Participation in export markets and the role of R&D: establishment-level evidence from the UK Community Innovation Survey 2005

There is a strong expectation in the literature that exporting and innovation activities (particularly R&D) are strongly related, and that the need to be innovative is increasing over time due to globalization. In this study, we find that R&D is endogenous in a model that determines which British establishments enter export markets, and when such simultaneity is taken into account the strength of the export–innovation relationship is generally quite weak (especially in the nonmanufacturing sector). Rather, we find that the size of establishments and firms, foreign ownership, the extent of international co-operation and, most importantly, the industry sector to which the establishment belongs, are the most significant in explaining which establishments are able to overcome entry barriers into overseas markets.


I. Introduction
It is often argued that the link between innovation activities, such as R&D, and exports has become increasingly interdependent as part of the process of globalization. Innovativeness is commonly taken as a proxy for productivity and growth, and exporting for the competitiveness of an industry or country. At the macroeconomic level, this relationship between trade and innovation often relates to several distinct paradigms, such as the Schumpetarian idea of creative destruction (Schumpeter, 1950), the government's pursuit of export-promotion policies, as well as the Prebisch-Singer model of the trade patterns between developed and less-developed countries (Prebisch, 1950;Singer, 1950).
From the perspective of the firm, several earlier theoretical studies maintain that innovating firms have incentives to expand into other markets so as to earn higher returns from their investment, as appropriability is improved when the product market widens (Teece, 1986). As the domestic market is usually limited in size, firms often face an increasingly strong need to expand their product market by different modes of internationalization, with exporting being one of the most conventional ways. In this process of international expansion, innovation activities play an important role in the development of competitive advantage as well as growth potential. For one thing, the advantage conferred by innovation will give firms an incentive to enter global markets and subsequently enhance their performance and international competitiveness in these new markets; in addition, the more competitive international environment that firms are exposed to per se may provide a source of new ideas spurring more and better innovations from them. A resource-based approach 1 has been employed in recent studies, offering new insights into this export-innovation relationship by focusing on the development of the firm's technological capacity, which is required for the firm to gain access and subsequently operate successfully in global markets (e.g. Dhanaraj and Beamish, 2003).
In comparison with the well-established tradeinnovation theoretical framework established as part of the macroeconomics literature, 2 the majority of the microeconomic evidence on the role of innovation in internationalization is empirics-led. Despite a number of theoretic attempts to study the firm's decision to export (particularly based on a framework of sunk costs and firm-level heterogeneity), these studies do not explicitly model innovation as one of the determinants of exporting (Bernard et al., 2003;Yeaple, 2005). Despite the lack of a solid theoretical foundation, studies at the micro level usually provide a useful way to disentangle this export-innovation relationship, by taking into account the heterogeneous characteristics of exporting and nonexporting firms. Thus, in this article, we concentrate on the export decision at the level of the establishment, 3 incorporating the role of innovation activities and other factors determining export-market entry (rather than export performance post entry). 4 Despite the importance of this area, there are still only a limited number of micro-based studies in the literature, especially with regard to the UK-based empirical analysis, and generally it is expected that exporting and R&D will be positively related without much evidence being in support.
In what follows, Section II reviews some relevant literature relating to the export-innovation nexus, while Section III discusses the datasets used and the construction of certain variables that enter our export model. Section IV sets out the model in more detail and discusses certain methodological issues surrounding estimation. Section V reports results for both manufacturing and nonmanufacturing sectors, whilst the last section concludes this article with a discussion of the limitations and some implications for policy makers.

