SEARCH

SEARCH BY CITATION

Keywords:

  • L1;
  • D1

Abstract

  1. Top of page
  2. Abstract
  3. I.Introduction
  4. II.R&D spillovers and the role of human capital
  5. III.The empirical strategy: the production function specification, the data sets and the variables
  6. IV.The empirical results
  7. V.Concluding remarks
  8. References
  9. Appendices
  10. Supporting Information

Economists have long agreed that the local availability of a more qualified workforce generates significant spillovers. This study suggests that these externalities may arise because plants by having access to a more qualified workforce at a regional level, can benefit more from R&D spillovers than those located in areas with less qualified workforce. This hypothesis is tested on a sample of British establishments drawn from the Annual Business Inquiry over the period 1997–2002. The main results are consistent with our expectations that the regional differences in the industry-level educational attainment of the workforce available to a plant will condition its capability of absorbing R&D spillovers.


I.Introduction

  1. Top of page
  2. Abstract
  3. I.Introduction
  4. II.R&D spillovers and the role of human capital
  5. III.The empirical strategy: the production function specification, the data sets and the variables
  6. IV.The empirical results
  7. V.Concluding remarks
  8. References
  9. Appendices
  10. Supporting Information

Economists have long agreed that the local availability (both at regional and county level) of a more qualified workforce generates significant spillovers for the firms located in that same area (Marshall, 1890; Ciccone and Peri, 2006; Moretti, 2004; Galindo-Rueda and Haskel, 2005). The quality of the pool of workers from where a firm draws its employees can affect its productivity in many ways. For instance, by locating in regions with a more educated workforce, firms can hire workers who are able to use the firm's tangible inputs in the most effective way. More qualified workers can as well enhance the productivity of their co-workers (as long as there are complementarities among their tasks) so creating an additional mechanism that allows a firm to experience productivity gains (Battu, Belfield and Sloane, 2003).

However, there are other channels through which larger concentrations of human capital in an area can help firms to experience productivity gains. In this study, we suggest that proximity to a more qualified workforce can facilitate the local firms’ absorption of R&D spillovers and through this route help increase their productivity. This may happen in several ways. For instance, informal contacts among ‘co-located’ workers with the same level of education may facilitate the flow of information not only in the industry but across industries as well (Jaffe, 1989; Audretsch and Feldman, 1996). In addition, a high proportion of graduates in an area may generate R&D spillovers from the local university to the firms in the same area. Finally, mobility of highly specialized workers from one company to another in the same geographical area is another established channel of creation and diffusion of R&D spillovers (Moen, 2005; Crespi, Geuna and Nesta 2007). However, no attempt has been made so far to quantify the productivity gains a firm experiences through this channel of absorption of R&D spillovers. Yet understanding the extent to which this mechanism of absorption of R&D spillovers can help firms to increase their productivity is quite important for policy-makers as it will allow them to identify more effective measures that facilitate knowledge transfer (see for instance European Commission, 2000 and BIS, 2008). Therefore, the purpose of this study is to quantify the extent to which a firm by having access to a more qualified workforce at the regional level, can benefit more (in terms of larger output) from R&D spillovers than firms located in areas with a less-qualified workforce.

Our empirical analysis is conducted on a sample of British plants, sourced from the British Annual Business Inquiry (ABI). Great Britain is an excellent case-study for our purpose. Indeed, a feature of the British economy is the existence of a substantial heterogeneity in the economic performance of its regions.1 These regional differentials appear to be rather persistent over time and can be mostly ascribed to regional variations in total factor productivity (TFP).2 Underlying these productivity differentials is a substantial regional variation in a number of economic factors, including different endowments of public, human and physical capital (Duranton and Monastiriotis, 2002; Rice and Venables, 2003). These differentials are important to explain the ‘productivity gap’ the British economy suffers from and not surprisingly, policy-makers have tried to devise policies that can address and reduce them. Among these, a special emphasis has been given to the regional differentials of educational attainment. The figures and the facts are well known: the average educational attainment in underperforming regions is below the average during the compulsory schooling years. These differentials are reinforced after graduation as graduates tend to relocate to regions with more productive jobs.3

For our empirical analysis, we estimate an augmented production function on a panel of plants drawn from the British ABI over the period 1997–2002 using the System-GMM estimator (Blundell and Bond, 2000). Lagged measures of both inter-industry R&D spillovers and regional differences in the educational attainment of the workforce across different industries (and their interactions) are introduced among the regressors to gauge the extent to which faster absorption of R&D spillovers can benefit the plants in terms of larger output. The ABI does not contain information on either workers’ educational attainment or plants’ stock of R&D. Therefore, we have to rely on additional data sources: the industry-specific educational attainment of workers is obtained by using information on their qualifications from the Quarterly Labour Force Survey (QLFS), while the measures of R&D spillovers have been computed using the information on the R&D activities conducted by private firms recorded by the Business and Enterprise Research and Development (BERD) data set.

Our results are consistent with the hypothesis that the educational attainment of the workforce available to a plant will condition its capability of absorbing R&D spillovers. Indeed we find that plants located in areas where the industry-specific educational attainment is closer to the ‘educational attainment’ frontier (or the highest value of the industry-specific educational attainment across all the regions) can absorb R&D spillovers faster and benefit more from them than plants located in areas where the distance from the ‘educational attainment’ frontier is larger. This result holds true for spillovers generated at the three different levels (national, regional and county level) of geographical disaggregation. However, this does not hold true for small and medium size firms which depend on the availability of educated workforce in the region where they are located for the absorption of spillovers at county level only. Different assumptions on the technology used by the plants give similar results; equally, our results are robust to alternative ways of constructing the R&D spillovers.

The rest of the article is structured as follows. In section II, we provide a brief overview of some of the themes and outcomes of the empirical literature on the spillover hypothesis and human capital. Our empirical strategy is explained in section III, along with the data sets and the details on how variables have been constructed. The main results are presented and discussed in section IV. Finally, some concluding remarks are offered in section V.

II.R&D spillovers and the role of human capital

  1. Top of page
  2. Abstract
  3. I.Introduction
  4. II.R&D spillovers and the role of human capital
  5. III.The empirical strategy: the production function specification, the data sets and the variables
  6. IV.The empirical results
  7. V.Concluding remarks
  8. References
  9. Appendices
  10. Supporting Information

