On the Measurement of real R&D - Divisia Price Indices for UK Business Enterprise R&D Gavin Cameron * Research Evaluation, vol. 6, no. 3, December 1996, pp. 215-219. In order to assess the contribution of R&D to economic growth it is necessary to measure R&D spending appropriately. One particular problem is the treatment of price changes. This paper argues that in order to measure the quantity of R&D undertaken it is necessary to deflate R&D by a measure of the cost of R&D. The paper constructs price indices for R&D spending in eight sectors of UK manufacturing and for manufacturing as a whole for the period 1970 to 1992. These indices are based on Divisia weighted averages of proxy price series that attempt to represent the diverse components of the R&D process. We argue that these indices are a better guide to the cost of performing R&D than the GDP deflator. Use of the GDP deflator overstates the rise in real Business Enterprise R&D (BERD) in the 1980s, although year on year changes are less distorted. We also argue that Divisia indices are theoretically and empirically better at capturing changes in the cost of R&D than are fixed weighted indices such as the Laspeyres or the Paasche indices. Finally, we investigate the rapidly rising level of pharmaceutical R&D in the UK, and conclude that the rises are genuine, to the extent that they have not been driven by rapid rises in research salaries in that sector. * Gavin Cameron is a Research Officer at Nuffield College, Oxford, OX1 1NF. Tel: +44 865 278653; Fax: +44 865 278621. This research was supported by ESRC grant R000234954. I am grateful to Alan Carter, Mary Dixon, Jeff Golland, Mary Gregory, Louise Kay, John Muellbauer, Steve Redding, Grahame Walshe, and an anonymous referee for helpful comments on an earlier draft. I would like to thank the CSO for providing breakdowns of BERD spending by component; for providing implicit deflators for fixed capital formation by industry; and for answering a variety of questions. I am also grateful to the Department of Trade and Industry and the Employment Department for help with the New Earnings Survey. Nevertheless, all errors and opinions are my own.
1. Introduction Most studies of the effect of R&D spending on economic growth have used the GDP deflator to convert current price spending into constant price spending. 1 However, it is not clear that the cost of R&D follows the path of prices in the economy as a whole. There are two main reasons that we might wish to deflate R&D spending. First, from the point of view of the firm, R&D should be deflated by the price of output, because that represents the amount of resources that the firm must give up in order to undertake R&D. If output prices rise rapidly relative to R&D costs, the firm has to devote a smaller share of output to carry out the same amount of R&D. The crucial point here is that the amount of R&D being undertaken has remained constant. However, from the point of view of an econometrician who wishes to assess how much contribution a constant quantity of R&D makes to productivity growth, such a measure will be far from ideal, because it will fall if output prices rise relative to the cost of R&D (say through a change in the level of competition in the product market) but the amount of R&D undertaken will still be the same. The only way to correct for this bias, from the econometrician s point of view, is to deflate R&D by a measure of its own cost. 2 Several studies have attempted to construct deflators for R&D spending. There have been two main approaches. Some researchers have used surveys to ask firms and institutions about their views of R&D price increases. 3 Other researchers have used weighted averages of proxy price indices of R&D components. 4 A study by Jankowski (1993) is perhaps the most comprehensive and reliable attempt to construct R&D price indices. He used a base weighted (Laspeyres) average of prices for five different factor inputs (engineering and scientist salaries, support personnel, materials and supplies, plant and equipment, and other inputs) to construct deflators for 12 US manufacturing industries between 1969 and 1988. His results suggest that the use of industry-specific R&D deflators produces very different results from the use of either the GDP deflator or an aggregate R&D deflator. The DTI (1979, 1980) used a similar five-factor input approach to that of Jankowski to compile R&D price indices for 7 UK manufacturing industries 2
