The impact of public and private expenditures
on technological innovation in the ICT
industry: A cross-national study.
Master Thesis of International Economics and Business
Faculty of Economics
Rijksuniversiteit Groningen
Landleven 5, Groningen
9747 AD, The Netherlands
Tel: +31 (0)503637098
15
thof August, 2007
The impact of public and private expenditures on technological
innovation in the ICT industry: A cross-national study.
Abstract
The impact of public and private expenditures on technological
innovation in the ICT industry: A cross-national study.
Introduction
Throughout the economic history of the past 100 years, most nations have grown to perceive technological progress as a major driver of economic growth. Technological innovation and the factors which lead to it have therefore increasingly become the central organizing principles for policy makers and firm managers. This tendency is reflected in industrial policies, supporting institutions, as well as in current literature about Science and Technology (S&T). The work of Steil et al. (2002), for example, illustrates that improvements in information technology account for more than one-third of US GDP growth during the last few years, as well as almost two-thirds of corporate capital investment in that same period.
There are two main reasons why technological innovation is believed to stimulate long lasting economic growth. First, new technological processes allow firms to increase the output per unit of capital or the output per worker; and secondly, new products contribute to improving the well-being of consumers (Guellec and Van Pottelsberghe, 2004).
As the forthcoming sections will demonstrate, attempting to identify all the sources and actors of innovation is a very complex task. The objective of this paper is therefore not to cover all interactions which explain innovation but to narrow down the input-output relationship of the innovation process to several crucial and measurable factors, with an important focus on government support.
Third and finally, “Learning-by-Interacting” is the positive outcome of interactions with other organizations, in the form, for example, of user-producer contacts. As Lundvall (1988) explains, in the case of certain complex innovation processes, firms may not be able to develop all the required knowledge and skills in-house. They sometimes rely on knowledge “spillovers”. A knowledge spillover is a positive externality which implies a transfer of knowledge from one actor to another through an interaction.
When exposed to the knowledge and expertise of other organizations, firms must of course be able to absorb information efficiently. A firm’s capacity to value, assimilate and apply new knowledge is referred to as its “Absorptive Capacity” (Cohen and Levinthal, 1990). Firms strongly rely on this skill in order to innovate. Absorptive capacity depends essentially on prior related knowledge. This means that firms develop their absorptive capacities largely through the accumulation of related knowledge that permits them to evaluate and exploit subsequent developments within a field. Other determinants are the establishment of close relationships with extramural knowledge sources (such as suppliers, buyers or universities) which create and strengthen information channels. Internal R&D activities as well as learning through experience and interactions will therefore determine a firm’s capacity to innovate.
As shall further be explained in part 6 of the literature review, governments play a leading role in stimulating these sources of knowledge. That is why, since R&D, Education and Telecommunication network expanding policies are very high on national government’s agendas, the main purpose of this paper is to test whether public expenditures in these areas can explain changes in innovation output in the ICT industry for a sample of 16 OECD countries over the period of 1985-2002.
The choice of a sample of 16 OECD countries has certain implications for this study. Firstly, as defined by the OECD convention, in order to become an OECD member, each country must satisfy numerous characteristics such as a strong orientation toward economic growth as well as research in scientific and technological fields. Since member countries are likely to be relatively more advanced from a technological point of view than non-members, a sample of OECD countries is certainly not representative for all countries. The knowledge base, the level of tertiary education and the quality of the telecommunication infrastructures are often relatively more developed in OECD countries than in many non-OECD countries. Government policies and expenditures will thus surely have different objectives and effects in OECD countries than in non-member countries.
In the literature review (Section 1), the first step will be to draw attention to the importance of innovation within current economic research. This will be followed by a description of the multiple inputs and outputs of the innovation process and an explanation of Lundvall and Nelson’s contribution to the National System of Innovation (NSI) theory. Thirdly, with the help of the NSI theory, we will highlight the numerous interactions between the different actors that take place in the innovation mechanism. The fourth step will explain the connection between innovation and economic growth according to the theory of Romer. This will clarify why technological progress is considered to be crucial to firms, industries as well as national governments and their policy makers. The fifth step will describe how industry innovativeness can be measured. The last part of the literature review will focus on the active role of national governments with regards to R&D, Education and Telecommunication infrastructure expenditures. Examples of previous studies will show the objectives of these public and private expenditures and demonstrate their influence on labourproductivity.
Section 1: Literature Review
1. Introduction to the concept of “ innovation”.
“If the firm or the country were to focus all efforts on allocating existing resources in a better way, and if every single unit kept producing the same product with the same techniques, it would not only stagnate. It would gradually become increasingly poor because its products would become less and less in demand. Therefore, when the focus is on economic development, successful innovation is more important than efficient allocation.” (Lundvall, 1998)
An innovation is the introduction of something new and useful or a significant improvement of an already existing product, process or service, destined to create value or increase productivity. Innovations can either be technological1 or non-technological. New and improved marketing techniques, as well as organizational innovations and services which apply technological innovations, are all examples of non-technological innovations.
While the ICT sector can be subdivided into manufacturing and service categories, innovation theory has nevertheless focused essentially on the analysis of technological innovation in manufacturing activities. Due to the fact that services are not usually producers of new or technically improved tangible artefacts, analysts have had difficulty applying the received understanding of innovation to services. In fact, most of the literature on innovation is concerned with services as adopters and users of new technologies rather than as creative innovators in their own right. These innovations will often determine the success of technological innovations. Due to this “fuzzy” nature of service outputs in ICT, as explained by Gallouj and Weinstein (1997) and to the importance of technological innovation (manufacturing) in theory, this paper will only concentrate on the latter.
The creation of knowledge or the invention of new products are not normally considered innovations until they have been truly incorporated into the enterprise’s activities. This illustrates that an innovative activity is something which doesn’t occur separately from a firm’s core activities. It implies the coordination of various inventive, implementation and learning skills (Rogers, 1998).
1 “Technological product and process (TPP) innovations comprise implemented technologically new products
While innovations in business enterprises may take place in numerous ways, most attention is usually given to R&D. Unfortunately, research projects do not always lead to tangible innovative products and investing in such activities can therefore be risky. Prior experience, an R&D team’s multidisciplinary character, timing, dedication and firm-innovation effort compatibility are all crucial factors, but a good R&D strategy does unfortunately not automatically lead to success. Serendipity and luck also play an important role. Teece (1996) explains the three types of uncertainty of R&D activities. First, primary uncertainties arise from “random acts of nature and unpredictable changes in concurrent preferences.” Secondary uncertainties arise from “lack of communication, that is, from one decision-maker having no way of finding out the concurrent decisions and plans made by others.” This can be affected by changing the boundaries of the organization (Koopmans, 1957). A third kind of uncertainty, called behavioural uncertainty (Williamson, 1975), is attributable to opportunism.