The hypothesis that technological knowledge acquired by a firm can spill over to other firms and so enhance their total factor productivity4 was first suggested by Arrow (1962) in his work on the effects of learning embodied in new capital equipment. According to this view, individual firms produce technological knowledge. At first, this is private to the firm; afterwards, it spills over to the rest of the economy as it can be copied immediately and at almost no cost by any number of firms, becoming social knowledge acting as an external effect in enhancing the productivity of all firms (Romer, 1986; Grossmann and Helpman, 1990). Since then, there has been a considerable theoretical and empirical debate on the extent to which a firm can benefit from spillovers, how much of its productivity growth such spillovers can explain and the factors that can facilitate their absorption (Jaffe, 1986; Romer, 1986; Bernstein and Nadiri, 1989; Cohen and Levinthal, 1989; Grossmann and Helpman, 1990; Raut, 1995). On this last point, Cohen and Levinthal (1989) suggested that the cost of fruitfully utilizing knowledge which is already in public domain is minimal only for firms which have accumulated sufficient ‘technological capability’ to absorb external knowledge (so-called absorption view). One obvious implication of the absorption view is that recipient firms need to carry out some R&D activity on their own before they can benefit from knowledge spillovers. However, there may be other factors that can help the assimilation of spillovers: for instance the stock of human capital of the recipient firms. Not surprisingly, a few empirical studies have tried to understand the extent to which the diffusion and the absorption of R&D spillovers are enhanced by firm-specific human capital. Early studies have concentrated on international R&D spillovers and therefore tested the role of countries’ human capital in helping the diffusion of R&D spillovers in a cross-country setting. Romer (1993) examines the interaction between imports of technologically advanced goods (machinery and equipment) and the level of human capital in a cross-country growth regression and finds that the term is significantly positive. Benhabib and Spiegel (1994) show some evidence suggesting that human capital can help the process of technology diffusion. Engelbrecht (1997) finds that human capital affects total factor productivity both directly as a production factor and indirectly by enhancing the effect of international spillovers.5 It is possible to conceive other mechanisms through which human capital affects the diffusion of R&D spillovers. For instance, theories of localized R&D spillovers emphasize the relationship between the geographical density of human capital and R&D spillovers. Their starting point is that geographical proximity among firms reduces the cost of accessing and absorbing R&D spillovers. Indeed firms located in agglomerations and spatial clusters will have higher profits than their isolated counterparts as they access external knowledge at a cost that is lower than the cost of producing it internally or of acquiring it externally from a geographical distance (Harhoff, 2000). The cost of transferring such type of knowledge is a direct function of geographical distance and therefore gives rise to localized externalities (Siegel, Westhead and Wright, 2003). The implicit assumption is that there is a specific type of knowledge, so called tacit knowledge, which cannot be patented and that therefore can only be transmitted through direct contacts between the source and the recipient. This is typically true for basic research that generates new fundamental ideas. In spite of the fact that the core work can be made available through normal public codified channels (e.g. scientific journals), there is still a considerable portion of the research that can only be conveyed via direct interaction and discussions with scientists (Poyago-Theotoky, Beath and Siegel 2002). Human capital affects the speed by which knowledge is diffused and absorbed across firms in several ways. Indeed, in the case of geographically bounded spillovers, knowledge diffuses through informal contacts among workers (such as industry conferences, talks and seminars) made possible because firms (and more importantly, individuals working for them) share the same location, something which decreases the cost of participation to these activities. On these occasions, potential adopters of innovations (who have limited information about costs and benefits of the innovations) come in contact with existing users, so the diffusion of intangible technological capabilities is promoted. Obviously the more qualified the users, the easier is the exchange and the absorption of new information, so suggesting that firms located in areas with a higher density of human capital will benefit more from existing R&D spillovers. Also localized R&D spillovers may be facilitated by workers’ mobility from one firm to the other (Geroski, 1995). Indeed, the R&D capital of high-tech firms is mostly embodied into its employees and the mobility of (for instance) technical personnel (along with the human capital embodied in them) across firms is a substantial source of knowledge externality (Moen, 2005). Another source of localized R&D spillovers is the mobility of graduates from the academia to local firms. Indeed, if we believe that in the case of tacit knowledge personal contacts are the main source for absorbing new knowledge, then the human capital embodied in students graduating from a research university may be another important mechanism through which R&D spillovers can be diffused (Saxenian,1994).6 In this case, R&D spillovers will be more beneficial to firms located in areas where the proportion of graduates is higher. From this survey we can draw the following conclusions: from a theoretical standpoint it seems uncontroversial human capital facilitates the absorption of R&D spillovers. This is particularly true for ‘geographically bounded’ spillovers where the local density of human capital can help the absorption of locally generated spillovers. This implies that in our empirical analysis we cannot simply focus on general, nationally generated spillovers but that in order to appreciate the role that human capital plays in this respect we need to focus also on geographically disaggregated spillovers.

III.The empirical strategy: the production function specification, the data sets and the variables

  1. Top of page
  2. Abstract
  3. I.Introduction
  4. II.R&D spillovers and the role of human capital
  5. III.The empirical strategy: the production function specification, the data sets and the variables
  6. IV.The empirical results
  7. V.Concluding remarks
  8. References
  9. Appendices
  10. Supporting Information

The production function specification

We assume that the technology used by our plants can be described by the following production function:

  • image(1)

where yitks is the (log) output of plant i, belonging to industry k, located in the region/county s/c, at time t. kitks, litks, mitks and ritks denote the set of (log) inputs (namely capital, labour, materials and the stock of R&D) while the error term ωitks includes a time-invariant fixed effect and an AR(1) error term ϑitks. aitks is a measure of the plant's time-varying total factor productivity (TFP). In turn, TFP is influenced by a whole host of additional factors. We assume that it depends on the lagged values of the inter-industry R&D spillovers across the two-digit industries (SPILL), the lagged value of the distance from the educational attainment frontier (henceforth the gap variable – GAP), the interaction between these two terms, a set of additional controls (age, ownership, etc.) and the unobserved productivity shocks. These include the permanent characteristics of the regions (such as the quality of the public infrastructure, the presence of research universities and the efficiency of local authorities), the industry characteristics, general trends in technology and so on. In our analysis, we decide to focus on inter-industry R&D spillovers (Bernstein and Nadiri, 1989). In other words, we do not consider spillovers that occur within a plant (as we expect these would be internalized) and spillovers that happen within the same industry. Therefore our estimates of the impact of the spillover variable can be considered as a lower bound on the magnitude of the total spillovers (Moretti, 2004). Also, we acknowledge that plants located in a region can benefit (in terms of larger output) from increases in the average educational attainment in the neighbouring regions and if this effect is not controlled for, increases in plants’ productivity could be erroneously attributed to improvements of the educational attainment in the region where the plant is located; hence, we introduce among the variables that may affect a plant's TFP the value of the industry-specific GAP variable of the neighbouring regions (lagged one period so as to be consistent with the other variables of interest). So the first empirical specification of TFP is as follows:

  • image(2)

where ɛs are the unobserved productivity shocks at the region, industry and year level, respectively, s* denote the neighbouring regions and Z collects the additional variables that affect TFP. In equation (2) we assume that as the average educational attainment in a region increases in each industry (in other words, the distance from the frontier reduces), plants can absorb faster the inter-industries R&D spillovers (generated across the country) and so experience an increase in productivity through this mechanism.

We also allow for the fact that plants can benefit more from the research activities carried out by plants located in the same geographical region or county (in addition to the national level R&D spillovers) as the links between these plants may be stronger.7 As mentioned in the literature survey presented in section II, the geographical density of human capital is particularly relevant in the case of geographically bounded spillovers and so we would expect this mechanism of absorption of R&D spillovers to have a positive impact on plants’ TFP. However, how to define the geographical boundaries of the R&D spillovers? We assume that geographical spillovers are delimited by the administrative boundaries of regions and counties and therefore, we have constructed measures of inter-industry R&D spillovers at both regional and county level with the expectation that a more substantial portion of productivity growth experienced by the plants may derive from these ‘geographically bounded’ spillovers. So the plant's TFP depends on the (lagged) across two-digit industries localized (namely regional and county-level) R&D spillovers, the lagged GAP variable of the region where the plant is located (together with the values of the GAP variable of the neighbouring regions), the interaction term between these two variables, the additional set of controls and the usual unobserved productivity shocks. However, we do acknowledge that this is not always the case as there may be spatial agglomerations that span across the administrative boundaries and therefore plants in a region (or a county) may benefit from the R&D spillovers generated by plants in the neighbouring regions or counties. So, as we did with the educational gap variable in the previous specification, we introduce among the determinants of plants’ TFP the R&D spillovers from the neighbouring regions (for the specification of TFP that focuses on R&D spillovers generated at regional level) and from the neighbouring counties (for the specification of TFP that focuses on R&D spillovers generated at county level). Both variables are lagged one period.

