Universidad del Zulia (LUZ)

Revista Venezolana de Gerencia (RVG)

Año 29 No. 108, 2024, 1776-1798

octubre-diciembre

ISSN 1315-9984 / e-ISSN 2477-9423

Como citar: Gutiérrez, C., Smith-Uldall, J., Ganga-Contreras, F., y González, P. (2024). Characteristics of innovation systems that lead to greater productivity and economic development. Revista Venezolana De Gerencia, 29(108), 1776-1798. https://doi.org/10.52080/rvgluz.29.108.19

Characteristics of innovation systems that lead to greater productivity and economic development

Gutiérrez Rojas, Cristián*

Smith-Uldall, Jerome**

Ganga-Contreras, Francisco***

González, Patricia****

Abstract

The national R&D system is a concept that has gained significant recognition; however, measuring it is challenging and not devoid of difficulties. In this paper the method of composite variables is applied to configure R&D systems and with this examines their characteristics and subsequent impact on fostering greater innovation by studying a group of OECD countries. Starting with many variables, 31 variables were selected for the second step and used in a factor analysis to create composite indicators or unobservable abstract variables. Each variable’s assignment to a single factor is clear, allowing the identification of five distinct and interpretable factors. Subsequently, a knowledge production function was estimated, considering the technological outcome of the R&D systems as the dependent variable; another function was configured reflecting scientific output. Finally, an additional model was estimated, using productivity as the dependent variable. In all models, the National R&D Effort and Innovative Firms Factor emerged as the most significant variable, underscoring the importance of reaching certain thresholds in terms of available human and physical resources for carrying out innovative efforts within an R&D system.

Key words: innovation; research policy; r&d intensity; organizations; OECD.

Received: 28.06.24 Accepted: 04.09.24

* Doctor en Economía, Universidad Complutense de Madrid. Académico de la Facultad de Ingeniería y Empresa en la Universidad Católica Silva Henríquez, Santiago, Chile. ORCID: 0000-0001-8607-1528. E-mail: cgutierrez@ucsh.cl

** Diploma for Graduates in Economics, London School of Economics. Licenciado en Matemáticas, Universidad de Chile. Estudiante de Magíster en Análisis Económico, Universidad de Chile, Santiago, Chile. ORCID: 0000-0003-0967-1663. E-mail: jsmithu@fen.uchile.cl

*** Doctor en Gestión Estratégica y Negocios Internacionales. Magíster en Administración de Empresas. Administrador Público. Profesor Titular en la Universidad Tarapacá, Arica, Chile. ORCID: 0000-0001-9325-6459. E-mail: franciscoganga@uta.cl (autor de correspondencia)

**** Doctoranda en Educación con Mención en Gestión Educativa, Universidad Privada de Tacna. Magíster en Educación con Mención en Gestión de Calidad y Psicóloga, Universidad Miguel de Cervantes. Profesora Adjunta, Universidad Miguel de Cervantes, Chile. ORCID: 0000-0003-1511-8442. E-mail: patricia.gonzalez@profe.umc.cl

Características de los sistemas de I+D de países OCDE que conducen a una mayor productividad y desarrollo económico

Resumen

El sistema nacional de I+D es un concepto que ha ganado un reconocimiento significativo; sin embargo, su medición es un desafío y no está exenta de dificultades. En este trabajo se aplica el método de variables compuestas para configuar los sistemas de I+D y con ello se examinan sus características y su posterior impacto en el fomento de una mayor innovación mediante el estudio de un grupo de países de la OCDE. Partiendo de muchas variables, se seleccionaron 31 variables para el segundo paso y se utilizaron en un análisis factorial para crear indicadores compuestos o variables abstractas no observables. La asignación de cada variable a un único factor es clara, lo que permite la identificación de cinco factores distintos e interpretables. Posteriormente, se estimó una función de producción de conocimiento, considerando como variable dependiente un resultado tecnológico de los sistemas de I+D, y se configuró otra función que refleja la producción científica. Finalmente, se estimó un modelo adicional, utilizando como variable dependiente la productividad. En todos los modelos, el factor de esfuerzo nacional en I+D y de empresas innovadoras emergió como la variable más significativa, lo que subraya la importancia de alcanzar ciertos umbrales en términos de recursos humanos y físicos disponibles para llevar a cabo esfuerzos innovadores dentro de un sistema de I+D.

Palabras clave: innovación; política de investigación; intensidad de i+d; organizaciones; OCDE.

1. Introduction

A vast academic literature indicates that one of the most recognised determinants of countries’ economic growth is innovation. Authors such as Schumpeter (1939), Solow (1956), Abramovitz (1956), Griliches (1986), Fagerberg (1988), Freeman (1995), Pradhan et al, (2020) and Hausman (2022) recognise innovation as a key factor for development and economic growth.

Innovation is a complex phenomenon and many models and perspectives have been developed to study and attempt to explain it. One of the most widely accepted theoretical frameworks on innovation is the systemic viewpoint, and in particular research and development (R&D) systems (Fagerberg, 2011).

The national R&D system (RDS) is a concept that has gained significant recognition, as evidenced by the numerous scholarly works published on the topic. The RDS embodies division of labour within the realm of innovation, involving a wide array of interconnected agents and institutions. These entities are expected to create synergies or reduce costs through their collaborative efforts. Innovation, being a multifaceted and interdisciplinary endeavour, requires the cooperation of numerous institutions, organisations, and businesses (Gutiérrez, 2018).

However, measuring R&D systems and their results is challenging and not devoid of difficulties. There are different methods such as: analysis with individual variables, network analysis, structural equations, input-output analysis, etc. (Suárez & Natera, 2024).

In this paper we apply the method of composite or synthetic variables to characterise R&D systems. We use this method to identify strengths and weaknesses of different national R&D systems. It follows that this method helps to answer an important question: What are the characteristics of countries, and more specifically, their national R&D systems, that lead to greater innovation and hence increased productivity and economic growth?

Such knowledge would clearly be important for public policy that aims to increase innovation activities, and ultimately, increased economic growth.