These uncertainties can have an impact on the outcome of investments in R&D. This means that some investments will lead to success and some will not. This study, however, will only focus on whether the amounts of public R&D expenditures over the years have had a positive impact on innovation output. The reason for this limitation is that uncertainty is very difficult to detect and include in a statistical model. Despite the uncertainty of their outcome, investments in R&D are obviously driven by the hope that they might lead to technological improvements. The use or rearrangement of the resulting products may then in turn lead to further knowledge creation.
2. Inputs and outputs of the innovation process
Various theories have been elaborated in order to encompass the multiple players and interactions which lead to innovation on a national scale. The most influential and frequently used concept is the National System of Innovation (NSI) theory. A generally accepted definition of the NSI would be the following, by Steil et al.(2002):
“An NSI is the cluster of institutions, policies, and practices that determine a nation’s capacity to generate and apply innovations.”
Lundvall (1992), on the other hand, takes a more theoretical perspective and focuses especially on the widely held belief that economic growth is increasingly driven by the ability to learn. Since the economic value of knowledge quickly becomes obsolete, he suggests that an increasingly proactive stance toward learning is needed.
The concepts of “learning” (by doing and interacting) are indeed critical to technological progress. A firm’s capacity to learn to do new things, to handle changes in its environment as well as to get access to information will determine its absorptive capacity. According to Lundvall (1998), specific information and knowledge that agents have at a certain point in time may be less important than their “learning capability”. For instance, since the capability to learn by doing improves with time and research experience, companies are usually better off investing in their own internal R&D activities, instead of simply purchasing another firm or laboratory’s results.
As a result of the authors’ very different approach, both their books may be used complementarily to understand the NSI. There are several reasons why Lundvall has nevertheless contributed more to the NSI theory than any other researcher. First of all, his work explains that the most significant parts of the knowledge base are tacit2 and emanate from routine, learning by doing, using and interacting and not just from search activities related to S&T. That is why the prevailing institutional structure is at the core of the NSI theory. Secondly, Lundvall´s approach is also more ambitious because the importance he attributes to certain policy implications is more far reaching. For instance, unlike many authors, he concentrates on policies of human resource development and on the institutional set up of markets for labour and finance.
In order to justify why the NSI theory concentrates on the nation state level (N from NSI), the two authors insist that industrial development in a country is greatly influenced by domestic resource endowments, legislation and culture, amongst other things. Nobody will deny that constantly improving technological infrastructures facilitate the flow of knowledge and information between spatially distant locations. Cairncross (1997) writes, for instance, that distance and much that is associated with it, such as geographic borders and time zones, will
2 Tacit knowledge is deeply rooted in an individual’s action and experience, it is personal intuitive knowledge. It
diminish in importance as the world becomes increasingly interconnected via our rapidly expanding communications systems.
According to Lundvall, however, localized user-producer learning processes, forming the building blocks of NSIs, are not menaced of being supplanted by the internationalization of economic activities. The reason for this is that skills are primarily accumulated in the work force and that labour is still the least mobile production factor, despite the growing tendency of students and graduates to cross borders. This explanation justifies the frequent interest to study innovation at a cross-national level.
As shown by Nelson, the case of Japan perfectly illustrates how cultural and historical circumstances will have an impact on innovative efforts. After World War 2, Japan was left devastated and thegovernment allocated its resources selectively and wisely to encourage the import of foreign technology and its domestic improvement. Due to import restrictions, the only way for foreign firms to exploit their technological superiority in Japan was to sell their technology to domestic firms. Empirical data shows that companies who imported technology also invested in R&D themselves. Domestic R&D was essential to enable firms to evaluate, adapt, and improve imported technology. In the automotive sector, for instance, Japanese firms initially acquired foreign technology through reverse engineering.
3. Who are the actors in the innovation process and how do they interact?
The complex nature of tacit knowledge as well as the great number of connections between the institutional pillars, make it particularly difficult to capture the impact of all interactions on innovation output in an industry. The reason is that the quality and outcome of these interactions, the resulting spillovers as well as learning depend on network-specific parameters such as the number of firms interacting, the frequency of these interactions, other institutions which are part of the network, nation-specific legal systems, etc.
This paper therefore focuses on three types of financial expenditures in order to capture their explanatory value to innovativeness. We do not observe the effect of these public expenditures on organizations nor their networks but directly isolate the connection between public financial input and industry innovation output. Since learning by doing and interacting determine a firm’s absorptive capacity through experience and interactions, we shall assume that public expenditures on tertiary education and telecommunication infrastructures are the best available indicators of AC. Part 6 of this literature review will explain the active role of the government in these areas. While it is not possible to include all NSI aspects in this paper’s statistical model, it is valuable however to look at governmental contributions to innovation and to test whether current yearly public expenditures have a positive and important impact on innovation output for a sample of countries over the years. This could help us weigh the effectiveness of these expenditures.
4. Innovation and economic growth
During the last few decades, a lot of attention has been given to the importance of technological innovation as the main driver to economic growth, which can be interpreted as the increase in value of the goods and services produced by an economy. Economic growth is often perceived as an indicator of the average standard of living and many researchers believe it to be enhanced by technological progress (Grossman and Helpman,1994).
Years later in fact, Romer (1990) describes economic growth as something which occurs whenever individuals take resources and rearrange them in ways that are more valuable. As a radical improvement to Solow’s legacy, his model of “Endogenous Technological Change” relies on the virtuous cycle of both the creation of new technologies and human capital. Firms and individuals have an incentive to invent in order to exploit an advantage over their competitors, thereby improving their own productivity (learning-by-doing). A part of this knowledge associated with the innovation "spills over" to other economic actors, which increases those actors' ability to innovate (learning-by-interacting). Knowledge creation is hence considered to be an interactive procedure during which learning by doing as well as by interacting are of utmost importance. The virtuous cycle arises through this mechanism.
5. How to measure innovativeness?
Innovative or scientific output can be measured in several manners. Among the different measures we find the number of patent applications or patents granted per industry; the number of published articles in scientific journals per university or industry, etc. (Los and Verspagen, 2002). Patents are unique in both the extent of the detail that they contain and in the breadth of their coverage. A patent document provides information on the characteristics of the underlying innovation such as its technological area or its citations to related innovations and its inventor, all of which are unavailable elsewhere. Such data is obtainable for all firms and individuals over a long period, which makes them well suited to studies of the efficacy of policies, etc (Lanjouw et al., 1998). R&D expenditure data has been the most commonly used alternative indicator of innovativeness but it is clearly much more related to inputs of the innovative process than to successful outputs.