Therefore the two additional specifications of the TFP are the following:

  • image(3)
  • image(4)

where the subscripts s and c denote the regions and the counties, respectively and c* denote the neighbouring counties. With these additional specifications, we assume again that the industry-specific educational attainment of the workforce available in the region may be particularly important for the absorption of locally generated R&D spillovers and that plants may not benefit completely from these spillovers if located in areas where the average educational attainment of the workforce in industry is below the maximum in the industry. Also, notice that in equation (4) we control for county-specific shocks. Plants’ TFP can also be affected by time-varying industrial and regional shocks that can be driven by changes in the structure of a specific industry over time, changes in the regional business conditions or changes in the R&D expenditure of Higher Education Institutions.8 Therefore, in the most robust specifications, we will also control for time-varying industry*region-specific productivity shocks.9

The econometric procedure we follow is quite simple. We first combine together equations (1) and (2), then equations (1) and (3) and finally equations (1) and (4). We then estimate the resulting three production functions (augmented by the TFP specification) with the Sys-GMM estimator.10 Afterwards we will test whether our hypotheses hold for different types of plants, first and the extent to which our results are robust to different assumptions on the technology and types of R&D spillovers, afterwards.

The data sets and the variables

For our empirical analysis we have focused on manufacturing and used a variety of data sets. Our starting data set is an unbalanced panel of single-unit plants, sourced from the ABI data set, 1997–2002. The ABI is very similar in structure to the US Longitudinal Research Database (LRD), being a population sample of large plants and a stratified random sample of smaller plants. The ABI contains all the basic information (namely the inputs – except the stock of R&D – and outputs) needed to estimate the production function. ABI allows to distinguish between plants that belong to multi-unit firms and plants that do not (so called single-unit plants). However, in the case of multi-unit firms, it does not allow to identify the plants where production takes place and therefore we have decided to focus on single-unit plants as this way we can identify with certainty where production takes place. The measures of the industry-specific educational attainment at regional level have been computed by using information drawn from the QLFS, 1997–2002. Finally the measures of R&D spillovers have been calculated from the BERD, 1997–2002. A full description of the data sets is in Appendix A2. In the remainder of the section we will explain the procedures we have used to construct the variables for our empirical analysis.

Production function variables

Output is measured by the gross output deflated by the four-digit PPI deflators 1995. Labour input is measured as the number of employees. As ARD does not record the number of skilled and unskilled workers that each plant hires, we decided that measuring the labour input just as the number of employees would bias the results in favour of plants that hire a more educated workforce. Therefore we control for the distribution of skills among the workers in each plant by introducing in the TFP specification the ratio between the plant-level average real wage and the average wage of the three-digit industry (see also, Wakelin, 1998).11 To avoid possible endogeneity we have decided to use the lagged value of this variable. Intermediate inputs are obtained by using the information on input purchases and are deflated by the materials deflator made available by the British Office for National Statistics (ONS). As for capital, the ABI does not contain information on capital stock. However, stocks of capital have been constructed by using the perpetual inventory method and plant-level information on investment (see Martin 2002, for more information) and have been made available at the ONS Virtual Lab in London. As mentioned above, among the inputs we have introduced the plant's own stock of R&D. This variable is not available in the ABI data set and therefore it has been sourced from elsewhere, namely from the UK BERD, 1997–2002. To this purpose, we have matched the plants from our data set with the ones contained in BERD and computed the stock of R&D for our sample of plants by using the perpetual inventory method (Bond et al., 2002). Information on the plants’ age (AGE) has been sourced from the ABI as well, along with information on whether a plant is owned by a foreign company or not. Finally, missing observations and zero values on relevant variables were removed; also 1% of the observations on both the upper and lower tails of the data distribution has been removed. So our final dataset is made of 8,617 observations over the period 1997–2002; of these 6,319 are small and medium size enterprises (SMEs).12 Finally all the data have been weighted so that they could be representative of the whole population of plants. Table 1 presents some selected descriptive statistics for the plants’ output and inputs. The data in the table are neither weighted nor deflated.

Table 1. Selected statistics for gross output and inputs (1997–2002)
  Plants (single-unit) Mean SD
  1. Notes: The figures are undeflated and unweighted.

  2. Source: ONS

  3. *Number of employees.

  4. In thousands pounds.

Employees*8,6174461,088
Gross output 79,8073,11,831
Materials 51,7182,33,039
Capital 75,2914,56,863
R&D exp. 1,51510,648
R&D spillovers

R&D spillovers have been computed by using the micro data underlying the annual UK BERD, 1997–2002, matched to the ABI. The methodology followed in this paper builds upon the seminal work of Terleckyj (1974) which used input-output data to measure inter-sectoral flows of technologies.13 We have constructed three different measures of inter-industry R&D spillovers at distinct levels of geographical disaggregation. The first measure proxies for inter-industry externalities from other two-digit industries; the second spillover variable tries to capture R&D spillovers from other two-digit industries located in the same region; while the third measure proxies for externalities derived from other two digit industries located in the same county. To obtain a proxy for domestic inter-industry spillovers, the micro data on intramural R&D spending14 have been first aggregated to the industry level using the industry and/or the region/county of the R&D reporting unit for the years 1997–2002. We have then calculated the industry R&D capital stock using the perpetual inventory method. Given the short data period, to obtain an estimate of the initial R&D capital stock in each industry, we have imposed that the initial R&D stock of an industry that is first observed is a share of the R&D stock of the two-digit industry, obtained from the STAN database.15 Finally, to obtain a measure of the embodied R&D spillovers, we have then weighted the real inter-industry R&D stock, using the output multipliers. The three proxies for inter-industry R&D spillovers can be then defined in the following way:

Inter-industry R&D spillovers
  • image(5)

Here, Bjt is the real R&D stock of industry j, at time t and ωjk is the (j,k) element of the output-to-output Leontief inverse.

Inter-industry regional R&D spillovers
  • image(6)

where Bjst is the real R&D stock of industry j, in region s, at time t and ωjk is the (j,k) element of the output-to-output Leontief inverse.

Inter-industry R&D spillovers at county level
  • image(7)

where Bjct is the real R&D stock of industry j, in county c, at time t and ωjk is the (j,k) element of the output-to-output Leontief inverse. Table A1 of the Online Appendix S1 shows the average growth rates of R&D and of the ‘national’ R&D spillovers for the two-digit industries over our sample period.

Educational attainment and the GAP variable

We decide to capture the differences in the educational attainment of the workforce across regions and industries by constructing an indicator of the (industry-level) educational attainment gap between two regions. More specifically, our measure of the gap (GAP) is the ratio between the (weighted) average educational attainment of the workforce in the two-digit industry k and region s (at time t) and the maximum (weighted) average educational attainment of the workforce in industry k across all regions at time t, where the weights are given by the coefficients of a Mincerian equation estimated for Great Britain (Harmon, Walker and Westergaard-Nielsen 2001).16 Formally:

  • image(8)

Where EdAtt is the weighted average educational attainment and EdAtt* is the maximum weighted average educational attainment across all the regions at time t. The denominator of the ratio can be considered as the industry-specific, regional ‘educational attainment’ frontier and therefore the ratio captures a region's distance from this frontier. In this formulation, the ratio varies between zero and one: so, the larger the ratio, the larger the industry-specific (weighted) average educational attainment in a region (with respect to the maximum in the industry); this implies that a plant located in this same region may have access to a more educated workforce and eventually be able to absorb R&D spillovers faster than plants located in regions where this ratio is smaller. In some sense this can be interpreted as a measure of the industry-specific regional capability to absorb new knowledge in the same spirit as Griffith, Redding and van Reenen (2004) and Girma (2005). Notice that in the actual regression analysis, we include the inverse of this variable to facilitate the interpretation of the sign of the coefficient.