II. Earlier Empirical Studies
Export orientation at the firm level has been extensively investigated in the literature: various empirical studies have emphasized the role of technology and innovation as one of the major factors facilitating entry into global markets and thereafter maintaining competitiveness and boosting export performance. For instance, studies covering the UK firms include: Bishop and Wiseman (1999), Bleaney and Wakelin (2002) and most recently, Harris and Li (2009); for Canadian manufacturing firms, Baldwin and Gu (2004); for Italian manufacturing firms, Basile (2001); for Spanish manufacturing, Cassiman and Martinez-Ros (2004); for German services, Blind and Jungmittag (2004); for comparative studies, Roper and Love (2002); for both the UK and German manufacturing firms, Dhanaraj and Beamish (2003) for the US and Canadian firms; in the context of the rest of the world, Alvarez (2001) for Chilean manufacturing and lastly, Ozcelik and Taymaz (2004) for Turkish manufacturing firms. Still evidence at this micro level does not seem to be conclusive, as inconsistent results have frequently been found (e.g. Sterlacchini, 2001).
With respect to the issue of causality associated with this linkage, the early consensus in the literature was that causality runs from undertaking innovation activities to internationalization. In line with the predictions of both more conventional product-cycle models as well as recently developed neo-technology 1 Refer to Penrose (1959) and Barney (1991) for more details on the resource-based view. 2 The macroeconomics literature offers at least two mainstream theoretical models to account for this relationship: neo-endowment models which concentrate on specialization on the basis of factor endowments, such as materials, labour, capital, knowledge and technology (Davis, 1995); and also neo-technology models which predict that innovative industries will be net exporters instead of importers (Greenhalgh, 1990). The latter type of models provides an extension of the conventional technology-based models such as the product life cycle theory (Krugman, 1979;Dollar, 1986) and technology-gap theory of trade (Posner, 1961). 3 Establishments were chosen as the level of analysis primarily because the Community Innovation Survey 2005 collected data at the establishment level (as opposed to the firm level). In the dataset, establishments can cover more than one plant (if the firm is a multi-plant organization), and thus we have the advantage of conducting analysis at this more disaggregated level, allowing for the characteristics of multi-plant firms to be taken into account. 4 Unfortunately, we are not able to measure export performance using the current dataset, as the information on export volume has been omitted in the Community Innovation Survey (2005, CIS4). In an earlier paper using the previous wave of UK Community Innovation Survey (2001, CIS3), Harris and Li (2009) provide an analysis of both export-market entry as well as performance, in a different setting. models (see footnote 2 for details), the intuition behind this causal chain is straightforward: product differentiation/innovation translates into competitive advantages that allows the firm to compete in international markets. 5 However, it can also be argued that causality may go from exporting to innovativeness, i.e. there exists a 'learning-byexporting' effect. This reverse direction of causation is in accordance with the theoretical predictions of global economy models of endogenous innovation and growth, such as those in Romer (1990) and Aghion and Howitt (1998). From a resource-based perspective, being exposed to a richer source of knowledge/technology that is often not available in the home market, exporting firms could take advantage of these diverse knowledge inputs and enhance their competence base, and hence such learning from global markets can foster increased innovation within firms.
In conjunction with the role of innovation, a number of other firm-specific factors have also been suggested in the literature that impact on export entry, and therefore moderate the way export and R&D activity affect (and interact with) each other. Above all, in a seminal paper, Cohen and Levinthal (1990) put forward the notion of Absorptive Capacity (AC) 6 and demonstrate that the ability to exploit external knowledge is a critical component of a firm's capabilities. Essentially, AC constitutes an analytical link between the firm's in-house resources and the external stock of knowledge in enhancing its resource base and generating competitive advantage.
There is also well-documented evidence on how the size of firms affects the probability of entering foreign markets, as larger firms are expected to have more (technological) resources available to initiate an international expansion (e.g. Roberts and Tybout, 1997;Bleaney and Wakelin, 2002). Higher productivity in general constitutes another significant factor determining the firm's internationalization decision. This positive impact of productivity on export-market entry is in line with the self-selection hypothesis, which assumes that the existence of sunk entry costs means exporters have to be more productive to overcome such fixed costs before they can realize expected profits (for recent surveys of the literature surrounding this export-productivity nexus, see Lo´pez (2005) and Greenaway and Kneller (2007)).