In order to allow cross-country comparability, the figures of this paper’s dataset concern only patent applications to the EPO (since disclosure rules and filing fees usually vary across legal systems). Patent applications of OECD-based ICT firms in the US or elsewhere have therefore not been included in this study. While patent application figures are certainly not a flawless measure of innovation output, they are nevertheless the best available for this paper. The methodology part (Section 3) will present more details regarding the dependent and explanatory variables.
6. R&D, Education and Telecommunication infrastructure funding: The active role of policy makers in the innovation process.
“The policies and programs of national boundaries… define an inside and an outside that greatly determines how technology diffusion proceeds” (Edquist, 1997)
Technological advance can not occur without certain intentional investments by profit seeking firms or entrepreneurs and national governments. Stiglitz et al. (2000) estimate that in the case of the US, social gains from innovation exceed private returns by between 35 and 60 percent. Given this differential, private markets will have the tendency to underprovide basic research. In an example concerning the development of the Personal Computer by IBM, Verspagen (2005) points out that technology may often present certain public good characteristics. For instance, once the knowledge to produce a PC had been generated by the IBM engineers, it could be used as many times as necessary by other firms (characteristic of non-rivalry). Also, IBM had no way to exclude other firms to use the knowledge (non-excludability). The fact that IBM’s invention has provided a lot of value to many people for which it hasn’t been paid for illustrates the possibility of social gains being higher than the private gains faced by the firm itself. Such inefficient situations call for a government intervention.
a) R&D and government support.
In the work of Navaretti and Venables (2004), we find two types of R&D policy directions that can be followed depending on the country’s size, its initial technological capabilities, absorptive capacity, etc. First, the governments of technology leading countries often choose to provide funds directly to local institutions such as universities and laboratories. By doing this, the nation’s leading technological position can be maintained and expertise may be internalized in domestic firms and organizations. Secondly, certain policy makers might instead try to attract foreign multinational R&D plants to their country, by offering subsidies or tax cuts. The aim of this strategy is to allow domestic firms to benefit from foreign expertise through knowledge spillovers and interactions, in the form of user-supplier relationships. This could, for instance, benefit less technology-advanced nations.
David (1993) has made a distinction between three mechanisms to stimulate domestic R&D: Patronage, Procurement and Patents (PPP). Patronage refers to the situation in which the government finances research without specific outcome expectancies. In the case of numerous universities or public research laboratories, knowledge that is generated is not immediately useful, but benefits are expected in the long run. Firms are likely to find this type of research too risky to invest in. Secondly, Procurement refers to specific pieces of research which are contracted out. In this case, the government has in mind a specific piece of knowledge or product which it wants to buy. It therefore establishes a contract with a private firm that is able to supply this knowledge from its own R&D activities. Thirdly, the Patent system is aimed more at the consumer market or the market for capital goods. It involves the market to a much larger extent than the two other methods.
b) Tertiary Education3 and government support.
“The capacity for countries to adopt, disseminate, and maximize rapid technological advances is dependent on efficient systems of tertiary education” (Saint, 2006)
Even extremely ambitious and productive R&D activities, as well as frequent exposure to valuable information, may not be beneficial, unless there are enough competencies in other parts of the organization to exploit the knowledge produced, and to enable communication and interaction between R&D, marketing, production and administration divisions. Innovation therefore requires much more than just corporate investments in R&D. The introduction of advanced technologies in an organization can, of course, only take place successfully when it is accompanied by an adequately skilled workforce and competence-building among employees.
Past research has shown that firms rely on their absorptive capacity to innovate. Since the presence of a higher educated workforce will enhance a firm’s ability to handle new information, knowledge as well as technology, education may be considered an indicator of absorptive capacity.
The rate at which each country’s education system produces advanced knowledge is shown by tertiary graduation rates (Blöndal et al. 2002). Countries with high graduation rates at the tertiary level are also the ones most likely to be developing or maintaining a highly skilled labour force. Specific skills and knowledge in science are of a particular interest as they increasingly represent a principal source of innovation and growth in knowledge-based economies. Tertiary education funding is thus necessary to innovation.
Furthermore, having previously explained the great importance of learning for organizations as well as the knowledge transfers between innovation actors, we must also highlight that graduates regularly carry knowledge from universities to business firms. In fact, academic publications and conferences assist industry personnel to monitor and exploit new knowledge produced by universities (Harman and Harman, 2004). In a study which encompasses a large sample of Finnish firms and industries, Leiponen (2005) writes that firms which have employees with technical or even research skills may have a greater ability to innovate.
3 Tertiary education broadly refers to all post-secondary education, including but not limited to universities.
The World Bank also emphasises the changes in modes of organization, the rise of market forces and above all, the need for a more holistic system of education. They write that tertiary education is necessary for the creation, dissemination and application of knowledge and that governments have a great responsibility to encourage innovation in their country’s tertiary education institutions. Public policies must therefore aim at promoting access to higher education, pursuing excellence in the work of students and in the scholarship and research of academic staff and finally, provide an environment to enable universities to fulfil their purposes (Kemp, 2000). These examples show the important role of the government, as well as public policies destined to improve tertiary education.
c) Telecommunication infrastructure and government support.
Egan and Wildman (1994) write that electronic applications which are intended to enhance and improve the flow of information between or within organizations are advantageous to labour productivity. Since telecommunication applications, such as broadband access, are a tool for rapid information sharing which facilitates learning-by-doing and interacting, we may assume that an efficient telecommunication infrastructure may hence be considered as an indicator of absorptive capacity. Investments in telecommunication infrastructures are therefore likely to induce increases in labour productivity.
For a sample of 30 countries, Dutta (2001) finds empirical evidence which confirms that improvements in telecommunication infrastructure lead to increases in economic activity in both industrialized and developing countries. Most importantly, he suggests that market mechanisms for expanding telecommunication infrastructures must be augmented by specific public incentives to guide telecommunications growth within a country in targeted ways. He especially insists on the frequent unevenness of the distribution of telecommunication infrastructure between rural and urban areas, within a same country. This can also be seen in industrialized countries such as Ireland, Germany and Spain. This spatial irregularity in infrastructure correlates well with spatial unevenness in economic activity levels, suggesting that differences in telecom availability are likely to influence firm location.
The working paper clearly states that before focusing on subsidies for the provision of internet services, subsidies should concentrate initially on supporting basic infrastructures necessary for internet access. Amongst further measures, we find: setting up preferential licensing schemes for rural providers or allowing operators flexibility in pricing, etc. These previous studies hence indicate that governments have an important responsibility regarding the enhancement of telecommunication infrastructures.
Section 2: Research Proposal
So far, we have illustrated the importance of innovation as a main catalyst to economic growth as well as the numerous inputs and the active role of the government in the innovation process. Innovation enhancing policies and expenditures are high on government agendas. That is why it seems interesting to investigate whether or not public expenditures can really explain changes in innovation output. The objective of this paper is thus to test whether annual expenditures on R&D, education and telecommunication infrastructure have an important positive impact on innovation output in the ICT sector, for a sample of 16 countries and over the period of 1985-2002.