The industry-specific educational attainment of the workforce at regional level has been computed by using the data from the QLFS. The Labour Force Survey questionnaire collects information about the respondents’ educational achievements and participation. Respondents are asked to list all of their qualifications and from this list the highest qualification is determined. We decide to focus on the academic qualifications (GCSE, A level and Degree or higher) of respondents between 16 and 64 years old (variable HIQUALD – see Appendix A1). We have dropped observations for which no information on education is available. For each qualification we have computed the number of schooling years necessary to get it. We have then averaged these schooling years across regions, two-digit industries and years, with each qualification being weighted by the parameters of the Mincerian equation.17 Information on the two-digit industry and the region where the respondent works is available in the QLFS. The classification of regions in the QLFS is different by the one used by the ABI data set. Therefore before matching the variables on regional human capital to the production function variables, we had to re-calibrate the regional classification for the human capital variables so that it could match the one contained in ABI inquiry (see also the Appendix A2). Table A2 of the Online Appendix S1 shows the mean and the standard deviation of the GAP variables by industry averaged over the sample period and regions. The largest value of the GAP variable is in the Office Machinery (SIC 30) sector while the sectors with the smallest values of the educational gap variable are the Rubber (SIC 25) and Leather (SIC 19) sectors. The dispersion of the GAP variable across the regions is the smallest in the Chemical sector (SIC 24) sectors while it is at its largest in the Tobacco sector (SIC 16).

Spatial lags

As mentioned above, in our analysis we are interested in the geographical scope of the R&D spillovers and of the GAP variable and of course we do not want to constrain this effect to purely within-region/county effects. Therefore, as in Acs et al. (2002) and Bode et al. (2009), we construct spatially lagged values of the regional and county-level R&D spillovers and of the GAP variable. To this purpose, we assume that R&D activity carried out in a region (county) may have a positive impact on the productivity of firms located in neighbouring regions (counties). Equally, we assume that the educational gap in a region may have a negative impact on the plants located in the neighbouring regions. Therefore for each region (or county), we define the spatially lagged regional (or county-level) inter-industry spillovers as the spatially weighted sum of the inter-industry R&D spillovers generated in regions (or counties) surrounding the region (or the county) of interest. Equally, we define the spatially lagged GAP variable as the spatially weighted average of the GAP variables in the regions surrounding the region of interest. In both cases the weights take the value of one for the regions (counties) that share a common border and 0 otherwise.

IV.The empirical results

  1. Top of page
  2. Abstract
  3. I.Introduction
  4. II.R&D spillovers and the role of human capital
  5. III.The empirical strategy: the production function specification, the data sets and the variables
  6. IV.The empirical results
  7. V.Concluding remarks
  8. References
  9. Appendices
  10. Supporting Information

The production function estimates

Our main results are shown in Table 2.18 More specifically, Table 2 presents the estimates of the parameters of the (restricted) production functions. The estimates in panel A control for time, sector and regional effects while in the specifications shown in panel B, we control for time-varying industry*region effects as well. In each panel, R&D spillovers are then measured at national (column 1), regional (column 2) and county level (column 3). Standard errors are clustered around region and industry. Across all specifications, the sum of the coefficients on the inputs is close to unit, or very close to constant returns to scale. The signs of the input variables are as expected and they are mostly significant at 5%. The R&D variable is significant at 10% in some specifications, but this appears to be consistent with the findings from previous studies (see Wakelin, 2001). The diagnostic tests suggest that the models are well specified. Indeed, the Hansen test for the overidentifying restrictions appears to validate the choice of instruments while the Arellano–Bond tests on the autocorrelation of the residuals support the choice of the System-GMM estimator.19 Finally, we have tested whether problems of weak instrumentation affect the quality of the instruments generated by the GMM estimator.20 At the moment, there is no formal test of weak instruments in a dynamic panel setting and therefore we have followed the procedure suggested by Bazzi and Clements (2010). The main idea is to estimate with 2SLS the main equation first in levels and then in first difference. In the former case, the endogenous variables will be instrumented by their lagged first differences while in the latter case they will be instrumented by their lagged levels. Afterwards, a test for weak exogeneity (for instance the Kleibergen–Paap (KP) (2006) test)21 can be used to test whether the instruments jointly explain enough variation of the endogenous regressors. If the tests suggest that both sets of instruments are weak, then this result casts great doubt on the ability of the system GMM estimators to yield strong identification. In our case, all our instruments pass the KP test and this suggests that the quality of our estimates is not affected adversely by weak instrumentation.

Table 2. Production function estimates
Dependent variable: gross output (ln Y) Panel A Panel B
  Coefficients Coefficients Coefficients Coefficients Coefficients Coefficients
  1. Notes: The dependent variable is the log of gross output. The time period is 1997–2002. Specifications reported in panel A include time, sectorial and regional dummies. Specifications reported in panel B include time, sectoral and regional dummies as well as industry*time and sector*time dummies. t-ratios are in the parentheses below the coefficients. Robust standard errors are used to compute the t-ratios. One step GMM results reported. All the specifications include age of plants, the skills variable and the dummy for foreign ownership. Serial correlation tests are LM tests of the first differenced residuals (see Arellano and Bond, 1991). Sargan–Hansen test of instrument validity is a test of the over-identifying restrictions. For the specification with county-level spillovers, the matrix of instruments has been collapsed to drop multicollinear instruments.

  2. Source: ONS.

Employment (ln L)0.1890.1880.1700.18330.18310.1733
(3.67)(4.78)(3.69)(4.46)(5.23)(2.87)
Materials (ln M)0.680.6890.690.61000.67360.6769
(3.90)(3.09)(12.89)(11.81)(14.14)(11.84)
Capital (ln K)0.1870.1350.1500.18600.13150.1411
(4.36)(4.78)(3.89)(3.11)(2.75)(2.38)
R&D (ln RD)0.0400.01550.01170.0380.01450.0104
(1.49)(1.50)(1.99)(0.81)(0.37)(2.41)
National spillovert−10.0072  0.0069  
(1.69)  (1.66)  
Gapt−1−0.010  −0.009  
(−1.78)  (−1.91)  
National spillovert−1*Gapt−1−0.0009  −0.0009  
(−1.98)  (−1.72)  
Regional spilloverst−1 0.0067  0.0063 
 (1.89)  (1.73) 
Gapt−1 −0.008  −0.006 
 (−1.89)  (−1.69) 
Regional spilloverst−1*Gapt−1 −0.0001  −0.0098 
 (−2.00)  (−1.92) 
County spilloverst−1  0.0059  0.0052
  (1.79)  (1.86)
Gapt−1  −0.009  −0.006
  (−2.68)  (−1.96)
County spilloverst−1*Gapt−1  −0.0007  −0.00064
  (−2.89)  (−2.00)
Spatially lagged Gapt−1−0.0014−0.0013−0.0050−0.0011−0.0012−0.0049
 (0.78)(0.07)(0.99)(0.78)(0.10)(0.89)
Spatially lagged Regional spilloverst−1 0.00018  0.00016 
  (1.64)  (1.60) 
Spatially lagged County spilloverst−1  0.0022  0.0020
   (1.66)  (1.69)
Hansen test (P-value)0.6780.1560.4630.4200.2630.309
AR(1) (P-value)0.0000.0000.0000.0000.0000.000
AR(2) (P-value)0.7980.4580.9810.8070.5150.781
KP test: Levels eq.PassPassPassPassPassPass
KP test: Diff. eq.PassPassPassPassPassPass
No. observations657865784890657865784890

