The livelihoods of hundreds of millions of rural families depend on farming. In many cases, farmers manage their crops, livestock and trees in ways that have not changed for decades. The practices—particularly in sub-Saharan Africa—often result in low yields, which jeopardize food security, nutrition, health and economic development. With so much riding on agriculture, transforming how agriculture is practiced and how food is produced is critical to future rural livelihoods.

Many levers are used to try to shift agricultural production, including policies, services, products and more. Although they act in different ways, all aim to affect farm- and field-management practices and technologies, a broad category that includes activities such as adopting drought-resistant crop varieties, reducing stocking rates of animals, changing harvesting and postharvest storage techniques, and many others. Regional and national policies are constructed around such practices, investments are structured to catalyze adoption of them, and the private sector develops products to deliver them. Implementation at massive scale is the key to transformative agricultural change.

The vast number of practices available and the equally vast range of possible results from using them create challenges in identifying what will work in which locations. Indeed, this has always been the central question for agriculture research. Given the need to transform agriculture, it is imperative that everyone has the best information available if we wish to design efficient and effective policies, programs and products.

ERA is a meta-dataset and analytical engine built to explore questions on the performance of agricultural technologies. ERA enables users to ask and answer questions about the effects of shifting from one technology to another on indicators of productivity, resilience and climate change mitigation outcomes. It was specifically created to uncover what the data reveal regarding what works where and for whom.

The search

We searched for evidence of the performance of agricultural technologies in the peer-reviewed literature using a systematic review protocol (Rosenstock et al. 2015). This search targeted information on 112 technologies used in crop, livestock and tree production and 58 indicators of performance (e.g., yield, net economic returns, soil carbon, etc). The search was conducted in Web of Science and SCOPUS and yielded about 150,000 articles across developing countries (49,000+ in Africa). A study was included in the resulting database if it contained primary, quantitative data and if the following criteria were met: (1) it contained data on both a conventional technology (a control) and an ‘improved’ agricultural technology, (2) it took place in a developing country and (3) it contained information on at least one of the selected performance indicators. Abstracts and titles were reviewed to see if they matched the inclusion criteria and, if so, we conducted a more in-depth analysis of methods. Applying these inclusion criteria resulted in nearly 7,000 articles with relevant data across the entire dataset and nearly 1,400 studies that took place in Africa.

The database

Currently ERA contains about 75,000 observations from about 1,400 agricultural studies that have taken place in Africa. Data have been compiled on more than 250 agricultural products. Management practices, outcomes and products are nested within respective hierarchies, allowing ERA to aggregate and disaggregate information. These data are linked to detailed covariates such as climate, soil and socioeconomics based on geographical coordinates reported in the studies and seasons reported or inferred from remote sensing. The beta version of ERA (currently online) contains information from the core agricultural dataset only; the environmental variables and analysis will be online in early 2020.

The analysis

ERA supports capacities to analyze the database in many ways. The fundamental analysis follows standard meta-analysis procedures. Meta-analysis is a statistical way to combine research results across studies (see below). Meta-analysis was first designed in medicine, then adopted in ecology, and more recently has been used in agricultural studies.

Specifically, ERA calculates response ratios and effect size statistics. The response ratio is calculated as the log ratio of the mean effect of the treatment (i.e., improved practice) against the mean effect of the control practice. These response ratios can then be combined to generate an ‘effect size’ for the relationship, which provides an overall estimate of the magnitude and variability of the relationship. When calculating effect sizes, results from different studies are weighted to reduce bias from any given study (e.g., one that had hundreds of observations). ERA weights the results of studies based on their level of precision. Because historical agricultural studies rarely report variance at the level necessary, ERA weighs results positively based on the number of replications and negatively based on the number of observations.

Figure. Meta-analysis combines research results coming from different studies (represented by fields on left) using statistics to generate an effect size for the

ERA uses the meta-analysis and ancillary data to compute additional types of analysis. These include but are not limited to costs, benefits and risks of changing practices; spatial scaling domains for agricultural technologies; the value (positive or negative) of bundling technology; resilience and resistance value of agricultural technologies for variable weather; and more.

