4 Induced Adaptive Invention/Innovation and Productivity Convergence in Developing Countries – Technological Change and the Environment

Chapter 4

Induced Adaptive
Invention/Innovation and
Productivity Convergence in
Developing Countries

Robert E. Evenson

4.1   Introduction

In the 1950s, economists concerned with economic development programs expected the remaining decades of the twentieth century to be characterized by a pattern of “convergence” in per capita income levels. Those countries with low per capita income in 1950 were perceived to enjoy the “advantages of backwardness” and to have the greatest scope for growth in subsequent decades. Development economists also did not generally expect the agricultural sector to be the leading sector in development experience. Many formal models (notably the Latin American structuralist models) were predicated on the expectation that agricultural producers were hampered by the limited visions of the peasant (or minifundia) farmer and that the scope for productivity gains in the sector was limited.1

By contrast, industrial development was widely expected to be the vehicle for modernization and convergence. Industrial technology was not regarded as being sensitive to local conditions (as was the case in agriculture) and could presumably be copied or mimicked at relatively low cost. Low-wage countries acquiring this technology would then be highly competitive in both domestic and international markets. Not only was industry expected to drive convergence, it was also expected to be the ultimate source of transformation of the traditional agricultural sector in developing economies.

At the beginning of the twenty-first century, we now have a considerable body of experience to evaluate and contrast with the expectations of the early development modelers. The evidence shows that a general pattern of per capita income convergence has not occurred for developing countries, even though it did occur for the countries of the Organisation for Economic Co-operation and Development (OECD).2 Many developing economies with low per capita income in the 1950s have made few, if any, gains in income in subsequent decades. By contrast, a small group of developing (newly industrial country, or NIC) economies has achieved very rapid economic growth.

The sectoral evidence shows that in most developing countries, to the degree that productivity convergence has occurred, it has occurred in the agricultural sector, not in the industrial sector. Moreover, while the rapidly growing developing economies did realize rapid industry-led growth, industrial growth did not “trans-form” the agricultural sectors of the NICs because they had already been transformed by biological invention. Industry-led growth certainly did not transform the agricultural sectors of the majority of developing economies, where most, if not all, productivity gains were realized in agriculture, not industry.

In Section 4.2, two alternative classes of mechanisms of productivity convergence are discussed. The two classes are mimicry mechanisms and induced adaptive invention mechanisms with international invention recharge. Section 4.3 develops the induced adaptive invention model in further detail. Section 4.4 explores evidence for these alternative mechanisms in the agricultural sector. This evidence clearly supports induced adaptive invention mechanisms as the dominant mechanism of convergence. Research organizations for agriculture have been designed to facilitate this form of convergence and have succeeded in doing so.

Section 4.5 explores industrial technology convergence in a setting where mimicry mechanisms have dominated technology policy. Evidence is reported, showing that the induced adaptive invention form of convergence with international invention recharge receives strong support from data on inventions patented. The section concludes that, to a considerable degree, the failure of industrial policy makers to recognize the importance of the adaptive invention/innovation mechanism has retarded industrial development.

4.2   The Mechanics of Convergence

In this section a distinction is made between two classes of convergence mechanisms: mimicry and adaptive invention/innovation.

4.2.1   Mimicry mechanisms

Mimicry mechanisms are appropriate when firms in a developing country can find cost-reducing techniques by simply mimicking current or past techniques used by firms in developed countries. This implies that the cost-reducing character of technological innovation in developed countries is effectively available to developing countries. Mimicking also applies to product change technology.

In this chapter we consider three classes of mimicry mechanisms:

•  Simple mimicry—with low tacit knowledge requirements and low transaction costs.

•  Transaction cost constrained mimicry—with low tacit knowledge requirements.

•  Tacit knowledge constrained mimicry—with low transaction costs.

Simple mimicry is what many development policy makers had in mind in 1950. It is also the model implicitly behind import substitution policies. Had this mechanism actually dominated, convergence in developing country incomes would have occurred.

Transaction cost constrained mimicry could, in principle, explain convergence failure. Countries with inadequate institutions (property rights, bankruptcy laws, etc.) could be prevented from attaining mimicry convergence. However, transaction cost constrained mimicry is not consistent with super-convergence. It could, however, explain the division between converging and non-converging economies.

Tacit knowledge constrained mimicry is perhaps the favored explanation for convergence failure. The basic idea is that the mastery of technology requires more than experience and normal skills. Some degree of engineering competence is required for mimicry. This mechanism is consistent with the technical assistance provisions in many foreign direct investment contracts. It is, at least in principle, consistent with convergence failure (where the engineering skills are insufficient) and super-convergence (where sufficient skills are available).3

4.2.2   Adaptive invention/innovation mechanisms

Adaptive invention/innovation mechanisms are appropriate when a developing country finds that the developed country technology available for mimicking has impaired cost-reduction effectiveness because of differences in production conditions between developed and developing countries. These differences may be differences in prices, institutions, or natural environments such as soil or climate conditions.

This impairment of cost reduction effectiveness opens up scope for adaptive invention/innovation activities to modify the developed country technology so as to reduce the impairment factor. Thus, in order to exploit this mechanism, the recipient developing country must have an invention/innovation capacity in place.4

In a globalized trading world, price differences between produced goods will not differ very much between countries. With cost-efficiency differences between developed and developing countries, the chief price difference will be the price of labor. This has two effects on convergence mechanisms. First, some developed economy technological improvements in machine processes may effectively be unavailable at all to developing countries with low wages. For example, rice harvesting equipment improvements may be of no value to economies where rice is harvested by hand because of low wages. Second, even where machine improvements are of value in developing countries, their value may be reduced because the improvements were induced by high wage conditions.

Production differences associated with natural conditions—soil, temperature, rainfall, day length, etc.—most clearly affect production requiring biological activity. Plants and animals perform differently in different environments. Darwinian natural selection produced plant and animal species reflecting comparative natural advantages. Farmer-produced improvements in cultivated crop species and in domesticated animals were also governed by Darwinian sensitivity. Similarly, today's modern plant breeding crop genetic improvements remain conditioned by these forces. Thus, the development of improved crop rice varieties in Japan had no direct value in India. But the genetic resources and the methodology behind the Japanese development were very important to the adaptive invention of crop varieties in India.