Our variables of interest – the educational attainment gap variable, the R&D spillovers at different levels of geographical disaggregation and the interactions between these terms – appear to be all significant and have the expected signs in the production function across all specifications. On average plants located in regions where the educational attainment of the available workforce in an industry is closer to its frontier tend to absorb R&D spillovers faster and experience an increase in productivity. Interestingly enough, this is true not only for inter-industry spillovers which are independent of the geographical distance but also for geographically bounded spillovers. Of course, the size of the impact on plants’ productivity of a 1% increase of the R&D spillovers varies with the size of the GAP variable and according to the type of R&D spillovers. To this purpose, consider first the case when the GAP variable is equal to one (i.e. the average educational attainment in region s, industry k, at time t, has caught up with the maximum regional-level average educational attainment in industry k, at time t). In this case a 1% increase of the R&D spillovers measured at the national level increases plants’ productivity by around 0.63% while for regional (county) based R&D spillovers, it may increase by 0.66% (0.58%). It is important to notice that the magnitude of these effects is not very large as in similar studies (see Higon, 2007). Also the impact on productivity of an increase of regional R&D spillovers is larger than for the other types of R&D spillovers and this implicitly confirms the finding that British plants rely on local network of suppliers as the main source of R&D spillovers. Assume now that the average educational attainment in industry k, in the best performing region s, at time t, doubles. In this case, if the average educational attainment in the industry k in the region where a plant is located stays constant, the GAP variable will double as well. We can easily compute the increase in productivity a plant experiences following a 1% increase in the each of three different types of R&D spillovers. This is equal to 0.54% for the national R&D spillovers, 0.65% for the regional ones and finally 0.57% for the county-based R&D spillovers. It is clear that the increase in the GAP variable has decreased the productivity gains plants experience following increases in R&D spillovers. These are not large decreases but still they show the importance of the mechanism of absorption of R&D spillovers we are analysing in this study. Spatial lags of regional – and county-level R&D spillovers appear to be significant on average suggesting that plant benefit from significant R&D externalities generated by neighbouring regions and counties, although the impact is not very large. However, the spatial lags of the GAP variable are generally not significant across the different specifications. Finally, these effects do not seem to be driven by time-varying industry and regional shocks because they are robust to the inclusion of region*year*industry dummies.

What conclusions can we draw from these results? First, we have shown that regional differentials in the distribution of human capital matter, as it was suspected both in policy and academic circles. They are important because they can explain the productivity differentials experienced by plants located in different parts of the country and they can explain part of the productivity gap the British economy suffers from. In addition, we have identified an additional channel through which the distribution of human capital across regions matters. Indeed, plants that have access to a more qualified workforce can introduce advanced production techniques faster and can take advantage of inter-industry R&D spillovers. We believe that it is possible to draw the following policy implications from our study:

  • (i)
     there is evidence of an externality firms are benefiting from by locating in areas where the educational attainment is rather high. Policies that want to rebalance the distribution of human capital across the regions need to take into account this fact and alter the incentives firms have to locate in the areas with higher density of human capital. The same argument applies to the workforce. Workers have the incentive to move to areas where they have access to more productive jobs and again trying to change the incentive structure in this respect is the key (although challenging) element of the these policies;
  • (ii)
     the design of policy measures which want to stimulate the overall rate of innovation across the country needs to take into account the existence of interdependences between R&D spillovers and the geographical density of human capital. For instance there is no need to develop sophisticated mechanisms to facilitate knowledge transfer processes in areas where the density of human capital is not very high and there are scarce incentives for workers to move there. This simply implies that innovation and education policies need to be coordinated and inform each other.

Robustness tests

To gauge the robustness of the results obtained in the previous section we decided to undertake several robustness tests whose results are shown in Table 3. First, we re-estimate our augmented production functions on a sample made of small and medium-sized plants only. Second, we test whether our results are sensitive to the way the R&D spillovers have been constructed by using an alternative set of weights (based on the distance between the source of the R&D spillovers and its beneficiaries). Finally, we investigate the robustness of our set of results to different assumptions on the production technology, by assuming a translog functional form for our plants’ production function (rather than a Cobb–Douglas).

Table 3. Robustness checks
Small and medium size plants
  1. Notes: See Table 2.

  Coefficients Coefficients Coefficients
National spillovert−10.0048  
(0.36)  
Gapt−1−0.016  
(−0.29)  
National spillovert−1*Gapt−1−0.0013  
(−0.16)  
Regional spilloverst−1 0.0061 
 (0.77) 
Gapt−1 −0.013 
 (−0.27) 
Regional spilloverst−1*Gapt−1 −0.0012 
 (−0.12) 
County spilloverst−1  0.0016
  (1.70)
Gapt−1  −0.004
  (−1.93)
County spilloverst−1*Gapt−1  −0.00018
  (−1.72)
Spatially lagged Gapt−1−0.030−0.033−0.069
 (−0.87)(−0.13)(−0.57)
Spatially lagged Regional spilloverst−1 0.00054 
  (0.23) 
Spatially lagged County spilloverst−1  0.0064
   (1.67)
    
Hansen test (P-value)0.7860.5680.298
AR(1) (P-value)0.0000.0000.000
AR(2) (P-value)0.2910.4720.829
KP test: Levels eq.PassPassPass
KP test: Diff. eqPassPassPass
No. observations4,5784,5782,464
Distance-based R&D spillovers
Distance-based spillover t−10.0025  
 (2.60)  
Gap t−1−0.0095  
 (−2.01)  
Distance-based Spillovers−0.0002  
 t−1 *Gap t−1(−1.92)  
Hansen test (P-value)0.398  
AR(1) (P-value)0.000  
AR(2) (P-value)0.189  
KP test: Levels eq.Pass  
KP test: Diff. eqPass  
No. Observations5,470  
Plants belonging to SMEs

Do we expect all types of plants to: (i) be able to absorb R&D spillovers faster thanks to the larger availability of a more educated workforce and (ii) to equally benefit from the absorption of the three types of R&D spillovers? Obviously, we expect this to be the case on average. However, we acknowledge that plants tend to be heterogeneous along many dimensions, where size (measured as the number of employees) is perhaps the most relevant. A few authors have pointed out that unlike large firm, small and medium sized (SMEs) tend to rely more on R&D spillovers (rather than investing directly in R&D) and therefore they may experience large productivity gains even if they do not invest directly in R&D (Acs, Audretsch and Feldman 1994; Audretsch and Vivarelli, 1996). Equally, they may rely more on R&D spillovers generated at local level rather than at national level (Simmie, 2002). If so, these plants will suffer from the shortage of qualified workforce at regional level as their capability of absorbing R&D spillovers generated locally will be hindered. Unsurprisingly the estimates of our coefficients of interest confirm this expectation (See Table 3, top panel). For these plants the availability of a more educated workforce at regional level is important only for the absorption of R&D spillovers generated at county level. The spatially lagged GAP variable is not generally significant across the different specifications while only the spatially lagged county-level R&D spillover variable is significant and positive, although the size of the coefficient is not very large. These results suggest that these plants tend to cluster with their suppliers in specific counties so to take advantage of the R&D spillovers generated locally (besides other benefits they can receive).

Alternative measure of R&D spillovers

The measure of R&D spillovers that we use for our main analysis has been computed under the assumption that spillovers produced by investment in R&D are embodied in the intermediate goods produced by each sector so that the weighted inter-sectoral flows of intermediate goods can mimic the flows of R&D spillovers among industries (where the weights reflect the relative importance of the sectoral contribution to the intermediate goods flows). One implicit assumption we have made when using this measure of R&D spillovers is that the distance between the source of the spillovers (i.e. the suppliers of intermediate goods) and its beneficiaries (i.e. the purchasers of the intermediate goods) does not influence the intensity of the R&D spillovers. However, this may not be necessarily the case. For instance, increasing spatial transaction costs (due to transportation costs etc.) may make plants less likely to buy intermediate goods from suppliers located far away with the result that the impact of R&D spillovers becomes less and less important as the distance between the source and the beneficiaries of the spillovers increases. One way to capture this is to assume that the smaller the distance between the source of spillovers (say s) and the beneficiary of the spillovers (say r), the greater the weight assigned to s with respect to its influence on r. Therefore, the new R&D spillovers measure is expressed as the distance-weighted average of the R&D expenditures of other plants in a given postcode where the geographical distances between postcodes are calculated by using the Code-Point data provided by Edina Digimap.22 The last panel of Table 3 reports the estimates of our augmented production functions with this new measure of R&D spillovers. Our variables of interest are still significant and with the right sign, suggesting that the local availability of qualified workforce matters for the absorption of distance-based R&D spillovers as well. The coefficients are not larger than the ones we get from the main specification, so suggesting that these R&D spillovers may not necessarily matter more for productivity growth than those based on the flows of intermediate goods.

