Foreword: We think our top charities do fantastic work, but we are always interested in other methods for helping the world’s poor. The Copenhagen Consensus and other authors have highlighted the potential of agricultural R&D as a high-leverage opportunity. This was enough to get us interested in understanding the area better, so we asked David Goll, a Giving What We Can member and a professional economist, to investigate how it compares to our existing recommendations.
Owen Cotton-Barratt, Director of Research for Giving What We Can
Around one in every eight people suffers from chronic hunger, according to the Food and Agricultural Organisation’s most recent estimates (FAO, 2013). Two billion suffer from micronutrient deficiencies. One quarter of children are stunted. Increasing agricultural yields and therefore availability of food will be essential in tackling these problems, which are likely to get worse as population and income growth place ever greater pressure on supply. To some extent, yield growth can be achieved through improved use of existing technologies. Access to and use of irrigation, fertilizer and agricultural machinery remains limited in some developing countries. However, targeted research and development will also be required to generate new technologies (seeds, animal vaccines and so on) that allow burgeoning food demand to be met.
Agricultural research and development encompasses an extremely broad range of activities and potential innovations. A 2008 paper issued by Consultative Group on International Agricultural Research (von Braun et al., 2008), an international organization that funds and coordinates agricultural research, identifies 14 ‘best bets’. These include developing hybrid and inbred seeds with improved yield potential, better resistance to wheat rust, increased drought tolerance and added nutritional value, but also encompasses the development new animal vaccines, better fertilizer use and improved processing and management techniques for fisheries.
Notable successes in seed development seem to have generated immense social benefit. The high-yielding varieties that spread through the ‘Green Revolution’ are often credited with driving a doubling of rice and wheat yields in Asia from the late 60s to the 90s, saving hundreds of millions of people people from famine (see, for instance, Economist, 2014). Given the prevalence of hunger and the high proportion of the extremely poor that work as farmers, agricultural research and development seems to offer a potential opportunity for effective altruism.
Existing benefit-cost estimates are promising, though not spectacular. The Copenhagen Consensus project ranked R&D to increase yield enhancements as the sixth most valuable social investment available, behind deworming and micronutrient interventions but ahead of popular programmes such as conditional cash transfers for education (Copenhagen Consensus, 2012).
The calculations that fed into this decision were based on two main categories of benefit. First, higher yield seeds allow production of larger quantities of agricultural output at a lower cost, bolstering the income of farmers. Around 70 per cent of the African labour-force work in agriculture, many in smallholdings that generate little income above subsistence (IFPRI, 2012). Boosting gains from agriculture could clearly provide large benefits for many of the worst off. Second, decreased costs of production lead to lower prices for food, allowing consumers to purchase more or freeing up their income to be spent elsewhere.
Projecting out to 2050, these two types of benefit alone are expected to outweigh the costs of increased R&D by 16 to 1 (Hoddinott et al., 2012). By comparison, the benefit-cost ratios estimated within the same project for salt iodization (a form of micronutrient supplement) range between 15 to 1 and 520 to 1, with the latest estimates finding a benefit-cost ratio of 81 to 1 (Hoddinott et al., 2012), and most of the estimates reported to the Copenhagen Consensus panel for the benefit-cost ratio of conditional cash transfers for education fall between 10 to 1 and 2 to 1 (Orazem, 2012). Using a very crude method, we can also convert the benefit-cost ratios into approximate QALY terms. Using a QALY value of three times annual income and taking the income of the beneficiaries to be $4.50 a day (around average income per capita in Sub-Saharan Africa), agricultural R&D is estimated to generate a benefit equivalent to one QALY for every $304.
Other types of benefit were not tabulated in the Copenhagen Consensus study, but should also be high. Strains that are resistant to drought, for instance, could greatly reduce year-to-year variation in crop yields. More resilient seeds could mitigate the negative effects of climate change on agriculture. Lower food prices may lead to better child nutrition, with life-long improved health and productivity. Finally, higher yields may decrease the potential for conflict due to the pressure on limited land, food and water resources resulting from climate change and population growth. Each of these benefits alone may justify the costs of research and development but, with our limited knowledge, they are not easily quantified.
The high benefit-cost ratio found by the Copenhagen Consensus team is broadly consistent with other literature. Meta-analysis of 292 academic studies on this topic has found that the median rate of return of agricultural R&D is around 44% (Alston et al., 2000). A rate of return, in this sense, indicates the discount rate at which the costs of an investment are equal to the benefits – rather like the interest rate on a bank account. More recent studies, focusing on research in Sub-Saharan Africa, have found aggregate returns of 55% (Alene, 2008).