General Research Question:
“Can public and private R&D, Education and Telecommunication infrastructure expenditures explain changes in innovation output of the ICT industry in the sample of 16 OECD countries over the period of 1985-2002?”
Hypotheses
From this, we can deduce that business R&D as an input to innovation has thus become increasingly important. Secondly, common sense tells us that only specific categories of R&D expenditures will have an impact on the innovativeness of specific industries. For example, R&D investments in automotive assembly procedures are unlikely to be positively related to the innovativeness of the ICT industry. It is crucial to therefore focus only on what is relevant to this particular study.
The OECD (1998) has classified the ICT manufacturing sector into 6 subcategories (appendix 1). Due to missing data, it is not possible to establish aggregate figures encompassing all yearly R&D expenditures in ICT manufacturing. When simultaneously including R&D expenditures in subcategories 3000 and 3220/32304 as two separate explanatory variables during initial statistical tests, both variables were non-significant with a high Adjusted R-square, suggesting a possible problem of multicollinearity. It was therefore not possible to include both of them into this study. That is why, with a nearly complete database of R&D expenditures regrouping subcategories 3220 and 3230 into aggregate figures (SourceOECD), the first hypothesis is the following:
H1: Public (and private) expenditures on Radio, T.V. and Communication Equipment R&D in the business sector, have an important positive impact on the innovativeness of the ICT industry for the selected sample of 16 OECD countries.
It has become clear from the literature review that a properly educated workforce is likely to improve a firm’s absorptive capacity which, in turn, is crucial to the innovation process. In the OECD’s 2005-2006 “Work on Education Paper”, Tertiary Education is given a prominent role as one of the main drivers to technological growth.
Given this substantial role of education in past technological innovation literature, especially with regards to technology-driven industries, the second hypothesis of this study is the following:
H2: Public expenditures on Tertiary Education have an important positive impact on the innovativeness of the ICT industry for the selected sample of 16 OECD countries.
Telecommunication infrastructures improve access to information and facilitate knowledge sharing within and between organizations. A good telecom infrastructure is likely to enhance the absorptive capacity of its users. The success of the numerous vertical and horizontal flows of information between innovation actors therefore depends on these infrastructures. This is particularly important for firms in high tech manufacturing or service industries such as ICT. Given the importance of information transfers in innovation literature and research, the third hypothesis concentrates on public expenditures:
H3: Public expenditures on telecommunication infrastructure have an important positive impact on the innovativeness of the ICT industry for the selected sample of 16 OECD countries.
Next, section 3 will consist of all the statistical steps which are necessary to test these three hypotheses. This research methodology part will begin with a description of the dataset, an explanation of the dependent and explanatory variables, as well as all the required adjustments in order for these figures to be applicable in this study. Next, all procedures designed to obtain the most accurate possible statistical model will be thoroughly examined in a chronological manner.
Section 3: Research Methodology
After introducing the three hypotheses, the precise relation that will be examined in this methodology section is the following:
PAT = f (RDCOM, EDUGOV, PTO). PAT: ICT patent applications
RDCOM: Public + Private expenditures on ICT-related R&D EDUGOV: Public expenditures on Tertiary Education
1. Data:
The dataset consists of one dependent and three independent variables for a sample of 16 OECD countries, over the period of 1985-2002 and all of which were collected from the SourceOECD database. Such a dataset, which is composed of both time-series and cross-section data, is called a panel dataset. Despite nearly complete SourceOECD tables, 19 figures of the collected data are missing (3 for R&D expenditures and 16 for Tertiary Education expenditures). This means that the dataset is unbalanced and, as shall be explained further, this will have consequences for the forthcoming statistical tests.
Prior to any statistical work, an industry-size adjustment is necessary in order to make the data of all countries comparable and functional within a same study. It is logical that a country such as the US has a more important ICT industry than Ireland, simply due to the country’s size and the number of workers in ICT. For instance, while the number of yearly patent applications (in absolute figures) will automatically be larger for the US, this does not immediately imply that the US is a better innovator in ICT. All figures must thus be corrected in order to obtain innovation output per hour worked in ICT. As shall be explained further, by doing this, the problem of heteroskedasticity will be removed considerably.
The 60-industry database5 provides data in accordance with ISIC Rev.3 on the total hours
worked in: Office Machinery, Insulated Wire, Other Electrical Machinery and Apparatus, Electronic Valves and Tubes, Telecommunication Equipment, Radio and Television Receivers, Scientific Instruments, Other (related) Instruments. To obtain an approximation of industry size, all these values must be added up per year and per country, as this gives us aggregate figures for the Total number of Hours Worked in ICT. The dataset may then be adjusted by dividing all figures (both dependent and explanatory variables) by the respective amount of hours worked in ICT. By doing this, all figures will be available per hour worked in ICT.
With regards to the explanatory variable of R&D expenditures, this will present a certain limitation. By dividing all figures of R&D expenditures (which only capture a part of the ICT manufacturing sector) by the total amount of hours worked in all sectors of ICT manufacturing, we obtain amounts of R&D expenditures per hour worked which are inferior to what they actually should be. This might impact the relation between R&D expenditures and patent applications but it is unfortunately a limitation which can not be avoided in this paper.
• Dependent variable:
PAT (Nr.): Number of ICT patent applications per country to the European Patent Office
(EPO) at the priority date / Approximation of the Number of hours worked in the ICT industry (1985-2002).
As shown previously in the part concerning innovation output, patent data is often used as a measure of innovativeness. The SourceOECD database of Main S&T indicators (2006) contains complete country tables of patent applications in ICT manufacturing to the EPO. The priority date implies the date of the international filing of a patent, which is the closest available to the date of invention. Patent grant figures, however, are only available for the US Patent Office (USPTO).
The choice to use patent applications in this paper is justified, since according to the OECD, the number of patent applications per country is as representative for industry innovativeness as the actual number of patent grants. Applying for an “internationally recognized” patent only becomes worthwhile for organizations anticipating returns in excess to the substantial costs. Firms are thus unlikely to go through a costly patent application procedure unless they expect considerable financial returns (Furman et al. 2002). That is why this study uses the number of yearly patent applications per country in ICT as a measure of innovativeness.
• Independent variables:
RDCOM (constant US$): Radio, T.V. and Telecommunication R&D Expenditures /
Approximation of the Number of hours worked in the ICT industry (1985-2002).
Since these figures are initially measured in current US$ values, it is necessary to convert them into constant US$ by using the GDP deflator which is also provided by the SourceOECD database.
