International comparison of R&D investment by European, US and Japanese companies

International comparison of R&D investment by European, US and Japanese companies M.R. Martínez-Torres Escuela Universitaria de Estudios Empresariales, University of Seville, Avda. San Francisco Javier s/n, 41018 Seville, Spain Fax: 34 954 55 69 89 Email: rmtor[email protected] S.L. Toral * E. S. Ingenieros, University of Seville, C/Enriquez de Ribera, 1, 41092, Seville, Spain Fax: 34 954 48 73 73 Email: [email protected] *Corresponding author Abstract: Looking across the rapidly developing world, it is even more crucial for companies to benchmark their R&D investment against best international practice in their sector and to understand the ways in which their R&D investment will affect future business performance. The globalisation of technology requires a global analysis involving companies all over the world. In this study, top R&D European, US and Japanese companies are analysed and compared using R&D investment scoreboard reports. The main objective consists of extracting clusters of companies with similar R&D policies and comparing the obtained clusters with the major developed economic areas. Results will characterise companies in accordance with different R&D profiles, highlighting some asymmetries in the three major economic areas analysed. Keywords: clustering; factor analysis; R&D investment; R&D policy; strategic groups; technology policy. Reference to this paper should be made as follows: Martínez-Torres, M.R. and Toral, S.L. (2010) ‘International comparison of R&D investment by European, US and Japanese companies’, Int. J. Technology Management, Vol. 49, Nos. 1/2/3, pp.107–122. Biographical notes: María del Rocío Martínez-Torres is an Associate Professor in Management and Business Administration at the Business Administration and Marketing Department, University of Seville. Her main research interests include Intellectual Capital and Knowledge Management and technological acceptance of e-learning tools. She has co-authored articles in many leading academic and professional journals, including Information and Management, IEEE Transactions on Education, Computers and Education and Behaviour and Information Technology. Sergio Toral is an Associate Professor in Digital Electronic Systems at the Department of Electronic Engineering, University of Seville. His main research interests include microprocessor and DSP devices, real-time systems and embedded operating systems. He has co-authored articles in many leading Int. J. Technology Management, Vol. 49, Nos. 1/2/3, 2010 107 Copyright © 2010 Inderscience Enterprises Ltd. academic and professional journals, including IEEE Transactions on Education, Electronics Letters, Computers and Education and Behaviour and Information Technology. 1 Introduction The concept of strategic groups has become an important, intermediate unit of analysis (between firm and industry) in developing theories of competition for the field of strategic management. The commonly accepted definition of a strategic group, according to Porter (1980, p.129), states that ‘a strategic group is a group of firms in an industry following the same or a similar strategy along the strategic dimensions’. McGee and Thomas (1986) and Thomas and Venkatraman (1988) provide a thorough review of both the conceptual framework and current research work in this area, while Harrigan (1985) proposes the application of clustering procedures for strategic group analysis using the ‘Mobility Barriers Paradigm’ (1985, p.56). Further, according to both Harrigan (1985) and McGee and Thomas (1986), it is important not only to identify the presence of inter-group mobility barriers but also to examine how and to what extent these barriers influence competitive activity. As McGee and Thomas state (1986, p.153) ‘Mobility barriers thus reflect the decisions of firms and are a way of defining the set of key strategies available to a firm’. Sudharshan et al. (1991) presented a procedure for identifying those key variables that act as mobility barriers in an industry. One of the most important mobility barriers in an industry refers to research and development policies. Two sets of studies regarding R&D policies can be found in the literature: the first analyses the valuation effects of R&D expenditures by measuring and explaining market price response