This paper attempts to answer this question, by studying a group of OECD countries and ascertaining the characteristics and mechanisms in their science, technology and innovation (STI) sectors that are most relevant or have good prospects to be applied as guidelines by emerging economies. This is the contribution to the literature that this paper intends to make.

In addition to this brief introduction, the second section presents a brief overview of R&D systems and the use of composite variables for their measurement; in the third, the theoretical basis for the variables used is explained. In the fourth section the results are shown and in the fifth the conclusions are presented.

2. Methodology: R&D systems and use of composite indicators

There is no doubt that there are clear differences between R&D systems of different countries but speaking of RDS it is implicitly assumed that there is a certain internal homogeneity between the regions that comprise them, although this constitutes an unrealistic abstraction (Lundvall, 1992).

This study subdivides the RDS, following Buesa & Heijs (2016) and Gutiérrez (2018), into four subsystems:

As discussed earlier and supported by Gutierrez et al, (2021), R&D systems and their subsystems are intricate constructs involving numerous participants and varying institutional structures. To accurately represent these systems, it is crucial to use a wide array of diverse variables, many of which are highly interrelated. When developing econometric models that include many correlated variables or indicators, it is necessary to condense the information from the original variables into a smaller set of abstract synthetic variables, referred to as factors using factorial analysis. These factors are identifiable with the components of the R&D subsystems (Gutierrez, 2018; Gutierrez et al, 2021; Alqararah, 2023).

From a conceptual point of view, synthetic variables are important because it is doubtful that individual variables correctly reflect the characteristics of an R&D system and its potential. On the other hand, composite indicators solve econometric problems (such as, among others, multicollinearity and the lack of degrees of freedom in regression models) or methodological problems (they smooth out the existence of ‘outliers’ or errors in the statistics).

Many individual indicators convey similar concepts and can often be used interchangeably. Despite their high correlation, these individual indicators sometimes present differing perspectives on the same aspect of the R&D system, potentially undermining the simultaneous or holistic nature of innovative behaviour. As Makkonen and Have (2013:251) note, “an individual indicator is only a partial indication of the total innovative effort made by a subject”. Thus, composite indicators would more accurately reflect the reality of the R&D system than individual indicators alone.

Despite the advantages of using composite indicators, there are also criticisms regarding their use, their usefulness, and the quality of their preparation (Hollenstein, 1996; Buesa et al, 2006; OECD, 2008; Grupp & Schubert, 2010; Makkonen & Have, 2013). However, these problems are far from being resolved unanimously and by consensus. The creation of composite indicators in the field of R&D systems is still a new phenomenon, and reaching a consensus and standardising the methodological model are both required to formulate the synthetic indices and weight of the variables included in them (Jiménez-Fernández et al, 2022).

2.1. Analysis units and study period for factor analysis

As in all empirical studies, the identification and selection of variables is of special importance in ensuring the quality of the results and their correct interpretation. This task was undertaken based on data compiled from the official statistics of the OECD and the World Bank. It comprises 56 variables, from 25 countries, for the period from 2000 to 2019, limited however by the availability of the data, in addition to being homogeneous and comparable. Finally, the factorial analysis itself led to the study of the R&D systems with 31 variables. These 31 indicators, in turn, can be summarised through principal component factor analysis, into a smaller number of synthetic variables —called factors— of an abstract nature, although identifiable with respect to the elements that make up the indicator.

The use of the statistical technique of factor analysis is very appropriate to render operational the information of the indicators of the R&D system, given its characteristics as a multidimensional reality, by representing it in a limited number of abstract elements. From a statistical perspective, this technique has the following advantages for the type of research carried out here:

2.2. Feasibility conditions for factorial analysis

In factorial analysis the variables are not assigned a priori to a factor but rather it is the statistical processing itself that groups them. Therefore, a factor analysis is only useful if the results are unequivocally interpretable from the conceptual framework provided by the theory. This interpretation will be possible if the following conditions hold simultaneously:

The variables whose concepts are described in this section, are introduced in the factorial analysis that serves to configure the R&D system of each one of the selected countries, as well as to obtain the inputs for the subsequent analysis of the optimisation process during the study period. These are reflected in Tables 1 - 5 in the results section which also include the basic statistics of each of them. These tables show the use of 31 relative variables to configure five factors that each represent a component of the R&D system: national effort in R&D and innovative firms, national environment, universities and human capital, public administration and degree of economic complexity.

Regarding the statistical tests and adequacy measures that validate the factorial model obtained, the four fundamental aspects that the factorial model must comply with are the following:

Finally, the factors are extracted by means of principal component analysis (PCA). This consists of extracting the factorial space and thus obtaining projections of the point clouds on a number of axes in such a way that the resulting factors are mutually perpendicular. It involves going from a set of variables correlated with each other, to a new set of variables, linear combinations of the original ones, that are uncorrelated.

3. Theoretical basis for the variables

As previously mentioned, R&D Systems are intricate entities involving the participation of numerous agents and exhibiting diverse institutional configurations. This complexity necessitates the utilization of multiple variables for effectively representing these systems.

3.1 Input variables of innovation processes

The variables that measure the input or effort of the R&D systems that were included in this analysis are described below, discussing their conceptual importance and their limitations. In this way, in the following pages the suitability of the variables used is discussed in the following order: (1) the effort or innovative ‘input’, (2) the socioeconomic context and (3) human capital. A basic description of the values of the variables is found in Tables 1 - 5. In addition, in section 3.2 the adequacy of the variables related to the results (the output of the R&D systems) is discussed.

Measurement of the effort or ‘input’ of the R&D systems

The input with the highest incidence according to different theoretical approaches is that which represents innovative effort, traditionally measured by spending on R&D (Boeing et al, 2022; Myers & Lanahan, 2022; Dai & Chapman, 2022). Expenditure on R&D includes all the financial means allocated to this activity and includes both current and capital expenses and is calculated as a percentage of Gross Domestic Product. In addition, both funding and execution variables have been incorporated.