Translog production function

In Table A3 of the Online Appendix S1, the assumption of a Cobb–Douglas technology is relaxed and a Translog production function is estimated. Our variables of interest are still significant and this suggests that our results are not driven by omitted variables. However, the size of the impact of our variables of interest is smaller than the corresponding ones obtained for the Cobb–Douglas production function.

V.Concluding remarks

  1. Top of page
  2. Abstract
  3. I.Introduction
  4. II.R&D spillovers and the role of human capital
  5. III.The empirical strategy: the production function specification, the data sets and the variables
  6. IV.The empirical results
  7. V.Concluding remarks
  8. References
  9. Appendices
  10. Supporting Information

This study has analysed the extent to which plants located in regions with a more educated workforce at industry level can benefit more from R&D spillovers than plants located in regions with a less qualified workforce. This issue is particularly relevant in the current British policy debate on the impact of the uneven inter-regional distribution of the human capital on the productivity gap. To this purpose we have estimated the impact that measures of R&D spillovers (computed from BERD) and of regional differences in the educational attainment of the workforce across industries (computed from the Labour Force Survey) – and their interactions – have on the productivity of a panel of plants drawn from the ABI over the period 1997–2002. From our empirical analysis, we get the following results. Plants located in regions with a more educated workforce at industry level tend to benefit more from inter-industry R&D spillovers and to be more productive. This is true for R&D spillovers generated at national, regional and county level. The results still hold after controlling for the plants’ stock of R&D, the skills of their workforce and other plants’ characteristics. These results are consistent with the theoretical view that the geographical density of human capital facilitates the absorption of R&D spillovers. Also, we find that this mechanism of absorption of R&D spillovers is not equally important for all types of plants. As we know from the relevant literature, there are many channels through which the larger availability of a more educated workforce at regional level may facilitate the absorption of R&D spillovers. Indeed knowledge may flow across plants thanks to informal contacts among ‘co-located’ workers or because of the mobility of highly specialized workers from one plant to the other in the same geographical area. Obviously with the data we have used for our analysis, it is impossible to identify exactly which mechanism is at work in the British manufacturing. However this is the first step of what could be a very promising research agenda whose next step entails understanding the exact mechanisms underpinning our results.

Footnotes
  • 1

    These are from South to North: London Area, South East, East Anglia, South West, West Midlands, Wales, East Midlands, Yorkshire and Humberside, North West, North, Scotland.

  • 2

    It seems that on average productivity differentials account for around 60% of the regional GDP per capita differentials.

  • 3

    For instance, in 2000, over 40% more graduates were employed in London than studied there (HM Treasury, 2001).

  • 4

    In the remainder of the study, total factor productivity and productivity will be used as synonyms.

  • 5
  • 6

    Varga (2000) shows that university graduates may be one of the most important channels for disseminating knowledge from academia to the local high-tech industry.

  • 7

    Indeed it is fair to assume firms can decide to locate in a specific region if there are already firms carrying out similar (or complementary) research activities.

  • 8

    This point is particularly relevant for this study. Indeed, in our analysis, we decide to focus on the spillovers generated by the investment in R&D of private firms rather than that performed by Higher Education Establishments (HEEs). This choice has been dictated by the lack of detailed data on R&D investment of HEEs for the time period we consider (indeed the only publicly available data on HEEs R&D investment is at regional level and is available only for a small number of years). We do acknowledge that the omission of this variable may bias the estimates of our variables of interest but the use of time-varying regional dummies should avoid this problem.

  • 9

    The following example can explain why we observe this correlation: workers of higher unmeasured abilities may move to regions that experience larger increases in the average educational attainment of the workforce.

  • 10

    GMM estimators are used to estimate production functions as they allow to deal with the potential endogeneity of the inputs and with those time-invariant, firm-level characteristics which cannot be controlled for. The first GMM estimator was developed by Arellano and Bond (1991) where the predetermined and endogenous variables in first difference are instrumented with the lagged values of their levels. However, lagged variables in levels can be poor instruments for first differences especially for variables that show some degree of persistence (Blundell and Bond, 2000). Therefore Blundell and Bond (2000) introduce the so-called system GMM where predetermined and endogenous variables in levels are instrumented with suitable lags of their own first differences.

  • 11

    We have also estimated a specification where the plant's total wage bill is used to measure the labour input to control for the fact that the internal skills’ distribution may vary across plants.

  • 12

    Our sample is an unbalanced panel. When using an unbalanced panel to estimate a production function, there is a risk that the attrition rate may not be random and that therefore the results may be contaminated by the so-called ‘survivor bias’ (see Habib and Ljungqvist (2005) and Contoyannis Jones and Rice (2004) for a discussion of the ‘survivor bias’ in unbalanced panels). Indeed, each year a certain number of plants leave the panel and it is reasonable to assume that some of the attrition rate can be directly attributed to their initial-period productivity levels; for instance, poorly performing plants may (after some time) exit from the market and therefore from the panel. So, the plants that remain in the panel are likely to be more productive than average and therefore the estimates of the relationship between productivity and our variables of interest may be misleading. To establish whether initial period plants’ total factor productivity may play a role in explaining the attrition rate in our panel we have estimated a probit model of exit/non-exit for each year of the panel using the full sample of firms that are observed in 1997 (first year of the sample). The dependent variable in these models equals 1 if the firms are present in year of the sample in question and 0 otherwise. The probability of exiting the sample is modelled as a function of the 1997 value of the Tornquist productivity index, the initial period size, age, regional and sectoral dummies. The estimates show there is no significant association between the probability a firm has of leaving the sample and its initial period total factor productivity. In addition, we have re-estimated our specifications by excluding the firms observed in 1997 and found out that the results do not change qualitatively.

  • 13

    This type of technology flow indicator focuses on R&D embodied in products purchased by a firm. The concept of ‘R&D embodiment’ relies on the fact that the flow of goods among industries is the channel for the transfer of the technology developed by supplying firms. The current R&D embodiment indicators have been formulated on the basis of a Leontief inverse (computed by using the UK 2000 Input–Output Tables), and more precisely, on the basis of the output multipliers, taking into account the cumulative nature of inter-industrial R&D flows.

  • 14

    Nominal R&D spending is converted to real spending using the implied GDP deflator.

  • 15

    The weights are constructed as the ratio between the regional/ county employment in industry j and the total aggregated employment of the 2-digit industry j.

  • 16

    Harmon et al. (2001) have estimated the returns to educational qualifications for men and women in a number of European countries around 1995 using a Mincerian equation. In Britain, the returns of the main educational qualifications in Britain are: 8% for GCSE, 17% for two or more A-levels and 19% for a degree while for men, these are: 10% for GCSE, 23% for two or more A-levels and 15% for a degree. Using these estimates we have constructed the average return to holding either a GCSE or an A-level or a Degree for the entire population by weighting the male and female estimates by the corresponding shares in total employment.

  • 17

    See previous footnote for details on how the weights were constructed.

  • 18

    In a preliminary stage, we have also compared our System GMM estimates with OLS and Fixed Effects estimates. Both sets of estimates (OLS and Fixed Effects) appear to be flawed as the inputs’ elasticities are too large and they reject the constant returns to scale restrictions.

  • 19

    Indeed recall that our model assumes that there should be a negative first-order serial correlation of the differenced residuals and no second-order serial correlation.

