Global networks of genetically modified crops technology: a patent citation network analysis

Abstract

This paper employs the patent data of four major genetically modified (GM) crops, soybeans, cotton, maize and rapeseed, to illustratee how the innovation of GM crop technology diffused and distributed globally over time. Data collected from the Derwent Innovation Index, were employed to construct country patent citation networks, from 1984 to 2015, and the results revealed that developed countries were early adopters, and the primary actors in the innovation of GM crop technology. Only seven developing countries appeared in the country citation network. Most developed countries were reluctant to apply GM crop technology for commercial cultivation. Private businesses stood out in the patent citation network. The early adoption and better performance of developed countries can be explained by the activities of large established private companies.

Introduction

Agro biotechnologies used to produce genetically modified (GM) food, are some of the most innovative fields in science. Since the early 1980s, genetic engineering has been applied to agricultural crops and heralded as a biological revolution, aimed at alleviating world hunger and poverty and contributing to the goal of global food security.

Two aspects of the distribution and adoption of a country’s GM agricultural products are worthy of examination, commercial cultivation and the patents of GM agricultural products. As shown in Fig. 1, commercialization, the planting of GM crops has diffused rapidly from developed nations, the United States, Canada and Australia, to developing countries, Argentina, Uruguay, China, South Africa and Brazil. The International Service for the Acquisition of Agribiotech Applications (ISAAA) reported statistics on the global commercialization of biotech crops from 1996 to 2015. 28 countries were growing biotech crops in 2014, and 26 countries in 2015 (James 2015). According to ISAAA, the top-10 biotech crop growing countries were the United States, Brazil, Argentina, India, Canada, China, Paraguay, Pakistan, South Africa, and Uruguay, accounting for 98% of the global GM hectarage in 2015 (James 2015). Among the top-10 countries, only the United States and Canada are developed countries. Countries in Europe and Africa have been very cautious in the adoption of the technology (Canadian Biotechnology Action Network 2015). GM crops were grown on approximately 0.14% of arable land in Europe, mainly in Spain, and only three countries (South Africa, Burkina Faso and Sudan) in Africa (Friends of the earth international 2014).

Fig. 1

Timeline of different countries initiation of GM crop production. The countries in the timeline are those growing GM crops in 2016 according to Global Status of Commercialized Biotech/GM Crops: 2016. The years shown on the timeline were collected from USDA Foreign Agricultural Service and other online sources

The global diffusion and distribution patterns of technological innovation and commercialization of GM crops may be different. New technologies must undergo the process of research, commercial event identification, safety evaluation, and finally, commercial authorization, which is lengthy, complex, and resource-intensive (Conko 2012; Grushkin 2013; Privalle et al. 2012). More importantly, widespread public reservation and concern shape the decision-making of GMO release regulation (Qaim 2009). As a result, overregulation keeps some GM agricultural technologies in the pipeline and hinders commercialization of the technology. On the other hand, GM seeds sometimes expand across borders without official authorization, and there is unauthorized cultivation around the world (Sinebo and Maredia 2016). Hence, the diffusion and distribution pattern of technological innovation may be distinct from commercialization. Understanding the innovation process of GM crop technology is crucial for governments, corporations, farmers, scientists and other relevant stakeholders in decision-making.

Much research has focused on the expansion and cultivation of GM crops, but the diffusion of the technology has not received the same attention. In this study, the global diffusion of GM crop technology is examined at the country level, complementing the assessment of the development and adoption of GM crop technology. To further understand the innovation diffusion process, we examined contributions from different social entities and organizations.

Literature review

Patent citation network

Although controversial, patents are used as representatives or proxies for innovation. Griliches (1990) concluded that patent data is a unique resource for the study of technical change. For strategic technology management, the information embodied in patent data can be employed for strategic planning. For instance, they can be used to monitor competition and manage the development of new technologies (Ernst 2003). Patent bibliometrics have also been used to investigate innovative and economic activity (Hicks et al. 2001; Lybbert and Zolas 2014), which can be used for government policy-making. Patent analysis has been applied in the many fields, such as semiconductors (Hall and Ziedonis 2001), fuel cells (Verspagen 2007), solar energy (Luan et al. 2012), and printed electronics technologies (Kim et al. 2014). Various techniques for analyzing patents have developed. Information including the inventors, assignees (organizations filing the patents) and citation information are included in structured patent data (Tseng et al. 2007), which provides a basis for text mining and analysis to fulfill a variety of different purposes.