Models of adaptive invention incorporate international invention recharge as a central feature. The next section develops this feature further.5

4.3   The Induced Invention/Innovation Convergence Model with an Application to Plant Breeding

The traditional induced innovation model postulates an “innovation-possibility frontier” (IPF) in economic dimensions or traits. The key insight of this model is that the economic traits have different values that should guide inventive effort. The model developed in this section is presented in terms of plant traits sought by plant breeders. These plant traits can be given economic values. An important part of this model, however, is the invention recharge specification, particularly the international recharge mechanism that is subjected to test in Sections 4.4 and 4.5.

Plant breeders have two alternative search strategies in their research programs. The first of these is the search for “quantitative” plant traits governing yields. Quantitative traits are controlled by multiple genes (or genetic alleles) and require complex strategies for crossing parental materials and selecting improved cultivars. The second strategy is the search for “qualitative” traits such as host plant resistance (HPR) to the tungro virus in rice. Qualitative traits typically are controlled by a single gene.

Both breeding strategies rely on searching for genetically controlled traits in collections of crop genetic resources (CGRs), which include landraces of the cultivated species (distinct types selected by farmers over centuries from the earliest dates of cultivation and diffused across different ecosystems), “wild” (related) species, and related plants that might be combined. CGR collections also include “combined” landraces, including varieties (officially recognized uniform populations of combined landraces, often with many generations of combinations). The systematic combining of landraces and evaluation is termed “pre-breeding.”6

4.3.1   The simple one-trait, one-period model

Here, we consider the following representation of the single-trait, one-period model.

In period 1, the existing breeders’ techniques and breeders’ CGR collections determine a distribution of potential varieties indexed by their economic value, x. Following Evenson and Kislev (1976), suppose this distribution to be an exponential distribution:


The cumulative distribution is


with mean and variance



The cumulative distribution of the largest value of x(z) from a sample of size (n) is the “order statistic” (Evenson and Kislev 1976),


and the probability density function for (z) is


The expected value and variance of hn(z) are



Evenson and Kislev discuss the applicability of expressions (4.7) and (5.8) to plant breeding research. Basically, expression (4.7) can be thought of as the breeding production function. The approximation ln(n) is a reasonable approximation for any symmetric distribution f(x) including the uniform distribution and the normal distribution. The marginal product of breeding effort is simply


Given a measure of the units over which (z) applies (e.g., the areas in a specific ecosystem), the value of the marginal product V can be computed and set equal to the marginal cost of search to solve for optimal n:


For two or more traits, each can be characterized by expression (4.7) with different parameters:


When these traits are qualitative traits, breeders typically search for them independently because there are techniques that enable the breeders to incorporate only the single trait in a cultivar (i.e., by back crossing the other methods, unwanted traits can be discarded). Thus, even if traits are highly correlated, the breeder will search independently for them.7

Figure 4.1 depicts distributions for two traits: x1 and x2. If we set N = N1 + N2 at some level (say, the optimizing level) where


we have the IPF depicted in Figure 4.1. (Note that the depiction is in terms of traits, but that these can be translated into economic units through values.) The point TD in Figure 4.1 is the technology determination point determined by the maximum value of the traits in each distribution (or an arbitrary stopping point in each distribution).

4.3.2   Multiple periods without recharge

Now consider periodicity. In practice, we do not observe the single-period optimal search implied by expressions (4.7), (4.11), and (4.12). Typically, we observe multiple-year research and development (R&D) programs even for narrowly defined objectives. Could we treat this multiple-year sequence as simply a long period instead of a sequence of periods? Certainly not in plant breeding. Plant crosses (genetic combinations) must be evaluated and selected over several generations. Plant “types” (quantitative) are built with multiple-generation crosses, where the crossing decisions for the second generation can effectively be made only after the first generation has been observed.8

In addition to this periodicity, two related types of shifts in the invention distributions are relevant. The first is periodicity associated with search field narrowing (elimination of unpromising search avenues). The second is recharge shift, discussed below.

The search field narrowing case is depicted in Figure 4.2, where a rightward shift in the mean of the distribution, but not the right-hand tail of the distribution, is depicted from period 1 to period 2. This shift can be thought of as the systematic elimination of unpromising search avenues. It may be possible to classify material into n groups. In period 1 sufficient sampling is undertaken to enable the estimation of the mean and variance for each of the n groups. On the basis of these estimates, several groups may be eliminated, with the resultant distributional shift depicted in Figure 4.2.9

The shift in the mean for both trait distributions is proportional to the period 1 optimal discovery as depicted in the IPF diagram for the two traits.

Figure 4.1. The Single-Period IPF.

The shape of IPF2 is affected by the period 1 search. Because the period 1 search was induced by prices to produce more n11 search for t1 than n21 search for t2, there is more exploitable search scope for t2 in period 2. Thus the resultant optimal point on IPF2 is not on the same ray from the origin as was the optimal point of IPF1. The search exhaustion phenomenon has moved the optimal point in the direction of the TD point.

Implications of this search exhaustion case are that multiple-period searches can take place but that they will eventually stop. During the multiple-period search, the ratio of inventions (proportional to t11t10,t12t11, etc.) to search resources (R&D) n will decline.

Figure 4.2. Multiple-Period Search with Search Field Narrowing.

4.3.3   Multiple periods with recharge

The second type of invention distribution shift is associated with several types of recharge mechanisms.10 These include the following:

•  Genetic resource collection and evaluation programs. These programs are designed to discover uncollected materials and make them available to breeders.

•  Pre-breeding programs where landrace materials are systematically combined into potential breeding lines by specialized research programs. These programs do not seek to develop “final products,” that is, new cultivars. Instead they seek to evaluate and produce “advanced lines” that are then used by final product inventors.

•  Wide-crossing programs where techniques for inter-specific combinations of genetic resources (between related species) are made possible. This expands the size and scope of the original materials that can be utilized in breeding programs.