between 1964 and 1975. The DTI study argued that between 1967 and 1975, the implied price deflator for R&D rose by 151 per cent, compared with a 118 per cent rise in the GDP deflator. A simpler methodology was adopted by Griliches (1984, following Jaffe, 1972), who used a weighted average of the US manufacturing hourly compensation index (49%) and the US non-financial implied deflator (51%). A similar approach was used by Schott (1976), who constructed a UK industrial R&D price deflator based on the CSO distribution of industrial R&D costs; these costs were for wages and salaries (47%), plant, materials and equipment and other (50%), and land and building (3%). These weightings did not appear to change significantly over the period 1948-70, so constant weights were assumed. Schott found that while the GDP deflator had risen by 135% between 1948 and 1970, her measure of R&D costs rose by 163%. The Frascati Manual (OECD, 1992) sets out guidelines for OECD countries to follow in the collection, interpretation, and publication of R&D statistics. The manual acknowledges the need for R&D deflators, and recommends the use of indices that allow for the changing weights of the components of R&D. This is the approach taken by this paper. We construct Divisia indices of the cost of R&D between 1970 and 1992 for seven broad sectors of UK manufacturing - chemicals, mechanical engineering, electronics (including computers), other electrical engineering, motor vehicles, aerospace, and other manufacturing, as well as for the pharmaceuticals industry (a component of the chemicals industry). We then constructed a Divisia index for manufacturing as a whole based on the weights of the seven sectors in real R&D spending, as well as a Laspeyres index for manufacturing as a whole based on the 1970 weights of the seven industries, and a Paasche index based on the 1992 weights. The paper is organised as follows. The diverse components of the R&D process and possible proxy price series are described in section 2. Section 3 discusses the method used 3
to construct Divisia indices of R&D costs between 1970 and 1992, and section 4 discusses the results. Conclusions are drawn in section 5. 3. Data Sources Mansfield, Romeo and Switzer (1983) argue that there are five principal inputs to the R&D process: (1) engineers and scientists, (2) support personnel, (3) materials and supplies, (4) plant and equipment, and (5) other inputs. The surveys of UK BERD that are conducted by the CSO have often included a breakdown of the R&D spending into its various components for each industry. 5 Table 1 shows that for manufacturing as a whole, the proportion of R&D spending devoted to wages and salaries has declined steadily between 1969 and 1989. This trend is largely repeated at an industry level. 6 Table 1. Components of UK BERD (percentages) Component 1969 1978 1981 1985 1989 Total 100 100 100 100 100 Current Spending 90 90 91 90 89 of which wages & salaries 50 48 44 42 42 materials 21 21 23 22 21 other 19 21 24 26 27 Capital Spending 10 10 9 10 11 of which land 3 3 3 2 3 plant & equipment 7 7 6 8 8 In order to compile an R&D price index it is necessary to compile time series data on each of these cost components and to weight them appropriately. We used a number of different data sources to compile the sectoral price indices. Some of these data are 4
