This result could indicate that having other managed systems nearby results in fewer source populations of pests in the immediate surrounding, whereas having widespread cropland in the landscape leads to a concentration of pests [i.e., resource concentration hypothesis ] or a lack of natural enemies [enemies hypothesis ] within the broader agricultural region. If so, the spatial location of additional cropland is important. In practice, this result implies that reducing insecticides per production area does not require reducing agriculture broadly, but coordinating crop types and crop locations in the nearby surroundings. Of course, farmers do not plant crops haphazardly, and policy mechanisms that sought to leverage this relationship by incentivizing coordination of crop area via agri-environmental programs would need to subsidize the lost revenue from planting crops with lower expected revenue. Further, as with diversity, the benefits of manipulating surrounding cropland extent would be crop-specific. Manipulating regional diversity and cropland extent requires coordination among multiple growers, and thus field size may be easier to modify through policy levers. We find large fields use more insecticides per area, even after controlling for crop type and regional attributes. From an ecological perspective, it is surprising to observe different effects of cropland extent compared with field size because both attributes are expected to increase the likelihood of pest problems by inhibiting natural enemy spillover or facilitating pest movement. However,growing blueberries from an economic or human behavior perspective, there are important underlying differences between large expanses of the same crop under multiple owners compared with a single owner.

For example, if pests are mobile and the population is shared between more than one farm, farmers may be more willing to spray on large fields because the influence of surrounding growers’ management decisions would be reduced . Thus, for a given pest distribution across space, the underlying spatial configuration of field size and ownership may be an important and understudied factor for how pests translate into pesticide use. Of course, landscape characteristics are dwarfed by individual crop attributes. For example, for diversity to drive application rates on oranges to be similar to application rates of carrots, diversity would need to increase by ∼10 SDs. Although improved seeds and integrated pest management may reduce application rates on a given crop, differences between crop types in their susceptibility to insect pests and the value lost due to pest damage are, at least in part, intrinsic characteristics. What ecologists are seeking to understand is not whether forcing substitution of carrots for oranges will reduce insecticides but rather, given demand for the suite of crops in production, what modifications can be made on farm and in the nearby surroundings such that pest numbers are reduced, thereby reducing insecticides and improving yields. Increasing agricultural production to meet future demand is a major challenge of the 21st century, with important consequences for human and environmental health. Ecological discussions on how best to achieve a sustainable and abundant global food supply centers, in large part, on how to allocate land to agriculture and how intensively to produce on agricultural land to balance high yields with acceptable levels of off-farm externalities, such as biodiversity loss and contamination of nearby human and natural communities. The simplified and contentious land-sparing/land-sharing dichotomy has persisted, in part, due to the difficulty of disentangling different components of a model landscape for a variety of different ecosystem services, including pest control. Here, we illustrate that, indeed, the effect of “landscape simplification” depends on different components of simplification, the spatial scale, and the focal crop.

Although land sparing/sharing may be a convenient division, to achieve major gains in insecticide reduction will necessitate crop-specific and spatially informed management.There has been a long and unique tradition in agriculture of using an experimental approach to establish rigorous causality between treatments and outcomes . It most often concerned new seeds released by breeders and new practices proposed by agronomists such as fertilizers and irrigation, and their impacts on yields. But it had been confined to the laboratory and the experiment station. When done in farmers’ fields for demonstration purposes, comparative plots were implemented following strict protocols designed by scientists. The Banerjee-Duflo-Kremer Nobelists made the important contribution of taking the experimental approach to the field under conditions where agency behavior becomes an important contributor to outcomes. Field experiments had been extensively used in medical research.They pioneered its application initially to issues of education and health. It quickly spread to all major questions in development economics, including rather naturally agriculture. This new adventure was endorsed by a multitude of research groups and sponsors. With this, a new culture and expertise in field research has emerged, with already major impacts on how to combat global poverty and in particular how to use agriculture for development. Use of field experiments in agriculture is happening at four levels running from the supply side to the demand side . A first level consists in making available to farmers what had been done by scientists under pre-specified conditions: assess the value of new technologies when used by farmers, particularly in terms of profit, food security, and risk. An example is experimentation with flood resilient new rice varieties, with not only a yield advantage under stress but also giving incentives to farmers to invest more in their crops, year-in-year-out, as downside risk has been reduced. Surprisingly, in Odisha where the experiment took place, yield gains from adjusted risk management turned out to be on average of equal magnitude as gains from better shock-coping . Countless other technologies have been experimented with under farmer agency including laser leveling, soil moisture retention, mechanization, etc.