•  Transgenic breeding programs where DNA insertion techniques allow traits associated with alien genes (i.e., from unrelated species) to be incorporated into cultivated plants.

These programs are “pre-invention” science or applied science programs. They provide recharge to the invention distributions by shifting both the mean and the right-hand tail of the invention search distribution. The actual mechanism of recharge, however, is often in the form of biological invention or varieties that serve as parents in the recharged invention distribution.

Figure 4.3 depicts the nature of these shifts for search distributions and IPFs with recharge. Note that the technological determinism point, TD, moves with recharge. The reader can readily see that one could have cases of “super-recharge” for a number of periods where inventions per inventor might increase over time (e.g., in sugarcane breeding; see Evenson and Kislev 1975). But recharge science itself is likely to be subject to diminishing returns, unless it is also recharged by the more basic sciences.

These ideas can be clarified with a little algebra. Describe the breeding (invention) process as


This system of equations describes the incorporation of traits as a function of germplasm, Gi, and breeding activity, Bi. The functional form is based on the search model. Note that the germplasm term enters linearly in this model. The implications of this are used in the empirical estimates reported below.

The first-order condition for allocating breeding research between any two traits when the marginal cost of Ti is equal to the marginal cost of Tj is


Figure 4.3. Multiple Search with Recharge.

where Vi and Vj are measures of the marginal contribution to crop value of traits i and j, respectively. (Note that each trait may appear in several varieties and that each variety may be planted in different areas.)

Now consider the production of germplasm Gi. This is characterized as being produced in a pre-breeding process:


In this pre-breeding process, pre-breeding activity converts evaluated genetic resources Gc into germplasmic breeding materials. This process is also a search process:


Evaluated germplasm is produced by the natural stock of genetic resources Gn and collection (C) and evaluation (E) activities.

The following features of this model can be noted:

•  If the marginal search coefficients are equal (λi = λj), plant breeding activity obeys “the congruence rule,” where inventive activity is proportional to the value of the units affected [see expression (4.13)].

•  Departures from congruence (a strong form of induced innovation) are justified when search parameters differ.

It can be further noted that the optimal conditions for pre-breeding or germplasmic recharge science [expression (4.14)] also imply that, if the germplasmic search coefficients are equal, then at least partial congruence occurs for both pre-breeding and breeding (partial, because of the common Gc term). This is a strong form of multiple-period induced innovation. The multiple-period invention path is a ray from the origin (if prices do not change) that is parallel to the TD expansion path. A change in prices (values) will result in a change in both the invention path and the TD path.

Agricultural experiment stations were developed to produce inventions for farmers. Extension systems were developed to diffuse these inventions. Over the years, these institutions have been continuously in tension over the relative weights to place on extension, invention, and pre-invention or recharge science. Figure 4.4 (Huffman and Evenson 1993) reflects the institutionalization of the level II recharge sciences in modern agricultural research systems.

Farmers express a demand for inventions (level III) and invention products (level IV). The agricultural experiment stations, however, have been able to convince state legislatures in the United States that level II pre-invention sciences are necessary to recharge the invention distributions. This represents a type of public sector competition to provide services to farmers at the state level. Comparable systems exist for medical technology but are weak for many other fields of technology.

Source: Huffman and Evenson

Figure 4.4. Institutional Specialization in R&D Systems for Agriculture.

Figure 4.5. Spatial Spillovers.

4.3.4   Technology spillovers

The problem of spillovers has been recognized in agricultural research systems and is reflected in numerous locations of agricultural experiment stations around the world (see discussion of rice below) with a high degree of germplasmic spillovers and adaptive invention for targeted ecosystems.11 It is well known that plants and animals perform differently in different ecosystems and that modern plant-breeding has only partially overcome the “Darwinian” adaptation to ecosystem niches in nature. It is also well known that relative prices affect the real value of an invention (an improved rice harvesting machine is valuable in Texas but has no real value in Bangladesh, where wages are low and rice is harvested by hand).

Figure 4.5 illustrates these issues for two types of technology, A and B.

•  I2 shows how non-price factors (ecosystem institutions, etc.) remediate the real performance of z1 and z2 in location 2 and lead to an interior IPF.

•  I3 is the real-value IPF in location 2 given that location 1 produced (z*1,z*2). This lies below I2. Location 2 will then have direct spill-in, shown as point A.

•  I4 is the IPF now available to location 2 should it choose to undertake research. Location 2 has a choice between no research (point A on I3) and conducting its own research (point B on I4).

•  Technology B has the same I1 as technology A but lower non-price remediation.

It is generally thought that agricultural technology is characterized by technology A, where non-price remediation is high and adaptive research potential is good. For biological traits, relative prices also may not differ between location 1 and location 2. This will lead to strong incentives to locate research capacity in both locations. These research programs “feed” off each other and sometimes on international recharge programs (see below).

Mechanical technology is thought to be more like technology B, where non-price remediation is low, price differences are great, and adaptive and germplasmic potential are low (at least at location 2’s prices).

4.4   Evidence from Agriculture in Developing Countries

Public sector agricultural research systems have been built in most countries of the world. These systems were among the earliest cases where governments recognized that incentive systems, chiefly intellectual property rights (IPRs) systems, were not sufficient to bring forth adequate invention from the private sector. In response, public sector colleges of agriculture and mechanics (A&Ms) were designed to train agricultural and engineering practitioners, and a system of public experiment stations was established to undertake biological invention. (See Figure 4.4 for further details.)

The seminal study of hybrid corn by Griliches (1957) provides insight concerning spillovers and recharge mechanisms. Figure 4.6, from the Griliches study, illustrates these factors. Griliches noted that the invention of hybrid corn was an invention of a technique or method of invention. It represented “recharge science.” In this case, the recharge science was undertaken in public sector agricultural experiment stations (notably the Connecticut Agricultural Experiment Station in New Haven). Most actual invention was undertaken by private firms, although many state experiment stations also produced hybrid corn varieties.