  • 20

    Bun and Windmeijer (2010) have shown that System GMM estimator can suffer from the weak instruments problem as this is the optimally weighted average of the difference and levels equations with the weights on the moments of the equation in levels increasing in the weakness of instruments of the equation in difference.

  • 21

    There exists a variety of tests for weak instruments within the context of 2SLS (see Clements and Bazzi for a short presentation of the different tests). We use the Kleibergen-Paap (KP) test as it is robust to the presence of heteroskedasticity. The critical values against which we need to compare the values of the KP test have been tabulated by Stock and Yogo (2005).

  • 22

    The Code-Point data provide a precise geographical location for each postcode in the UK determined by its National Grid co-ordinates.

References

  1. Top of page
  2. Abstract
  3. I.Introduction
  4. II.R&D spillovers and the role of human capital
  5. III.The empirical strategy: the production function specification, the data sets and the variables
  6. IV.The empirical results
  7. V.Concluding remarks
  8. References
  9. Appendices
  10. Supporting Information
  • Acs, Z. J., Audretsch, D. B. and Feldman, M. P. (1994). ‘R&D spillovers and recipient firm size’, The Review of Economics and Statistics, Vol. 76, pp. 336340.
  • Acs, Z. J., Anselin, L. and Varga, A. (2002). ‘Patents and innovation counts as measures of regional production of new knowledge’, Research Policy, Vol. 31, pp. 10691085.
  • Arellano, M. and Bond, S. (1991).‘Some tests of specification for panel data: Monte Carlo evidence and an application to employment functions’, Review of Economic Studies, Vol. 58, pp. 277297.
  • Arrow, K. J., (1962). ‘The Economic Implications Learning by Doing’, Review of Economic Studies, Vol. 29, pp. 15573.
  • Audretsch, D. B. and Feldman, M. P. (1996). ‘R&D spillovers and the geography of innovation and production’, American Economic Review, Vol. 86, pp. 630640.
  • Audretsch, D. B. and Vivarelli, M. (1996). ‘Firms size and R&D spillovers: evidence from Italy’, Small Business Economics, Vol. 8, pp. 249258.
  • Battu, H., Belfield, C. R. and Sloane, P. J. (2003). ‘Human capital spillovers within the workplace: evidence for the Great Britain’, Oxford Bulletin of Economics and Statistics, Vol. 65, pp. 575594.
  • Bazzi, S. and Clements, M. A. (2010). Blunt Instruments: a Cautionary Note on Establishing the Causes of Economic Growth, Working Paper 171, Center for Global Development, University of Maryland.
  • Benhabib, J. and Spiegel, M. M. (1994). ‘The role of human capital in economic development: evidence from aggregate cross-country data’, Journal of Monetary Economics, Vol. 34, pp. 143173.
  • Bernstein, J. I. and Nadiri, M. I. (1989). ‘Research and development and intra-industry spillovers: an empirical application of dynamic duality’, Review of Economic Studies, Vol 56, pp. 249267.
  • BIS (2008). Innovation Nation, White Paper, London, UK.
  • Blundell, R. and Bond, S. (2000).‘GMM estimation with persistent panel data: an application to production functions’, Econometric Reviews, Vol. 19, pp. 321340.
  • Bode, E., Nunnenkamp, P. and Waldkirch, A. (2009). Spatial Effects of Foreign Direct Investment in US States, Working Paper 1353, Kiel Institute for the World Economy.
  • Bond, S., Harhoff, D. and Van Reenen, J. (2002). Corporate R&D and Productivity in Germany and the UK, Department of Economics, University of Oxford, Oxford.
  • Bun, M. and Windmeijer, F. (2010). ‘The weak instrument problem of the system GMM estimator in dynamic panel data models’, The Econometrics Journal, Vol. 13, pp. 95126.
  • Ciccone, A. and Peri, G. (2006). ‘Identifying human capital externalities: theory with applications’, Review of Economic Studies, Vol. 73, pp. 381412.
    Direct Link:
  • Cohen, W.  M. and Levinthal, D.  A. (1989). ‘Innovation and learning: two faces of R&D’, Economic Journal, Vol. 99, pp. 569596.
  • Contoyannis, P., Jones, A. and Rice, N. (2004). ‘The dynamics of health in the british household panel’, Journal of Applied Econometrics, Vol. 19, pp. 473503.
  • Crespi, G., Geuna, A. and Nesta, L. (2007). ‘The mobility of university inventors in Europe’, Journal of Technology Transfer, Vol. 32, pp. 195215.
  • Criscuolo, C. and Haskel, J. (2003). Innovations and Productivity Growth in the UK: Evidence from CIS2 and CIS3, CERIBA Working Paper, ONS.
  • Duranton, G. and Monastiriotis, V. (2002). ‘Mind the gaps: the evolution of regional inequalities in the UK 1982–1997’, Journal of Regional Science, Vol. 42, pp. 219256.
  • Engelbrecht, H. J. (1997). ‘International R&D spillovers, human capital and productivity in OECD economies: an empirical investigation’, European Economic Review, Vol. 41, pp. 14791488.
  • European Commission (2000). Innovation Policy in a Knowledge-based Economy, A MERIT Study Commissioned by the European Commission, Office for Official Publications of the European Communities, Luxembourg.
  • Galindo-Rueda, F. and Haskel, J. (2005). Skills, Workforce Characteristics and Firm-Level Productivity: Evidence from the Matched ABI/Employer Skills Survey, Discussion Paper 1542, IZA.
  • Geroski, P. A. (1995).‘Do spillovers undermine the incentive to innovate?’ in Dowrick S. (ed.), Economic Approaches to Innovation, Edward Elgar, Aldershot, pp. 7697.
  • Girma, S. (2005). ‘Absorptive capacity and productivity spillovers from FDI: a threshold regression analysis’, Oxford Bulletin of Economics and Statistics, Vol. 67, pp. 281306.
  • Griffith, R., Redding, S. and van Reenen, J. (2004). ‘Mapping the two faces of R&D: productivity growth in a panel of OECD industries’, The Review of Economics and Statistics, Vol. 86, pp. 883895.
  • Grossman, G. and Helpman, E. (1990). ‘Trade, innovation and growth’, American Economic Review, Vol. 80, pp. 8691.
  • Habib, M. A. and Ljungqvist, A. (2005). ‘Firm value and managerial incentives: a stochastic frontier approach’, Journal of Business, Vol. 78, pp. 20532094.
  • Harhoff, D. (2000). ‘R&D spillovers, technological proximity and productivity growth – evidence from German Panel Data’, Schmalenbach Business Review, Vol. 52, pp. 238260.
  • Harmon, C., Walker, I. and Westergaard-Nielsen, N. (2001). Education and Earnings in Europe, Edward Elgar, Cheltenham, UK.
  • Higon, D. A. (2007).‘The impact of R&D spillovers on UK manufacturing TFP: a dynamic panel approach’, Research Policy, Vol. 36, pp. 964979.
  • HM Treasury (2001). Productivity in the UK: The Regional Dimension, London, UK.
  • Kleibergen, F. and Paap, R. (2006). ‘Generalized reduced rank tests using the singular value decomposition’, Journal of Econometrics, Vol. 133, pp. 97126.
  • Kneller, R. and Stevens, P. A. (2006). ‘Frontier technology and absorptive capacity: evidence from OECD manufacturing industries’, Oxford Bulletin of Economics and Statistics, Vol. 68, pp. 121.
  • Jaffe, A. (1986). ‘Technological opportunity and spillovers of R&D: evidence from firm's patents, profits and market value’, American Economic Review, Vol. 76, pp. 9841001.
  • Jaffe, A. (1989). ‘Real effect of academic research’, American Economic Review, Vol. 79, pp. 957970.
  • Martin, R.(2002). Building the Capital Stock, CERIBA Working Paper, ONS.
  • Marshall, A. (1890). Principles of Economics, Macmillan, New York.
  • Mastromarco, C. and Gosh, S. (2009).‘Foreign capital, human capital, and efficiency: a stochastic frontier analysis for developing countries’, World Development, Vol. 37, pp. 489502.
  • Moen, J. (2005). ‘Is mobility of technical personnel a source of R&D spillovers?’, Journal of Labour Economics, Vol. 23, pp. 81113.
  • Moretti, E. (2004). ‘Workers’ education, spillovers and productivity: evidence from plant-level production functions’, American Economic Review, Vol. 94, pp. 656690.
  • Poyago-Theotoky, J., Beath, J. and Siegel, D. S. (2002). ‘Universities and fundamental research: reflections on the growth of university-industry partnerships’, Oxford Review of Economic Policy, Vol. 18, pp. 1021.
  • Raut, L. K. (1995). ‘R&D spillover and productivity growth: evidence from Indian private firms’, Journal of Development Economics, Vol. 48, pp. 123.
  • Rice, P. and Venables, A. J. (2003). ‘Equilibrium regional disparities: theory and British evidence’, Regional Studies, Vol. 37, pp. 675686.
  • Romer, P. M. (1986). ‘Increasing returns and long-run growth’, Journal of Public Economics, Vol. 94, pp. 10021037.
  • Romer, P.  M. (1993). ‘Idea gaps and object gaps in economic development’, Journal of Monetary Economics, Vol. 32, pp. 543573.
  • Saxenian, A. (1994).Regional Advantage: Culture and Competition in Silicon Valley and Route 128, Harvard University Press, Cambridge, MA.
  • Siegel, D. S., Westhead, P. and Wright, M. (2003). ‘Assessing the impact of science parks on the research productivity of firms: exploratory evidence for the United Kingdom’, International Journal of Industrial Organisation, Vol. 21, pp. 12171225.
  • Simmie, J. (2002). ‘Knowledge spillovers and reasons for the concentration of innovative SMEs’, Urban Studies, Vol. 39, pp. 885902.
  • Stock, J. H., and Yogo, M. (2005).‘Testing for weak instruments in linear IV regression’, in Andrews D. W. K. and Stock J. H. (eds), Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg, Cambridge University Press, Cambridge, pp. 80108.
  • Terleckyj, N. E. (1974). Effects of R&D on the Productivity Growth of Industries: An Exploratory Study, National Planning Association, Washington, DC.
  • Varga, A. (2000). Local academic knowledge transfers and the concentration of economic activity’, Journal of Regional Science, Vol. 40, pp. 289309.
  • Wakelin, K. (1998). ‘Innovation and export behaviour at the firm level’, Research Policy, Vol. 26, pp. 829841.
  • Wakelin, K. (2001). ‘Productivity growth and R&D expenditure in UK manufacturing firms’, Research Policy, Vol. 30, pp. 10791090.