The external position of the firm is also generally found to determine export behaviour, in terms of sectoral, regional effects or market structure. For instance, as industries are neither homogeneous in their technological capacity nor exporting patterns, the sectoral effect (reflecting technological opportunities and product cycle differences) is usually expected to be significant (e.g. Bleaney and Wakelin, 2002, for empirical evidence). The role of spatial factors are also important; for example, see Overman et al.'s (2003) survey of the literature on the economic geography of trade flows and the location of production. If information on costs and foreign-market opportunities is asymmetric, then it is reasonable to expect firms to cluster within the same industry/region so as to achieve information sharing and therefore minimize entry costs; such co-location provides better channels through which firms distribute their goods (e.g. Greenaway and Kneller, 2004). Lastly, market concentration is also expected to positively impact upon a firm's propensity to export. A high level of concentration of exporters within an industry may improve the underlying infrastructure that is necessary to facilitate access to international markets or to information on the demand characteristics of foreign consumers.

III. Dataset
The dataset employed for the current study comprises the merged Community Innovation Survey 2005 (henceforth CIS4) and the 2003 Annual Respondents Database (henceforth ARD 2003). The features of the two datasets as well as the matching procedure are discussed in more detail in the Appendix. Table 1 sets out the list of variables used in this study, along with the source of the datasets. Note that the establishment's R&D activity is used to proxy for its innovation activities, with R&D spending defined as intramural R&D, or acquired external R&D or acquired other external knowledge (such as licenses to use intellectual property). 7 Of particular importance is the AC of the establishment (cf. Section II for more details). No direct 5 For empirical evidence on this causation, see Bleaney and Wakelin (2002), Roper and Love (2002) and Barrios et al. (2003). 6 They argue that '. . .prior related knowledge confers an ability to recognize the value of new information, assimilate it and apply it to commercial ends. These abilities collectively constitute what we call a firm's 'AC'.' 7 There is other spending that is categorized in the CIS4 database relating to innovative activities, such as the acquisition of machinery and equipment (including computer hardware) in connection with product and process innovation, but we chose to exclude these from our narrower and more traditional definition of R&D after some initial analysis of the data and by comparing the Community Innovation Survey (CIS) totals with those obtained from the Business Enterprise Research and Development (BERD) data source. information on this variable is available, but CIS4 contains information on key elements of internal and external knowledge that can be related to AC. 'Internal' AC is proxied using data on the impact on business performance of the implementation of new or significantly changed corporate strategies, advanced management techniques (e.g. knowledge management, Investors in People), organizational structures (e.g. introduction of cross-functional teams, outsourcing of major business functions) and marketing concepts/strategies. 8 'External' AC was proxied using data on the relative importance of different sources of information used for innovation-related activities. These are grouped under the following sub-headings with associated elements: market -suppliers, customers, competitors, consultants, commercial labs/R&D enterprises; institutional -universities, government research organizations; other -professional conferences, meetings, professional and trade associations, technical press, fairs, exhibitions, technical, industry or service standards. 9 To obtain a single index of AC, a factor analysis was undertaken using all the 14 variables listed above. We found that the first principal component had an eigenvalue of 6.2 and could explain 44% of the combined variance of the variables. Thus, we use this principal component (with a mean of zero and a SD of one) as an adequate proxy of AC. Note, our measure of AC indicates mostly the ability of the establishment to internalize external knowledge for innovation activities rather than for a range of other activities, such as overcoming barriers to exporting. That is, we assume (as detailed in Section IV) that our measure directly determines whether R&D is undertaken or not, while undertaking R&D is then used as a determinant of whether the establishment exports or not. Thus, AC impacts on exporting, but indirectly through R&D activities. We empirically test whether the model structure that we impose is supported by the data, by considering the significance of our AC indicator when directly entered into a model which tests for the determinants of exporting. If AC is significant as a direct determinant of R&D activities but insignificant as a direct determinant of whether an establishment exports, this is taken as support for our approach.