Griliches also noted that corn plants have a high degree of genotype– environment interactions, that is, their performance is sensitive to soil type, day length, etc. (non-price remediation in Figure 4.5). Accordingly, the rapid adoption of hybrid corn varieties produced for Iowa and Illinois did not transfer to rapid adoption in Alabama, because the varieties suited to Iowa were not suited to Alabama. It was only after breeding programs designed for Alabama conditions were developed that Alabama farmers had access to the hybrid technology (and it was only after similar programs were developed in the Philippines in the 1980s that the technology was available there). Sub-Saharan Africa did not have access to the hybrid technology until the 1990s.

Figure 4.6. Adoption Patterns: Hybrid Corn in the United States, Percent of Total Corn Acreage Planted with Hybrid Seed.

Sources: USDA (various years); Griliches (1957).

4.4.1   Investment strategy in developing countries

During the first half of the twentieth century, plant breeding programs were established in most developed countries and the specializations depicted in Figure 4.4 were established. Agricultural experiment stations were located in US states, and public competition to serve farmers emerged as the motivating force for investment. The lessons implicit in the hybrid corn experience regarding adaptive innovation and pre-breeding were incorporated into system design.

For the developing countries, agricultural research programs were not as well developed. Most developing country systems were guided by colonial politics. This did produce effective research programs for crops destined for export to the mother country (tea, coffee, sugar, spices), but for most food crops little real research capacity was in place.

During the 1950s, considerable emphasis was placed on extension programs providing farmers with technical advice. These programs were motivated by the “easy mimicry” model that also guided industrial policy. By the end of the 1950s, however, the broad outlines of the population expansion in developing countries were becoming clear. Most countries were faced with the prospect of a doubling, and in some cases a tripling, of population in the next four decades. This called for an unprecedented expansion of food production.

The response of development agencies, both bilateral and multilateral, was to support National Agricultural Research System (NARS) development and to support NARS programs by developing a system of commodity-focused International Agricultural Research Centers (IARCs). The design of the IARC system (supported by a consortium of donors—the Consultative Group for International Agricultural Research) was based on two predecessor projects, each of which facilitated early “green revolution” achievements.

The first of these was a special project of the Rockefeller Foundation in Mexico. Beginning in the early 1940s, Norman Borlaug instituted a wheat breeding program designed to incorporate genetic improvements in temperate zone wheats into wheat varieties suited to the subtropical wheat regions in developing countries. By the early 1960s this program had achieved considerable success, and by the late 1960s wheat varieties from this program were credited with creating a green revolution in Asia.

The second program was the program for Japonica–Indica rice development supported by the Food and Agriculture Organization of the United Nations (FAO). This program was dedicated to incorporating features of temperate zone Japonica rice improvements into tropical zone Indica rice cultivars. While the program was discontinued at the end of the 1960s, it did lead to the development of several important rice varieties and established the foundations for the early development of semi-dwarf plant type rice varieties in the 1960s at the first IARC, the International Rice Research Institute (IRRI). This development was also credited with creating a green revolution.

The design of the IARCs was guided by past experience with agricultural experiment stations. The features of recharge science and the provision of germplasm to NARS in the crop-focused IARCs were guided by induced innovation perspectives. In a sense, the IARCs were built to remedy the lack of level II pre-invention science in developing country NARSs.

The IARCs were also designed to be governed by independent boards instead of member countries. Again, this organization was based on experience gained from regional and United Nations (UN) programs, where the member country model of control was clearly shown to be inconsistent with the effective conduct of science. This enabled the IARC system to recruit highly qualified scientists into the system.

4.4.2   Varietal production and germplasm

Early accounts of the green revolution in developing countries are incomplete in two dimensions. First, they are based primarily on crop genetic improvement (CGI) in the form of “modern” varietal (MV) development in wheat and rice. Second, they are based on varietal development for the period from 1965 to 1980. A more complete picture of the green revolution is reported in Table 4.1, where a summary of the production of modern improved crop varieties for 10 food crops is presented. These 10 crops account for 85 to 90 percent of food crop production in the developing world. (Rice is the most important crop, followed by wheat and maize.) Table 4.1 also presents data for five-year periods extending through the 1980s and most of the 1990s. Summaries of varietal releases by region are also reported.12

Table 4.1. Average Annual Varietal Releases by Crop and Region, 1965–1998.

*Numbers in parentheses are simple repetitions of 1991–1995 rates because of insufficient data.

Note: IX = variety based on IARC cross; IP = variety based on NARS cross with at least one IARC parent; IA = variety based on NARS cross with at least one non-parent IARC ancestor; IN = variety based on NARS cross with no IARC ancestors.

IARC content in released varieties is summarized by crop and region as well. For all crop varieties, 36 percent were produced in an IARC breeding program where the cross leading to the varietal release was made in the IARC program (IX). The remaining varieties were based on crosses made in a NARS program.

NARS-crossed releases can be further classified according to whether one or both parents in the cross was an IARC release. For all NARS-crossed varieties, roughly 17 percent had at least one IARC-crossed parent (IP). (These varieties accounted for 25 percent of MV acreage in 1998.) This attests to a strong germplasmic recharge effect, since NARS breeders found success in using IARC parental material. When grandparents and other ancestors of NARS-crossed varieties are considered (IA), IARC-crossed germplasm appears in 23 percent of all NARS-crossed varieties. This IARC germplasm proportion is also rising over time.

There are significant differences in these patterns among crops. There are also differences in what might be termed the “maturity” of the breeding programs by crop. In wheat, the total number of releases has stayed relatively constant since 1985, but with a high proportion of varieties based on IARC crosses or parents. The varieties/breeder ratio has been constant since 1985. Wheat is produced over a narrower range of climate and biotic diversity than is rice and has a relatively high level of multiple releases. This might be termed a mature pattern.

Rice also exhibits a mature pattern, but of a different type. Total releases have also been roughly constant since the mid-1980s, but the IARC cross proportion has declined from the “green revolution” levels of the 1970s. This appears to be a case of maturing and strengthened NARS programs. Previous work by Gollin and Evenson (1997) supports this interpretation.

Maize, the third most important cereal grain in developing countries, offers another pattern. Public sector releases appear to be rising, with a relatively low IARC germplasm component. Private sector varietal production, primarily of hybrids, is clearly increasing. It is also clear that IARC germplasm has been useful to private breeders, along with NARS germplasm. This is a case of public sector research creating a “platform” on which the private sector can be productive. (Note that this also provides a platform for modern biotechnology products.)

