On the origins of food loss

In this paper, we try to understand what the main causes of food losses (FLs) are. Our results show that producers' education and experience and the number of years in which a producer has been involved in the production of a specific crop are significantly correlated with reduction in FL. Unfavorable climatic conditions, pests, and diseases, as well as limited knowledge and access to equipment, credit, and markets, are also chal-lenges to increasing production of higher quality and therefore reasons for FL. Policies to reduce and prevent FL need to be targeted to specific commodities and contexts. Why does so much food get lost along the food value chains? Little is known about what causes food loss (FL) in developing countries and how best to reduce them. It would be too simplistic to blame it on the carelessness of producers or vendors in the pre- or post-harvest handling of produce. FL can occur at different nodes of the value chain: production, harvest, or post-harvest stages, involving storage, transport, handling, or processing. Gaining insight into the causes of FL can help develop the right interventions. Even though it would be impossible to completely eliminate FL and waste, experts agree that there is room for reducing FL and waste.

Why does so much food get lost along the food value chains? Little is known about what causes food loss (FL) in developing countries and how best to reduce them. It would be too simplistic to blame it on the carelessness of producers or vendors in the pre-or post-harvest handling of produce. FL can occur at different nodes of the value chain: production, harvest, or post-harvest stages, involving storage, transport, handling, or processing. Gaining insight into the causes of FL can help develop the right interventions. Even though it would be impossible to completely eliminate FL and waste, experts agree that there is room for reducing FL and waste.
[The copyright line for this article was changed on 08 April 2021 after original online publication.] We implemented specially designed surveys to capture FL along five staple food value chains in seven countries: potato in Peru and Ecuador; maize and beans in Honduras and Guatemala; maize in Mozambique; teff in Ethiopia; and wheat in China. Cereal grains, such as wheat, maize, potatoes, and beans, are the world's most popular food crops and form the basis of the staple diet in most developing countries. Our aim is to gain a better understanding of the links between FL and a rich set of socioeconomic features, agricultural production, and postharvest treatment characteristics, as well as climatic conditions.
Methodologically, we use two alternative econometric models: the model of classical maximum likelihood estimation is used to assess the relationship between the right-hand side variables and the binary FL variable; fractional response models (GLM) are used on the share of product loss to account for the boundedness of the dependent variable. We use these models to estimate the relationship among these variables, using FL data. FL is defined through the "attributes method" (see details in Delgado et al., 2020). The results reveal specific areas that require investments to reduce FL. They also show considerable heterogeneity of FL. The causes of FL appear to be highly specific to the context and type of the commodity. This paper is organized as follows. The introduction is followed by a literature review (Section 2) on the causes of FL and waste in developing countries. Section 3 presents the data and empirical approach. Section 4 presents descriptive statistics and key findings for Ethiopia, Ecuador, Honduras, Guatemala, Peru, China, and Mozambique. Section 5 discusses the findings with respect to the scientific literature. The paper ends with conclusions and policy recommendations.