Appendices

  1. Top of page
  2. Abstract
  3. I.Introduction
  4. II.R&D spillovers and the role of human capital
  5. III.The empirical strategy: the production function specification, the data sets and the variables
  6. IV.The empirical results
  7. V.Concluding remarks
  8. References
  9. Appendices
  10. Supporting Information

Appendix A1

List of regions
  • 00
    London Area
  • 01
    South East
  • 02
    East Anglia
  • 03
    South West
  • 04
    West Midlands
  • 05
    East Midlands
  • 06
    Yorkshire and Humberside
  • 07
    North West
  • 08
    North
  • 09
    Wales
  • 10
    Scotland
List of qualifications considered in the QLFS (variable: HIQUALD)
  • HIQUALD – highest qualification (detailed grouping)

  • 1
    Degree or equivalent
  • 2
    Higher education
  • 3
    GCE A level or equivalent
  • 4
    GCSE grades A*–C or equivalent
  • 5
    Other qualification
  • 6
    No qualification
  • 7
    Don't know

Appendix A2

The Annual business inquiry

The ARD data set consists of microdata from the Annual Census of Production (ACOP) up to 1997 and then the Annual Business Inquiry (ABI) afterwards. It covers both the production sector (including manufacturing) and the non-production sector (services). However the time series dimension varies across the twos: indeed for the production sector it is possible to have information available up to 1980 (and early 1970s for some industries) and before in some cases, while the data for the services sector is available only after 1997. The information is assembled from the replies to the Census forms: as this is a mandatory requirement for UK-based business, the response rates to the ABI are rather high and this makes the ARD data set highly representative of the underlying population. Each establishment has got a unique reference number that does not change over time and so allows us to build up a panel data set.

The ABI is a stratified random sample where sampling probabilities are higher for large establishments: indeed for establishments with more than 250 employees, the sampling probability is equal to one. The sampled business form the so-called selected sample and they account for 80% of total employment in manufacturing (Criscuolo and Haskel, 2003). The rest of units on the register are not sampled and they form the ‘non-selected’ sample: of these units, only basic information is recorded in ARD (namely industrial classification, region and employment). This sampling structure requires the data to be weighted by sampling weights derived from both the selected and non-selected samples.

When working with the ARD data set, it is important to define the correct level of aggregation at which the analysis has to be carried out. The ABI data set has got information on enterprises, reporting units (RU) (or the decision-making unit) and local units (LU). It is a well-known fact that the statistical ‘reporting units’ do not correspond to the economic notion of a firm. Typically the RU is the plant or unit that replies to the questionnaire and so it may correspond to the decisional centre of the firm. RU may coincide with local units if the firm is a single-plant unit while in a multi-plants firms, the RU is a group of local units. This problem arises because multi-plants companies can choose how to paper information to the ABI. Indeed, they can paper each plant individually or various groups of plants. Each RU has its own unique identification number, an enterprise and an enterprise group identification number. The problem of multi-plants is particularly relevant for geographical analysis as only in the case of single local unit (which is the RU as well) there will be no ambiguity with regard to the specific location of an RU. Our solution to this problem has been to carry out the empirical analysis on single-unit plants.

Business enterprise research and development

The Business Enterprise Research and Development survey covers all business R&D expenditure undertaken in the UK, along with its total R&D employment and sources of funds (business themselves, government and overseas). The first inquiry of this type was carried out in the early 1970s; however, since 1993 the ONS has moved to an annual inquiry based on a continually updated register of R&D performers. Estimates are made for the R&D activity of unsampled and non-responding businesses. The sample and survey results only cover ‘business enterprises’ as defined in the OECD ‘Frascati’ Manual. The sample of firms covered includes both UK and foreign-owned enterprise. However, it excludes R&D performed overseas by UK subsidiaries and by sectors such as government organizations, higher education establishments and charities.

Labour force survey

The QLFS is a sample survey of households living at private addresses in Great Britain. The population covered is all people resident in private households, all persons resident in National Health Service accommodation and young people living away from the parental home in a student hall of residence or similar institution during term time. The sample design currently consists of about 55,000 responding households in Great Britain every quarter, representing about 0.2% of the population. Its main purpose is to provide information on the UK labour market which can then be used to develop, manage and evaluate labour market policies. So the survey seeks information on respondents’ personal circumstances and their labour market status during a specific reference period, normally a period of 1 or 4 weeks (depending on the topic) immediately prior to the interview.

Supporting Information

  1. Top of page
  2. Abstract
  3. I.Introduction
  4. II.R&D spillovers and the role of human capital
  5. III.The empirical strategy: the production function specification, the data sets and the variables
  6. IV.The empirical results
  7. V.Concluding remarks
  8. References
  9. Appendices
  10. Supporting Information

Appendix S1: Additional descriptive statistics and regression output.

FilenameFormatSizeDescription
obes724_sm_online-appendix.doc99KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.