The pattern for sorghum is roughly similar to that for maize, again with a growing proportion of hybrid sorghum varieties being produced by private sector breeders. The pearl millet pattern indicates relatively weak NARS production in early years. This is a case where the IARC program not only provided germplasm to NARS but initiated expanded CGI work generally. Until ICRISAT began its CGI work, there was little useful raw material for NARS programs to work with.13

The pattern where the IARC programs effectively initiated breeding work on a crop holds for barley, lentils, and cassava, as well as for millet. In each of these crops the IARC cross proportion is high and total varietal production is generally rising. The NARS CGI programs for these crops are not very mature at this point.

For beans, IARC programs have also stimulated increased varietal production, with IARC crosses accounting for high proportions of released varieties. In sub-Saharan Africa, the IARC programs are dominant in beans, cassava, and potatoes.

The conditions for potato CGI differ from those of other crops because of different taste and management factors. IARC cross shares are low but have been rising.

Table 4.1 also reports release data for all crops by region. These data show that the highest rate of increase in varietal production in the 1980s and 1990s occurred in the Middle East/North Africa and sub-Saharan Africa regions. These regions were also the most dependent on IARC crosses and germplasm.

A specific study of rice varieties (Gollin and Evenson 1997) reported that, of the rice varieties released over the 1965–1991 period, only 6 percent were crosses made in one NARS system that were released in another NARS system. However, for parents of releases, 18 percent were crossed in one NARS program and served as a parent in another NARS program. IRRI was responsible for 17 percent of varietal crosses and 24 percent of parental crosses. Only 8 percent of all modern rice varieties released since 1965 did not have international landrace content in their genealogy.

Total varietal releases for all crops show a steady increase over time. Annual varietal releases in the 1990s were more than double the releases during the 1970s.

The induced innovation model outlined earlier provides a basis for testing for a germplasm impact on NARS breeding programs. The “breeding with recharge” function actually imposes a specific functional form for the germplasm impact, and this form can be tested against a more general form. This test is carried out using NARS data for three periods, 1965–1975, 1976–1985, and 1986–1996, for varietal releases in wheat, rice, maize, beans, and potatoes.

The specific functional form implied by the induced innovation model is


where VN is the number of varieties produced by a NARS CGI program in a given period, BN measures the NARS plant breeding resources employed during the period, and GI is a measure of IARC germplasm available to the program.

The principle of diminishing returns dictates the logarithmic specification. The GI variable is not in logarithmic form because it is not part of the NARS search per se. That is, IARC germplasm affects NARS productivity, but (except indirectly) NARSs do not produce IARC germplasm. GI is thus a linear shifter of the search distribution for NARS and has the specific form noted in expressions (4.17) and (4.13).

A specification that does not rely on search theory would use a general production function. Perhaps the most widely used production functional form is the Cobb-Douglas form:


The variables are defined as follows:

•  VN is the number of varietal releases based on NARS crosses over the period, where N indexes countries.

•  BN is the number of scientists engaged in CGI research on the crop at the beginning of the period. BN was estimated in two stages. First, the total number of senior agricultural scientists for the period and country was computed from the International Service to National Agricultural Research (ISNAR) database (Pardey and Roseboom 1989). Then a search of the FAO Agrostat database was conducted for publications on plant breeding and related activities by crop, as well as on social science research, animal and pasture research, and other fields of agricultural science. Publication shares for plant breeding on the crops in question were then formed for each country. These shares were then multiplied by the ISNAR scientist data to obtain our measure of BN .

•  GI is measured as the cumulated number of IARC crosses released as varieties in the country. This definition of germplasmic input attempts to correct for the fact that only a subset of IARC-crossed material is relevant in a given country. If the country actually released an IARC cross as a variety, this is taken to be an indication of relevance.

Table 4.2. Estimates: Indirect Impacts of IARC Germplasm on NARS CGI Programs.

Dependent variable = NARS-bred varietal releases by period: 1965–1975, 1976–1985, 1986–1997 (t ratios in parentheses).

Table 4.2 reports a goodness of fit test (adjusted R2) for the two specifications. In all cases, specification (4.17) fit the data significantly better than specification (4.18). This supports the interpretation of the coefficients as real research effects. This is a case where model-guided specification is possible and supported by the data.

Coefficients for specification (4.17) are reported in Table 4.2. The key germplasm impact variable is ln(Bn)GI. The coefficients for this variable are positive and highly significant in all specifications. The coefficients are similar for crops separately and pooled. The net effect of the variable GI depends on both the coefficient of GI and on the coefficient of ln(Bn)GI. Since GI is growing over time, the negative coefficient on GI is adjusting for this to some degree. However, the net effect of GI is positive and large. The production elasticity of GI is approximately .3, evaluated at the mean of the data in the pooled estimates.

The varietal production elasticity of the NARS breeding effect depends on the coefficient of ln(Bn) as well as on the coefficient of ln(Bn)GI. For the pooled estimate, the production elasticity of Bn is approximately .7 (note the dependent variable is not in logarithms, so this is not a constant elasticity).

Thus, NARS breeding effects are subject to diminishing returns. NARS programs have approximately doubled over the periods studied. This would have produced a 70 percent increase in varietal production in the absence of GI effects. The GI effects were quite large and contributed roughly 30 percent more varietal production than would have occurred in their absence. Thus, IARC germplasm impacts on NARS CGI programs were sufficient to offset the diminishing returns to NARS breeding effects over the periods covered.

Evenson (2001) also estimated the net effect of IARC germplasm on NARS plant breeding investments. Since IARC programs both compete with and complement NARS programs, IARC germplasm could either crowd out or stimulate NARS investments. The study estimated NARS investment specifications and concluded that IARC germplasm stimulated NARS investments.

4.4.3   Modern variety adoption and location specificity

As noted in the previous section, roughly 35 or 36 percent of varieties released were crossed in an IARC program, and many of these crossed international borders in the form of releases. An additional 8 to 10 percent of these varieties were crossed in one NARS program and released in another. (Data on these international flows will be compared with data on patented inventions in the next section.)