to R&D increase announcements (Chan et al., 2001), while the second explores the determinants of corporate R&D expenditures (Greeve, 2003). Some other studies have been devoted to comparing firms strongly involved in R&D activities (Martínez-Torres, 2006). Financial policies of R&D intensive firms from the US, Japan, the UK and some other European countries have been analysed by Bah and Dumontier (2001), concluding that they exhibit homogeneous financial policies in spite of their different institutional structures. The impact of globalisation and internationalisation is another topic of interest. As part of the process of globalisation, companies place more emphasis on the global management of technology. As a result, technology and product development processes are increasingly subject to internationalisation (Chiesa, 1996). Several strategic grouping studies (e.g. Cool, 1985; Cool and Schendel, 1987; Dess and Davis, 1984; Fiegenbaum, 1987; Fiegenbaum and Thomas, 1990) have used cluster analysis to form strategic groups. Typically, cluster analysis leads to the development of strategic groups by grouping the companies in an industry based on their ‘scores’ on a set of important strategic variables representing key competitive resources typically chosen by the researcher on the basis of industry studies, expert opinion and judgement. In our paper, we have developed a factor analysis to extract the different strategic groups in top R&D investment companies from Europe, the US and Japan and to identify the kind of firms included in each group. Next, we have compared the results from these economic areas. 108 M.R. Martínez-Torres and S.L. Toral International comparison of R&D investment 109 2 Context of study The strategic and contingent perspectives state that technology intensive companies should use different approaches from traditional organisations. Technology intensive companies have been found to have several common characteristics that differentiate them from traditional enterprises, namely: •a product that is highly advanced technologically •greater priority placed on R&D •frequent innovations •high geographic concentration (e.g. Boston, Nice, Toulouse, Sun Valley) •a high mortality rate •a relatively high percentage of scientists and engineers in the workforce •a high level of intellectual work •an abnormally high turnover rate among technical personnel (Campbell and Guttel, 2005; Cardi and Dobbins, 1995; Gomez-Mejia et al., 1990a,b; Milkovich et al., 1991; Rogers, 2001). Tremblay and Chênevert (2004) hypothesised that technology intensive firms must adapt their strategies to their specific environment to reach their objectives. As outlined above, strategic groups are made up of all those firms following the same or a similar strategy along strategic dimensions. We are going to analyse top R&D investment companies in the three major economic areas of the world, Europe, the US and Japan, in order to identify strategic groups. We hypothesise that we can identify similar patterns of behaviour in these different economic areas as a consequence of globalisation. Ten variables considered in previous studies have been selected to identify the different strategic groups (Schoenecker and Swanson, 2002). Six of them (V1, V2, V4, V7, V9 and V10) show the absolute data of companies not only related to R&D, but to some other features of companies, such as net sales, employees, operating profit, market capitalisation and capital expenditures. Two of them (V6 and V8) are ratios related to the R&D effort measured over net sales and number of employees. Finally, the last two variables (V3 and V5) are related to the evolution of the companies in term of net sales and number of employees. These variables are: •V1: R&D investment •V2: net sales •V3: net sales change 04/03 •V4: employees 2004 •V5: employees change 04/03 •V6: R&D/net sales ratio •V7: operating profit 2004 (% of net sales) •V8: R&D/employees •V9: market capitalisation 2004 •V10: capital expenditures 2004 (% of net sales). The selection of these depended on the availability of data, their homogeneity and the quality and reliability of the information sources. In this sense, it was important to note that the quality of the output depended on that of the input. 