Among the taxonomy of patent analysis techniques, patent citation has experienced a dramatic increase in social science research (Jaffe and de Rassenfosse 2016). Patent citation indicates that the citing patent conveys part of existing knowledge in the cited patent (Érdi et al. 2013). Patent citation reflects the transferring of knowledge between different patents and organizations. Trajtenberg (1990) highlighted the power of applying patent citation to measure the value of innovations instead of utilizing just patent counts. Patent citation analysis provides a representation and evaluation of technological innovation, tracks the trajectory of technological knowledge flow and explores the origins of the technology (Jaffe et al. 1993). In this study, patent citation is employed to measure the diffusion and distribution of GM crop technology. Although using patent citation analysis to measure innovation and knowledge flow has been challenged by scholars (Alcacer and Gittelman 2006; Gambardella et al. 2008), Barberá-Tomás et al. (2011) corroborated the validation of patent citation methodologies.

In this study, the diffusion and distribution of GM crop technology are represented by country and organization citation networks. The application of network analysis methods provides a novel perspective and a much richer source of information. Based on its connectivity, the relationship between patents, the understanding of innovation and knowledge flows is facilitated (Brantle and Fallah 2007; Pyka and Scharnhorst 2010; Nam and Barnett 2011). In the patent citation network, the nodes (the unit of analysis) could be patents, institutions or even countries (Huang et al. 2003; Choe et al. 2013), and the edges (the links between nodes), between them are usually a direct citation relationship. Networks help describe and analyze the patent system and complement existing econometric studies of patent citations, which typically rely on the number of patent citations. In this paper, two types of networks are constructed: the organization citation and country citation networks. In organization citation networks, the nodes are the organizations that acquire ownership interest in patent applications and the edges the direct citation relationships between them. In country citation networks, the nodes are the countries, which refer to the patent applicants’ countries of affiliations and the edges the direct citation relationships between the countries. The influence of countries or organizations is measured through degree centrality, the number of links with other nodes, which reflects the location and influence of countries or organizations. The structural properties of networks, their technological relationships, facilitates the analysis of innovation distributions and dynamics of the citation system.

Innovation diffusion of GM crop technology at the country level

The disparity between consumers’ attitudes towards genetically modified foods worldwide is well known, and it is same with the discrepancy between countries’ overall attitudes. Some countries have been very cautious in the application of GM crop technology, including most countries in Europe that have large biotechnology companies. Hence, we question whether the global diffusion and distribution patterns between technological innovation and commercialization of GM crops are the same, leading to our research question:

RQ1

What is the global innovation diffusion pattern of GM crop technology among countries?

Based on product and industry life cycles, Vernon (1966) has shown that new products and processes often originate in the world’s richest countries, such as the United States, and are then exported to countries with similar revenue levels. Keller (2004) argued that the pattern of worldwide technical change is determined in large part by international technology diffusion, as only a few rich countries are responsible for the majority of the world’s new technologies. Many factors affect a country’s adoption of technological innovation. Caselli and Coleman (2001) found that human capital plays an important role in computer technology adoption, and that developed countries have the advantage of a large human capital pool, which promotes the adoption of new technology. According to Hall and Khan (2003), the regulatory environment and governmental institutions have a significant effect on technology adoption, and rich countries with sound policies are better prepared for adoption. In general, innovation starts in the richest nations and is easily adopted by them, which makes them more influential in technology development.

Biotechnology was born in the United States, immediately diffused to Canada, Japan and Western Europe, and then slowly diffused to less developed countries (Niosi et al. 2013). GM crop technology is an application of biotechnology. Rich nations with skilled personnel are more likely to lead in its development and adoption. Thus, we suggest the following hypotheses:

H1

In the country citation network of GM crop technology, developed countries entered the network earlier.

H2

In the country citation network of GM crop technology, developed countries are more influential.

Innovation diffusion of GM crop technology at the organization level

The GM agricultural technological innovation activities of a country are the aggregation of different social entities. There are diverse organizations and stakeholders involved in the research of GM crop technology, including universities, industry and government institutes (Etzkowitz and Leydesdorff 2000; Ho and Cheo 2014). To better understand the innovation diffusion of GM crop technology, we aggregated to the organizational level and assigned organizations into different categories, trying to determine which organizational types contributed more to the technology development, and how they perform in different countries. Our second research question is:

RQ2

How do different organizations perform in the research and development (R&D) of GM crop technology?