Farmers actually have to adopt modern varieties if they are to have a production impact. Data on MV adoption are presented by crop and region in Table 4.3. These data are characterized by two features. First, adoption rates differ by region for the same crop, attesting to high degrees of location specificity. Second, adoption rates are correlated with varietal production data, but the correlation is far from perfect. In particular, both the Middle East/North Africa and sub-Saharan Africa regions have low levels of MV adoption in 1970 and 1980, even though there were significant levels of varietal release in preceding periods. Many earlier releases were not widely adopted because of susceptibility to plant diseases, insect pests, and abiotic stresses. It is when varieties are developed in response to these problems that high levels of MV adoption are observed.

Table 4.3. Modern Variety Diffusion: Percent area Planted with Modern Varieties, 1970, 1980, 1990, and 1998.

MV adoption rates can be converted into growth contributions with estimates of the production gain associated with MV adoption. The CGI study from which these data were drawn included three country studies (India, China, and Brazil) where these production gains were estimated. Estimates were based on cross-section variability in adoption. Adoption was treated as an endogenous variable in these studies. The resultant growth contributions are reported by decade and region in Tables 4.4 and 4.5.

Table 4.4 reports annual CGI contributions for the crops in the study. These rates differ by crop because of differences in MV adoption rates. For all crops combined, 35 percent of the area under MVs in 1998 was based on an IARC cross (IX). Another 30 percent was based on a NARS cross with an IARC-crossed ancestor (IA).

Table 4.5 reports CGI production growth components for all crops by region. This table essentially describes the “transformation” of developing country agriculture. Comparisons are often made between countries in the sub-Saharan Africa region and to a lesser extent in the Middle East/North Africa region showing low rates of productivity growth relative to Asian economies. Tests of these CGI component estimates with actual yield changes indicate that, for countries with low levels of research investment, the CGI component makes up more than half the total factor productivity (TFP) growth measured for the region.

Given this, Table 4.5 explains the poor growth performance of the sub-Saharan Africa and Middle East/North Africa regions. In the 1960s and 1970s, sub-Saharan Africa experienced virtually no CGI growth. Even in the 1980s, the region experienced only one-third the growth seen in Asian economies. The Middle East/North Africa region also lagged in CGI growth until the 1990s. Note, however, that the IARC content in both released and adopted varieties was highest in these regions, indicating a strategy of IARC compensation for weak NARS programs.14

4.5   Industry Evidence

As noted in the introduction to this chapter, most economists concerned with industrial development have adopted a “mimicry” policy philosophy and hence have not stressed the development of an industrial R&D capacity in developing countries except in advanced stages of development. This is in sharp contrast to agriculturalists, who, by and large, conclude that technology–ecosystem interactions produce a high degree of location specificity calling for a research (especially CGI) capacity in all important regions at all stages of development.

Table 4.4. Crop Genetic Improvement Contributions to Yield Growth, by Crop.

Note: IX = varietal cross made in IARC program; IA = varietal cross in NARS program with IARC ancestor.

Table 4.5. Crop Genetic Improvement Contributions to Yield Growth, by Region.

Note: IX = varietal cross made in IARC program; IA = varietal cross in NARS program with IARC ancestor.

Because industrial technology policy is dominated by the mimicry perspective, it has stressed direct foreign investment as a vehicle for acquiring tacit knowledge and has downplayed the building of domestic R&D capacity. Experience with industrial R&D and invention in the chemical, electrical, and mechanical fields has been quite different regarding public research organizations than has been the case for biological invention. Public sector research programs have been effective in biological inventions. They have not been as effective in other fields of invention. Nor have public sector extension programs for industry been as effective as they have been for agriculture.

As a consequence of the mimicry-based technology policy for industry, only 25 or so developing countries have a bona fide R&D capacity in producing industrial firms. Of these, only 15 to 20 have effective intellectual property rights systems (Evenson and Westphal 1994).

4.5.1   Indirect evidence on location specificity: International patent data by industry

International patent data afford a related measure of location specificity and enable a comparison between agricultural inventions and industry inventions. International patent classifications (IPCs) are given to patents for virtually all countries maintaining patent protection systems. The International Patent Documentation (INPADOC) database includes IPCs. The Yale Technology Concordance (YTC), based on Canadian Patent Office industry assignments, enables the assignment of patents to industries of manufacture (IOMs) and sectors of use (SOUs). This database also records the country of origin (the priority country) and the granting country for each patent.

Table 4.6 illustrates the priority country versus granting country for seven countries for the 1975–1988 period for six IPC-defined fields of agriculturally related inventions. According to Paris Convention rules, a priority inventor has a relatively short period (one year) to obtain patent protection in another member country while maintaining the priority date. The decision to obtain protection abroad depends on several factors, including the degree of location specificity of the invention. These factors also include the perceived importance of the invention and the size of the granting country market.15

Table 4.6 illustrates the differences in these decisions by field of technology. For non-fertilizer agricultural chemicals, the off-diagonal elements are higher relative to the diagonal elements than is the case for harvesting machinery. This reflects differences in location specificity as well as differences in the distribution of invention values. It is arguable that the patent field comparisons control for value distribution differences, so that the ratio of off-diagonal elements to diagonal elements in these matrices is an indicator of location specificity.

Table 4.7 reports ratios of off-diagonal to diagonal elements for patents granted in all fields in eight OECD countries over the 1969–1987 period. The patents in this case are classified by IOM and SOU using the YTC. The agricultural sector is not an important IOM, but it is an important SOU of inventions. While these indices are not entirely comparable with crop varietal data, they are instructive for comparison purposes. They suggest that inventions intended for use in transport equipment, fabricated metals, wood and furniture, other machinery, mining, construction, finance, and other services may be as location specific as inventions intended for use in agriculture.

Table 4.6. International Spill-In Comparison, by Number of Patents, 1975–1988.

Source: Evenson and Westphal (1994).

Table 4.7. Invention Spill-In Indexes by Industrial Class for Eight OECD Countries, 1969–1987.