Unlike the price index, the GDP deflator is not based on a fixed basket of goods and services. The GDP deflator of a specific year is based on the set of all goods that were produced domestically, weighted by the market value of the total consumption of each good. Therefore, consumption patterns or the introduction of new goods and services are allowed to show up in the deflator as people respond to changing prices. The advantage of this approach is that the GDP deflator measures changes in both prices and consumption. For this reason, the GDP deflator is in most ways a more accurate of pure price changes in the overall economy. This also makes the GDP deflator less than desirable from a political and policy standpoint, as it cannot be manipulated in any way to reflect subjective preferences regarding what goods are most important to measure when figuring an inflation rate.
EDUGOV (constant US$): Government Expenditures on Tertiary Education / Approximation
of the Number of hours worked in the ICT industry (1985-2002).
As mentioned previously, tertiary education is a particularly relevant input to technological progress and innovative industries. Since the appropriate education expenditure figures are initially measured in national currencies at current values, it is necessary to first convert these values into current US$. As explained previously, constant US$ may then be obtained by applying the GDP deflator.
PTO (constant US$): Public Telecommunication Operator Investment in infrastructure /
Approximation of the Number of hours worked in the ICT industry (1985-2002).
2. Methodology:
The methodology part of this study consists of a panel data analysis for 16 OECD countries6 and the period of 1985-2002. The combination of cross-sections with time-series data will enhance the quality of the model in a way that would be impossible when using only one of these two dimensions. Panel studies can provide a rich and powerful study of a set of countries. The economic model of the number of patent applications for the ith country in the tth time period may be expressed as:
PATit = f (RDCOMit, EDUGOVit, PTOit).
In their work, Hill et al. (2001) clearly describe the following very flexible general regression model. As will be shown further on, certain simplifying assumptions are necessary to actually apply it:
yit = β1it+ β2itx2it + β3itx3it + β4itx4it + eit
yit = PATit
x2it = RDCOMit
x3it = EDUGOVit
x4it = PTOit
eit = error term with two dimensions, one for the country and one for the time period.
Step1: The choice between OLS, FE and RE
The initial step to take when working with a pooled dataset is to choose the best statistical model at our disposition. We must therefore decide between the Ordinary Least Squares
model (OLS), the Fixed Effects model (FE) and the Random Effects model (RE). These three
statistical models differ mainly in their assumptions regarding the intercept and error terms. It is assumed that the error termeit may be decomposed into two independent elements:
eit = ui + vit
where ui is time-invariant and accounts for any unobservable country specific effects not
captured by explanatory variables. The term vit represents the remaining disturbance, and
varies over cross-sections and time.
6 Australia, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, The Netherlands,
First, the OLS model, also referred to as the “pooled regression model”, consists of having the same coefficients over all selected countries as well as years. In this model, the ui’s are
assumed to take the same value for all cross-sections. The OLS is suitable in the event that we are confronted with neither significant country nor significant temporal effects. It is most likely to be inappropriate for such a dataset, since either one of these effects are usually present. Secondly, Hill et al. (2001) explain that both the FE and RE models assume that only the intercept parameter (β1i) varies, not the response parameters; and the intercept varies only
across countries and not over time:
β1it = β1i β2it = β2 β3it = β3 β4it = β4
The model for FE and RE therefore becomes: yit = β1i + β2x2it + β3x3it + β4x4it + eit
The constant term β1i is the individual effect or individual heterogeneity for country i.
All behavioural differences between individual countries and over time are captured by this intercept.
In the FE model, ui’s are fixed parameters to be estimated with dummies for cross-sections,
while in the RE model, the ui’s are assumed to be random, independently and identically
distributed. The RE model views the chosen sample of countries as a random sample from a larger population of countries. This means that the intercepts are treated as random drawings from the population distribution of country intercepts. For a study which would, instead, cover the entire population of countries in the world, the FE model would be more appropriate. Choosing the wrong model would of course lead to untrustworthy results and that is why the next step must be to apply the tests which are necessary to pick the most appropriate one.
The key issue in panel estimations is whether β1i is correlated or not with the observed
explanatory variables. If the correlation between β1i and the explanatory variables is 0, that is
Cov (χit, β1i) = 0, t = 1,2,…,T, then the RE procedure is appropriate. Otherwise, if β1i is
In his work, Greene (1997) explains the tests to apply in order to choose between OLS, FE and RE:
a) The Likelihood Ratio (LR) test for the OLS model against the FE model. The null hypothesis in this case is the following:
H0: u1 = u2 = un (OLS is the correct specification) H1: FE model is the correct specification.
Table 1: Likelihood Ratio test:
Redundant Fixed Effects Test Pool: POOL01
Test cross-section fixed effects
Effects Test Statistic d.f. Prob.
Cross-Section F 26.994824 (15,233) 0.0000
Cross-Section Chi-square 270.930914 15 0.0000
Both Chi-square and F are legitimate test statistics to show whether we must either use OLS or FE across both time and countries. The large value of the LR statistic (270,9) allows us to reject the null and favour FE over OLS.
b) The Hausman Specification (HS) test for the RE model against the FE model. The null hypothesis in this case is the following:
H0: RE model is the correct specification. H1: FE model is the correct specification.
Table 2: Hausman Specification tests:
Test Summary StatisticChi-Sq. Chi-Sq. d.f. Prob. Cross-section random 6.80 3 0.0785
Test Summary StatisticChi-Sq. Chi-Sq. d.f. Prob.
Period random 13.71 3 0.0033
After running both LR and HS tests, the conclusion of this first step is that the cross-section Fixed Effects estimation method is the most appropriate, given the dataset that has been compiled.
Step 2: Diagnostic tests: Heteroskedasticity, Autocorrelation and Multicollinearity.
Having previously chosen the Fixed Effects model, it is now important to further test and adjust the model in order to obtain the most appropriate model available.
1. Heteroskedasticity.
The first obstacle to overcome is heteroskedasticity. According to Hill et al. (2001), heteroskedasticity is referred to as the violation of the constant variance assumption that at each level of the various explanatory variables, we are equally unsure about how far the number of ICT patent applications will fall from their mean value. The consequences of heteroskedasticity are the following:
! The least squares estimator is still a linear and unbiased estimator but no longer the best one.
! The standard errors usually computed for the least squares estimator are incorrect. Confidence intervals and hypothesis tests that use these standard errors may be misleading.
The reason is that with such a dataset we deal with economic units of different sizes, or in the case of this particular study, countries with ICT industries of varying sizes. Hill et al. (2001) write that, frequently, the larger the firm, industry or country, the more difficult it becomes to explain the variation of the dependent variable by the set of independent variables. Secondly, heteroskedasticity also often occurs in time series data, due to external shocks or changes in circumstances leading to more uncertainty about the dependent variable. By correcting all variables for industry size at an early stage, the heteroskedasticity problem has thus been mainly removed.