LITERATURE REVIEW ON THE CAUSES OF FL
A review of the evidence suggests a wide range of possible causes, categorized into six groups: levels of human capital (education, experience); climatic conditions, insects, or pest attacks; access to infrastructure and post-harvest infrastructure (especially storage facilities); access to technology, post-harvest crop management techniques, and handling; economic incentives (standards); and market access (mainly roads to markets). In practice, multiple factors are at play and reinforce one another. For instance, heat and humidity tend to damage perishable food products. It is more likely to be a problem in places where there is no temperaturecontrolled storage and transportation. The literature review is summarized in Table 1. Kaminski and Christiaensen (2014), Maziku (2020), Doki et al. (2019), and Gebretsadik et al. (2019) find that human capital, or education and experience, to be negatively correlated with reduction of losses: that is, the higher the education, the lower the level of FL.
Climatic conditions, such as high heat and humidity and post-harvest rainfall, have been found to be a major cause of post-harvest FL in many contexts. 1 In African countries, there is high dependence on sun-drying of crops among small-holder farmers. Post-harvest rainfall could lead to substantial losses if crops are not dried properly before being stored or taken to the market.
Insects and pest attacks on produce have also been identified as important causes, typically compounded by heat or moisture and poor storage conditions (Chegere 2018). John (2014), for instance, finds that rodents are a major factor for post-harvest loss (PHL) of rice in Southeast Asia. Abdoulaye et al. (2016) report that more than 75% of farmers in Ghana, Tanzania, and Benin identified insects as the major cause for PHL, while most farmers in Ethiopia, Uganda, and Nigeria reported rodents and moisture as the main causes for PHL. Finally, Compton et al. (1997) and Baoua et al. (2014) show that each percentage point of insect infestation results in 0.6%-1% depreciation in the value of maize. Certain climatic conditions, especially heat and  Negative moisture, tend to increase the prevalence of insects, pests, and other bio-deterioration factors, especially when proper storage and transportation structures that control temperature and humidity are lacking. Rosegrant et al. (2015) find that electricity, roads, and railways have an important role in PHL reduction. After getting the estimates of infrastructure on PHL reductions, the authors use the cost of infrastructure development to estimate a number of investment scenarios. These scenarios are later implemented in the IMPACT global food supply and demand model from the International Food Policy and Research Institute to simulate the impact of PHL reduction on food prices, security, consumer and producer surplus, net welfare gains, and benefit-cost ratios to the investment. Overall, the authors find that reduction in PHL is not a low-cost alternative; rather it requires large investments and should be part of long-term investments to achieve food security. Kasso and Bekele (2016), Macheka et al. (2018), Kumar and Kalita (2017), Folayan (2013), Paneru et al. (2018), and Maziku (2020) also identify lack of storage as an important factor behind the losses of horticultural crops, and lack of transportation facilities for losses of maize crops.
The risk of FL is further escalated by poor post-harvest crop management techniques and handling. The techniques that constitute proper handling may vary from case to case. Tefera (2012) finds that improper post-harvest crop management and harvesting techniques account for between 14% and 36% of losses in maize grains in Africa. Insufficient drying, excessive drying, and missing grains are some of the problems of the harvesting and drying stages. Other problems such as improper threshing and shelling, which can cause grain breakage and grain cracking, are predominant in this stage; transportation to storage facilities, on-farm storage, transportation to markets, and marketing are identified as other critical areas where losses occur in maize. Studies also point to credit constraints as a main bottleneck to technology adoption, preventing FL reduction. 2 Economic incentives affect PHL in a number of ways, although evidence is mixed. Goldsmith et al. (2015) demonstrate how poor market incentives lead producers of both soybeans and maize in tropical Brazil to accept significant PHLs during the intercropping season. Farmers cannot afford any delay in harvesting soybeans because they must ensure timely plantation of maize, a high-value crop, on the same land. Any delay in planting would expose maize cultivation to higher risk of loss. Since the opportunity cost of delayed plantation of maize is higher, it may lead farmers to harvest and handle soybeans hastily. This is especially so if the cost of hired seasonal farm labor is high relative to the market price. Therefore, this could lead to greater PHL for soybeans. Rosegrant et al. (2015) find that better infrastructure facilitating transportation of products to markets reduces PHLs, but that the impact will be stronger if farmers have better education, as it would enable them to adopt proper crop handling and processing techniques. The authors also find that PHLs are correlated with farm size. Larger farms are more likely to incur PHLs but experience fewer losses in the intensive margin. The overall impact suggests a negative relationship between the share of PHLs and farm size.

Data
We developed and implemented detailed surveys that allowed us to quantify the extent of FL at the producer level, using approaches that are comparable across commodities and regions. The survey enabled us to characterize the nature of FL, specifically during the production and particular processing stages. The same surveys were conducted in seven countries (Ecuador, Peru, Honduras, Guatemala, Ethiopia, China, Mozambique) for five crops (potato, maize, beans, wheat, teff). We adapted our instrument for the specifications of each crop and country (for more extensive information on the survey, see Delgado, Schuster, & Torero 2017, 2020.
In all the countries, the surveyed sample was based on pre-census registration of producers who had produced the specific crop of interest in the last cropping season, which formed our baseline. The representative sample extracted from the baseline comprises 302 potato farmers in Honduras, 411 potato farmers in Peru, 1209 maize and beans farmers in Honduras, 1155 maize and beans farmers in Guatemala, 1203 teff farmers in Ethiopia, 1114 wheat farmers in China, and 774 maize farmers in Mozambique.
The survey captures both quantitative losses and qualitative deterioration of the product, from pre-harvest to sale to an intermediary or end-user. While the survey instrument allows different ways to estimate FL along the commodity value chains, in this paper we adhered to what has been defined the "attribute method" (see Delgado, Schuster, & Torero 2020). The method is based on the evaluation of a crop according to inferior visual, tactile, and olfactory product characteristics. It leads to results that are comparable to alternative methods to estimate FL that have been used in other studies (Compton & Sherington 1999;Delgado, Schuster, & Torero 2020).