The private sector (private businesses) and the public sector (research institutes and universities) are the basic social entities involved in biotechnology research. Schumpeter argues that firms with larger market shares have a tendency to adopt novel technology because they have a greater ability to profit from adoption (Hall and Khan 2003). Private businesses are profit-driven which leads them to early adoption of GM crop technology.

Research from private businesses is more practical and will result in patents, while research from the public sector is more exploratory and mainly results in scientific publications (Aghion et al. 2010). Private businesses, are motivated by the development of the technology; they need to earn profits. Therefore, as we evaluate the practical development of GM crop technology, we predict that:

H3

In the organization citation network of GM crop technology, private businesses are more influential than other organizations.

What about the performance and diffusion patterns among private businesses? Feldman and Audretsch (1999) argued that innovation often originates from and would be first adopted by metropolitan cities, where many large firms and companies are located. The initiation and adoption of innovation are often related to firm size. Rogers (1995) argued that firm size is an important organizational attribute for innovation diffusion: “[Firm] size is probably a surrogate measure of several dimensions that lead to innovation: total resources, slack resources, technical expertise of employees” (Rogers 1995, p. 379). Large firms are usually early adopters of technological innovation (Ahuja et al. 2008), which is consistent with the Schumpeter’s hypothesis that larger firm confers an advantage to exploit R&D and market resources in innovation activity (Cohen 2010).

There are a number of studies specifying the relationship between innovation and firm size. As firms grow large, the loss of either managerial control or excessive bureaucratic control will result in less R&D efficiency. Moreover, with larger firm size, the incentives of individual scientists and entrepreneurs may be blunted, which suggests a counterargument to the relationship between innovation and firm size (Cohen 2010, p. 133). Except for the very largest firms, innovative activity is still positively related with firm size (Cohen 2010). The innovation of GM crop technology should be facilitated by financial services, specialized talents and managerial resources, advantages held by large firms, suggesting that they are more likely to adopt and invest in R&D of GM crop technology. Large firms should be more productive and influential in the development of the technology. Therefore, we predict that:

H4

In the organization citation network of GM crop technology, large firms would be principal participants, and locate in the core (center) of the network.

H5

In the organization citation network of GM crop technology, large firms would be the early adopters entering the network.

Method

Data collection

To study the innovation of GM crop technology, we examined four GM products: soybeans, maize, cotton and rapeseed. According to ISAAA’s GM Approval Database, 29 crops have been approved for commercialization and planting (“ISAAA’s GM Approval Database,” n.d.). Among these crops, soy, maize, cotton and rapeseed account for almost all commercial GMO production (Ricroch and Hénard-Damave 2016). Patents of these four crops were used to measure the innovation of genetically modified technology. The patents were collected from Derwent Innovation Index by Boolean operators: TOPIC: (“genetically modified” OR transgenic OR “gene edit*”) AND TOPIC: (soybean OR soyabean OR “soya bean” OR maize OR corn OR cotton OR rapeseed OR “oilseed rape” OR rapa OR “rape seed” OR canola) from 1984 to 2015. As for the rules of topic search for Derwent Innovation Index, the keywords were searched through the field of the Title and Abstract within a patent record. The first GM crop related patent was issued in 1984.

The collected patents were shown as a function of time in Fig. 2. 5616 patents were downloaded from Derwent Innovation Index. Generally, the granted patents increased from 1984 to 2015, except for the period between 2000 and 2006; during which genetically modified foods were hotly debated.

Fig. 2

Numbers of patents issued as a function of time

Network construction

The first step in the construction of the citation network was to clean the patent data. For a single patent record, only patent numbers, assignees (organizations filing the patents), and cited patent numbers were extracted from structured meta-data of patent documents. For this research, we studied the innovation diffusion at the organizational and country levels, so organization and country citation networks are constructed.

Unit of analysis

The networks were constructed for each year. Suppose that time t, containing patents from 1984 to t, is the year the first citation network formed, then the second citation network contains patents from 1984 to t + 1, so the citation networks to be studied include patents: 1984 ~ t, 1984 ~ t + 1, 1984 ~ t + 2, …1984–2015.