3 Methodology for analysis The required information was provided by The 2005 EU industrial R&D investment Scoreboard database (European Commission, 2005). All the firms with available information on all the selected variables were included in the sample. Table 1 presents the number and the percentage of included firms in each economic area considered in the analysis. 110 M.R. Martínez-Torres and S.L. Toral Table 1 Top industrial R&D investment companies from Europe, the US and Japan Industry Europe USA Japan Number % Number % Number % Aerospace and defence 19 2.71 9 2.26 0 0.00 Automobiles and parts 37 5.29 15 3.77 23 11.62 Banks 3 0.43 0 0.00 0 0.00 Beverages 3 0.43 0 0.00 2 1.01 Chemicals 38 5.43 22 5.53 35 17.68 Construction and building 18 2.57 0 0.00 12 6.06 Diversified industrials 10 1.43 8 2.01 3 1.52 Electricity 10 1.43 0 0.00 7 3.54 Electronic and electrical 50 7.14 18 4.52 28 14.14 Engineering and machinery 81 11.57 17 4.27 15 7.58 Food and drug retailers 0 0.00 2 0.50 0 0.00 Food producers 24 3.43 3 0.75 4 2.02 Forestry and paper 7 1.00 3 0.75 2 1.01 General retailers 7 1.00 3 0.75 0 0.00 Health 23 3.29 20 5.03 1 0.51 Household goods and textiles 25 3.57 8 2.01 5 2.53 Insurance 1 0.14 0 0.00 0 0.00 IT hardware 49 7.00 112 28.14 18 9.09 Leisure and hotels 3 0.43 1 0.25 1 0.51 Life assurance 1 0.14 0 0.00 0 0.00 Media and entertainment 13 1.86 4 1.01 5 2.53 Mining 6 0.86 0 0.00 0 0.00 Oil and gas 13 1.86 10 2.51 1 0.51 Other financials 6 0.86 0 0.00 0 0.00 Personal care and household 8 1.14 8 2.01 2 1.01 Pharma and biotech 94 13.43 63 15.83 20 10.10 Software and computer services 83 11.86 60 15.08 4 2.02 Steel and other metals 12 1.71 1 0.25 5 2.53 Support services 26 3.71 7 1.76 0 0.00 Telecommunication services 16 2.29 1 0.25 2 1.01 Tobacco 2 0.29 2 0.50 1 0.51 Transport 4 0.57 0 0.00 2 1.01 Utilities – other 8 1.14 0 0.00 2 1.01 TOTAL 700 100.00 398 100.00 198 100.00% Source: European Commission, 2005. International comparison of R&D investment 111 The sample comprises companies of 33 global sectors from the thirty-five sectors of the FTSE Global Classification System. The aim of the performed analysis consists of determining a small set of underlying dimensions able to summarise the information contained in the set of selected variables using factor analysis. 3.1 Factor analysis Factor analysis is a way to fit a model to multivariate data, estimating their interdependence. It addresses the problem of analysing the structure of interrelationships among a number of variables by defining a set of common underlying dimensions, the factors (Hair et al., 1995), which are not directly observable, segmenting a sample into relatively homogeneous segments (Aguila-Obra and Padilla-Meléndez, 2006; Rencher, 2002). Since each factor may affect several variables in common, they are known as ‘common factors’. Each variable is assumed to be dependent on a linear combination of the common factors and the coefficients are known as loadings. Mathematically, the factor analysis model expresses each descriptor as a linear combination of underlying common factors f1, f2, . . ., fm, with an accompanying error term to account for that part of the variable that is unique (not in common with the other variables). For y1, y2, . . ., yp in any observation vector y, the model is as follows: (2) (1) Equation (1) can be written in matrix notation as in Equation (2), where is the factor loadings matrix. Ideally, mshould be substantially smaller than p, otherwise we have not achieved a parsimonious description of the variables as functions of a few underlying factors. The coefficients ij are called loadings and serve as weights, showing how each yiindividually depends on the underlying factors. With appropriate assumptions, ij indicates the importance of the jth factor fjto the ith variable yiand can be used in interpretation of fj. For instance, f2could be interpreted by examining its coefficients, 12, 22, . . ., p2. The larger loadings relate f2to the corresponding ys. From these ys, a meaning or description of f2could be inferred. It is expected the loadings will partition the variables into groups corresponding to factors. Factor analysis can be used for either exploratory or confirmatory purposes: exploratory analyses do not set any a priori constraints on the estimation of factors or the number of factors to be extracted while confirmatory analyses do. In our case, we have developed an exploratory analysis as we did not know the number of underlying