Source: Evenson and Westphal (1994).

Note: IOM = industry of manufacture; SOU = sector of use.

4.5.2   Indirect evidence for location specificity: International patterns

Table 4.8 reports a more complete pattern of international choices regarding obtaining patent protection abroad. The table includes inventions in all fields; almost all inventions, however, are produced in the industrial sector. Table 4.8 essentially covers three blocks of countries: the developed market economy block, the recently industrialized countries (RICs) block, and the newly industrialized countries (NICs) block. The table does not include the countries of the former Soviet Union and the substantial majority of developing countries that do not have functioning patent systems.16

Table 4.8 thus has several “blocks.” The upper portion of Table 4.8a shows invention flows between developed market economies. These flows are extensive. The upper portion of Table 4.8b shows flows from developed countries to RICs and to NICs. These flows are also substantial. Developed market economies protect their inventions in these “downstream” countries.

The lower portion of Table 4.8a shows upstream invention flows from RICs and NICs to developed countries. These upstream flows are minor for RICs and negligible for NICs.

The lower portion of Table 4.8b shows minor flows from RICs to NICs and negligible flows from NICs to RICs. This portion of the table also shows negligible flows from RICs to other RICs and from NICs to other NICs. These data then do not support the proposition of technology “cascading” for industrial invention. In other words, it does not appear that RICs are adapting developed country inventions and selling them further downstream in the NICs.

The diagonal elements in both the RIC and NIC blocks, however, are of special interest to the analysis of convergence mechanisms. Are these adaptive inventions? If so, they will be related to the invention flows from developing countries (upper portion of Table 4.8b) as well as to domestic R&D in much the same way as the IARC germplasmic inventions are.

4.5.3   International recharge in industrial invention

An international recharge test for industrial invention that is roughly comparable to the test for agricultural inventions (in the form of varieties produced by NARS) is reported in Table 4.9. This test requires variables measuring the following:

•  Domestic invention

•  Domestic R&D or scientist and engineering resources

•  International recharge

Domestic invention (DOMINV) was measured by numbers of domestic patents obtained in 1980, 1985, 1990, and 1995.17

Table 4.8a. International Patent Flows, All Sectors, 1990.

Table 4.8b. International Patent Flows, All Sectors, 1990.

Note: RIC = recently industrialized country; NIC = newly industrialized country.

Table 4.9. Estimates: Industrial Adaptive Invention Specifications.

Dependent variable = domestic patents.

Note: All specifications include time and country dummy variables.

Domestic R&D was measured by two alternative variables:

•  SC is the total number of scientists and engineers in all economic sectors according to United Nations Educational, Scientific, and Cultural Organization (UNESCO) data.18

•  SCBE is the reported proportion of R&D financed by the business enterprise sector reported by UNESCO.19

International recharge is measured by three alternative variables:

•  FPATSTK is the current period number of foreign origin patents granted in the country.

•  CFPATSTK is the cumulative number of foreign origin patents granted in the country.20

•  FPAYSTK is the cumulative royalty payments (in US dollars) to foreigners for technology rights.21

Table 4.9 reports five specifications incorporating the alternative variables. Observations are for four periods: 1980, 1985, 1990, and 1995. All specifications include country and year fixed effects.

•  Specifications (1) and (2) provide a comparison between a general specification (1) and the search-based recharge specification (2). The data clearly prefer the search-based recharge specification.

•  Specifications (2) and (3) provide a comparison of current foreign patents and cumulated foreign patents as the recharge measure. The data prefer specification (3).

•  Specifications (3) and (4) provide a comparison between SC as a measure of scientists and SCBE as a measure of scientists. The data prefer SCBE as a measure of scientists.

•  Specifications (4) and (5) provide a comparison between FPATSTK and FPAYSTK as measures of foreign recharge. Interestingly, both measures appear to be relevant indices of foreign technology.

The negative coefficients for FPATSTK (FPAYSTK) combined with the positive coefficients for ln(SCBE) × FPATSTK (FPAYSTK) can be interpreted as both a competition effect (the negative effect) and a germplasm or recharge effect (the positive coefficient). The partial elasticities of domestic invention, with respect to SCBE or SC, are .8 to .85 in the absence of foreign recharge and rise to 1.15 to 1.25 when mean foreign recharge is present. The partial elasticities of domestic patents to foreign patenting or technology payments are actually negative for countries with low SC or SCBE levels (below the 40th percentile) and rise as SC or SCBE rises.

Thus, it appears that the competitive effect of foreign technology payments discourages small countries from investing in R&D even though the complementary effect encourages it.

4.6   Conclusion

Empirical tests of the induced adaptive invention model with international recharge are reported for both agricultural biological invention in developing countries and for industrial invention for a subset of developing countries. Both sets of tests support the proposition that induced adaptive invention/innovation is a vehicle for convergence. Technology policy for agriculture is based on this proposition. Technology policy for industry is generally based on mimicry as the chief mechanism for convergence. In recent decades, convergence failures in industrial development have moved policy in the direction of mimicry with high tacit knowledge requirements as the basis for policy design.

Does the evidence for the adaptive invention/innovation model imply a glaring inconsistency in industrial development policy? Or is the evidence also consistent with the mimicry with high tacit knowledge mechanism? More work is required to settle this issue, but we do know that virtually all cases of rapid industrial productivity growth have taken place in economies with an R&D capacity in industry (Evenson and Westphal 1994). We also know that all cases of modest industrial productivity growth have taken place in economies purchasing foreign technology. The evidence in this paper indicates a strong complementarity between domestic R&D and foreign technology purchase.

While agricultural research institutions were built in response to food security issues associated with the population expansion of the post–World War II period, the longer-term investment strategy for development requires more explicit attention to convergence mechanisms. Regardless of whether the adaptive invention/innovation mechanism or the mimicry with high tacit knowledge mechanism dominates, it is relatively clear that investment in research and engineering capacity is required to achieve convergence.


1.   Many of these models also stressed institutional constraints that were perceived to be more important for agriculture than for industry. Gerschenkron (1952) provides a discussion of economic backwardness. Landes (1990) provides a more recent historical perspective.