Empirical approach
We used a statistical framework to assess the association between different socioeconomic and production factors and FL at the producer level. It is important to mention that our analysis does not provide evidence on causal impacts, as this would require our explanatory variables to be strictly uncorrelated with other characteristics that are either omitted from the regression framework or unobservable.
With this in mind, our main goal is to determine the correlation between producer FL and socioeconomic characteristics, market access, agricultural production techniques, on-farm postharvest practices, and climatic and geographic variables (e.g., weather, pest, etc.). Given the uncertainties on the origins of loss, we believe that the intensity of correlations can provide insight into the causal effectiveness of targeted interventions for future studies.
For each commodity and country, we estimate regressions of the following type: where FL i,c,x is an indicator of FL of producer i in country c and for commodity x. FL is either a discrete outcome (0 if no loss; 1 if at least some loss) or the share of the lost production, as estimated with the "attribute method," and including both quantity and quality degradation. X i,c,x are a set of socioeconomic characteristics, Z i,c,x are agricultural production characteristics, and N i,c,x are post-harvest managing and handling techniques, including storage. W i,c,x is a proxy for production issues highlighted by the producer during the growing process or post-harvest stages (e.g., unfavorable climatic conditions, limited knowledge or information). While the first three sets of variables intend to capture characteristics, knowledge, and instruments available at the farm level, W i,c,x captures external growing conditions and limitations. Finally, location-fixed effects ᴨ v are included to control for common district, municipality, or village effects, and ρ AE are the agro-ecological zone dummies, which control for climatic conditions that could be correlated with farm loss. ε i,c,x is the unobservable error term. We use classical maximum likelihood estimation to assess the parameters β. Probit regressions are used to estimate the relationship between the right-hand side variables and the binary FL variable. Fractional response models (GLM) are used on the share of product loss to account for the boundedness of the dependent variable (Papke & Wooldridge 1996, 2008. 3 We calculate the estimated marginal effects for both models. Because estimation errors between different countries and commodities in the same geographical areas are correlated with the same idiosyncratic shocks, we cluster the standard errors at the geographic level disaggregation in each survey. Table 2 shows summary statistics of the producers across the different countries and commodity groups. Around 90% of all sampled producers are male in all the countries and across value chains. On average, they are 47 years old and have between 17 and 30 years of experience in growing the analyzed crops. Most producers have primary education. In Peru and China, almost half of the producers have also completed secondary education. Producers are rural smallholders. They cultivate between 0.35 ha (beans in Guatemala) and 3.5 ha (potato in Ecuador) of land. On average, they live 2.5 hours away from the closest village market.