$P = \left\{ {p_{1} ,p_{2} , \ldots p_{n} } \right\}$, n = 5616. $\forall {\text{i}} \in \left\{ {1, \ldots n} \right\}$. We extracted the patent numbers, organization names and codes, and patents cited by inventors/examiners. So $P = \left\{ {p_{1} ,p_{2} , \ldots p_{n} } \right\}$, $\forall {\text{i}} \in \left\{ {1, \ldots n} \right\}$, $p_{i}$ has 3 data sets as follows:

$$\left\{ {\begin{array}{*{20}c} {A_{i} = \left\{ {a_{1}^{i} ,a_{2}^{i} , \ldots a_{{r_{i} }}^{i} } \right\}} \\ {B_{i} = \left\{ {b_{1}^{i} ,b_{2}^{i} , \ldots b_{{s_{i} }}^{i} } \right\}} \\ {C_{i} = \left\{ {c_{1}^{i} ,c_{2}^{i} , \ldots c_{{t_{i} }}^{i} } \right\}} \\ \end{array} } \right.$$

$A_{i}$ represents patent numbers, $B_{i}$ represents organizations (individual-excluded assignees, also referred to assignee), and $C_{i}$ represents cited patent numbers. If the intersection between data sets $A_{i}$ and $C_{j}$ is not an empty set, then patent $P_{i}$ is cited by patent $P_{j}$. An arbitrary element in $B_{i}$ has a citation relationship with arbitrary element in $B_{j}$. The weighted edges are the number of patent citations between the organizations. The first organization citation network formed in 1990, and a total of 26 weighted organization citation networks were constructed (Fig. 3).

Fig. 3

The flowchart for the construction of organization citation networks

Based on the organization citation networks, country citation networks were constructed. First, we identified the organization’s country, based on the contacted address of the organizations. Then, the MapReduce Paradigm was applied to construct the country citation networks.

Figure 4 illustrates the procedure for constructing the country citation network. First, downloaded patent records are input. As there may be several organizations included in a patent, splitting was conducted, followed by mapping and shuffling. At reducing phase, we have listed all the key-value pairs. The final step is the summary of the reducing results.

Fig. 4

The MapReduce Paradigm

Types of the organizations

We are also interested in the how the innovation of GM crop technology flows between different social entities, so all the organizations were coded and assigned to different types based on the information provided on organizations’ official websites. Based on the owners, basic objective and funding, the organizations are classified into nine social types: public sector (including government agency, public research institute and public university), private sector (private business), private university, third sector, public private partnership (PPP), semi-state and government-owned companies, see “Appendix”.

Two-mode network

Two-mode networks focus on two sets of actors, or one set of actors and one set of events. Relations in a two-mode network measure tie between the actors in one set and actors in a second, which means ties existing only between nodes belonging to different sets (Wasserman and Faust 1994, p. 39).

To answer the question “How do different organizations perform in different countries?” two-mode networks were constructed. There are two types of nodes in the network: country and types of organizations. For these networks, there are only connections between countries and social entities. There are no connections among countries or social entities.

Results

Innovation distribution of GM crop technology at the country level

To answer the research question: “What is the pattern of global innovation diffusion at the country level?” 26 country citation networks were constructed from 1990 to 2015, one for each year. While 1984 was the first year that a patent was filed, 1990 was the first year that a patent citation network formed. There were 28 nodes and 162 edges in the network in 2015. The number of nodes increased over time, indicating that more countries were adopting this technology, although the rate of change was very slow (Fig. 5).

Fig. 5

Number of nodes and edges of the country citation network as a function of time

Timeline for the countries entering the network

To test H1, we examined when countries entered the citation network. Figure 6 shows the first time a country either cited or cites others in the network. There were two countries in 1990, and 28 in 2015. Germany and Switzerland were the first to appear in 1990, followed by the United States, Netherlands, Austria, UK, Belgium, Australia, France, and Japan in 1990s. All the nations that entered network early were developed countries. Nearly 10 years later (2004), developing countries began to appear in the network, indicating a lag in GM crop technology development. Accordingly, H1 is supported. The seven developing countries involved in GM crop technology R&D were India, China, Brazil, Uzbekistan, Malaysia, Mexico, and South Africa.

Fig. 6

Time line showing countries’ time of entry into the technology diffusion network

Countries involved in the technology’s development were mainly economically rich and scientifically advanced, while those engaged in growing commercial GM crops were mainly developing countries. In 2015, while the top-ten biotech crop growing countries were the United States, Brazil, Argentina, India, Canada, China, Paraguay, Pakistan, South Africa, and Uruguay, only the United States and Canada were developed countries (James 2015). For the R&D of GM crop technology, there were 21 developed countries and only seven developing countries.