Meta-analysis is a powerful tool to support decision-making in agriculture. There have been more than 650 meta-analyses conducted in agriculture since 2012. There have been so many meta-analyses that there are now meta-analyses of meta-analyses. Despite the intended purpose of informing decisions, these evaluations rarely provide targeted information for specific uses. This is in part due to the resolution, scope, access and one-off (static) nature of results. ERA changes that situation. Through three innovations, ERA makes data and tools on the performance of agricultural technologies under climate change available on demand.

Hierarchical mapping

Users inherently need information at different levels of aggregation. Policymakers, for example, often talk about management practices and outcomes such as agroforestry and productivity, respectively. Extension agents, by contrast, refer to management practices in greater detail, describing, for example, the difference between intercropping with leguminous and non-leguminous trees—both types of agroforestry—and their relative impacts on yield and gross economic returns, both of which are indicators of productivity. ERA can serve both needs. It developed and employed a hierarchical structure for management practices, outcomes and products. This structure, when combined with meta-analysis, allows research results to be aggregated and disaggregated to any level demanded. These hierarchical structures are in the process of being mapped to major agricultural ontologies to increase interoperability with other datasets and efforts.

Meta-analysis on demand

Data in journals are not readily available to the vast majority of scaling partners, either because of paywalls or because of the time and knowledge it takes to access them. Useability of meta-analyses for decision-making is further constrained because they typically focus on one technology and one outcome (e.g., organic agriculture and yield), though that is changing (see Kuyah et al. 2019), or disproportionately focus on a few technologies, outcomes or products (e.g., conservation agriculture). The consequence is a large gap between the information available in the scientific literature and the information smallholders use to make agricultural decisions. ERA, by contrast, estimates the impacts of hundreds of practice and outcome relationships—and it is growing every day. The reason for the ambition is simple. Data users need information not only on the major practices and products but on the minor ones as well. The data available in ERA is limited only by the scientific literature itself. Users are at the controls as they analyze and discover the relationships that are useful to them. ERA then compiles the available information and delivers it in digestible, decision-ready packets of information, complete with links to original sources for further reading and context.

A living database

New data become available every day, and this new information should affect our understanding of agriculture. Yet scientific research can take years to publish and then gets filed away, unavailable for use. Only in limited cases are the lessons applied to decision-making. ERA aims to disrupt the data–decision relationship. ERA is creating the functionality for users, team members and the community to input new data. These data will be immediately integrated into the database and analysis, ensuring that information is based on the most recent data and reducing the time between data and decision.

With these innovations, and when combined with state-of-the-art analytics and the extensive and transparent search, ERA helps the research community meet FAIR data standards (findable, accessible, interoperable, and reusable).

Looking back

ERA started as the Climate-Smart Agriculture Compendium in 2012. Since then, the project has gone through many iterations and improvements to clarify the underlying practices, outcomes and product descriptions and hierarchies; bolt on external databases; prototype data products with users; and develop online capacity. The CSA Compendium (and subsequently ERA) has been principally funded by the CGIAR's Research Program on Climate Change, Agriculture and Food Security (CCAFS) Flagship on Practices and Technologies. It has received supplementary funding from the European Union, International Fund for Agricultural Development (IFAD), Food and Agriculture Organization of the United Nations, United States Department of Agriculture–Foreign Agricultural Service (USDA–FAS), CCAFS Flagship on Low-emission Development and Center for Forestry Research’s (CIFOR) Evidence-Based Forestry. The Web portal was specifically funded by the EU–IFAD project: Building Livelihoods and Resilience to Climate Change in East and West Africa: Agricultural Research for Development (AR4D) for large-scale implementation of Climate-Smart Agriculture (#2000002575).

Moving forward

Evidence for Resilient Agriculture (ERA) is the outgrowth of early efforts and lessons from the CSA Compendium. The next steps with ERA include but are not limited to the following (time and resources permitting):

• Integrate analysis and visualization apps on costs, production risks and climate change
• Develop training materials and short videos to demonstrate functionality
• Create interoperable datasets with other agricultural meta-data projects
• Extract data from papers in our library that took place in Asia and Latin America (N = 4,500)
• Search for new papers in non-English- language literature
• Collaborate with users to develop new analytical tools for specific uses and questions

ERA is always looking for new opportunities for collaboration, both scientific and development. Please contact us with any questions or interest in the project.