The R&D effort variables, in turn, are broken down for each of the three main types of agents in the R&D system, in accordance with the recommendations of the Frascati Manual: firms, higher education (universities) and public administration.

Variables of the socioeconomic context of the R&D Systems

The notion of global environment includes various aspects that indirectly influence the technological capacity of a country, such as the educational system (Junge et al, 2012; Biasi et al, 2022), the level of human capital (Fonseca et al, 2019; Sun et al, 2020), the financial system (venture capital) (Leogrande et al., 2021), the sophistication of consumers of goods and services, the culture, and the standard of living. Thus, various variables have been introduced that reflect the socioeconomic context (Puertas et al, 2020).

The first of these —which is included indirectly— is size. When working with very heterogeneous countries, their size must be considered. For this reason, it is advisable to correct the different variables by population or economic size, which has been done opportunely through the annual average number of inhabitants or the Gross Domestic Product (GDP). In addition, variables have been incorporated that describe the economic reality of the countries. For the above, variables such as GDP per capita and apparent labour productivity have been added.

Another crucial aspect of the environment is the country’s relative wealth and productive capacity, which is represented by two variables. The first variable is GDP per capita, reflecting the standard of living and indirectly indicating the technological sophistication of consumer demand. Higher GDP per capita levels suggest that consumers seek higher quality and more feature-rich products, prompting companies to boost their innovative efforts (demand pull). Additionally, a higher standard of living and higher wages attract new talent and top researchers or inventors. The second variable, which is directly correlated with GDP per capita and related to the innovative capacity of a region or industry, is apparent productivity. This metric tends to rise with the technological advancement of a country or specific industry, being significantly higher in medium and high technology sectors compared to traditional industries (technology push).

As a last aspect of the socioeconomic environment, the degree of commercial openness of the economies was included, particularly exports (Onetti et al, 2012; Filipescu et al, 2013; Yang, 2018).

Human capital indicators

Another very important aspect for innovation is human capital. It is the researchers and engineers —with their talent, experience, and quality— who lead the innovation process and largely determine its level of success and efficiency (Jansen et al, 2016; Meissner & Shmatko, 2019). Measurement of human capital is not easy, and the data tend to be approximations. Nonetheless, the available indicators are generally accepted and can be considered quite accurate. As the OECD states in the Frascati Manual, R&D personnel is not enough to measure the technological performance of a country since it only represents a part of the human input of an R&D system. Scientific and technical personnel equally contribute to technological advancement through their involvement in production, quality control, management, or education.

In addition to using these variables in absolute terms (number of people), relative variables are also provided with respect to the total number of workers in the economy (as a percentage of the labour force). In addition, in the factorial model, another variable that was adequately incorporated in this dimension is the rate of higher education (percentage of the population between 25 and 64 years of age with a university education).

3.2. Output variables of innovation processes

The variables used as output were patents filed with the European Patent Office (EPO), patents filed with the United States Patent Office (USPTO) and scientific papers published.

Business intellectual property (patents)

The use of patents as a measure of output is justified by an extensive literature on the subject that highlights its advantages and disadvantages, establishing a balance in favour of the former. Thus, patents are for the moment the best measure of the national innovative capacity that we have (Hall, 2022; Nguyen et al, 2020).

In short, patents far from being a perfect measure of technological output are for the moment, the best and most complete measure available. Their drawbacks only entail a series of restrictions that must be considered when interpreting the results of the model (Baumert, 2006).

The patents variable has been incorporated into the study using patent applications in Europe via the EPO and in North America via the USPTO, corrected per million inhabitants. The location of the domicile of the inventor (or research group that obtains the patented knowledge) has been considered and not the domicile of the owner of the rights protected by those patents. This makes the use of this statistic the most appropriate for the research presented here.

Results of scientific research (publications)

To address the issue of relative unawareness regarding the significant portion of innovation systems’ outputs that include scientific research activities, this study has incorporated statistics on publications in academic journals (Ganga et al, 2016; Castaneda & Cuellar, 2020).

4. Characteristics of I&D systems in OCDE countries: Results and discussions

In this section of results, the composite variables resulting from the factor model are first presented, the statistical robustness of the model is demonstrated, and the result obtained is associated with that indicated by the theory of innovation systems. Then, using the previously obtained factors, the production function of knowledge applied to national R&D systems is estimated.

4.1 Factor analysis results

The factorial model resulting from the application of the CPA technique to the battery of indicators available to describe the R&D systems includes five factors. The following tables show the results for each factor in terms of the variables that constitute the factor.

As can be seen in the tables above, each factor is associated with a subsector of the R&D system. Table 1 indicates the innovative effort made by firms, in addition to grouping the results of the innovation process, publications and patents.

Table 1

National R&D Effort and Innovative Firms

Variable

Mean

Standard

Deviation

Maximum

Minimum

Execution of R&D by Firms (% of GDP)

0.0113

0.0073

0.0300

0.0011

Intensity in R&D

0.0180

0.0089

0.0387

0.0036

Funding of R&D by Firms (% of GDP)

0.0099

0.0067

0.0278

0.0007

R&D Workers in Private Sector (% of Workforce)

0.0060

0.0036

0.0141

0.0003

Patents Applied for by Inventor (per million inhabitants)

245.0

231.5

875.3

0.5

Researchers in Private Sector (% of Workforce)

0.0036

0.0025

0.0134

0.0002

Execution of R&D by Firms (% of Total Expenditure on R&D)

0.5702

0.1480

0.7942

0.1686

Funding of R&D by Firms (% of Total Expenditure on R&D)

0.4963

0.1312

0.7906

0.1674

Expenditure R&D per Worker R&D

120,272

46,149

217,955

22,552

R&D Workers (% of Workforce)

0.0060

0.0036

0.0141

0.0003

Researchers (% of Workforce)

0.0076

0.0031

0.0157

0.0028

Publications (per million inhabitants)

1,221

546

2,704

122

Source: Compiled by the authors from OECD data

Table 2 presents indicators that account for the macroeconomic context or the environment for innovation.