dimensions. Our analysis involved several decision-makings. First, we had to decide the method for extracting the factors. Second, the number of factors to be extracted has to be determined, evaluating how well they fit the original data. Finally, the extracted factors should be named in order to make easier their interpretation. There are several extraction methods (Jolliffe, 1986; Spicer, 2005): principal components and principal axis factoring (or principal factor analysis) are among the most widely used. According to Hair et al. (1995), the former is used when the objective is to summarise most of the original information in a minimum number of factors, whereas the latter is used to identify the underlying dimensions reflecting what the variables share in common. In most applications, both methods arrive at essentially identical results. Since our objective was to obtain a small set of patterns of behaviour in top R&D investment companies, principal components seemed to be the appropriate method. As we wanted to reduce the complexity of our problem, we had to decide on the number of factors to be extracted. There are several criteria for doing this, the most extensive being the eigenvalue and percentage of variance criteria. For the eigenvalue criterion, all of them should be at least higher than 1 to be retained. This guarantees that any factors accounts for at least the variance of a single variable. Meanwhile, the percentage of variance criterion considers all factors accounting for at least 60% (typically but sometimes 80%) of the variance of the original variables (Vicente Cuervo et al., 2006). In our case both criteria suggested retaining three factors, which explained more than 70% of the total variance of the original variables. Once the number of factors has been determined, the next step is to interpret them according to the factor loadings matrix. The estimated loadings from an unrotated factor analysis fit can usually have a complicated structure. Fortunately, an interesting property of loadings is that they can be multiplied by an orthogonal matrix preserving the essential properties of the original loadings. Let Tbe an arbitrary orthogonal matrix, TTI. Inserting TTinto the basic Equation (2): 112 M.R. Martínez-Torres and S.L. Toral (3) (4) Associating Twith and Twith f, the model becomes: It can be demonstrated that the new loadings *Treproduce the covariance matrix (Rencher, 2002). This property is frequently used to facilitate the interpretation of factors. If we can achieve a rotation in which every point is close to an axis, then each variable loads highly on the factor corresponding to the axis and has small loadings on the remaining factors. In this case, there is no ambiguity. The rotated factor analysis fit ensures that factors represent unidimensional constructs. We applied the varimax rotation, introduced by Kaiser in 1958. 4 Results The main applications of factor analysis techniques are to classify and to reduce the number of variables, structuring the relationships between them. Factor analysis attempts to identify underlying variables or factors, which can explain the pattern of correlations within a set of observed variables. In our analysis, each factor represents a distinguished pattern of behaviour of R&D intensive companies. The analysis has been performed for each of the considered economic areas, in order to compare the obtained results. Factor analysis has been performed using the principal component method. The eigenvalues of the sample covariance matrix are shown in Table 2. In factor analysis, it is usual to consider a number of factors equal to the number of eigenvalues higher than 1. We have also considered that the number of factors is able to explain 70% of our variable. In our results, up to three factors satisfy this condition. Table 2 also illustrates the percentage of variance explained by each factor and the cumulative variance. The considered three factors account for more than 70% of the total sample variance and, therefore, represent the ten starting variables sufficiently accurately. International comparison of R&D investment 113 Table 2 Total variance explained Europe USA Japan Initial eigenvalues Initial eigenvalues Initial eigenvalues Total % of Cumulative Total % of Cumulative Total % of Cumulative Factors variance % variance % variance % 1 3.116 31.158 31.158 