farmers are willing to pay more than 40% more -and 40% more than the prevailing market price at the time -for urea fertilizer that has been lab tested and found to be pure.Ashour et al. (2019a) also find that 69% of farmers avoid buying fertilizer because they worry about quality.A single study (Maertens et al., 2023) has used a randomized controlled trial to show that information can change beliefs about urea quality and by changing beliefs also change purchasing and use.They use a randomized information campaign of posters and flyers to inform farmers and agri-dealers that urea tested in the markets two years previously was of good quality.They find that the information exposure significantly improves farmer beliefs about urea quality and increases purchasing.The effect is driven by changes at the extensive margin, by farmers who were not previously using fertilizer.

Farmer beliefs about herbicide quality
Haggblade (2021) notes that farmers are generally unable to distinguish the registered herbicides from those that are fakes in the marketplace.Ashour et al. (2019b) use a field experiment to measure beliefs about glyphosate herbicide.They asked 1,390 households to imagine that "10 farmers like themselves" were to go to their local market and purchase one bottle of glyphosate herbicide each.They were asked how many of those 10 purchased bottles would be counterfeit or adulterated.82% of surveyed farmers who had previously used herbicide believe that glyphosate quality in their markets is likely to be tampered with.Farmers believe that 41% of glyphosate in their local market was counterfeit or adulterated.They find that farmers living in areas with worse quality glyphosate (based on the test results) on average believe that quality is worse.The relationship is statistically significant but economically small in magnitude: beliefs about the prevalence of bad herbicide are only 4.8 percentage points lower in the worst market than in a "perfect market" with no quality problems in their data.Their results suggest that farmers may have only limited ability to identify quality problems on average.