Producer characteristics
Mechanization and technology adoption in production and post-harvest activities is low on average, but considerable variation exists across countries and crops. Around two-thirds of all farmers use improved seeds for teff in Ethiopia and for wheat in China. However, less than 20% use improved seeds to grow beans and maize in Guatemala and Honduras. Resistant crop varieties are not widely common in Peru, Ecuador, and Mozambique. Machine-driven production methods, such as soil preparation, sowing, pest control, fertilizer application, weeding, mulching, cutting, and harvesting, are most widely used in the Chinese wheat value chain and Peruvian potato value chain. However, they are almost nonexistent in the bean value chain in Guatemala, the maize value chain in Mozambique, and the teff value chain in Ethiopia. Mechanization in post-harvest activities is even less common. Only in Honduras do farmers engage in mechanical threshing of beans and maize; very few farmers in Honduras and Guatemala mechanically dry and winnow beans and maize. On average, producers use 2.5 different types of inputs to grow their crops (fertilizers, insecticides, herbicides, and fungicides), but there is a large variation between countries, ranging from almost no input (maize in Mozambique) to more than four different types of inputs (wheat in China).
In six out of the nine value chains, almost all producers store their grain as food reserves and seed for the next season for an average of 5 months (beans and maize value chains in Guatemala and Honduras; teff value chain in Ethiopia). About 50% of all wheat farmers in China and 30% of all potato farmers in Peru store their produce for an average of 1 month. Only farmers in Ecuador rarely store the potatoes they grow. Around 63% of all farmers store their produce in their house in bulk or in bags. Around 14% of all farmers store them in traditional storage facilities. Less than 10% of farmers use metal or plastic silos, with the exception of maize farmers in Honduras.  Finally, across all countries and commodities, on average, about 50% of the crops are sold by farmers. The share is around 80% for the potato value chains in Ecuador and Peru and for wheat value chain in China. The share is considerably lower at around 30% in Guatemala, Honduras, Ethiopia, and Mozambique. The product is sold directly to an intermediary on the farmers' plot. Table 3 provides insight into the likelihood of FL and the magnitude of losses across the different value chains and countries. As mentioned above, loss figures are estimated with the "attributes approach" described in Delgado et al. (2020). The methods assume that the producer evaluates the produce based on a specific number of quality attributes and defines the share of total production affected by the inferior damage attribute. The product attributes are identified and validated prior to the survey implementation in collaboration with commodity experts and local valuechain actors. 4 The quantity and quality degradation at the farm level are thus defined by the sum of the total produce loss (equal to the total amount that completely disappeared from the value chain between harvest and sale or consumption) and the share of product affected by a damage attribute (meaning not totally lost and can still be used, but the quality is degraded). This degradation can be expressed either in weight or in economic value (Table 3). Table 3 shows that most farmers suffered at least some weight or value loss in the previous harvest season. The figure ranges from 64% of all teff farmers in Ethiopia to 95% of all wheat farmers in China and 97% of all maize farmers in Guatemala. On average, 20% of the farmers' produce was lost. Figures range from 14% of all products lost in the potato value chain in Ecuador to 31% loss in the teff value chain in Ethiopia. Percentage losses expressed in value tend to be 4% smaller on average than those expressed in weight, indicating that some quality degradations at the farm level do not seem to be penalized by the market.

Likelihood and magnitude of FL
T A B L E 3 Total quantity and quality degradation at producer level (expressed in weight and value of total production) Note: Estimation of the loss through the "attribute method" (see Delgado, Schuster, & Torero 2020). Share of product lost, in weight = Quantity of product that disappeared from value chain + quantity of product affected by a damage attribute. Share of product lost, in value = Economic value of the product lost.

Regression results
Tables 4-8 presents the Probit and GLM regression results, respectively, on the probability of incurring a loss and on the share of produce lost. We classified the potential origins of FL in five groups: socioeconomic characteristics of the farmer; market access; mechanization and technology; storage facilities; and growing conditions (pests and disease) and climatic conditions. Overall, we notice that there is a considerable heterogeneity in the determinants of FL across commodity and country contexts. It is important to highlight that the models do not provide evidence on causal impacts; yet, they can be helpful for future hypothesis tests for causality.

Socioeconomic characteristics
Most farmers are men, but there is no clear gender pattern in FL across countries and commodities. For example, being a male farmer tends to be correlated with 4.9% to 10.9% less points share of beans loss, but it is associated with, respectively, about 10% points more likelihood to incur in a loss of maize and 5% more points share of maize loss in Guatemala and Honduras.
No correlation with gender is detected in other commodity value chains. Age, education, and experience tend to be negatively correlated with the probability and share of FLs. In particular, being older is associated with about 3% points less likelihood to incur in a loss in the maize value chain in Guatemala and Honduras. Formal education, such as primary, secondary, or higher education, significantly correlates with 5%-30% points reduction in losses in the potato value chain in Ecuador and Peru, the bean value chain in Honduras, and the wheat value chain in China. The number of years in which a producer has been producing a specific crop significantly correlates with the reduction in losses in the potato value chain in Ecuador, the bean and maize value chain in Guatemala, and the maize value chain in Mozambique. We have the farmers' income data only for Peru and Ecuador. In addition, we find that in Peru and Ecuador when a producer's main income stems from an agricultural activity, it is correlated with lower losses that is statistically significant (all else equal, a producer's main income that stems from an agricultural activity is associated with 47% points less likelihood of any loss in Peru and with 14% and 68%, respectively, points less share of FL in Ecuador and Peru). This result is in line with the outcome we find on crop cultivation experience.