Centrality for the country

While rich advanced nations were first in technological innovation, developing countries started to catch up. Generally, there were two paths for catching up: path-following and path-skipping. Sometimes developing countries that catch up become more influential than developed countries.

Are developed countries more influential in the country citation network of GM crop technology (H2)? Do developing countries still lag in innovation? The influence of a country can be measured through its weighted in-degree and out-degree. The weighted in-degree and out-degree of countries are shown in Fig. 7. Weighted in-degree is defined as the number of times a node is cited by other nodes, and weighted out-degree is number of times it cites other nodes. Countries with high in-degree are highly cited by others, indicating that they are influential in technology development. Countries with high out-degree frequently cite other countries, which suggests that they have a large number of patents.

Fig. 7

a weighted in-degrees and b weighted out-degrees of the countries from 1990 to 2015. (Different color represents different year)

The United States had the largest weighted in-degree and weighted out-degree for almost all times, followed by wealthy European nations: Germany, Switzerland, Belgium, Netherlands, UK, France, and Austria, which were also the early adopters of the innovation. China, and Brazil entered the citation network very late but surpassed some developed countries in both citing and being cited.

Innovation diffusion of the technology at the organization level

A country’s GM crop technology innovation comes from different organizations, depending on a nation’s various stakeholders. To answer the question “How do different organizations perform in the R&D of GM crop technology?” we constructed organization citation networks and assigned them into different categories. Also, we explored the specific organizations engaged in the history of GM crop technology development. To test the innovation diffusion of GM organizations, 26 organization citation networks were constructed; this is shown in Fig. 8. The number of nodes and edges kept increasing; reaching 417, and the number of edges reached 1546 in 2015.

Fig. 8

Number of nodes and edges of the organization citation network as a function of time

Social entity analysis

The organizations constructed in the network were assigned to nine categories as described in the method section, including government agency, public research institute, public university, private university, private business, third sector, public private partnership (PPP), semi-state and government-owned companies. A sequence of six networks corresponding to different times (1984–1990, 1984–1995, 1984–2000, 1984–2005, 1984–2010, 1984–2015) is shown in Fig. 9. In the networks, the nodes are the organizations and the edges the citations between them. The size of the node is proportional to its degree and the colors represent different types social institutions.

Fig. 9

The organization citation networks, as a sequence of 6 static networks corresponding to different times. Node colors indicate types of organization

The number of nodes and edges increased over time, indicating that the types of organizations became more diverse as more entities were engaged in the citation network. Figure 10 shows the percentage of the different types of organization in the citation network. There were two categories in 1990, four in 1995, and eight after 2000.

Fig. 10

The percentage of the entities of organizations at 6 stages

Private businesses entered the citation network early. The pink color representing private business remains in the core of the networks over time. The proportion of private business declined 75% in 1990, 87.1% in 1995, 55.4% in 2000, 55.7% in 2005, 54.7% in 2010, and 46.4% in 2015, indicating increased activity by other organizational types.

How do the different types of organizations perform in different countries? To answer this question, a two-mode network of country and social entity types was constructed for 2015, summarizing the different organizational types involved in different countries (See Fig. 11). Private business connected most of the countries and was the major participant for most nations. For the most influential countries, the United States, Germany, Netherlands, UK, France, and Switzerland, technological innovation mainly came from private business. Public research institutes and public universities also established connections for many countries. In China, public research institutes and public universities contributed a great deal. The participation of other types of social entities was less. They were connected to only a few countries in the network.

Fig. 11

Two-mode network for social entity and country. Pink nodes represent countries and blue nodes represent nine social entities involved in innovation. The sizes of the nodes indicate the number of the participants of a country in the network

The performance of the major participant, private business is reflected in two aspects: the volume of the production and the influence of the technology. The weighted in-degree and weighted out-degree of the private business and non-private business (social entities without private business) were compared in Table 1.

Table 1 T test of the in-degree and out-degree for organization citation network—2015

As shown in Table 1, both the weighted in-degree and out-degree of private businesses were significantly higher than that of non-private businesses. Thus, H3 is supported. In the R&D of GM crop technology, innovation from private business was more highly cited by others, indicating they are more influential than other entities. Also, private businesses produced more of innovations.

Power law distribution of the network

Generally, private businesses performed better in the innovation of GM crop technology. Did these companies contribute equally? We explored the influential organizations engaged in GM crop technology development. To understand the performance of the organizations, we determined the distribution of the weighted in-degree and weighted out-degree for 2015. The power law distribution model was fit to the data. Figure 12 shows the frequency of in-degree and out-degree in log-scale and the linear fits.