3.254 32.536 32.536 3.516 35.155 35.155 2 2.797 27.967 59.125 2.770 27.701 60.237 1.938 19.384 54.539 3 1.313 13.133 72.258 1.199 11.989 72.226 1.677 16.767 71.307 4 0.855 8.551 80.809 0.871 8.714 80.940 1.128 11.279 82.586 5 0.725 7.255 88.064 0.778 7.777 88.717 0.546 5.463 88.049 6 0.589 5.895 93.958 0.499 4.986 93.703 0.527 5.268 93.317 7 0.325 3.250 97.208 0.396 3.964 97.667 0.324 3.240 96.557 8 0.154 1.544 98.752 0.158 1.577 99.245 0.218 2.182 98.738 9 0.123 1.230 99.982 0.065 0.653 99.897 0.080 0.796 99.534 10 0.002 0.018 100.000 0.010 0.103 100.000 0.047 0.466 100.000 Using the associated eigenvectors, factor loadings can be estimated. Sometimes, it is difficult to perform the right interpretation of factors using the estimated loadings. Fortunately, factor loading can be rotated through the multiplication by an orthogonal matrix. The rotated loadings preserve the essential properties of the original loadings. The varimax method is an orthogonal rotation method that minimises the number of variables that have high loadings on each factor. This method simplifies the interpretation of the factors. Table 3 reports the rotated factor loadings with varimax rotation for each of the economic areas analysed. To extract the meaning of each factor, we move horizontally through Table 3, from left to right, across the three estimated loadings of each variable, identifying the highest loading and the corresponding factor. To assess significance of factor loadings, a threshold value of 0.6 was considered (Rencher, 2002). The resulting aggregation of variables leads to the following latent factors for each economic area, included in Table 4. 114 M.R. Martínez-Torres and S.L. Toral Table 3 Rotated component matrix Europe USA Japan Component Component Component Variable 1 2 3 1 2 3 1 2 3 VAR1 0.006 0.752 0.007 4.41E- 0.827 0.009 0.111 0.944 0.024 005 VAR2 0.028 0.918 0.030 0.036 0.879 0.073 0.103 0.969 0.030 VAR3 0.030 0.013 0.828 0.017 0.032 0.767 0.024 0.033 0.756 VAR4 0.046 0.837 0.054 0.068 0.854 0.167 0.108 0.904 0.027 VAR5 0.038 0.050 0.775 0.083 0.099 0.627 0.170 0.005 0.769 VAR6 0.977 0.006 0.062 0.991 0.029 0.010 0.883 0.042 0.141 VAR7 0.965 0.000 0.086 0.985 0.037 0.002 0.458 0.032 0.633 VAR8 0.567 0.080 0.176 0.485 0.140 0.470 0.929 0.004 0.001 VAR9 0.015 0.830 0.021 0.024 0.830 0.023 0.076 0.883 0.132 VAR10 0.933 0.024 0.042 0.928 0.006 0.118 0.144 0.271 0.286 Table 4 Latent factors Europe USA Japan (Factor 1) V6: R&D/net V6: R&D/net V6: R&D/net Knowledge intensive sales ratio sales ratio sales ratio companies which V7: operating V7: operating V8: R&D/employees need a high effort profit 2004 profit 2004 in R&D (% of net sales) (% of net sales) (negative) (negative) V8: R&D/employees V8: R&D/employees V10: capital V10: capital expenditures 2004 expenditures 2004 (% of net sales) (% of net sales) (Factor 2) V1: R&D investment V1: R&D investment V1: R&D investment Large companies V2: net sales V2: net sales V2: net sales which consider R&D V4: employees 2004 V4: employees 2004 V4: employees 2004 a strategic value for V9: market V9: market V9: market competition capitalisation 2004 capitalisation 2004 capitalisation 2004 (Factor 3) V3: net sales V3: net sales V3: net sales Quickly growing change 04/03 change 04/03 change 04/03 companies V5: employees V5: employees V5: employees change 04/03 change 04/03 change 04/03 V7: operating profit 2004 (% of net sales) The first factor describes knowledge intensive companies in which R&D spending is the primary way in which firms differentiate themselves. Basically, this factor exhibits a high value of V6 and V7 loading, corresponding to the R&D effort. In Europe and the US, as a difference from Japan, the factor shows a low operating profit margin and high capital expenditures. The low operating profit means a low value compared with companies belonging to the other two factors. The association of R&D and capital expenditures has been tested by Amir et al. (2007) in R&D intensive industries. The second factor is explained by exactly the same variables in the three studied economic areas. This factor comprises large companies where R&D is one of the strategic values for competition. The high value of net sales, employees and market capitalisation confirm that, although R&D investment is higher than in the rest of the factors, the ratios of R&D effort (R&D intensity and R&D per