industry-specific, and some of them are for manufacturing as a whole. Each cost component will be dealt with in turn. The largest component of R&D cost is wages and salaries. CSO surveys distinguish between three different kinds of personnel working on R&D projects - (1) scientists and engineers, (2) technicians, laboratory assistants and draughtsmen, and (3) administrative, clerical, industrial and other staff. In 1992, scientists and engineers accounted for half of all full-time equivalents engaged in BERD, and the other two categories accounted for one quarter each. New Earnings Survey data on the earnings of these three occupational groups suggest that we should give a weight of two-thirds to scientists and engineers and one-third to support staff. Data on salaries of scientist and engineers in each industry are available in the New Earnings Survey. 7 The salaries of support personnel were proxied by industry-specific data on the earnings of administrative, clerical and technical staff available in the Employment Gazette. The cost of materials was proxied by a simple average of two industry-specific components - the producer price output index and the producer price input index. The category 'other current' relates primarily to administrative and overhead related costs and are proxied by two aggregate indices - the retail prices index (excluding food) and the manufacturing producer prices output index. Lastly, to proxy the cost of the capital component of R&D spending we used the industry-specific gross fixed capital formation deflator. 8 Table 2 summarises the proxy variables used. 5
Table 2. Proxy variables used for R&D Cost components R&D Component Proxied by Industry- Specific? Wages and Salaries Index of earnings of Yes scientists and engineers. Source New Earnings Survey & Employment Department Index of average Yes Employment Gazette earnings of admin, technical and clerical employees. Materials PPI (input) Yes Annual Abstract of PPI (output) Yes Statistics & CSO Other Current RPI (ex food) No Annual Abstract of PPI (input) for all No Statistics & CSO manufacturing Capital Fixed Investment Yes CSO deflator 3. Methodology The simplest way to calculate a price index for R&D is to apply fixed weights to the appropriate proxy price indices. Taking p t as the price at time t, and q t as the quantity at time t, examples of such indices are: (1) P ( 0 p, 1 p ; 0 La q, 1 q ) 1 0 p q / 0 0 p q [Laspeyres price index] (2) P ( 0 p, 1 p ; 0 Pa q, 1 q ) 1 1 p q / 0 1 p q [Paasche price index] 6
When the shares of the cost components of an aggregate are changing over time, an alternative approach to a fixed weight index is to use a Divisia index (see Deaton and Muellbauer, 1980, pp. 174-175, and Diewert, 1976). Instead of comparing discrete situations, a Divisia index analyses the continuous effect of price changes. The Divisia index is a weighted sum of its components growth rates, where the weight for each component is the expenditure on that component as a proportion of total expenditure. Providing there exists a well-defined aggregate and the aggregator function is linearly homogenous, the Divisia index has several desirable properties, the most important of which is that it is consistent with the original optimisation problem faced by the representative consumer. The Törnqvist-Theil discrete time approximation to the continuous time Divisia index (where w 0 and w 1 are the expenditure shares in the two situations) is given by: 0 1 1 1 0 1 0 (3) ln P( p, p, T) = Σ ( wk + wk ) ln( pk / pk ) k 2 We divided manufacturing into seven broad sectors - chemicals, mechanical engineering, electronics (including computers), other electrical engineering, motor vehicles, aerospace, and other manufacturing. We constructed Divisia indices for the cost of R&D between 1970 and 1992 for each of these industries and also for the pharmaceuticals industry (a component of the chemicals industry). We then constructed a Divisia index for manufacturing as a whole based on the weights of the seven sectors in real R&D spending. Furthermore, we constructed a Laspeyres index for manufacturing as a whole based on the 1970 weights of the seven industries, and a Paasche index based on the 1992 weights. 7