Fig. 12

Power law fit of the weighted degree of the organization citation network in 2015. (Left) weighted in-degree. (Right) weighted out-degree

The typological analysis of the organizations gives us a picture of the network structure. Many patent citation networks are scale free networks characterized by a power law distribution (Bilke and Peterson 2001; Chen and Hicks 2004; Choe et al. 2013). Barabási and Albert (1999) proposed scale free networks, such that the nodes’ degrees follow the power law distribution. For these networks, typological properties are determined by a few highly connected nodes acting as hubs and accounting for the majority of the connections. The majority of the nodes have a relatively small number of edges. Organizations with high in-degree, located in the hub of the network, are highly cited by others, indicating that they are influential in technology development. Whereas, organizations with high out-degree cite many other patents and tend to have higher levels of production.

Both the distributions of weighted in-degree and weighted out-degree followed a power law, indicating that there are few central organizations with high in-degree and out-degree in the network. The influence of organizations in GM crop technology development was not evenly distributed and the production of GM crop technology varied among the organizations. A few organizations with high in-degree created fundamental innovations, seminal patents, and were very influential in GM crop technology development. While those few organizations with high out-degree cited others many times, implying they produced many patented innovations. In conclusion, the power law distribution indicates that there was a core or the hub in the network.

Core of the organization citation network

The power law distribution indicates uneven influence of organizations. Which ones were located in the core of the network? H5 asked whether the large firms would be principal participants and located in the center of the organization citation network.

Table 2 lists the top-20 organizations with the largest weighted in-degree centrality weighted out-degree centrality and betweenness centrality. The top-20 organizations in 2015 were Monsanto, Dupont, Bayer, DEKALB, Syngenta, Novatis, Ciba Geigy, Biotechnica International, BASF, Plantec Biotechnologie GMBH, Hoechst Schering Agrevo GMBH, Pioneer Hi Bred Int, DOW, AVENTIS, DSM, Stine Seed Farm Inc, Limagrain Euro SA, Cornell University, Agric Genetics Co, and Sandoz. Most are large private firms. H4 is supported. Most organizations with high in-degree centrality also have high weighted out-degree and enjoy high betweenness centrality.

Table 2 Centrality of the top organizations in the citation network 2015

Large firms were at the center of the organization citation network. Did they also become part of the citation network earlier? Figure 13 shows the timeline when the organizations entered the citation network.

Fig. 13

Timeline of core organizations entering the technology citation network. Only organizations that appeared in the core of the 2015 citation network are shown

According to H5, the large firms were the early adopters. Figure 13 lists the year when the organizations became part of the citation network. Among the top-20 organizations, 19 are large firms. More than half of the large firms began GM crop technology before 2000. The industrial giants began innovating before 1995. H5 is supported. Besides private businesses, only Cornell University entered the organization citation network (1997).

Discussion

Global diffusion and distribution patterns between technological innovation and commercialization of GM crops

Although the development and adoption of a country’s GM crop technology can be evaluated commercially from its growing status and technological innovation, the diffusion and distribution patterns of agricultural innovations and the commercialization of GM crops are inconsistent. Figures 1 and 6 show the timeline when countries became involved in the commercialization of GM crops and when they developed innovations. The GM technological innovation diffused from developed to developing countries. Developed countries started innovating in the 1990s, and it was nearly 10 years later that developing countries began appearing in the patent citation network. Commercially genetically engineered crops did not follow the same pattern of innovation and diffused from advanced nations, the United States, Canada and Australia, directly to developing countries, Argentina, Uruguay, China, and South Africa faster than the rate of patent diffusion. Also, the commercialization didn’t diffuse to other developed countries, such as Europe.

The distributions of the adoption of technological innovation and commercial GM crops are different. For the R&D of GM crop technology, the majority of countries were wealthy and scientifically advanced, with only seven developing countries, India, China, Brazil, Uzbekistan, Malaysia, Mexico, and South Africa in the network. In contrast, the commercialization of GM crops occurred mainly in developing countries. There are several factors that contribute to the difference between the two diffusion patterns.

