and there are alternative plausible explanations. Relating an impact indicator with a specific research investment is only valid in the absence of other effects on indicators, such as markets and policies (Ekboir, 2003). In addition, many ROR estimates often fail to account for the effects of work done by others in the research development continuum. Tempo­ral aspects of the attribution problem would result when assuming a specific time lag between research results and their implementation. At times, the period over which re­search affects productivity may be overestimated. A number of strategies can be used to address attribution, called con­tribution analysis (Mayne, 1999), which may enhance the validity of the estimates, but do not eliminate the problem.
          In many estimates the spill-over effects are not usually included as benefits (see 8.2.7). In others, the effects result­ing from changes in rural employment, health and educa­tion policies and programs are excluded. Environmental impacts, both negative and positive, are often ignored (see 8.2.5) as well as those costs arising from institutional mar­keting arrangements. This issue can be addressed through estimations of the ROR for research and complementary services, as well as research. Increasingly it is likely that valuation of nonmarket impacts of agricultural AKST will be incorporated into economic analysis. Past studies did not address these issues because of measurement difficulties and the fact that research impacts directly observable in com­modity and factor markets were abundantly available. Re­search systems are now increasingly called upon to provide positive nonmarket environmental or health benefits as well as to mitigate past negative impacts. A number of economic tools are now available to measure nonmarket environmen­tal benefits and costs such as environmental quality, longev­ity, and health effects (Freeman, 1985; Feather et al., 1999; Dolan, 2000; Hurley, 2000).
          There has been little empirical work in the area of assess­ing the ROR for social science research, and organizational and institutional innovations including capacity strengthen­ing. This lack is associated with the difficulty of attributing any change in policy, institutions or process and the linked economic growth or poverty reduction to research infor­mation generated by social science or other factors (Alston and Pardey, 2001). Attempts to quantify the ROR to social science research have used esoteric methods, utilizing "in­credible identifying assumptions" that cannot be robustly defended (Gardner, 2003; Schimmelpfenning and Norton, 2003; Schimmelpfennig et al., 2006). In addition, social and applied science research is often undertaken jointly and dif­ficult to isolate.

8.2.4 Empirical evidence
There are many studies on the ROR of investment in ag­ricultural development. This critique draws heavily upon a number of meta-analyses (Alston et al., 2000a; Evenson, 2001; Evenson and Rosegrant, 2003; Thirtle et al., 2003), which cover 80% of published materials (more than one thousand studies). Most of these studies are associated with production technologies, far fewer address other research outputs e.g., post harvest processing, marketing, policy, or­ganizational or institutional innovations. We recognize the shortcomings in these data but these reviews provide use­ful insights into impacts of AKST investments across con-

tinents, commodities and components (research, extension, private sector research.

8.2.4.1 Rate of returns to national AKST investments
One meta-analysis estimated the economic impact of agri­cultural R&D investment at the national level for 48 se­lected developing countries in Africa, Asia, and Latin Amer­ica (Thirtle et al., 2003) The analysis revealed that R&D expenditures per unit of land have an elasticity of 0.44 in terms of productivity. It was also noted that the elasticity of agricultural R&D is positive and highly significant in all cases and is slightly larger for Africa than Asia and both are over 50% greater than the Latin America's elasticity. The elasticity of value added per unit of land with respect to agricultural R&D was used to calculate ROR to agricul­tural R&D at the country and the continental level (Table 8-7). The estimated ROR for the sample countries in Africa ranged between -12 and 58%. In only three cases the gains were less than the expenditures. For the Asian countries, the estimated ROR ranged between -1 and 50%; and appears to be less varied and generally higher. The mean of the coun­try RORs for Asia (26%) is better than for Africa (18%) and the weighted mean (31%) is still higher. These means are dominated by the huge agricultural sectors of China and India, both of which seem to have done well in economic terms. In the case of Latin America, only five of the thirteen countries had positive RORs. The estimated ROR ranged between -22 and 40%. The poor results for the Latin Amer­ican countries are at least partly due to the limitation of data availability (Thirtle et al., 2003).

8.2.4.2 Rates of return to crop genetic improvement in­vestments
Over the years, a significant amount of AKST resources have been devoted to genetic improvement. A second assessment of economic consequences of crop genetic improvement estimated the economic impact of 17 commodities and 35 country/regions using a "global market equilibrium" model (Evenson and Rosegrant, 2003). Benefit/cost ratio (using 6% as the external interest rate) and IRRs of crop genetic improvement programs by region have been computed for both national agricultural research systems (NARS) and in­ternational agricultural research centers (IARCs) (Table 8-8). The IRRs for the NARS ranged between 9 and 31%, which are considerably lower than the ones reported in individual studies. This is primarily because most individual studies tend to ignore the research costs to build the germplasm stock that is required to reach the stage where benefits are produced. The lowest IRR was observed for SSA (9%). The IRRs for the IARC programs are very high and ranged be­tween 39 and 165%. The lowest IRR was observed for Latin America. These high RORs reflect the leveraging associated with the high production of IARC crosses and high volume of IARC germplasm (Evenson and Rosegrant, 2003).

8.2.4.3 Economic impacts of research and extension investments
A number of economic impact studies were assessed to evaluate the contribution of agricultural research and exten­sion programs both public and private, using the estimated ROR on investment to index economic impacts (Evenson,