2. Autocorrelation.
We have seen that in the case of heteroskedasticty, the constant variance assumption (error variance being the same for each observation) is not respected. The second important diagnostic test, which also involves the error terms, is the test for autocorrelation, or the correlation between the error terms. Since a panel dataset includes time-series data, there might be a chance for error terms to be correlated with each other. That would imply that in any one period, the current error term would not only contain effects of current shocks but also of previous shocks. We may illustrate this with the following equation:
et = ρet-1 + vt
-1 < ρ < 1
in which the error term et depends on its lagged value (et-1) plus another random component
that is uncorrelated over time. ρ is a parameter that determines the correlation properties of et.
The larger the value of ρ, the greater the carryover from one period to another and the more slowly the shock spreads over time. The consequences of autocorrelation on a model are essentially the same as those of heteroskedasticity: 1) least squares estimator no longer the best + 2) incorrect formulas for the standard errors.
Since both the Durbin Watson statistic (0.287387) in table 3 and the Lagrange Multiplier (LM) test7 did not provide satisfactory results, GLS across time was implemented in order to obtain uncorrelated error terms.
While the GLS regression output8 shows a slightly improved Durbin-Watson statistic of
0.547463, it is still very far from the ideal result. In fact, a second LM9 test confirms that there is little improvement, suggesting that the Error Correction model might be a good alternative in order to curb autocorrelation.
Table 3: Pooled Least Squares with cross-section fixed effects:
Dependent Variable: PAT Method: Pooled Least Squares Sample: 1985 2002
Included observations: 18 Cross-sections included: 16
Total pool (unbalanced) observations: 269
Variable Coefficient Std. Error t-Statistic Prob. RDCOM 3.26E-07 3.07E-08 10.64008 0.0000 EDUGOV 2.86E-08 1.18E-08 2.434988 0.0156 PTO 1.17E-07 1.42E-08 8.242533 0.0000 C -9.32E-07 1.48E-07 -6.280530 0.0000 Fixed Effects (Cross)
Effects Specification Cross-section fixed (dummy variables)
Adjusted R-squared 0.999684 Durbin-Watson stat 0.287387
The table above illustrates the pool of dependent and three explanatory variables (total of 269 observations) for the period of 1985-2002. At this stage, only the measures of the table marked in bold letters will be looked at. The Adjusted R-squared statistic indicates the goodness of fit of the model. It is a correction of R-squared for the number of independent variables in the model. The current Adjusted R-squared of 0.999684 illustrates that expenditures on RDCOM, EDUGOV and PTO possibly explain almost all variability in PAT.
variable is caused by a change in another one. While further tests are still required to establish conclusive results, we may however already identify a positively significant impact of the three chosen categories of private and public expenditures on the number of ICT patent applications for the sample of 16 OECD countries.
The Error Correction model is presented by Hill et al. (2001) with respect to time series and with the necessary conditions of nonstationarity and cointegration. As a general statistical rule, nonstationary time series variables should not be used in regression models, in order to avoid the problem of spurious regressions10. As shown in the table above, spurious regressions exhibit a low value of the Durbin-Watson statistic and a high Adjusted R-squared. There is one exception, however, with regards to the estimation of a relationship in the case of nonstationarity.
If the dependent and independent variables (yt and xt) are nonstationary variables, then we
would expect their difference, or any linear combination of them (et = yt – β1 – β2xt), to be so
as well. If, however, this is not the case, then we may say that they are cointegrated. If two nonstationary variables are cointegrated, their long-run relationship can be estimated via a least squares regression. The authors therefore refer to the EC model as a tool to capture both short and long term dynamics and it can be applied in order to curb the model for autocorrelation. The first step must therefore be to test the model for cointegration.
Step 4: Dickey-Fuller test for cointegration.
Cointegration implies that yt and xt share similar random trends and since their difference, et 11, is stationary, they never diverge too far from each other. The cointegrated variables yt
and xt display a long-term equilibrium relationship defined by yt = β1 + β2xt.
H0: No cointegration H1: Cointegration
10 A spurious regression is a regression which will lead to apparently significant results from unrelated data when
using nonstationary series in regression analysis. (Hill et al., 2001)
11 e
The Dickey-Fuller test allows us to check whether the relation between the dependent and independent variables presents stationarity (this would imply cointegration). The table below is based on the following Dickey-Fuller test:
∆êt = α0 + γêt-1 + vt ;
where : ∆êt = êt – êt-1
and it shows that the lagged error term is significant (p-value = 0.0068). This illustrates that dependent and independent variables are, in fact, cointegrated (Null is rejected). Since this is a case of both nonstationarity and cointegration, the EC model may now be estimated.
Table 4: Dickey-Fuller test for cointegration:
Dependent Variable: ERROR-ERROR(-1) Method: Pooled Least Squares
Sample (adjusted): 1986 2002
Included observations: 17 after adjustments Cross-sections included: 16
Total pool (unbalanced) observations: 253
Variable Coefficient Std. Error t-Statistic Prob. C 7.49E-09 3.28E-08 0.228198 0.8197
ERROR(-1) -0.109482 0.040077 -2.731791 0.0068
Adjusted R-squared 0.027615 Durbin-Watson stat 1.659483
Step 2: Error Correction model.
The EC model relates changes or corrections in ∆yt to departures from the long run
equilibrium in the previous period (yt-1 – β1 – β2xt-1). It can be written as:
∆yt = α1 + α2 (yt-1 – β1 – β2xt-1) + vt
The shock vt leads to a short-term departure from the cointegrating equilibrium path; then,
there is a tendency to correct back toward the equilibrium. The coefficient α2 governs the
One way to estimate the EC model is to first use least squares to estimate the cointegrating relationship yt = β1 + β2xt (table 3) and to then use lagged residuals as the right-hand-side
variable in the EC model, estimating it with a second least squares regression. This allows us to separate the long-term and short-term effects. The EC model table bellow does unfortunately not show a significant relationship between ∆yt and the lagged residuals. We are
therefore unable to make any useful conclusions concerning the short-term effects. Table 5: Error Correction model (short term effects):
Dependent Variable: PAT-PAT(-1) Method: Pooled Least Squares Sample (adjusted): 1986 2002
Included observations: 17 after adjustments Cross-sections included: 16
Total pool (unbalanced) observations: 254
Variable Coefficient Std. Error t-Statistic Prob. C 1.56E-07 5.60E-06 0.027781 0.9779
ERROR(-1) 1.335940 6.846249 0.195135 0.8455
Fixed Effects (Cross)
Effects Specification Cross-section fixed (dummy variables)
Adjusted R-squared -0.067337 Durbin-Watson stat 3.186527
More importantly, however, the long-run effects, which are captured by the cointegrating relationship (yt = β1 + β2xt) have been estimated previously via least squares with cross-section
fixed effects (table 3). The final step before discussing the results of this cointegrating relationship, must be to test the model for multicollinearity.