The level of R&D is one factor. Wealthy advanced countries with adequate financial support, human resources, and scientific management systems are capable of GM crop technology development, while it is difficult for developing countries without these resources to engage in these activities. However, the commercialization of GM crops does not require those resources. Even poor countries could promote the commercial production of GM crops. For technological innovation, only rich nations adopted early, and only a few developing nations adopted later, which lead to the difference between development and production.

Another factor is the regulation of the release of GMOs, which varies widely by country. Stringent regulations sometimes restrict the application of the technology. Although Germany, Switzerland, Belgium, and the UK are very influential in GM crop technology R&D, the cultivation of GM crops is limited in these countries. The regulatory differences will lead to different adoption times. For example, in 2015, the Ministry of the Environment and Natural Resources of Vietnam issued a bio-safety certificate for Monsanto’s MON 89094 corn variety, enabling farmers to cultivate the crop commercially. It was still banned in China and Europe (“Vietnam approves commercial crops of GMO corn to cut imports,” 2015; “MON89034 | GM Approval Database- ISAAA.org,” n.d.). Regulations contribute to the divergence of the technology development and commercialization patterns.

Other factors that influence the decisions of policy makers about release regulations could cause the inconsistency between the two patterns. Negative public perceptions of GMOs influence the consumption of GM food, which results in restriction or the decrease in GM crop plantings and import. Even though GM crop technology is highly developed in Europe, China and Japan, consumers’ negative attitudes lead to stringent regulation of GMO release, which increases the gap between the two patterns. Factors like global prices, reserve stocks, profitability, and environmental stress will affect regulations relating to the release of GMOs, but will not affect investment in the R&D of the technology in the developed world.

Private businesses are more influential in the global innovation of GM crop technology

There were nine types of organizations involved in the organization citation network. They are government agencies, public research institutes, public and private universities, private businesses, third sector, public private partnerships (PPP), and semi-state and government-owned companies. The proportion of private business was greater than the other social entities in the organization citation network, and it retained this position from 1990 to 2015.

The high in-degree and out-degree of developed countries stem from the accomplishments of private businesses, especially large firms. In developing countries like China, South Africa, and Brazil, public research institutes and public universities contributed to R&D of GM crop technology. Private business, public universities, and public research institutes are the main actors in innovation activity, but other social entities are joining them. It will be a more diverse environment for the technological innovation in the future.

Although private business performed much better than other organizational types, the contributions of private business’s technological innovation were mostly from a few large firms like Novartis, Bayer, Monsanto, BASF, Dow, Syngenta, BioTechnica International and Dupont. The distribution of in-degree and out-degree of the citation network follows a power law with a few large firms located at the network’s center. Moreover, they also entered the citation network and started GM R&D activity earlier. Also, in the early networks, we did not see many small businesses or start-ups. Large established firms have an abundance of capital and specialized talents (Rogers 1995, p. 379), which facilitates seminal patents resulting in high weighted in-degree.

Future study

There are a few limitations to consider in interpreting the results. The landscape of GM crop technology diffusion and distribution may not be fully shown. First, applying for patents, especially international patents, can be complicated and expensive, and not all organizations are capable of patenting, especially non-private organizations and small businesses. Second, only patents are measured while scientific publications are not included in this paper. Usually research from private businesses is more practical and will result in patents, while the research from public research institutes and universities is more exploratory and mainly results in publications rather than patents (Aghion et al. 2010). The performance of the public sector may be underestimated. In the future, we will examine scientific publications to better describe the performance of public universities and research institutes.

Future research may also explore the factors that influence the formation of the GM crop technology citation network at the country and organizational level. For example, we want to explore whether the technology diffusion is geographically localized within countries and organizations. Does technological diffusion have a geographic boundary? Does international trade of GM crops correlate with the diffusion of technology between countries (Xanat et al. 2018)? Is the technology more likely to diffuse between two countries that share similar GMO release regulations? Are there any preferences for the countries or organizations to cultivate GM crops with specific traits (Parisi et al. 2016)? We have studied the technology diffusion and distribution network in this paper, and will explore the influence on the formation of the network structure in the future to help us better understand the diffusion process.

Appendix: Methodology

For this research, we studied the innovation diffusion at the organizational and country levels, so organization and country citation networks are constructed (Table 3). Since all the networks are constructed based on the citation relationship, all the networks are fully connected and there are no isolated nodes in the networks.

Table 3 Steps

Step 1: coding of the organizations

We took the mergers and reorganizations of the companies into consideration. The patents belonging to which company were coded based on unique 4-letter code assigned by Derwent (https://images.webofknowledge.com//WOKRS529JR13/help/DII/hs_assignee.html). If the companies share the same 4-letter code, then we coded the companies as one organization (see Table 4).

