The impact of research and development on productivity growth and its economic benefits

Research involves an expenditure of effort to increase the stock of knowledge. Ideas are likely to be generated from the larger stock of knowledge and can be converted (through commercialisation and adoption) into production technologies, good and services, and other forms of innovation. The use of public resources to support agricultural research and development (R&D) begs a critical question of how expenditures on R&D affect the long run competitiveness of the sector.

Estimation issues of R&D impacts on productivity

While there is a good deal of evidence that links R&D expenditure to agricultural productivity growth, quantifying the relationships is challenging and is subject to the availability of suitable methodology and data (Alston, 2010).¹ Some of these data and measurement issues are related to the measurement of R&D expenditure and TFP, while some are related to the often complex problem of relating productivity growth to R&D expenditure. Moreover, measures of agricultural inputs (especially capital, but also farm labour) and outputs are sometimes found problematic. Unpaid owner and family labour, is often an approximation of labour at best. Changes in soil quality or changes in the use of ground water, or influences of changing climate can have important impact but are seldom measured carefully. Limitations on the types and quantities of data that are available, combined with some misunderstanding or misuses of the measures, are likely to have contributed to weaknesses in some studies linking agricultural R&D to productivity.

There are also many data issues in measuring R&D effort. Data on private research expenditures is particularly difficult to obtain as firms often protect this strategic information.² Even finding data on public research expenditures in a useful form is often an arduous task because of how agricultural research expenditures are recorded over time. In addition, the data issues with respect to TFP and agricultural R&D expenditures are also confounded by the need to find a series of data which are long enough to reliably estimate the long lags involved in agricultural research particularly in countries which have conducted basic research for extended time periods.

Besides the data issues, attribution problems have bedevilled studies of the effects of research on agricultural productivity (Alston and Pardey, 2001). The principle areas of difficulty are: 1) in determining how much productivity growth is attributable to organised R&D; 2) in attributing responsibility among alternative public and private providers of R&D, and 3) in identifying the research lag structure. Many studies assume implicitly or explicitly that all measured agricultural productivity growth is attributable to R&D. This implicitly assumes that other important drivers of productivity such as education, or infrastructure development, scale economies, clustering effects³ and changing weather patterns would not have increased productivity growth in the absence of the R&D expenditure. There is also an implicit assumption that productivity would not have decreased due to disease and pest pressure, weather changes or resource depletion in the presence of R&D.

Research usually takes a long time to affect production, and then it affects production for a long time. One element of the attribution problem, then, is to identify the specific dynamic structure linking research spending, knowledge stocks accumulation, and productivity growth. A large number of previous studies have regressed a measure of agricultural output or productivity and variables representing agricultural research and extension, often with a view to estimating the rate of return to research.⁴ The specification of the determinants of the lag relationship between research investments and production, which involves the dynamics of knowledge creation, depreciation, and utilisation, is crucial. Only a few studies have presented much in the way of formal theoretical justification for the particular lag models they have employed in modelling returns to agricultural research.

Table 5.1 summarises some key features of research lag distribution models applied in studies of agricultural productivity in OECD countries. Until quite recently, it was common to restrict the lag length to be less than 20 years. In the earliest studies, available time series were short and lag lengths were very short, but the more recent studies have tended to use longer lags. Since the time span of the data set is usually not much longer than the assumed maximum lag length, and the individual lag parameter estimates are unstable and imprecise, most studies have restricted the lag distribution to be represented by a small number of parameters.⁵

Table 5.1. Research lag structures in studies of agricultural productivity

Note: This table is based on the full sample of 292 publications reporting 1 886 observations.

• 1. Unspecified estimates are those for which the research lag length is not made explicit.
• 2. Lag length is unclear.

Source: Alston et al. (2009b), as adapted from Alston et al. (2000).

In their application using long-run, state-level data on US agriculture, Alston et al. (2009a) found evidence in favour of a gamma lag distribution model with a much longer research lag than most previous studies had found — for both theoretical and empirical reasons.⁶ Their empirical work supported a research lag of at least 35 years and up to 50 years for US agricultural research, with a highest correlation in year 24.⁷ This comparatively long lag has implications both for econometric estimates of the effects of public R&D on productivity and the implied rate of return to research. It should be noted, however, that lags are likely to depend on the type of research (general or applied, scientific or organisational, by sector, etc.) and the starting point. For example, basic research most likely takes more time to affect productivity gains than applied or adaptive research. Research lags are likely to be longer in OECD countries, which spent significant resource on basic research, than in developing countries, which adopt or adapt existing technologies from international research centres or other countries.

More recently, agricultural economists have been paying increasing attention to the fact that knowledge created within a particular geopolitical entity can have impacts on technology elsewhere, with implications that may matter to both the creators of the spillouts and the recipients of the spillins. For example, Huffman and Evenson (1993) and Alston et al. (2010) found that a sizable share of the benefits from research conducted in US State Agricultural Experiment Stations was earned as interstate spillovers. Given the size of these spillovers studies that did not allow for spillovers probably have overestimated the local benefits of research, while underestimating the regional benefits.

Studies that have examined research spillovers have found that knowledge created in neighbouring jurisdictions, or in similar agroclimatic regions can have large impacts on productivity (e.g. Huffman and Evenson, 1993; Pardey et al., 1996 and Alston et al., 2010). Similarly, the varieties and germplasm created in the international research institutions find their way into varieties around the world. Upstream basic research or downstream expenditures on extension can also impose the spillover impacts. Finally private and public research can create spillovers across organisational boundaries and can not only affect research outcome but can also affect research investment decisions by “crowding out” or “crowding in” other research activities. Being able to estimate the spillover effects requires that expenditure data be collected from each potential source of spillover, which further compounds the difficulty of data collection.