Most of the variables included in Table 1 are self-explanatory. In particular, industrial agglomeration is included to take account of any Marshall-Romer external (dis)economies of scale (cf. Henderson, 2003). The greater the clustering of an industry within the local authority in which the plant operates, the greater the potential benefits from spillover impacts. Conversely, greater agglomeration may lead to congestion, and therefore may lower the productivity. The diversification index is included to pick up urbanization economies associated with operating in an area with a large number of different industries. Higher diversification is usually assumed to have positive benefits to producers through spillover effects. Note, agglomeration is measured as the percentage of industry output (at the five-digit Standard Industrial Classification (SIC) level) located in the local authority district in which the establishment is located; diversification is measured as the percentage of five-digit industries (from over 650) that are located in the local authority district in which the establishment is located. The Herfindahl index of industrial concentration is measured at the five-digit 1992 SIC level to take account of any market power effects (which are expected to be associated with the propensity to both export and to undertake R&D). The variable that measures if the establishment belongs to an enterprise operating in more than one (five-digit) industry (41 SIC multiplant) is included to proxy for any economies of scope. The data on the age of (manufacturing) establishments and their capitallabour ratios were obtained from updating the series on plant and machinery capital stocks computed by Harris and Drinkwater (2000). Capital stock estimates are not available for the nonmanufacturing sector (as gross investment data are only available from 1998); however, we can provide an indicator of the age of nonmanufacturing establishments based on a question asked in the CIS4 questionnaire (i.e. whether the enterprise was established after 1 January 2000).
All the data are weighted to ensure it is representative of the UK distribution of establishments (i.e. it is not biased towards the CIS4 sample). Initial inspection shows that 25.1% of manufacturers engaged in both exporting and spending on R&D; while only 8.5% undertook both activities in the nonmanufacturing sectors.

IV. Model Specification
Separate models are estimated for manufacturing and services (given the known differences between these two sectors in terms of their propensities to export). For each sector, we estimate a model of what determines whether exporting is undertaken or not, where undertaking R&D is included as a likely major determinant 10 ; thus, to account for simultaneity between exporting and R&D we use the structural probit model approach first presented by Maddala (1983). 11 This involves using instruments to replace 9 For each element, respondents were asked to rank from zero 'not used' to three 'high importance '. 10 In this study, R&D activity is employed to proxy for an establishment's innovation behaviour. Note that we have also attempted to include in our model other proxies from the output side such as product/process innovations; nevertheless, the estimation results obtained using alternative measures are considerably less robust, and therefore we only report results based on the current model using the R&D measure. 11 Other simultaneous approaches have been employed in several empirical studies treating innovation and exports as endogenous (e.g. Cassiman and Martinez-Ros, 2004;Lachenmaier and Woessmann, 2006). the endogenous variables and thus obtain unbiased estimates of the impact of R&D on exporting. There is also a procedure for correcting the resulted SE when the instrumented R&D is incorporated in the exporting model (since this instrument is generated from a different model).
In choosing the likely determinants of whether exporting (or R&D spending) takes place or not, we include those variables that have been shown to be important in the literature and that are available in the CIS4-ARD database. Thus, all the variables listed in Table 1 are included in our empirical models.
We begin with the following probit model determining whether an establishment exports or not where EXP is a 0/1 dichotomous variable which takes on the value 1 if the establishment exports; R&D is a 0/1 dichotomous variable which takes on the value 1 if the establishment spends on R&D; X 1 is a vector of other (exogenous) variables that determine exporting and u 1 is an error term that includes all other random effects.