Table 2

National Environment

Variable

Mean

Standard Deviation

Maximum

Minimum

Workforce

11,350,223

15,616,485

69,045,529

169,444

GDP (US$ PPP x 1 million)

894,532

1,233,893

5,361,159

11,020

Population

22,690,287

30,524,511

128,083,960

281,200

Investment (US$ PPP x 1 million)

197,484

288,802

1,405,994

1,864

R&D Workers

129,684

199,774

912,202

2,645

Manufacturing (US$ PPP x 1 million)

152,325

234,284

1,123,949

1,682

Researchers

88,465

146,162

684,884

1,719

Expenditure R&D (US$ PPP x 1 million)

19,230

35,104

172,589

123

Exports (US$ PPP x 1 million)

306,826

339,921

2,030,439

3,783

Source: Compiled by the authors from OECD data

Tables 3 and 4 present variables associated with other relevant actors in the national R&D system, universities and public administration.

Table 3

Universities and Human Capital

Variable

Mean

Standard

Deviation

Maximum

Minimum

Researchers in Universities (% of Workforce)

0.0030

0.0011

0.0057

0.0011

R&D Workers in Universities (% of Workforce)

0.0038

0.0013

0.0075

0.0014

Higher Education Rate (% Population)

0.3057

0.0984

0.5937

0.0884

Workforce with Higher Education (% of Workforce)

0.6023

0.1767

1.0755

0.1739

Source: Compiled by the authors from OECD data

Table 4

Public Administration

Variable

Mean

Standard Deviation

Maximum

Minimum

R&D Workers in Public Sector (% of Workforce)

0.0014

0.0008

0.0048

0.0002

Researchers in Public Sector (% of Workforce)

0.0008

0.0005

0.0029

0.0001

Execution of R&D by Public Administration (% of GDP)

0.0019

0.0011

0.0069

0.0002

Source: Compiled by the authors from OECD data

Finally, Table 5 presents the variables that account for the level of economic complexity of the country.

Table 5

Economic Complexity

Variable

Mean

Standard Deviation

Maximum

Minimum

GDP per capita (US$ PPP)

39,280

12,398

83,874

12,182

Productivity (US$ PPP/hour)

51.1

16.0

102.3

16.7

Manufacturing (% of GDP)

0.1598

0.0648

0.4344

0.0736

Source: Compiled by the authors from OECD data

Moreover, it is interesting that the variables are saturated in the different factors so that these can be interpreted simply and clearly. This is the purpose pursued by the Varimax rotation, which also maximises the orthogonality of the factors —or minimises their correlation—, thus, avoiding multicollinearity problems when the factors are used to estimate econometric models.

The relevant statistics that validate this model are indicated below:

● The KMO measure is equal to 0.78.

● The null hypothesis of Barlett’s sphericity test is rejected with a confidence level of 99%.

● A percentage of 90.3% of the total variance of the sample is preserved.

● All communalities are above 80%, except four.

As can be seen in the Table 6, the communalities (correlation of each variable with respect to the set of other variables that make up that factor) of the variables are relatively high, most of them higher than 0.75 (with the exception of the enrolment rate of workforce in higher education (0.731) and exports (0.733)) which guarantees the reliability of the results and indicates the high degree of conservation of their variance.

Table 6

Communalities

VARIABLES

Initial

Extraction

Population

1.000

.965

GDP

1.000

.982

R&D Workers

1.000

.990

R&D Workers Private (% Total)

1.000

.943

R&D Workers Public (% Total)

1.000

.975

R&D Workers Universities (% Total)

1.000

.820

Researchers

1.000

.967

Researchers Private (% Total)

1.000

.873

Researchers Public (% Total)

1.000

.971

Researchers Universities (% Total)

1.000

.807

Researchers (% Workforce)

1.000

.926

Workforce

1.000

.980

Intensity R&D

1.000

.961

Finance R&D from Firms (%GDP)

1.000

.969

Finance R&D from Firms (% Total)

1.000

.839

Execution R&D Firms (% GDP)

1.000

.978

Execution R&D Firms (% Total)

1.000

.796

Execution R&D Public Administration (% GDP)

1.000

.936

Expenditure R&D

1.000

.966

Expenditure R&D per Worker R&D

1.000

.887

Exports

1.000

.733

Higher Education (% Population)

1.000

.770

Investment

1.000

.976

Manufacturing

1.000

.983

GDP per capita

1.000

.961

Productivity

1.000

.933

Patents Applied Inventor (millions of hab.)

1.000

.835

Publications (millions of hab.)

1.000

.824

Manufacturing (% Total)

1.000

.760

Workforce with Higher Education (% Total)

1.000

.731

R&D Workers (% Total)

1.000

.959

Source: Compiled by the authors from OECD data.

Table 7 shows the result: five factors were extracted by the method of principal component analysis (PCA). Therefore, it is considered that the model with five factors is supported by two facts: firstly, it is the result of objective processing (principal component analysis). Secondly, as will be seen below, the model allows for easy interpretation (since the variables are not saturated in more than one factor); the factors obtained are consistent with the theory of innovation systems, and the model is extremely robust, in addition to maintaining a high percentage of the original variance, as can be seen in Table 7. This shows the total explained variance in three sections: the first indicates the initial eigenvalues, the second indicates the sum of the squared saturations of the extraction, and the third presents the sum of the squared loadings after rotating the factors.