employee) do not reach the values of companies grouped in factor 1. Finally, the third factor is explained by variables related to big changes during the last year, such as net sales change and employees change. As a result, the third group has been defined as quickly growing R&D companies. In the case of Japan, variable V7 (operating profit) is also included in this factor. The mean values of the variables per strategic group and economic area are shown in Table 5. To assess the significance of these results, the null hypothesis of equal population means should be rejected. An analysis of variance (ANOVA) has been performed for this purpose and the obtained results allow us to reject the null hypotheses in all the cases with a significance value below 0.05. International comparison of R&D investment 115 Table 5 Mean values of selected variables per strategic group and economic area Europe USA Japan Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3 Var 01 24.36 756.3 33.90 77.65 1115.51 143.37 208.37 1318.04 132.75 Var 02 432.18 26461.8 1191.18 70.87 28714.83 1240.04 2040.07 31101.20 4556.96 Var 03 11.40 5.6 47.19 11.36 12.09 61.81 1.80 2.11 12.33 Var 04 1459.91 93355.3 6068.32 565.33 85250.95 4857.60 6086.41 102514.05 17389.07 Var 05 4.36 1.4 28.09 1.72 1.79 25.17 3.35 0.21 10.45 Var 06 245.39 4.0 14.95 615.21 6.02 33.41 11.15 3.91 4.09 Var 07 2087.92 9.3 0.27 851.76 12.18 13.79 9.69 6.00 10.40 Var 08 126.66 9.61 21.71 156.89 17.52 65.14 37.69 11.64 10.55 Var 09 1119.24 28322.32 1420.61 691.79 44108.58 5759.46 3758.83 19395.63 4551.28 Var 10 28.31 6.35 6.48 147.93 4.27 7.67 3.67 6.47 6.23 Looking carefully at Table 5, it can be observed that companies from the US included in factor 1 are characterised because their R&D/net sales is three times higher than in Europe and they also have the highest value of capital expenditures. However, companies from Europe, also included in this factor, show very negative operating profit. This has been defined as gross profit minus marketing, engineering and administration costs and Fiegenbaum, A. and Thomas, H. (1990) ‘Strategic groups and performance: the case of the insurance industry, 1970–1984’, Strategic Management Journal, Vol. 1, No. 1, pp.197–216. Greeve, H.E. (2003) ‘A behavioural theory of R&D expenditures and innovations: evidence from shipbuilding’, Academy of Management Journal, Vol. 46, No. 6, pp.685–702. Hair, Jr. J.F., Anderson, R.E., Tatham, R.L. and Black, W.C. (1995) Multivariate Data Analysis with Readings, London: Prentice Hall International. Harrigan, K.R. (1985) ‘An application of clustering for strategic group analysis’, Strategic Management Journal, Vol. 6, pp.55–73. Jolliffe, I.T. (1986) Principal Component Analysis, Springer Series in Statistics, New York: Springer-Verlag New York Inc. Martínez-Torres, M.R. (2006) ‘A procedure to design a structural and measurement model of intellectual capital: an exploratory study’, Information and Management, Vol. 43, No. 5, pp.617–626. Mcgee, J. and Thomas, H. (1986) ‘Strategic groups: theory, research and taxonomy’, Strategic Management Journal, Vol. 7, No. 2, pp.141–161. Milkovich, G.T., Gerhart, B.A. and Hannon, J.M. (1991) ‘The effects of research and development intensity on managerial compensation in large organizations’, The Journal of High Technology Management Research, Vol. 2, No. 1, pp.133–150. Porter, M.E. (1980) Competitive Strategy, New York: Free Press. Rencher, A.C. (2002) Methods of Multivariate Analysis, 2nd edn., Wiley Series in Probability and Statistics, John Wiley & Sons. Rogers, E.W. (2001) ‘A theoretical look at firm performance in high-tech organizations: what does existing theory tell us?’, Journal of High Technology Management Research, Vol. 12, pp.39–61. Schoenecker, T. and Swanson, L. (2002) ‘Indicators of firm technological capability: validity and performance implications’, IEEE Transactions on Engineering Management, Vol. 49, No. 1, pp.36–44. Spicer, J. (2005) Making Sense of Multivariate Data Analysis, California: Sage Publications. Sudharshan, D., Thomas, H. and Fiegembaum, A. (1991) ‘Assessing mobility barriers in dynamic strategic groups analysis’, Journal of Management Studies, Vol. 28, No. 5, pp.429–438. Thomas, H. and Venkatraman, N. (1988) ‘Research on strategic groups: progress and prognosis’, Journal of Management Studies, Vol. 25, pp.537–556. 122 M.R. Martínez-Torres and S.L. Toral