3. Multicollinearity
Multicollinearity refers to any linear relationship amongst explanatory variables in a regression model. This can affect two or more of them. Consequences are the difficulty to reveal the separate effects of the explanatory variables on yt and the lack of guarantee that the
According to Moore and McCabe, 1993, one should preferably not accept correlation coefficients which are above 0.7 in an analysis. In order to avoid having one single table in which correlation coefficients for all countries are aggregated, this study will concentrate on separate country-specific tables (see appendix 5) as this will allow us to observe any potential problems per country. We can see in these tables that there are several countries (such as Japan) for which coefficients are above 0.7. We may thus conclude that there is some multicollinearity between the explanatory variables and this must of course be considered when interpreting the final results.
After applying all necessary diagnostic tests and adjustments to the statistical model, table 3 shows us the most appropriate and fairest model available. Considering the fact that the chosen variables are both nonstationary and cointegrated (see table 4 for Dicky-Fuller test), we may look at this model to study the long-term relationship. The following step will be to discuss the results and the contribution of the empirical section to the paper’s hypotheses and research question.
4. Results and Conclusion
Past innovation literature has shown that expenditures in Business R&D contribute to the creation of knowledge and technological expertise within firms and their networks, especially in an innovation-driven industry such as ICT. Public expenditures in tertiary education also clearly contribute to the development of the skills and knowledge of the workforce. In addition, they also enhance the ability to learn and therefore improve absorptive capacity, which is a requirement for innovating firms. Finally, expenditures in the maintenance and expansion of telecommunication networks facilitate communication and information flows within and between firms (especially in high tech industries), as well as universities and other institutions, therefore also improving absorptive capacity.
100 million US$ leads to an increase of 2.86 patent applications (coefficient of 2.86E-08 for a 1 US$ expenditure) and a PTO expenditure increase of 10 million US$ leads to an increase of 1.17 patent applications (coefficient of 1.17E-07 for a 1 US$ expenditure). R&D expenditures hence appear to have a relatively more important impact on PAT than the two other explanatory variables. The cointegrating relationship suggests that the following three hypotheses which capture the separate dependent-independent variable relations may therefore be confirmed. These findings must, of course, be interpreted very carefully, since the precision of the statistical results may have been negatively influenced by the problem of multicollinearity:
a) Public (and private) expenditures on Radio, T.V. and Communication Equipment R&D in the business sector, have an important positive impact on the innovativeness of the ICT industry for the selected sample of 16 OECD countries.
b) Public expenditures on Tertiary Education have an important positive impact on the innovativeness of the ICT industry for the selected sample of 16 OECD countries. c) Public expenditures in telecommunication infrastructure have an important positive
impact on the innovativeness of the ICT industry for the selected sample of 16 OECD countries.
The Adjusted R-squared in table 3 illustrates that on the long-run, approximately all of the variation in innovation output can be explained by the variation of the combination of the three explanatory variables. Consequently, we have shown statistically that public and private R&D, Education and Telecommunication infrastructure expenditures explain almost all of the changes in innovation output of the ICT industry in the sample of 16 OECD countries over the period of 1985-2002. While the current three explanatory variables provide satisfying statistical results, further research is needed, however, in order to mitigate multicollinearity and to improve the precision of the model.
Section 4: Limitations and Discussion
that perform R&D must be considered. Other potentially influential factors such as firm organization, firm proximity to complementary organizations and the financial system will also be explained in this last section. A country’s ICT industry innovativeness truly relies on its respective NSI (country specific inter-firm, university-firm, government-university, government-firm linkages, etc.).
In their work, Guellec and Van Pottelsberghe (2004) have studied Business, Public and Foreign R&D separately. First, while the R&D explanatory variable of this paper simultaneously covers business-performed R&D funded both by business itself and governments, they find that government support has a negative impact on business R&D. They show that when disaggregated according to socioeconomic objectives (civilian vs. defence-related aims), defence-related public funding has a negative impact on the effectiveness of business R&D. This is due to the fact that the performer of defence R&D is generally not the owner of its technological outcome. It will therefore be informative to separate both categories of business R&D into two explanatory variables: publicly funded business R&D and privately funded business R&D. On top of this, while it was not the case in this paper, Business R&D expenditures should cover all subcategories of ICT manufacturing. This should lead to more accurate results.
Secondly, the authors find that the impact of public R&D is positively affected by the share of universities in public research. While much government laboratory R&D is aimed at public missions, universities and other tertiary education institutions provide the basic knowledge that might eventually be used in later stages by industry to perform technological innovations. One explanation might be the difference in fund allocations: An important share of funding for university research is based on project evaluations whereas government laboratories have an institutional funding. The two following explanatory variables: tertiary education-performed R&D (funded publicly) + government laboratory-education-performed R&D (funded privately) should thus also be included in order to illustrate their separate effect on ICT industry innovativeness.
result underlines the importance of looking at business funded university R&D. Public expenditures in tertiary education could also be subdivided in order for tertiary education R&D and other education expenditures to be separated.
Thirdly, they write that the knowledge generated in other countries is a third source of new technology for national economies. There are many ways for technology to cross borders, as knowledge coming out of a given country’s research is used by another country’s enterprises. Companies can interact with foreign competitors who invested in their country (foreign direct investment), read scientific and technological literature, or have direct contacts with foreign engineers in conferences or fairs. The impact of foreign produced knowledge on a country’s productivity may depend on the recipient country’s absorptive capacity. The spatial range of knowledge spillovers is a controversial topic. Empirical evidence strongly supports that knowledge often exhibits a certain degree of tacitness. Geographical proximity and a common language which permit informal communication are therefore often important to allow the appropriation of such non-excludable knowledge (Döring et al. 2006). Foreign knowledge must, nevertheless, be included in the future as it might improve the model.
In the NSI literature, Lundvall (1992) explains that differences in historical experience, language and culture will be reflected in national idiosyncrasies which are relevant to innovation: The international organization of firms, the scope of activities of workers, inter-firm relations and the institutional set-up of the financial sector might all influence industry innovativeness. Teece (1996) writes that firm boundaries (the level of integration), the structure of financial markets, as well as the formal and informal organizational structure of firms must be recognized as major determinants to innovation. He insists on the importance of inter-firm agreements linking firms with complementary capabilities. The opening up of financial markets and the emergence of a lively venture capital industry have also provided new forms of finance for innovative activity and entrepreneurship across the industrialized world.