Table 7

Total explained variance

Component

Initial eigenvalues

Sums of Squared Extraction Charges

Sums of squared charges of rotation

Total

% of

variance

%

accumulated

Total

% of

variance

%

accumulated

Total

% of

variance

%

accumulated

1

13,728

44,284

44,284

13,728

44,284

44,284

9,185

29,630

29,630

2

7,752

25,005

69,289

7,752

25,005

69,289

9,024

29,111

58,740

3

3,083

9,945

79,234

3,083

9,945

79,234

4,026

12,987

71,727

4

2,146

6,924

86,158

2,146

6,924

86,158

2,969

9,579

81,306

5

1,285

4,145

90,303

1,285

4,145

90,303

2,789

8,997

90,303

6

0,862

2,779

93,082

7

0,480

1,549

94,631

8

0,428

1,379

96,010

9

0,277

0,893

96,904

10

0,231

0,744

97,648

11

0,171

0,551

98,199

12

0,122

0,393

98,592

13

0,083

0,268

98,859

14

0,078

0,251

99,110

15

0,061

0,196

99,306

16

0,058

0,187

99,493

17

0,040

0,130

99,623

18

0,029

0,094

99,717

19

0,023

0,074

99,791

20

0,016

0,052

99,844

21

0,012

0,039

99,883

22

0,010

0,033

99,916

23

0,010

0,031

99,947

24

0,007

0,023

99,970

25

0,003

0,009

99,978

26

0,002

0,007

99,986

27

0,001

0,005

99,990

28

0,001

0,004

99,994

29

0,001

0,003

99,997

30

0,001

0,002

99,999

31

0,000

0,001

100,000

Source: Compiled by the authors from OECD data.

The initial eigenvalues reflect the percentage of the variance explained by each variable and it is by this value that the system is governed when incorporating variables in the model. Obviously, by including all the variables (each variable would be a factor), 100% of the variance is explained but this would not have achieved the objective of reducing the number of variables with which we worked. The second section shows the percentage of the variance explained by each of the five factors extracted according to the previous specifications as well as the accumulated percentage before the rotation. As can be seen, with five factors the model maintains 90.3% of the variance, that is, when going from 31 variables to five factors, less than 10% of the information is lost.

However, for the purpose of this study, the percentages of variance explained by the factors after rotation are more interesting. As can be seen, the percentage of the variance accumulated by the set of factors remains the same after rotation. However, what is altered is the specific contribution of each factor to the total. Rotation consists of rotating the axes at the origin until reaching a certain position to maximise the load or saturation of the variables in one factor, simultaneously minimising them in the rest, thus allowing a more interpretable solution. There are different rotation procedures —orthogonal rotation and oblique rotation— although in this case only the former was used, since it maintains a 90-degree angle between the axes, thus guaranteeing orthogonality between the factors. Specifically, we carried out a Varimax-type rotation, since the factorial pattern obtained by this procedure tends to be more robust than that obtained by alternative methods.

As shown in Table 7, the assignment of each variable to a single factor is now clear, allowing the identification of five distinct and interpretable factors. These are the following:

  1. The national environment.
  2. National R&D efforts and innovative firms (including the specific activity of creating technological knowledge).
  3. Higher education institutions (universities) and human capital (reflecting the specific generation of scientific knowledge).
  4. Public administration.
  5. The degree of economic complexity (in a technological sense).

These results from the factor analysis align closely with the determinants highlighted by the theory.

In summary, the estimated factorial model provides an accurate representation of the R&D systems for the selected sample of countries, meeting all necessary statistical and conceptual criteria. Consequently, the factors derived from this model—which represent the resources, organization, and interrelationships characterizing R&D systems—can be used to analyse the activities related to the creation and dissemination of technological knowledge within these countries.

The adopted solution includes five factors whose name and participation in the variance explained by the model have been represented in Diagram 1.

Diagram 1

The Final Factorial Model (in parentheses the percentage of the total initial variance explained by each resulting factor)

1. National Effort in R&D and Innovative Firms (29.63%)

2. National Environment (29.11%)

3. Universities and Human Capital (12.98%)

4. Public Administration (9.57%)

5. Economic Complexity (8.99%)

Source: Compiled by the authors from OECD data.

4.2 Estimation of a knowledge production function

As stated in the introduction, finally we identify the determinants of national innovation for the selected countries and their degree of incidence on the technological outcomes of their R&D systems.

The objective consists therefore, in detecting the determinant factors of innovation and their degree of incidence, based on —according to the theoretical assumptions— the hypothesis that all the elements of the R&D system should positively influence its results, albeit with different intensities.

For this, the configuration methodology of the R&D systems explained in the last section is used, considering the factorial scores obtained from a new factorial analysis, this time only including the elements of effort and system environment, discarding the output factors. Technological output (patents per capita) is one of the dependent variables in the regression of the knowledge production function. In addition, another function was configured considering scientific publications per capita as a dependent variable, thus reflecting the scientific output of the R&D systems. Finally, one additional model was estimated, with economic output represented by productivity as the dependent variable.

In this stage of the analysis, the previously calculated ‘synthetic’ variables were used to estimate a knowledge production function from panel data. An additive model was proposed, being common in this type of study, according to the following specification:

(Equation 1)

The output variable refers, on the one hand, to new economically valuable knowledge, both in technological terms (Kit = number of patents per capita) and scientific terms (Kit = number of scientific publications per capita), and on the other, to national economic performance (Kit = productivity) while the explanatory variables are the five factors of effort and environment previously calculated: National environment (NENV), National Effort in R&D and Innovative Firms (FIR), Universities and Human Capital (UNI), Public Administration (ADM) and Economic Complexity (ECOM). 1

It is important to highlight that the results presented here are intended to identify the relative importance of the determinants of innovation and knowledge through an “explanation” function, rather than to predict future outputs as a “prediction” function. This distinction carries significant methodological implications. Specifically, it means that there is no need for a lag structure between inputs and outputs. Additionally, using regression techniques in combination with other statistical methods like factor analysis is less suitable for forecasting. This is because the resulting (non-standardised) regression coefficients reflect the elasticity of the factor score, which depends on changes in all the variables included in the factor, rather than the elasticity of a single variable.

Moreover, working with diverse national contexts introduces greater errors and non-uniform variance across the regression plane, which are crucial assumptions for predictive regression models. This does not imply that the models have not been thoroughly optimised; appropriate transformations such as robust errors, stationarity tests for panels, and the Hausman test have been applied to ensure their robustness.

The general results are presented in Tables 8 - 11. According to the results of the Hausman test, the fixed effects model corrected for autocorrelations and heteroscedasticity was preferred (Table 8). The global adjustments are acceptable with an R2 between 30% and 64% in the models.