Discussion
We summarize and synthesize this review in four key findings and one concluding reflection, each with an implication for future research.We discuss each of these in this section, along with associated research recommendations and priorities.
First, the existing literature on agricultural input quality is narrowly focused on a few key inputs, predominantly in the SSA region.While the inputs that have received the most attentione.g., urea fertilizer and glyphosate herbicideare indeed important, there is a clear need to prioritize quality testing of a broader set of agricultural inputs.Further testing of urea fertilizer in SSA, for example, may not be especially insightful given the documented lack of variation in its quality, the difficulty and lack of economic incentive for urea adulteration, and the scarcity of manufacturing problems for single-nutrient prills.Instead, future testing might more constructively focus on blends including NPK, DAP, and CAN as well as blends including micronutrients including zinc and boron which are also widely used by small farmers and more prone to quality issues due to manufacturing problems or issues with granule segmentation in transport and storage.Testing of agricultural lime, fungicides, and other agri-chemicals as well as on regions outside SSA is similarly promising.In particular, given the consequences of agri-chemical for local environments and also for human health (Sheahan et al., 2017), additional evidence should be gathered with special attention to the health and environmental effects of counterfeit and adulterated herbicides.Future work could focus on the way that farmers understand these environmental and health risks, and to what degree their concerns about the special dangers of counterfeit herbicides might interact with and inform the adoption decision.The results we review showing variable quality for some inputs, a lack of evidence for others, but widespread suspicions among farmers, have implications for the host of initiatives related to digital agronomy and digital extension (Aker et al., 2016;Birner et al., 2021;Fabregas et al., 2019), which include in some cases tailored site-specific recommendations for fertilizers and crop management based on assessments of local soils (Harou et al., 2022;Corral et al., 2020).The potential of these spatially calibrated recommendations requires that the quality of the nutrient contents of recommended fertilizer blends and compounds be correct, known, and trusted by farmers.Work reviewed in this article suggests that this not may not always be the case.
Second, we recommend standardization and documentation of rigorous testing in reports and academic outputs making claims about planting material, fertilizer and pesticide quality.Maize seed, especially hybrid seed, has (rightly) been the focus of many studies that use genotyping to test for varietal purity, but these results are often difficult to interpret because of reference library complications that stem from genetic complexities with these varieties.In contrast, some of the most reliable genotyping evidence is for clonally-propagated crops like cassava (e.g., Wossen et al., 2019aWossen et al., , 2019bWossen et al., , 2022) ) because the method is more straightforward for cuttings than for seeds.Yet, these critical differences in the quality of the evidence produced by genotyping planting material does not come through clearly from these studiesprimarily because serious complications in the case of maize hybrids are often glossed over or poorly articulated.Future research on the quality of fertilizers and herbicides using lab-based measures should likewise carefully document details related to the laboratory used, the certifications of that laboratory (whether for example the lab is ISO certified), and a measure of the analytical error in the result.For all inputs, informed reviewers should request these details as a part of their peerreview, and editors may need to search a bit more and a bit differently to find qualified reviewers of these methods and measures that remain novel to most economists and other social scientists. 34 Third, research should continue to measure farmer and agri-dealer beliefs alongside lab-based quality assessments and to design and evaluate interventions that enable farmers and agri-dealer to update their beliefs.Work to understand the origins, the persistence, and the evolution of these concerns is also important.For example, given the frequent use of blended fertilizers by small farmers and evidence that suggests more quality issues occur in blends, it may be that farmer beliefs about fertilizer quality more broadly originate in experience with bad blends.Similarly, research documents that farmers are also suspicious regarding glyphosate quality and seed quality, inputs where evidence so far has identified high prevalence of quality issues.It may easier for farmers to learn about true quality for some inputs than others.In particular, the quality of pesticides, which are used in land preparation and to deal with incipient insect outbreaks or other problems, may be more easily and 34 For example, a handful of studies include speculation about quality problems in fertilizer or discuss market circumstances being conducive to fraud or fakes in purchased input value chains; some of these cite newspaper reports or undocumented government reports as evidence of quality problems (Liverpool-Tasie et al., 2010;Khor and Zeller, 2012).We tried to track down many of these newspaper reports and government documents.Other studies describe test results without presenting the data or describing the testing protocol and/or sampling.Jacob et al. (2012), for example cite unpublished data on 11 zinc sulfate samples in the Philippines, eight of which they report they found to be considerably short in zinc content.
H. Michelson et al. more immediately assessed by farmers.Further investigation of such beliefs and their evolution may be important in their own right for understanding adoption frictions, but may also lead to broader insights about learning, the evolution of suspicions over time, the social dynamics of input beliefs, and documenting and interpreting patterns of spatial heterogeneity.However, research strongly suggests that elicited beliefs are unlikely to provide a good proxy for actual quality available in the local market (Maertens et al., 2023;Michelson et al., 2021;Hoel et al., 2022;Ashour et al., 2019).
Fourth, for inputs with documented quality issues, quality testing and analysis should be conducted further upstream in supply chains to better understand the origin and nature of the problem.Such testing could lead to important work at the intersection of development, industrial organization, and information economics.Research in this area might focus on testing incentives or policy strategies to raise quality.For example, in the context of planting material, the incentives faced by the seed multipliers and nurseries can drastically reduce quality due to a combination of carelessness and lack of awareness and training (e.g., Diiro et al. forthcoming).Producing hybrid seed at scale demands a great deal of expertise and precisionand lapses at these upstream stages could introduce major quality issues downstreambut without evidence about where quality is leaking out of the system it is impossible to even formulate policy recommendations.Similarly, quality issues with fertilizer blends may originate in the manufacturing stage and could be addressed with very targeted interventionsbut only if there was an evidence base on which to design them.
Finally, a concluding reflection: input quality in low-income countries is among a range of factors that shape small farmer production outcomes.We have presented and synthesized evidence on the quality of single inputs in isolation and we have provided guidance regarding established and emerging methods to measure quality.But in practice, the critical question for policy and research is whether a given input reaches its full potential in a given production context.This question is harder to answer, harder to measure and harder to assess.But it is the question that matters to farmers.
Focusing on that broader question -how small farmers can attain the full potential of agricultural inputs -requires consideration of numerous relevant complexities of agricultural production in concert with input quality (Just and Pope, 2001).These relevant complexities include understanding how and when inputs are applied, establishing complementarities between inputs, and analyzing threshold effects in application (Burke et al., 2022).
Potentially reinforcing relationships between application practices, low yields, and beliefs about input quality and efficacy suggests the need to better understand not merely the amounts of inputs farmers apply but also when and how they apply them.This will involve improving measurement of farmer application quantity and timing.A small but growing literature has studied the effects of the timing of non-labor inputs application (Jagani et al., 2021;Islam and Beg, 2021).Similarly, only a modest amount of work as yet has focused on experimenting with methods to improve the accuracy of input application quantity measurement (Beegle et al., 2012;Wollburg et al., 2021;Mueller et al., 2022). 35Conceptually, we need models of farmer decision-making faithful to these complementarities to inform analysis and data collection.For example, a farmer's expected marginal productivity for a given input will influence decision-making and behavior; but how and when and in what quantity the farmer applies the input will itself contribute to the observed marginal productivity (the "quality" signal) of the input in the field.This in-the-field quality assessment will then inform the farmer's decisions regarding whether, when, and how much of an input to apply in the next season.Dimensions of the farmer decision-making process related to complementarities between belief, expectation, and agronomy have not been well-captured in agricultural household surveys in the past and deserve future attention.collection efforts, head-to-head within-survey experiments to compare methods for measuring quantity of input application do not seem to have been undertaken as yet.Within-survey experiments about agricultural non-labor input timing could also be conducted.We encountered no research empirically exploring methods for measuring the quantity of fertilizer or herbicide applied using within-survey variation in recall periods or input application diaries.
H. Michelson et al.

Fig. 2 .
Fig. 2. Share of crop farming households using inorganic fertilizers.Left panel presents Sub-Saharan African countries.Right panel presents non-SSA countries.Source: RuLIS.
H.Michelson et al.

Fig. 3 .
Fig. 3. Share of crop farming households using agri-chemicals.Left panel presents Sub-Saharan African countries.Right panel presents non-SSA countries.Source: RuLIS database.

Table 1 Overview of urea fertilizer quality sampling test results: published studies and reports (a version of this table is also in Hoel
et al. (2022)).