Market access
The cost or time to reach markets has a significant correlation with increased losses in five of the seven countries. In Peru, Guatemala, Mozambique, Ethiopia, and China, the absence of markets can represent important limitations for farmers. Farmers in these countries decide not to market (or even harvest) all produce because of their high costs relative to the market price (an increase of the cost of a kilogram of produce to reach a market or the time-in 10 hoursto reach a market, can increase the share of produce loss by an average of 0.4% points). Mechanical transport with a car is associated with a significant increase of these costs through additional losses during travel in beans and maize value chains in Guatemala. The farmers in our survey mention the lack of access to markets and credits as a challenge to increasing production of high-quality products.
Surprisingly, mechanization and technology in production and post-harvest activities have negative correlations with loss across value chains and countries, highlighting the importance of adequate knowledge. In particular, the number of machine-driven activities, including cleaning, sowing, herbicide application, pest control, fertilizer application, and harvest, correlates with about 8% and 20%, respectively, points less likelihood of a loss in the bean value chain in Honduras and the teff value chain in Ethiopia, and with 3%-5% points reduced share of FL in the potato value chain in Peru and the maize value chain in Mozambique. On the other hand, the number of machine-driven activities correlates with increased losses in the Ecuadorian potato value chain (16% points more likelihood of a loss), Guatemalan maize value chain and Chinese wheat value chain (respectively, about 3% and 15% points more share of produce loss). The mechanization of harvesting tools considerably affects losses. For example, traditional hoes damage potatoes during the harvest. In Peru, new mechanized tools are used to reduce this damage: the use of both the tractor and the "lampa" has a correlation with a significant reduction of the share of potatoes lost during harvest (all else equal, 30% points less). Similarly, in Mozambique, mechanization reduces the likelihood of incurring a loss of maize (− 4% points). The potato value chain in Ecuador is more traditional, with very few mechanical tools used. Finally, resistant varieties or improved seeds have a consistent correlation with reduction in losses, with the correlation being significant in the potato value chains in Ecuador and Peru and the maize value chains in Guatemala and Honduras (between 4% and 8% points less likelihood of a loss and around 3% points less share of produce loss). Mechanical post-harvest activities are not widespread, with mechanical drying, winnowing, and threshing activities being observed only in the maize and bean value chains in Honduras and Guatemala. Increased mechanization in the drying and winnowing activities reduce loss in the bean value chain in Guatemala and the maize value chain in Honduras, but mechanical threshing increases losses in the bean value chain in Honduras and the maize value chain in Guatemala. Farmers likely incur grain damage, cracks, and lesions when mechanically (instead of manually) stripping the grain from the plant. This makes the grain more vulnerable to insects and visually less appealing.
Most of the harvesting is still performed manually in these countries, making it laborintensive and slow. During the harvest season, countries may face labor shortages, which can be resolved by hiring external labor. The hired labor force is mostly correlated with reduction of losses. This is significant in the bean value chains in Guatemala and Honduras (between 4% and 19% points less likelihood of a loss), the potato value chain in Ecuador and the maize value chain in Mozambique (between 4% and 8% points less share of produce loss).

The lack of adequate storage techniques
This can lead to FL due to biotic factors (pests, insects, fungi, and rodents), abiotic factors (rain, temperature, humidity), or spillage when filling or emptying storage space. Potato producers in Ecuador and Peru rarely store their product. However, other products included in our survey are grown seasonally, and after harvest the grains are stored as food reserves and seeds for the next season. All else equal, post-harvest storage has a correlation with increased loss in the bean value chains in Guatemala and Honduras, the maize value chain in Mozambique, and the wheat value chain in China (across all value chains and countries, we see between 14% and 37% points more likelihood of incurring any loss, and between 3% and 13% points more share of produce loss). In Honduras and China, the storage duration correlates with increased share of produce loss (between 0.9% and 1.1% points more). In most countries, grains are generally stored as bulk or in bags in the farmer's house or simple granaries built with locally available materials (mud and bricks). Improved storage infrastructure (silos or improved granaries) is associated with mitigation of these risks in the bean value chain in Guatemala, the maize value chain in Guatemala and Honduras, and the wheat value chain in China (between 18% and 31% point less likelihood of incurring in a loss; and between 2% and 16% points less share of produce loss). It is also the case in the teff value chain in Ethiopia, where "pits" are used instead of other traditional storage facilities. This is because they reduce the probability of insect infestation and mold growth. Storage conservation activities, such as chemical or natural fumigation or increased ventilation, are correlated with reduced losses of stored food in Honduras and Ethiopia.