Table 8

Final Estimation Results for the total sample: Fixed effects corrected for autocorrelation and with standard errors corrected for heteroscedasticity

FACTORS

Patents

Publications

Productivity

National Environment

55.35

(0.000)

-80.88

(0.168)

1.18

(0.098)

National R&D Effort and Innovative Firms

129.84

(0.000)

213.80

(0,000)

6.72

(0.000)

Universities and Human Capital

59.05

(0.000)

228.63

(0.000)

3.89

(0.000)

Public Administration

-16.82

(0.018)

-27.32

(0.153)

-0.78

(0.024)

Economic Complexity

72.14

(0.000)

193.48

(0.000)

Constant

221.04

(0.000)

1166.68

(0.000)

50.74

(0.000)

Rho

0.83

0.94

0.95

Wald Test

382.93

(0.000)

224.73

(0.000)

173.23

(0.000)

R2

0.39

0.30

0.64

Note: In parentheses the p values. Source: Compiled by the authors from OECD data

In the technological model (patent as output) all the variables positively affect technological production and are statistically significant at 1%, with the exception of the Public Administration variable. In the model whose output is productivity (in this model the 5th factor disappeared developing a new factor model), only two variables are significant with positive signs.

In all models, the National Effort in R&D and Innovative Firms Factor is the most important variable, highlighting the relevance that it would have for a R&D system to reach certain thresholds in terms of the amount of human and physical resources eventually available to execute innovative efforts. Regarding the actors that make the effort, all the models highlight the role of firms (contained in the National Effort in R&D factor) in innovative processes, and Universities as a fundamental subsystem in the configuration of the knowledge base of the R&D system as well as the articulating axis of the transfer of this knowledge to the productive sector.

In the case of Public Administration, its direct role as executor of R&D is not relevant since its coefficient in the models is not significant. In this sense, it is important to understand the role of the public sector in the R&D system, and how to differentiate cases where the State plays a direct role in the effort and execution of R&D, from cases where its role is to promote scientific and technological policies, as a generator of economic structural conditions, especially the formation of human capital (educational system) and/or coordinator of the rest of the actors that make up the system.

Dividing the total sample into 3 subsamples of national R&D systems according to their level of productivity, the results differ among the respective models2. In the case of the technological output of patents and the scientific output of publications, the model that best fits the group of emerging R&D systems is fixed effects corrected for heteroscedasticity but without autocorrelation, unlike the groups of emerging and developed R&D systems that must be corrected for first order autocorrelation.

In the case of the model with technological output (Table 9), among the emerging R&D systems, only the variables of National Effort in R&D and Innovative Firms, plus Universities and Human Capital, are significant. In the case of medium-developed R&D systems, the Economic Complexity variable is added and in the case of developed R&D systems, all are significant, including Public Administration, which enters but with a negative sign.

Table 9

Estimation Results by Clusters, Output: Patents

FACTORS

Emerging

N=8

Medium development

N=9

Developed

N=8

National Environment

-22.50

(0.456)

14.81

(0.625)

77.92

(0.000)

National R&D Effort and Innovative Firms

21.31

(0.005)

67.61

(0.001)

262.49

(0.000)

Universities and Human Capital

12.03

(0.001)

48.39

(0.000)

86.83

(0.000)

Public Administration

2.72

(0.379)

-12.92

(0.230)

-61.80

(0.000)

Economic Complexity

20.42

(0.034)

49.42

(0.000)

144.67

(0.000)

Constant

51.13

(0.000)

183.28

(0.000)

135.82

(0.000)

Sigma u

10.18

Sigma e

7.08

Rho

0.674

0.890

0.744

F Test

26.27

(0.000)

Wald Test

49.40

(0.000)

149.84

(0.000)

R2

0.183

R2 within

0.676

R2 between

0.334

R2 overall

0.497

Note: In parentheses the p values. Source: Compiled by the authors from OECD data

In the case of the model with scientific output (Table 10), among the emerging R&D systems, only the variables Economic Complexity plus Universities and Human Capital, are significant. In the case of medium-developed R&D systems, the Economic Complexity variable is not significant, but added to the National R&D Effort and Innovative Firms, it is. Finally, in the case of developed R&D systems, all are significant, including National Environment, which enters but with a negative sign, except for the Public Administration variable.

Table 10

Estimations Results by Clusters, Output: Publications

FACTORS

Emerging N=8

Medium development N=9

Developed N=8

National Environment

-930.94

(0.027)

34.90

(0.692)

-166.41

(0.000)

National R&D Effort and Innovative Firms

230.05

(0.043)

150.07

(0.005)

175.44

(0.000)

Universities and Human Capital

404.47

(0.000)

211.39

(0.000)

2541.87

(0.000)

Public Administration

-50.72

(0.359)

-17.90

(0.558)

-72.39

(0.016)

Economic Complexity

428.99

(0.001)

47.16

(0.257)

177.05

(0.000)

Constant

1003.32

(0.000)

1258.58

(0.000)

1393.55

(0.000)

Sigma u

312.98

Sigma e

109.39

Rho

0.891

0.911

0.878

F Test

37.41

(0.000)

Wald Test

55.27

(0.000)

269.11

(0.000)

R2

0.177

0.707

R2 within

0.837

R2 between

0.171

R2 overall

0.409

Note: In parentheses the p values. Source: Compiled by the authors from OECD data

Finally, as can be seen in Table 11, in the case of productivity, Universities and Human Capital is the only significant variable for all subsamples of R&D systems, thus becoming the subsystem that best explains productivity increases in the long term. The National Effort in R&D and Innovative Firms is a relevant variable for countries with less developed systems, being less important in the case of mature systems, where stationary states are seen both with regard to their growth in productivity as well as in the stagnation of its innovative performance (Park et al, 2023).

These results make it possible to identify those elements of national R&D systems that allow countries to increase both their scientific and technological outputs, as well as their economic performance, guiding the design and implementation of their scientific and technological policies.