A second level, also on the supply side, is for the new potentially profitable technologies to be known and understood by farmers so they can decide on adoption. This is difficult in agriculture as farmers are widely dispersed and heterogeneous in the conditions under which they operate, making any extension service of limited effectiveness. This led to search for better entry points in selecting contact farmers to facilitate social learning , ways of creating incentives for contact farmers in diffusing information , approaches to induce community members to seek information from informed farmers , use of IT services such as phone consultation platforms, customized recommendations based on soil testing, and SMS reminders on when and how to act . A third level is to explore the demand-side constraints to adoption of technologies known to farmers. Most remarkable has been to experiment with the design of multiple institutional innovations in pursuit of removing these constraints . Major constraints are financial liquidity, exposure to uninsured risks, high costs in accessing markets, and inconsistent decision-making. Examples of such institutional innovations included progress with the design of credit products better customized to the seasonality of agriculture ,square plant pots linking credit and insurance and making index-based weather insurance into a better product with more effective demand , reforming markets to improve quality recognition of inputs and products and reduce transaction costs , and offering options to contract for nudges facilitating time-consistent decision making in purchasing inputs . A fourth level is to take a broader look at what determines labor productivity in agriculture and well-being in rural areas. The first requires looking at labor calendars and farming systems that keep the land and family labor occupied throughout most of the year in spite of seasonality. This is the agricultural transformation, with the development of value chains, contracts, and links to high value markets . The second is the densification of activities in a rural non-farm economy that allows multiple sources of income. This is the rural transformation, with the development of local enterprises processing agricultural products, supplying agriculture with inputs, and delivering consumption goods and services to the rural population. Experimental research on these transformations is incipient and creates new challenges.Taking the experimental approach from the lab to the field has to confront challenges specific to agriculture which is simultaneously a dimension of nature, a sector of economic activity, and a way of life. This includes, among others, issues of spatial heterogeneity, seasonality and long lags, exposure to random weather shocks and risks, market failures and household decision making, and difficulties of measurement . We briefly address these in turn to illustrate how the method has been enriched to handle these difficulties. Spatial heterogeneity. Local agro-ecological and market conditions vary in space, especially under rainfed agriculture that characterizes 95% of Sub-Saharan Africa farming. With heterogeneity, the design of location-specific technological packages is expensive as it limits economies of scale. Implications for the design of experiments include: experimenting over a well-defined and potentially large range of heterogenous contexts; eventually making the observability of heterogeneity directly available to the decision-maker, either directly with leaf color charts that reveal soil conditions and fertilizer deficits, or through third-party services; and when heterogeneity is third-party unobservable, inducing self-selection to help farmers reveal at a cost their types in the dimensions of heterogeneity that matter to the experimenter .

Seasonality and long lags. Agriculture follows the seasons and most crops require several months and sometimes years between planting and harvesting. Decisions have to be taken based on expectations at planting time, and can be adjusted as the seasons unfold. Long delays have to be incurred between buying inputs and selling products, creating liquidity problems. Long lags are an invitation to time inconsistencies in decision-making. Implications for the design of experiments include: not isolating crops from the longer cycle in which they are imbedded; experimenting over at least a year which is costly and risky as conditions can easily change while the experiment unfolds ; and adapting experiments to changing conditions, which challenges the design of blueprint proposals and makes the registration of experiments difficult. Weather shocks and risk. Outcomes in agriculture depend on weather realizations that can overwhelm and thus obfuscate the experimental determinants. We typically only have limited observations both across space and over time about weather realizations. And weather itself is multidimensional and difficult to characterize. As a consequence, farmers’ understanding of the value of an innovation is necessarily highly incomplete, making decision-making about adoption difficult and potentially inducing costly errors. Implications for the design of experiments include: assessing a treatment conditional on the distribution of weather realizations which requires spreading experiments over wide geographic areas and sustaining them over a large number of years ; and if learning from others is important in adoption, facilitating information-sharing across locations and time, with a good characterization of prevailing weather and other local conditions. Other specificities of field experimentation in agriculture. To mention just a few, an important aspect of experimenting with agriculture in farmers’ fields is that decision-making is at the household level, not at the plot level. If markets are failing, and household decision-making implies non-separability between consumption and production decisions, understanding behavior in agriculture requires inserting it into household decision making with all its multi-dimensionality of objectives and constraints, and its agency complexity. In this case, field experiments need to collect data not only on the practice under investigation, but also on the many other features of households that affect decision-making on production and consumption. Surveys thus become multidimensional, with risks of inaccuracies, and high cost. The other specificity is difficulty of measurement. This includes such obvious dimensions as inputs used, yields achieved, prices paid and received, the opportunity cost of family labor, the valuation of self-provided inputs and services, and the measurement of quality.Water resources development is increasingly important for meeting nutritional needs in low- and middle-income countries, but agricultural intensifcation is also a driver of some human infectious diseases. Exposure to pathogens transmitted through environmental pathways is often greater for those undertaking resource dependent livelihoods. With few alternatives for reducing exposure, the resulting disease can impair economic development. As agriculture intensifes to meet human needs, it is important to understand unintended consequences of infrastructure designed to improve food security. Irrigation infrastructure is often implicated in the increased occurrence of human schistosomiasis, a snail borne parasitic disease second only to malaria in its global burden.