2.   Jones (1998) provides evidence for this. Barro and Sala-i-Martin (1996) summarize this evidence. Mankiw et al. (1992) also report estimates.

3.   Most endogenous growth models are predicated on the tacit knowledge constrained mechanism. Lucas (1988), for example, postulates human capital as the key factor. Jones (1995) has a learning-by-doing model. Mokyr (1996) provides a broad historical perspective on technological change.

4.   The models of Romer (1986, 1994) and to some extent the Jones model (1995) recognize the importance of research and development and inventions but do not model international recharge mechanisms. Arrow (1962) provided the original model of “learning by doing.”

5.   Invention recharge is implied by the term adaptive inventions.

6.   The pre-breeding activities of the International Agricultural Research Centers are an important part of the recharge mechanism discussed in Section 4.4.

7.   This independent search for traits is dictated by the fact that traits have different genetic sources.

8.   The selection process following crossing requires multiple-period evaluation.

9.   The elimination of unpromising search avenues is one of the products of research programs.

10.  Note that, for inventors receiving recharge, the recharge elements shift the invention distribution linearly. The diminishing returns to pre-invention or recharge activities are not incorporated into the recharge recipient's invention functions. This implication is tested in the application in Sections 4.4 and 4.5.

11.  Spillovers can be of two types, direct and indirect. Direct spillovers occur when an invention made in one location or industry is directly used in another. Indirect spillovers occur through the invention recharge mechanism.

12.  These data are based on a CGI study commissioned by the International Agricultural Evaluation Group—a body of the CGIAR. The author was the principal investigator of that study. The three country studies were commissioned to provide consistent estimates of the impact of MVs on production.

13.  ICRISAT is the International Center for Research in the Semi-Arid Tropics, one of the 16 IARCs.

14.  Related work on technology capital (or infrastructure) evaluates other sources of TFP growth in acquisitions and shows that direct technology spillovers from industry are an important component of TFP growth (Evenson 2000, 2001).

15.  In many cases inventors consider political factors in decisions to obtain patent protection abroad. Most developing countries exhibit hostility toward IPRs generally. The perception of widespread “piracy” of IPRs by developing countries was a factor in the inclusion of IPRs in the Uruguay round of the General Agreement on Trade and Tariffs.

16.  The classification of developing countries in the RIC and NIC categories is somewhat arbitrary. It should be noted that Zambia is not generally regarded as an NIC and that Indonesia, Thailand, and Chile are regarded as NICs (or near NICs) but do not have functioning patent systems.

17.  These data are tables for the INPADOC database and are reported in Johnson and Evenson (2000).

18.  These data are from the UNESCO database: http://unescostat.unesco.org/stat sen/st

19.  These data are also from UNESCO.

20.  These data are reported in Johnson and Evenson (2000).

21.  This variable is from the 2000 World Development Indicators CD-ROM, World Bank.


Arrow, K.J., 1962, The economic implications of learning by doing, Review of Economic Studies, 29(June):153–173.

Barro, R.J., and Sala-i-Martin, X., 1996, Convergence Across States and Regions, Brookings Papers on Economic Activity, Brookings Institution, Washington, DC, USA, pp. 107–158.

Evenson, R.E., 2000, Agricultural production and productivity in developing countries, in The State of Food and Agriculture, Food and Agriculture Organization of the United Nations, Rome, Italy.

Evenson, R.E., 2001, Economic impacts of agricultural research and extension, in B. Gardner and G. Rausser, eds, Handbook of Agricultural Economics, Vol. 1, Elsevier Science B.V., Amsterdam, Netherlands.

Evenson, R.E., and Kislev, Y., 1975, Agricultural Research and Productivity, Yale University Press, New Haven, CT, USA.

Evenson, R.E., and Kislev, Y., 1976, A stochastic model of applied research, Journal of Political Economy, 84(2):265–282.

Evenson, R.E., and Westphal, L., 1994, Technological change and technology strategy, in T.N. Srinivasan and J. Behrman, eds, Handbook of Development Economics, Vol. 3, North Holland Publishing Company, Amsterdam, Netherlands.

Gerschenkron, A., 1952, Economic backwardness in historical perspective, in B.F. Hoselitz, ed., The Progress of Underdeveloped Areas, University of Chicago Press, Chicago, IL, USA.

Gollin, D., and Evenson, R.E., 1997, Genetic resources, international organization, and rice varietal improvement, Economic Development and Cultural Change, 45(3):471–500.

Griliches, Z., 1957, Hybrid corn: An exploration in the economics of technological change, Econometrica, 25:501–522.

Huffman, W., and Evenson, R.E., 1993, Science for Agriculture, Iowa State University Press, Ames, IA, USA.

Johnson, D.K.N., and Evenson, R.E., 2000, R&D spillovers to agriculture, measurement and application, Contemporary Economic Policy, 17(4):432–456.

Jones, C.I., 1995, R&D-based models of economic growth, Journal of Political Economy, 103(August):759–784.

Jones, C.I., 1998, Introduction to Economic Growth, W.W. Norton & Company, New York, NY, USA.

Landes, D.S., 1990, Why are we so rich and they so poor?, American Economic Association Papers and Proceedings, 80(May):1–13.

Lucas, R.E., Jr., 1988, On the mechanics of economic development, Journal of Monetary Economics, 32(July):3–42.

Mankiw, N.G., Romer, D., and Weil, D., 1992, A contribution to the empirics of economic growth, Quarterly Journal of Economics, 107(May):407–38.

Mokyr, J. (1996), The Lever of Riches, Oxford University Press. New York, NY, USA.

Pardey, P.G., and Roseboom, J., 1989, A Global Data Base on National Agricultural Research Systems, ISNAR Agricultural Research Indicator Series, Cambridge University Press, Cambridge, UK.

Romer, P., 1986, Increasing returns and long-run growth, Journal of Political Economy, 94(October):1002–1037.

Romer, P., 1994, The origins of endogenous growth, Journal of Economic Perspectives, 8(Winter):3–22.

USDA, various years, Agricultural Statistics, United States Department of Agriculture, Washington, DC, USA.