Table 11

Estimation Results by Clusters, Output: Productivity

FACTORS

Emerging

N=8

Medium development

N=9

Developed

N=8

National Environment

4.92

(0.049)

5.48

(0.069)

-3.03

(0.000)

National R&D Effort and Innovative Firms

4.92

(0.000)

5.97

(0.000)

0.563

(0.604)

Universities and Human Capital

3.46

(0.000)

3.36

(0.003)

2.13

(0.000)

Public Administration

-1.00

(0.049)

-1.23

(0.014)

0.31

(0.655)

Constant

39.97

(0.000)

57.44

(0.000)

62.03

(0.000)

Rho

0.871

0.958

0.938

Wald Test

71.80

(0.000)

28.07

(0.000)

35.89

(0.000)

R2

0.466

0.623

0.85

Note: In parentheses the p values. Source: Compiled by the authors from OECD data

5. Conclusions

In this paper, a configuration of national R&D systems for a sample of 25 OECD countries has been developed by means of the elaboration of composite variables through factor analysis. This allowed us to identify 5 variables that summarise the main characteristics of the R&D sectors, namely: innovative firms, universities, public administration, as well as two variables that identify structural elements of the R&D systems: the socio-economic environment and economic complexity. Using the aforementioned variables, three econometric models were calculated, each measuring different results of these systems: technological output through patents, scientific output through publications and economic productivity.

In all models, the National Effort in R&D and Innovative Firms factor is the most important variable, highlighting the relevance that it would have for a R&D system to reach certain thresholds in terms of the amount of human and physical resources eventually available to execute innovative efforts. Regarding the actors that make the effort, all the models highlight the role of firms (contained in the National Effort in R&D factor) in innovative processes and universities as a fundamental subsystem in the configuration of the knowledge base of the R&D system as well as the articulating axis of the transfer of this knowledge to the productive sector. According to the results obtained, the important thing is to understand that the role of public investment in R&D is varied, and its effectiveness will depend on the structural conditions of the R&D system in question.

The main limitations of this work relate to the sample of countries selected, all developed or emerging countries, which implies that these results cannot be extrapolated to countries that are backward in both economic and technological terms. Furthermore, the lack of publicly available statistical data has prevented the addition of more variables to the configuration of R&D systems. In particular, environmental variables of some importance could not be included due to factors such as the quality of universities, the level of cooperation, etc.

Beyond the characteristics of their R&D systems, countries have recently steadily increased their innovative efforts, especially public spending on R&D. However, given the budgetary and financial restrictions faced by governments, universities and firms, it is important, in addition to a greater innovative drive, to ensure an efficient allocation of resources (public and private), optimising results and minimising costs. For this reason, new research should be aimed at measuring the efficiency of R&D spending, considering the systemic nature of innovation.

References bibliographic

Abramovitz, M. (1956). Catching up. forging ahead and falling behind. Journal of Economic History, 46(2), 385-406. https://doi.org/10.1017/S0022050700046209

Alqararah, K. (2023). Assessing the robustness of composite indicators: the case of the Global Innovation Index. Journal of Innovation and Entrepreneurship, 12(61). https://doi.org/10.1186/s13731-023-00332-w

Baumert, T. (2006). Los determinantes de la innovación. Un estudio aplicado sobre las regiones de la Unión Europea. [Tesis Doctoral. Universidad Complutense de Madrid. España]. https://produccioncientifica.ucm.es/documentos/5d1df64029995204f7665f77

Biasi, B., Deming, D., & Moser, P. (2022). Education and innovation. In M. J. Andrews, A. Chatterji, J. Lerner, & S. Stern (Eds.), The role of innovation and entrepreneurship in economic growth (pp. 537-551). University of Chicago Press. [invalid URL removed] https://www.nber.org/books-and-chapters/role-innovation-andentrepreneurship-economic-growth/education-and-innovation

Boeing, P., Eberle, J., & Howell, A. (2022). The impact of China’s R&D subsidies on R&D investment, technological upgrading and economic growth. Technological Forecasting and Social Change, 174(121212), 121212. https://doi.org/10.1016/j.techfore.2021.121212

Buesa, M. & Heijs, J. (2016). Manual de economía de innovación. Tomo I. Teoría del cambio tecnológico y sistemas nacionales de innovación. [Instituto de Análisis Industrial y Financiero. Universidad Complutense de Madrid. España].

Buesa, M., Heijs, J., Martínez Pellitero, M., & Baumert, T. (2006). Regional systems of innovation and the knowledge production function: the Spanish case. Technovation, 26(4), 463–472. https://doi.org/10.1016/j.technovation.2004.11.007

Castaneda, D. I., & Cuellar, S. (2020). Knowledge sharing and innovation: A systematic review. Knowledge and Process Management, 27(3), 159–173. https://doi.org/10.1002/kpm.1637

Dai, X., & Chapman, G. (2022). R&D tax incentives and innovation: Examining the role of programme design in China. Technovation, 113(102419), 102419. https://doi.org/10.1016/j.technovation.2021.102419

Fagerberg, J. (1988). International Competitiveness. Economic Journal, 98(391), 355-374. https://econpapers.repec.org/RePEc:ecj:econjl:v:98:y:1988:i:391:p:355-74

Fagerberg, J., & Sapprasert, K. (2018). National innovation systems: the ermergence of a new approach. En Innovation, Economic Development and Policy. Edward Elgar Publishing.