4. UK Real Business Enterprise R&D 1970 to 1992 Table 3 shows the resulting price indices. The first four columns represent deflators for manufacturing R&D (the GDP deflator, the manufacturing Divisia index, and the Laspeyres and Paasche indices for manufacturing). The final eight columns show industry-specific Divisia indices of the cost of R&D. Table 3 shows that while the aggregate R&D deflator rises in a similar manner to the GDP deflator, this conceals a wide degree of variation between industries. For example, the cost of R&D in motor vehicles and other manufacturing has risen particularly rapidly since 1985. The Laspeyres and Paasche indices are shown to be a poor guide to trends in aggregate manufacturing. The impression given by table 3 is confirmed by chart 1, which shows UK real manufacturing BERD deflated by two of the aggregate manufacturing indices from table 3 (the GDP deflator and the Divisia index for manufacturing as a whole). The differences between the methods are interesting although not huge. The implicit R&D deflator shows that real manufacturing BERD rose by 7.33 per cent between its trough in 1983 and 1992, while the GDP deflator suggests a rise of 11.43 per cent In 1992, manufacturing (including petroleum and mineral oil refining) accounted for 88 per cent of total Business Enterprise R&D. The remaining 12 per cent was divided between the extractive industries, construction and services and utilities. Although our aggregate index is calculated for manufacturing only, it seems likely that a price index for total BERD would approximate that for manufacturing. Table 4 compares the CSO estimates of real total BERD in the UK (current prices BERD deflated by the GDP deflator) with our estimates of real total BERD (current prices BERD deflated by our aggregate manufacturing R&D deflator). The table shows the differences between the two methods of deflation. The CSO estimated a rise in real total BERD between 1981 and 1992 of 14.8 per cent, while we estimate a rise of 9.7 per cent. Similarly, while the CSO estimated a rise in real BERD of 6.5 per cent between 1985 and 1992, we estimate a rise of 4.8 per cent. 8
Table 3 1985=100 Divisia Price Indices for UK Manufacturing Business Enterprise R&D Manufacturing Deflators Industry-specific Deflators GDP Simple Aggregate Chemicals Mech. Electronics Other Motor Aero- Other Pharmaimplicit R&D R&D Eng. Elec Veh. space Manuf. ceuticals deflator deflator deflator Eng. 1970 19.39 17.97 19.91 19.01 20.13 20.46 20.37 19.94 19.96 19.83 19.18 1971 21.20 19.71 21.71 21.03 22.07 22.18 22.09 21.70 21.66 21.67 21.20 1972 22.93 21.12 23.53 22.82 23.72 24.04 23.91 23.89 23.37 23.49 22.76 1973 24.56 23.04 25.51 24.92 25.45 26.07 26.15 25.95 25.26 25.49 24.77 1974 28.21 27.73 29.78 29.51 29.42 30.11 30.26 29.80 29.71 29.79 29.30 1975 35.89 34.51 36.48 35.89 36.26 37.06 36.78 36.47 36.67 36.14 35.59 1976 41.33 39.83 41.98 41.13 42.11 42.12 42.33 42.25 42.54 41.82 41.12 1977 47.09 46.06 47.35 47.35 47.52 47.01 47.21 47.20 48.47 46.60 47.32 1978 52.43 51.51 53.07 52.91 53.42 53.06 52.93 52.72 54.22 51.76 52.76 1979 59.97 58.24 60.64 60.14 60.52 60.66 60.79 60.58 61.98 59.26 59.53 1980 71.66 67.61 70.49 70.24 69.57 70.50 70.06 71.47 71.09 69.95 69.47 1981 79.84 75.18 77.48 78.49 77.12 76.83 76.48 77.84 78.18 76.69 77.83 1982 85.91 81.81 83.34 83.95 83.36 82.96 82.63 83.55 83.72 82.84 83.60 1983 90.43 87.30 88.52 89.06 88.33 88.25 88.09 88.13 88.54 88.98 88.87 1984 94.60 93.51 94.49 95.13 94.21 94.39 94.37 94.17 94.25 94.36 94.85 1985 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 1986 103.77 106.94 104.90 104.48 105.90 104.33 106.05 106.67 105.15 105.78 106.28 1987 109.03 112.84 110.88 109.85 109.88 110.85 111.41 114.26 110.51 112.14 111.75 1988 115.62 120.06 116.68 115.75 117.44 115.92 117.85 121.34 116.28 119.13 117.91 1989 123.78 129.26 124.78 124.11 126.22 123.75 125.94 130.56 123.72 127.51 124.36 1990 131.68 139.61 133.75 133.46 135.41 132.02 134.46 140.96 133.22 136.36 133.52 1991 140.03 149.21 141.76 141.14 143.74 139.77 142.61 150.61 141.10 144.65 140.76 1992 145.40 156.83 147.75 148.02 148.59 145.45 148.48 155.51 146.45 151.34 148.92 9
5500 Chart1 UK Manufacturing Business Enterprise spending on R&D ( 1985m) 5000 4500 Divisia R&D deflator GDP deflator 4000 3500 3000 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 Year 10