Unfavorable climatic conditions, pests, and diseases
These are often mentioned as problems farmers face during production. In Honduras, Guatemala, Mozambique, and Ethiopia, unfavorable climatic conditions, as assessed by farmers, are positively correlated with the likelihood of incurring losses and the share lost (all else equal, climatic conditions are associated with 8%-18%, respectively, more likelihood of FL, and 3%-7% more share of produce loss). In addition, farmers mention pests, diseases, and rodents as major production problems.

DISCUSSION
We break down our results by five groups of potential origins of FL and compare them with those of other studies.

Socioeconomic characteristics
Our results on the impact of gender on FL are contradictory. Similar findings have been reported by Chegere (2018), who found that being male is correlated with reduced losses in the sub-Saharan maize value chain. On the other hand, Folayan (2013) and Ngowi and Selejio (2019) find that being male is correlated with an increase in losses in the maize value chain in Nigeria and Tanzania. Our results that age, education, and experience tend to be negatively correlated with losses are in line with most of the literature. Yet, in some contexts, they seem to have opposite correlations. Education has been found to have a positive correlation with losses in the maize value chain in Mozambique (Ansah, Tetteh, & Donkoh 2017;Doki et al. 2019). Ngowi and Selejio (2019) and Shee et al. (2019) analyzed the maize and white-fleshed sweet potato value chains in Tanzania and Uganda and found that age is positively correlated with losses.

Market access
In line with most studies, our results find that transportation is positively associated with FL owing to the additional costs imposed on the farmer and complexities in transporting food commodities. Chegere (2018) finds that maize farmers in sub-Saharan Africa experience more losses if they transport maize themselves. Gebretsadik et al. (2019) found increased losses due to the distance between the farm and the residence as well as the distance between the stacking place and the threshing place. The mode of transportation positively affects post-harvest grain losses in sesame in Ethiopia (Gebretsadik, Haji, & Tegegne 2019), maize in Tanzania (Ismail & Changalima 2019) and sweet potato in Uganda (Shee et al. 2019). These findings directly support previous studies' findings, which highlight the importance of road to reduce FL across the value chain (Rosegrant et al. 2015).

Mechanization and technology in production and post-harvest activities
The literature is full of conflicting effects of mechanization and adoption of technology on reducing FL. For example, Ahmed et al. (2015) find that losses are lower for fruits picked with scissors, rather than by hand, when it comes to the Pakistani kinnow value chain. Khatun and Rahman (2019) also find that shifting from traditional packaging to improved packaging decreases losses in the eggplant value chain in Bangladesh. Our findings on the Peruvian potato value chain are consistent with those findings. On the contrary, Shee et al. (2019) find that mechanization of harvesting considerably increased losses for maize and sweet potatoes in Uganda. These mixed results highlight the importance of adequate knowledge and training that must accompany the use of new tools.

The lack of adequate storage techniques
Post-harvest storage significantly increases the likelihood of losses (Ngowi & Selejio 2019), and our results on storage techniques confirm this finding by previous studies on FL. Previous studies have also found that losses significantly increased during longer storage periods (Aidoo, Danfoku, & Mensah 2014;Ismail & Changalima 2019). At the same time, the lack of modern storage facilities is positively correlated with losses (Folayan 2013;Maziku 2020;Paneru, Paudel, & Thapa 2018), demonstrating that improved storage infrastructure mitigates the risks of FL.