Filipescu, D. A., Prashantham, S., Rialp, A., & Rialp, J. (2013). Technological innovation and exports: Unpacking their reciprocal causality. Journal of International Marketing, 21(1), 23–38. https://doi.org/10.1509/jim.12.0099

Fonseca, T., de Faria, P., & Lima, F. (2019). Human capital and innovation: the importance of the optimal organizational task structure. Research Policy, 48(3), 616–627. https://doi.org/10.1016/j.respol.2018.10.010

Freeman, C. 1995. Innovation and Growth in M. Dodgson & R. Rothwell (Ed.), The Handbook of Industrial Innovation (chapter 7). Edward Elgar Publishing. https://econpapers.repec.org/RePEc:elg:eechap:143_7

Ganga, F., Castillo, J., & Pedraja-Rejas, L. (2016). Factores implicados en la publicación científica: una revisión crítica. Ingeniare. Revista Chilena de Ingeniería, 24(4), 615–627. https://doi.org/10.4067/s0718-33052016000400007

Griliches, Z. (1986). Productivity, R and D, and basic research at the firm level in the 1970’s. American Economic Review, 76(1), 141–154. http://www.jstor.org/stable/1804132

Grupp, H., & Schubert, T. (2010). Review and new evidence on composite innovation indicators for evaluating national performance. Research Policy, 39(1), 67–78. https://doi.org/10.1016/j.respol.2009.10.002

Gutiérrez, C. (2018). Eficiencia de los sistemas regionales de innovación en Europa y análisis econométrico de sus determinantes. [Tesis para optar al grado académico de Doctor. Universidad Complutense de Madrid. España]. https://eprints.ucm.es/49890/1/T40533.pdf

Gutiérrez, C., Roa P., V., & Smith, J. (2021). The Chilean sectoral innovation systems: An approach from the national innovation survey. Journal of Technology Management & Innovation, 16(1), 3–13. https://doi.org/10.4067/s0718-27242021000100003

Hall, B. H. (2024). Patents, innovation, and development. International Review of Applied Economics, 38(1–2), 17–42. https://doi.org/10.1080/02692171.2021.2022295

Hausman, N. (2022). University innovation and local economic growth. Review of Economics and Statistics, 104(4), 718-735. https://doi.org/10.1162/rest_a_01027

Hollenstein, H. (1996). A composite indicator of a firm’s innovativeness. An empirical analysis based on survey data for Swiss manufacturing. Research Policy, 25(4), 633-645. https://doi.org/10.1016/0048-7333(95)00874-8

Jansen, S., Hunter, E. & Currall, S.C. (2016). Managing the innovators: Organizational and professional commitment among scientist and engineers. Research Policy, 45(2016), 1247-1262. https://doi.org/10.1016/j.respol.2016.03.009

Jiménez-Fernández, E., Sánchez, A. & Ortega-Pérez, M. (2022). Dealing with weighting scheme incomposite indicators: An unsupervised distance-machine learning proposal for quantitative data. Socio-Economic Planning Sciences, 83, 1-11. https://doi.org/10.1016/j.seps.2022.101339

Junge, M., Severgnini, B., & Sørensen, A. (2012). Evidence on the Impact of Education on Innovation and Productivity. Department of Economics. Copenhagen Business School. Working Paper / Department of Economics. Copenhagen Business School No. 2-2012 https://research.cbs.dk/en/publications/evidence-on-the-impact-of-education-on-innovation-and-productivit

Leogrande, A., Costantiello, A., & Laureti, L. (2021). The impact of venture capital expenditures on innovation in Europe. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.3930697

Lundvall, B. (1992). National Systems of Innovation: Towards a Theory of Innovation and Interactive Learning. (1rst edition). Pinter Publishers. https://www-jstor-org.uchile.idm.oclc.org/stable/j.ctt1gxp7cs

Makkonen, T., & van der Have, R. P. (2013). Benchmarking regional innovative performance: composite measures and direct innovation counts. Scientometrics, 94(1), 247–262. https://doi.org/10.1007/s11192-012-0753-2

Meissner, D., & Shmatko, N. (2019). Integrating professional and academic knowledge: the link between researchers skills and innovation culture. The Journal of Technology Transfer, 44(4), 1273–1289. https://doi.org/10.1007/s10961-018-9662-8

Myers, K. R., & Lanahan, L. (2022). Estimating spillovers from publicly funded R&D: Evidence from the US Department of Energy. American Economic Review, 112(7), 2393–2423. https://doi.org/10.1257/aer.20210678

Nguyen, C. P., Schinckus, C., & Su, T. D. (2020). The drivers of economic complexity: International evidence from financial development and patents. International Economics, 164, 140–150. https://doi.org/10.1016/j.inteco.2020.09.004

OECD (2008). Handbook on Constructing Composite Indicators: Methodology and User Guide. OECD Publications.

Onetti, A., Zucchella, A., Jones, M. V., & McDougall-Covin, P. P. (2012). Internationalization, innovation and entrepreneurship: business models for new technology-based firms. Journal of Management & Governance, 16(3), 337–368. https://doi.org/10.1007/s10997-010-9154-1

Park, M., Leahey, E., & Funk, R. J. (2023). Papers and patents are becoming less disruptive over time. Nature, 613(7942), 138–144. https://doi.org/10.1038/s41586-022-05543-x

Pradhan, R. P., Arvin, M. B., Nair, M., & Bennett, S. E. (2020). The dynamics among entrepreneurship, innovation, and economic growth in the Eurozone countries. Journal of Policy Modeling, 42(5), 1106–1122. https://doi.org/10.1016/j.jpolmod.2020.01.004

Puertas, R., Marti, L., & Guaita-Martinez, J. M. (2020). Innovation, lifestyle, policy and socioeconomic factors: An analysis of European quality of life. Technological Forecasting and Social Change, 160(120209), 120209. https://doi.org/10.1016/j.techfore.2020.120209

Schumpeter, J. A. (1939). Business cycles: theoretical, historical, and statistical analysis of the capitalist process. McGraw-Hill.

Solow, R. M. (1956). A contribution to the theory of economic growth. The Quarterly Journal of Economics, 70(1), 65. https://doi.org/10.2307/1884513

Suárez, D. & Natera, J.M. (2024). Métodos para el análisis de los procesos de ciencia, tecnología e innovación. Herramientas para el estudio del desarrollo de América Latina. Vol.2 Métodos cuantitativos. Ediciones UNGS.

Sun, X., Li, H., & Ghosal, V. (2020). Firm-level human capital and innovation: Evidence from China. China Economic Review, 59(101388), 101388. https://doi.org/10.1016/j.chieco.2019.101388

Yang, C.-H. (2018). Exports and innovation: the role of heterogeneity in exports. Empirical Economics, 55(3), 1065–1087. https://doi.org/10.1007/s00181-017-1312-8