With regards to inter-firm connections in ICT, the potential influence of high-tech industrial clusters12 on innovativeness is important as well. Through the actions of key individual
12
The term ‘innovative cluster’ is used to refer to a geographically confined collection of firms working in related or supporting technologies. An infrastructure of institutions and social relationships provide resources and
change agents, the configuration of a cluster and the industry’s technological trajectory may be jointly determined. This suggests that as companies, industries and regions benefit from the same factors and decisions, their evolution may be intertwined. This is likely to influence entrepreneurship, which facilitates the realization of innovation (Feldman et al., 2005). An example of this phenomenon is Silicon Valley, for which success is perceived as an outgrowth of intense technology transfers and interactions between industry and universities due to geographical proximity in that region. The inclusion of the impact of geographical proximity between firms might also improve the model.
This last section has shown that there are numerous aspects to consider when studying the input-output relation of innovation thoroughly and when attempting to compare the influence of several factors on industry innovativeness. Including these elements and subdividing the current variables is likely to increase the precision and hence improve the value of the model.
Bibliography:
Blöndal, S., S. Field, N. Girouard. 2002. Investment in Human Capital Through Upper Secondary and Tertiary Education. OECD Economic Studies, no.34, 2002/1
Cairncross, F. 1997. The Death of Distance. Boston, Mass.: Harvard Business School Press.
Cohen, W.M., D.A. Levinthal. 1990. Absorptive Capacity: A New Perspective on Learning and Inovation. Administrative Science Quarterly, vol.35, pp.128-152.
David, P.A. 1993. Intellectual Property Institutions and the Panda's Thumb:
Patents, Copyrights, and Trade Secrets in Economic Theory and History. In Global Dimensions of Intellectual Property Rights in Science and Technology, edited by M.B. Wallerstein, M.E. Mogee and R.A. Schoen, 19-64. Washington, D.C.: National Academy Press, 1993.
Döring, T and J. Schnellenbach. 2006. What do we know about geographical
knowledge spillovers and regional growth?: A Survey of the Literature. Regional Studies, vol. 40., Issue 3, May 2006, pp.375-395.
Dutta, A. 2001. Telecommunications and Economic Activity: An Analysis of
Granger Causality. Journal of Management Information Systems. Vol. 17. no.4., pp.71-95.
Egan, B.L., S.S. Wildman. 1994. Funding The Public Telecommunications
Infrastructure. Telematics and Informatics, vol.11, no.3.
Edquist, C. 1997. Systems of Innovation; Technologies, Institutions and
Organizations. Science, technology and the international political economy series. Pinter, London.
Feldman, M.P., J. Francis, J. Bercovitz. 2005. Creating a Cluster While Building a Firm: Entrepreneurs and the Formation of Industrial Clusters. Regional Studies, vol.39.1, pp.129-141.
Furman, J.L., M.E. Porter, S. Stern. 2002. The Determinants of National Innovative
Capacity. Research Policy, nr. 31, pp. 899-933.
Gallouj, F., O. Weinstein. 1997. Innovation in Services. Research Policy. Vol.26. no.4/5, pp. 537-556
Greene, W. Econometric Analysis, Third Edition, Prentice Hall 1997
Guellec, D., B. Van Pottelsberghe de la Potterie. 2004. From R&D to Productivity Growth: Do the Institutional Settings and the Source of Funds of R&D Matter? Oxford Bulletin of Economics and Statistics, vol.66, no.3, 0305-9049. Hall, B. and J. Van Reenen. 2000. How effective are fiscal incentives for R&D? A
review of the evidence. Research Policy, no.29, pp. 449-469.
Harman, G., K. Harman. 2004. Governments and Universities as the Main Drivers of Enhanced Australian University Research Commercialisation Capacity. Journal of Higher Education Policy and Management, vol. 26, no.2, July 2004. Hill, R.C., W.E. Griffiths, G.G. Judge. 2001. Undergraduate Econometrics.
Second Edition, John Wiley & Sons, Inc.
Kemp, D.A. 2000. Higher Education Report for the 2000 to 2002 Triennium, DETYA, Canberra.
Koopmans, T.C. 1957. Three Essays on the State of Economic Analysis. New York. MacGraw-Hill.
Lanjouw, J.O., A. Pakes and J. Putman. 1998. How to Count Patents and Value Intellectual Property: The Uses of Patent Renewal and Application Data. The Journal of Industrial Economics, vol. 46, issue 4, pp.405-432.
Leiponen, A. 2005. Innovative Objectives, Knowledge Sources, and the Benefits of Breadth. Working Paper, Cornell University.
Los, B., B. Verspagen. 2002. An Introduction to the Analysis of Systems of
Innovation: Scientific and Technological Interdependencies. Economic Systems Research, vol.14, no.4.
Lundvall, B.A.1988. Innovation as an Interactive Process: From User-Producer Interaction to the National System of Innovation. Technical Change and Economic Geography. London: Pinter Publishers, 1988.
Lundvall, B.A. 1992. National Systems of Innovation: Towards a Theory of Innovation and Interactive Learning. London: Pinter Publishers, 1992. Lundvall, B.A. 1998. Why Study National Systems and National Styles of
Innovation? Technological Analysis & Strategic Management, vol.10, no.4., pp. 407-421.
. Moore, D., G. McCabe. 1993. Introduction to the Practice of Statistics. W.H. Freeman and Company, New York, 854p.
Navaretti, G.B., A.J. Venables. 2004. Multinational Firms in the World Economy. Princeton University Press. Princeton and Oxford.
OECD Science, Technology and Industry Scoreboard, OECD Publishing, 2005. Polanyi, M. 1962. Personal Knowledge. Chicago, IL: University of Chicago Press. Rogers, M. 1998. The Definition and Measurement of Innovation. Melbourne
Institute Working Paper, no.10.
Romer, P.M. 1990. Endogenous Technological Change. Journal of Political Economy, vol.98, no.5, pp.71-102.
Saint, W. 2006. Innovation Funds for Higher Education: A User’s Guide for World Bank Funded Projects. Education Working Paper Series, no.1, World Bank. Solow, R.M. 1956. A Contribution to the Theory of Economic Growth. Quarterly
Journal of Economics, vol.70, pp.65-94.
Solow, R.M. 1957. Technical Change and the Aggregate Production Function. Review of Economics and Statistics, vol.39, pp.312-320.
Steil B., D.G. Victor, R.R. Nelson. 2002. Technological Innovation & Economic Performance. A Council on Foreign Relations Book. Princeton University Press.
Stiglitz, J.E., P.R. Orszag, J.M. Orszag. 2000. The Role of Government in a Digital Age. Study Commissioned by the Computer & Communications Industry Association, October 2000.
Teece, D.J. 1996. Firm Organization, Industrial Structure, and Technological Innovation. Journal of Economic Behaviour & Organization, vol.31, pp. 193-224.
Verspagen, B. 2005. Innovation and Economic Growth. The Oxford Handbook of Innovation. Oxford and New York: Oxford University Press.
Williamson, O.E. 1975. Markets and Hierarchies: Analysis and Antitrust Implications. Free Press: New York.