Unfavorable climatic conditions, pests, and diseases
Our finding that unfavorable climatic conditions increase the likelihood of incurring losses is in line with the literature. In particular, Ambler et al. (2018), Gebretsadik et al. (2019), and Maziku (2020) find this correlation when it comes to post-harvest rainfall in the value chains of maize, groundnuts, and soy in Malawi. This correlation was also found between wind and rain during harvesting to threshing time in the sesame value chain in Ethiopia and between rain and post-harvest activities in the maize value chain in Tanzania. Our results confirm previous findings, highlighting that the lack of rainfall causes significant pre-harvest losses for crops like potato, maize, beans, and teff in Ecuador, Peru, Honduras, Guatemala, and Ethiopia (Delgado, Schuster, & Torero 2020).

CONCLUSION
Identifying the causes and costs of FL across the value chain is critical for setting priorities for action. Analyzing the factors affecting FL at the micro-, meso-, and macro-levels can help in identifying effective reduction interventions.
Our results show that socioeconomic characteristics, such as education and experience, positively correlate with reduction of losses. In four out of the nine value chains studied, the association of education and the number of years a producer has grown specific crops with reduction of losses is significant. Unfavorable climatic conditions are positively correlated to losses in most countries, and the major production problems mentioned by farmers are pest, diseases, and rodents.
The techniques that constitute proper handling of produce may vary from case to case. For example, mechanical production activities increase losses in Ecuadorian potato value chain, Guatemalan maize value chain, and Chinese wheat value chain. On the contrary, it was traditional harvesting tools, such as hoes, that accounted for an important share of losses in Peru's potato value chain. Likewise, in Mozambique, mechanization reduced losses of maize. The number of inputs applied follows similar mixed trends. This emphasizes the critical need for knowledge and training in addition to adopting technology to effectively decrease losses. The lack of appropriate storage techniques is consistently correlated with higher losses: longer storage durations tend to exacerbate the losses. Improved storage infrastructure can mitigate these risks.
Finally, the cost of accessing markets has a significant correlation with increased losses in five out of the seven countries. This indicates that the absence of markets represents a critical limitation for farmers. This directly supports the findings of previous studies that show the importance of better roads to reduce FL across the value chain.
While there are commonalities, FL is very context-specific. The heterogeneity suggests that policies aiming at the reduction and prevention of FL need to be developed with specific commodity and context in mind.
More research is needed to identify the drivers behind losses. For example, disentangling the role of farmers' demography, education, producer experience, and gender is needed. It is necessary to analyze the factors related to production-access to technology and agricultural assets, infrastructure-geography, and climate. Furthermore, experimental studies on different storage techniques and mechanizations, as well as targeted training programs, can confirm the effectiveness of specific interventions on FL reduction.
These findings should be used to inform policies. Governments should ensure that public and private sector investments facilitate reductions in FLs by identifying the main causes of FL in specific commodities and contexts. Such investments cover a broad gamut of areas related to food systems, including food safety, education, infrastructure, regulations and standards, and market failures.
Small holders, who produce only small surpluses, often face substantial market failures that contribute to FL. Public sector investment can address some of these shortcomings, such as the need for appropriate storage facilities, efficient transport systems, policies that improve access to credit, support for market incentives for improved food safety as in the case of aflatoxins, and access to crop varieties resistant to weather shocks. Reducing FL can generate profits. For example, choosing appropriate crop varieties, dealing with pre-harvest pests, and making processing and retail decisions may be best addressed by the private sector. There is a clear need to build an evidence base on the efficacy of these reduction interventions, particularly when combined with training, changes in handling practices, and access to finance.

ACKNOWLEDGMENTS
This work was undertaken as part of the CGIAR Research Program on Policies, Institutions, and Markets, which is led by IFPRI and funded by the CGIAR Fund Donors and under the Food Loss and Waste program of FAO. 1 Ambler, De Brauw, and Godlonton (2018) and Tefera (2012), for instance, emphasize post-harvest rainfall as a main cause of food loss in Malawi and other Africa countries, while Kaminski and Christiaensen (2014), Basavaraja, Mahajanashetti, and Udagatti (2007), Arah et al. (2016), and Kasso and Bekele (2016) identify high heat and moisture as the main causes of food loss in sub-Saharan Africa and India. 2 HLPE, 2014. [This needs complete citation and should be added under references, rather than as a footnote.]