Table 4. Total Business Enterprise R&D at Current and 1985 Prices 1981 1985 1989 1990 1991 1992 % incr. 81-92 % incr. 85-92 At Current Prices 3792 5122 7650 8099 7767 7930 109.1 54.8 Deflated by GDP deflator 4750 5122 6180 6151 5548 5454 14.8 6.5 Deflated by R&D deflator 4894 5122 6131 6055 5479 5367 9.7 4.8 A major trend in UK technological performance over the last decade or so has been the rising importance of R&D in the pharmaceutical industry - which accounted for 9.1 per cent of total BERD in 1985, rising to 16.8 per cent in 1992. Using the GDP deflator, real pharmaceutical R&D rose by 95.5 per cent over that period. Such a major shift of resources into pharmaceuticals R&D might be expected to have led to increased wages and salaries for R&D workers in that sector, as demand for scarce resources bid up their price. We constructed a Divisia index of the cost of pharmaceutical R&D from 1970 to 1992, 9 see table 3 Between 1985 and 1992, the average earnings of scientists, engineers and technologists in the pharmaceutical industry rose by 87.7 per cent, while the average earnings of those workers in all industries rose by 77.4 per cent. Using our Divisia index, real BERD in the pharmaceutical industry is shown to have risen by 90.9 per cent between 1985 and 1992, compared with 95.5 per cent using the GDP deflator. This suggests that the rising importance of pharmaceuticals BERD is being driven only slightly by increased rents to scarce factors in that industry. 11
6. Conclusion This paper has derived Divisia price indices for UK Business Enterprise R&D spending between 1970 and 1992 in eight manufacturing sectors (chemicals, pharmaceuticals, mechanical engineering, electronics, other electrical engineering, motor vehicles, aerospace, and other manufacturing), and hence constructed a Divisia index for manufacturing as whole. The majority of BERD is carried out by manufacturing industry (88 per cent in 1992). It therefore seems likely that our results are also relevant to total BERD. We have four main conclusions. First, we have argued that price indices specifically for R&D provide a better guide to the quantity of R&D undertaken than does the GDP deflator. This is not to say that for certain purposes it might be relevant to look at R&D deflated by some other measure of prices, but no other measure of prices gives a better guide to the quantity of R&D undertaken. Trends in the cost of R&D and in the GDP deflator tend to be similar, as is to be expected, but differences still emerge. For example, our implicit R&D deflator shows that real manufacturing BERD rose by 7.3 per cent between 1983 and 1992, while the GDP deflator suggests a rise of 11.4 per cent. Second, we have shown that the cost of R&D in individual industries can rise at very different rates from that in manufacturing as a whole. These two conclusions agree with those of Jankowski (1993) for the USA. Third, unlike Jankowski, we have also argued that Divisia indices reflect the changing components of the R&D process in a way that is both empirically and theoretically superior to the use of fixed weight indices. Fourth, we also examined the rising share of pharmaceutical R&D in total BERD, and concluded that this rise was probably genuine and not particularly due to R&D workers in pharmaceuticals gaining higher wage increases than those in other industries. 12
Notes 1. Griliches (1980) has the classic discussion of the problems of measuring R&D inputs. 2. There still may be problems, to the extent that the prices of inputs into the R&D process (such as computing power) may not reflect their quality. See Baily and Gordon (1989) for further discussion of these issues. 3. See Cohen and Ivins (1967) for the UK, and Mansfield (1987) for the USA. 4. See Schott (1976), Bosworth (1977), DTI (1979, 1980), Griliches (1984), Jankowski (1993) and Aspden (1994). 5. These breakdowns of spending by each industry were available for 1978, 1981, 1985, 1989 and 1991, and were kindly provided by the CSO. 6. Proportions of spending by component for each of the eight industries are available from the author. 7. These data were supplemented, where necessary, by data kindly supplied by the Employment Department and the Department of Trade and Industry. 8. Capital spending as measured by the CSO BERD survey refers to actual expenditure, rather than to capital consumption. 9. Data on the average earnings of scientists, engineers and technologists in the pharmaceutical industry were supplied by the Employment Department from 1985 onwards. They were not available before 1985, so data from the whole chemicals industry were used for those years instead. 13