As written, Equation 1 assumes that R&D is an exogenous determinant of exporting. To allow for a simultaneous relationship, we also estimate a (probit) model to determine whether an establishment spends on R&D or not where AC measures the AC of the establishment in terms of its ability to internalize external sources of knowledge related to its (potential) innovation activities; X 2 is a vector of other (exogenous) variables that determine R&D (with some variables in X 2 also likely to be included in X 1 , but X 2 = 2 X 1 ) and u 2 is a random error term.
The reduced-form model that determines R&D is therefore obtained by substituting Equation 1 into Equation 2 and rearranging Estimating Equation 3 and obtaining predicted values for R&D allows us to replace R&D in Equation 1 by its instrument (R&D), thus obtaining the following structural model of the determinants of whether an establishment exports or not (allowing for endogeneity between exporting and R&D): We have assumed that AC enters the model determining R&D, but does not directly determine exporting (although AC enters the reduced-form model for exporting, through Equation 2 which shows that R&D -as a determinant of exporting -is itself determined by AC). In addition to formally testing whether R&D is endogenous in Equation 1, using a form of the Smith-Blundell test for exogeneity, we also estimate another version of (1) that includes AC to test whether this variable is significant or not, with nonsignificance being an indication that AC only determines exporting indirectly through its impact on R&D.

Manufacturing
The results for the reduced-form models for the manufacturing sector are presented in Table 2. Table 3 presents the results of exporting modelling for the manufacturing, treating R&D as either exogenous (cf. Equation 1) or endogenous (cf. Equation 4), respectively. Note that stepwise (probit) regression models are estimated, 12 and marginal effects are reported. Also the z-statistics reported in Table 3 have been 'corrected' when R&D is instrumented, since the latter is generated from the reduced-form model. Maddala (1983, pp. 246-7) provides the relevant formula for the variance-covariance matrix, which we have adopted to use in STATA (although we find very little difference between the SE obtained using the standard and corrected variance-covariance matrices).
Comparing the reduced-form results for exporting and R&D (Table 2) shows that certain variables are only significant in one or other of the two equations estimated, and it is these (unique) variables that separately identify structural models for exporting (and R&D). 13 We use this information on which variables are unique to the exporting and R&D reduced-form models in order to test for exogeneity 12 Variables were retained in each model that had associated parameter estimates significant at the 15% or better level. In addition, the null hypothesis that the variables dropped had jointly coefficients equal to zero was not rejected at better than the 10% significance level. The full list of variables that entered each model is shown in Table 1. 13 The presence of such (unique) variables is necessary in order to obtain a valid instrument for R&D in Equation 4. Note that we also treat AC as an instrument, given our modelling approach (cf. Equations 1-4) and the fact that (see further) this variable is not significant in the export model (based on the estimation of Equation 1). using the Smith-Blundell test available in STATA. 14 As shown in Table 3, the null hypothesis that R&D is exogenous is rejected at better than the 1% significance level.
As to the issue of whether AC should enter the structural model determining exporting, we found this variable to be insignificant when estimating Equation 1. The reduced-form models for exporting and R&D (Table 2) also show that AC has a much larger impact in determining R&D and a relatively small effect upon exporting. Thus, we take this as evidence that this variable impacts on whether an 14 The probexog routine in STATA is used. This is based on the methodology devised by Smith and Blundell (1986), revised for probit (and logit) models. In practice, the EXP variable is regressed on both those variables that are significant in determining the reduced-form export model ( establishment exports (or not) via its (highly significant) impact on R&D, rather than through any direct impact of its own. In terms of the determinants of exporting, when R&D is treated as exogenous, the impact of an establishment undertaking R&D spending is to significantly increase the probability that it also exports by almost 17%. However, this impact of R&D on exporting drops to less than 7% when we take into account the simultaneous relationship between exporting and R&D. 15 Thus incorrectly treating R&D as exogenous leads to an upward bias in the strength of the exporting-R&D relationship, and in general undertaking R&D has only a relatively small impact on the decision to export at the establishment level in Great Britain. Notes: Population weights are available in CIS4 data. a R&D was replaced by its predicted value (R&D) in the endogenous model. These values were obtained from estimating Equation 3 (cf. Table 2). b The z-statistics reported here have been 'corrected' when R&D is instrumented, using the formula for the variancecovariance matrix provided in Maddala (1983).