Table 4 Coding of organizations

If company X is merged or demerged, then there are two categories of codes for company X like X (AAA-C) and X (BBB-C), then owner of the patents will be coded according to the standard codes of company (see Table 5).

Table 5 Coding of organizations

As we are also interested in the how the innovation of GM technology flows between different entities, so all the organizations were coded and assigned to different types based on the information provided on organizations’ official websites. Based on the owners, basic objective and funding, the organizations are classified into 9 types: public sector (including government agency, public research institute and public university), private sector (private business), private university, third sector, public private partnership (PPP), semi-state and government-owned companies, as is shown in Table 6.

Table 6 Types of organizations

Step 2: coding of the countries

In country citation networks, the nodes are the countries, which refer to the patent applicants’ countries of affiliations and the edges the direct citation relationships between the countries (see Table 7).

Table 7 Coding of countries

Step 3: construct the organization citation network

The first step in the construction of the citation network was to clean the patent data. The full record for a patent structured metadata is shown in Table 8. For a single patent record, only patent number(s), patent assignee name(s) and code(s) (organizations filing the patents) and cited patent numbers were extracted from structured meta-data of patent documents (see example in Fig. 14).

Table 8 The full record for a patent structured metadata

Fig. 14

Example for a patent record. For the set of assignees, only organizations are considered and extracted

Unit of analysis

The networks were constructed for each year. Suppose that time t, containing patents from 1984 to t, is the year the first citation network formed, then the second citation network contains patents from 1984 to t + 1, so the citation networks to be studied include patents: 1984 ~ t, 1984 ~ t + 1, 1984 ~ t + 2, …1984–2015.

Construction of the organization citation networks

$P = \left\{ {p_{1} ,p_{2} , \ldots p_{n} } \right\}$, n = 5616. $\forall {\text{i}} \in \left\{ {1, \ldots n} \right\}$. We extracted the patent numbers, organization names and codes, and patents cited by inventors/examiners. So $P = \left\{ {p_{1} ,p_{2} , \ldots p_{n} } \right\}$, $\forall {\text{i}} \in \left\{ {1, \ldots n} \right\}$, $p_{i}$ has 3 data sets as follows:

$$\left\{ {\begin{array}{*{20}c} {A_{i} = \left\{ {a_{1}^{i} ,a_{2}^{i} , \ldots a_{{r_{i} }}^{i} } \right\}} \\ {B_{i} = \left\{ {b_{1}^{i} ,b_{2}^{i} , \ldots b_{{s_{i} }}^{i} } \right\}} \\ {C_{i} = \left\{ {c_{1}^{i} ,c_{2}^{i} , \ldots c_{{t_{i} }}^{i} } \right\}} \\ \end{array} } \right.$$

$A_{i}$ represents patent numbers, $B_{i}$ represents organizations (individual-excluded assignees, also referred to assignee), and $C_{i}$ represents cited patent numbers. If the intersection between data sets $A_{i}$ and $C_{j}$ is not an empty set, then patent $P_{i}$ is cited by patent $P_{j}$. An arbitrary element (organization) in $B_{i}$ has a citation relationship with arbitrary element (organization) in $B_{j}$. The weighted edges are the number of patent citations between the organizations. The first organization citation network formed in 1990, and a total of 26 weighted organization citation networks were constructed (Fig. 3).

Step 4: construct the country citation network

Based on the organization citation networks, country citation networks were constructed. First, we identified the organization’s country, based on the contacted address of the organizations. Then, the MapReduce Paradigm was applied to construct the country citation networks.

Figure 4 illustrates the procedure for constructing the country citation network. First, downloaded patent records are input. As there may be several organizations included in a patent, splitting was conducted, followed by mapping and shuffling. At reducing phase, we have listed all the key-value pairs. The final step is the summary of the reducing results.

Step 5: construct the two-mode network

Two-mode networks focus on two sets of actors, or one set of actors and one set of events. Relations in a two-mode network measure tie between the actors in one set and actors in a second, which means ties existing only between nodes belonging to different sets (Wasserman and Faust 1994, p. 39).

To answer the question “How do different organizations perform in different countries?” two-mode networks were constructed. There are two types of nodes in the network: country and types of organizations. For these networks, there are only connections between countries and social entities. There are no connections among countries or social entities.