Bibliography Aspden, C. (1994) Constant Price Estimation of Research and Experimental Development Expenditure as Practised by the Australian Bureau of Statistics presented at Measuring Research and Innovation for Policy Purposes symposium, Canberra, 1994. Baily, M. and Gordon, R. (1989). 'Measurement Issues, the Productivity Slowdown, and the Explosion of Computer Power' CEPR Discussion Paper No. 305. Bosworth, D. (1977) 'Price Indices for Research and Development in Private Manufacturing Industry', Loughborough University, Dept. of Economics, Occasional Paper No. 18, November 1977. Central Statistical Office (1993) Business Enterprise Research and Development 1992. Cohen, A. and Ivins, L. (1967) 'The Sophistication Factor in Science Expenditure' Department of Education and Science Policy Studies, no. 1. Deaton, A. and Muellbauer, J. (1980) Economics and Consumer Behaviour (Cambridge: Cambridge University Press). Department of Trade and Industry (1979) 'Expenditure at constant prices and employment - some international comparisons' Trade and Industry, 6 April 1979, pp. 34-36. Department of Trade and Industry (1980) ' Industrial expenditure and employment on scientific research and development in 1978' British Business, 8 August 1980, pp. 619-622. Diewert, W. (1976) Exact and Superlative Index Numbers Journal of Econometrics, vol. 4, pp. 115-145. 14
Griliches, Z. (1980) 'Issues in Assessing the Contribution of Research and Development to Productivity Growth' Bell Journal of Economics, pp. 92-116. Griliches, Z. (1984) 'Comment' on Mansfield in Griliches, Z. ed. R&D, Patents and Productivity, (Chicago: University of Chicago Press). Jaffe, S. (1972) 'Construction of a Price Index for Industrial R&D Inputs' (Washington, DC: National Science Foundation). Jankowski, J. (1993) 'Do we need a price index for industrial R&D?' Research Policy, vol. 22, pp. 195-205. Mansfield, E. (1984) 'R&D and Innovation: Some Empirical Findings' in Griliches, Z. ed. R&D, Patents and Productivity, (Chicago: University of Chicago Press). Mansfield, E. (1987) 'Price Indexes for R&D Inputs, 1969-83' Management Science, January 1987, pp. 124-129. Mansfield, E., Romeo, A. and Switzer, L (1983) 'R&D Price Indexes and Real R&D Expenditures in the United States' Research Policy, pp. 105-112. OECD (1992) Frascati Manual: The Measurement of Scientific and Technical Activities: Proposed Standard Practice for Surveys of Research and Experimental Development, (Paris: OECD). Schott, K. (1976) 'Investment in Private Industrial Research and Development in Britain' Journal of Industrial Economics, vol. 25, no. 2, pp. 81-99. 15
1 Griliches (1980) has the classic discussion of the problems of measuring R&D inputs. 2 There still may be problems, to the extent that the prices of inputs into the R&D process (such as computing power) may not reflect their quality. See Baily and Gordon (1989) for further discussion of these issues. 3 See Cohen and Ivins (1967) for the UK, and Mansfield (1987) for the USA. 4 See Schott (1976), Bosworth (1977), DTI (1979, 1980), Griliches (1984), Jankowski (1993) and Aspden (1994). 5 These breakdowns of spending by each industry were available for 1978, 1981, 1985, 1989 and 1991, and were kindly provided by the CSO. 6 Proportions of spending by component for each of the eight industries are available from the author. 7 These data were supplemented, where necessary, by data kindly supplied by the Employment Department and the Department of Trade and Industry. 8 Capital spending as measured by the CSO BERD survey refers to actual expenditure, rather than to capital consumption. 9 Data on the average earnings of scientists, engineers and technologists in the pharmaceutical industry were supplied by the Employment Department from 1985 onwards. They were not available before 1985, so data from the whole chemicals industry were used for those years instead. 16