Unfortunately, the rate of return on investment is not directly comparable to a benefit-cost ratio; the methodology applied often deviates from the welfare based approach applied by the Copenhagen Consensus team and the two numbers cannot be accurately converted into similar terms. Nonetheless, a crude conversion method can be applied to reach a ballpark estimate of the benefit-cost ratio implied by these studies. Assuming a marginal increase in spending on research is borne upfront and that research generates a constant stream of equal benefits each year from then on, the benefit-cost ratio for an investment with a 44% rate of return at a 5% discount rate is 9 to 1.
There are, however, at least two reasons to treat these high benefit-cost estimates with skepticism.
First, estimating the effect of research and development is difficult. One problem is attribution. Growth in yields can be observed as can spending on research and development, but it is much more difficult to observe which spending on research led to which increase in yields. If yields grew last year in Ethiopia, was this the result of research that occurred two years ago or ten years ago? Were the improved yields driven by spending on research within Ethiopia, or was it a spillover from research conducted elsewhere in the region or, even, research conducted on another continent? Estimating the effect of R&D spend requires researchers to adopt a specific temporal and spatial model dictating which expenditures can effect which yields in which countries. Teasing out causality can therefore be tricky, and some studies have suggested that inappropriate attribution may have led to systematic bias in the available estimates (e.g. Alston et al., 2009).
Another problem is cherry picking. Estimates garnered from meta-analysis are likely to be upwardly biased because studies are much more likely to be conducted on R&D programmes that are perceived to be successful. Failed programmes, on the other hand, are likely to be ignored and, as a result, the research may paint an overly optimistic picture of the potential impact of R&D.
Second, for new technologies to have an impact on the poor, they need to be widely adopted. This step should not be taken for granted. Adoption rates for improved varieties of crops remain low throughout Africa; farmer-saved seeds, which are unlikely to be improved, account for around 80 per cent of planted seeds in Africa compared to a global average of 35 per cent (AGRA, 2013). To some extent, this is because previous research has been poorly targeted at regional needs. The high-yield varieties developed during the Green Revolution require irrigation or very high levels of rainfall. New seed development was focused on wheat and rice, rather than alternative crops such as sorghum, cassava and millet. High yielding varieties required extensive fertilizer use. All of these features rendered them unsuitable for the African context, and explain why it was not easy to replicate the Asian success story elsewhere (Elliot, 2010).
However, there are more structural features of many developing countries that will limit adoption. Lack of available markets for surplus production can mean that smallholders can see limited benefit from larger harvests, especially when new seeds are costly and require additional labour and expensive fertilizer. Weak property rights undermine incentives to invest, given that farmers may be unable to hold on to their surplus crop or sell it at a fair price. Unavailability of credit means that, even when it makes good economic sense for farmers to invest in improved seeds, they may not be able to raise the initial capital required. The benefit-cost estimates discussed above, based on a synthesis of evidence from a diverse set of contexts, may underestimate the difficulties with adoption in more challenging countries.
Even in Asia during the Green Revolution, high-yield varieties were adopted first and foremost by large agricultural interests rather than smallholders (Wiggins et al., 2013). If this was the case for newly developed seeds, the impact on the poorest would be more limited than suggested in the Copenhagen Consensus study. They could still benefit from lower food prices and increased employment in the agricultural sector, but in extreme scenarios smallholders may even lose out due to low cost competition from larger farms that adopt new seeds.
In combination, the difficulties with estimating the effects of R&D and the potential barriers to adoption suggest that the estimated benefit-cost ratios reported earlier are likely to be upwardly biased. The benefit-cost ratios estimated are also lower than those associated with Giving What We Can’s currently recommended charities. For instance, the $304 per QALY estimate based on the Copenhagen Consensus benefit-cost ratio, which appears to be at the higher end of the literature, compares unfavourably to GiveWell’s baseline estimate of $45 to $115 per DALY for insecticide treated bednets (GiveWell, 2013). The benefit-cost ratios also appear to be lower than those associated with micronutrient supplements, as discussed earlier. While there are significant benefits that remain unquantified within agricultural R&D, the same is also true for interventions based on bednet distribution, deworming and micronutrient supplements. As a result, while this area could yield individual high impact opportunities, the literature as it stands does not seem to support the claim that agricultural R&D is likely to be more effective than the best other interventions.