In what follows, we concentrate on the results where R&D is taken to be endogenous, although generally there is little difference in estimated parameter values for the other (non-R&D) determinants of exporting. First and foremost, establishment-specific characteristics seem to play a major role in determining the probability of entering into international markets. For instance, the size of the establishment had a major impact on whether any exporting took place; vis-a`-vis the baseline group (establishments employing less than 20), moving to 20-49 employees increased the probability of exporting by 8.4%; while having 50-199 workers increased the probability by 13.1%. This confirms the results found in the literature that size and the propensity are positively related to export (see Section II for a discussion). In addition to establishment size, the size of the enterprise was also important in determining which establishments had nonzero exports. A doubling of enterprise size (from the mean of 34 employees) resulted in a 5.1% increase in the likelihood of the establishment exporting in 2004. Meanwhile, doubling the age of establishments from its mean value of nearly 11 years increased the probability of exporting by just over 3%.
Establishments with higher labour productivity were also more likely to enter export markets; a doubling of this variable (from its mean value of just over £68k turnover per worker) increased the probability of exporting by nearly 8%. Establishments that belonged to enterprises that operated plants in more than one region were (ceteris paribus) over 8% less likely to export. In contrast, foreign-owned establishments had significantly higher export propensities (e.g. being US-owned resulted in a 34.2% higher probability of exporting, while other foreign-owned establishments were 23% more likely to export, compared with their UK-owned counterparts).
Selling to national markets had a significant impact on increasing the likelihood that the establishment was also engaged in exporting abroad (the probability of exporting was over 32% greater for such establishments). Data are also available on whether establishments had engaged in co-operation on innovative activities with overseas organization; the results presented in Table 3 show that (ceteris paribus) those that did were some 17 % more likely to export. Moreover, as indicated by the insignificant estimated parameters of some variables related to market failures, in the context of the establishment's innovation activities (e.g. a lack of information on technology, high innovation costs, impact of regulations, etc.), none of these factors capturing obstacles to innovation seemed to constitute barriers to export-market entry. However, establishments that received government support for innovation activities were also more likely to export (an increase of nearly 8%).
As far as the market or industry is concerned, the results in Table 3 indicate that industry/market concentration and agglomeration were both linked to a greater probability of exporting. Increasing the Herfindalh index of market concentration from its mean value of 0.06 to 0.16 (the latter being the average value for the 90th decile group in manufacturing), increased the (ceteris paribus) probability of exporting by around 14%. Similarly, a five-fold increase in the percentage of industry output (at the five-digit SIC level) in the local authority district, in which the establishment was located, would have resulted in a 7% increase in the probability of exporting.
Regional impacts are significant as well -being located in the South East region resulted in a higher propensity to export by some 7.3%; the other regions included in the results were between 6% and 8% less likely to export. Sector also mattered; those with the highest propensities to export were (ceteris paribus) chemicals, basic metals and machinery and equipment. None of the other entered variables (Table 1) proved to have a significant impact on establishment's entry into international markets (e.g. industry diversification, whether the establishment belonged to an enterprise operating in more than one industry or the Greater South East).
Nonmanufacturing Table 5 presents the results for nonmanufacturing, again based on different models involving the different treatment of R&D. We use the information on which variables are unique to the exporting and R&D reduced-form models (based on their statistical significance levels in Table 4) in order to test for exogeneity using the Smith-Blundell test. 16 As shown, the null hypothesis that R&D is exogenous is rejected.
As to the issue of whether AC should enter the structural model determining exporting, we find this variable to be insignificant when estimating Equation 1. The reduced-form models for exporting and R&D (Table 4) also show that AC had a much larger impact in determining R&D and a relatively small effect upon exporting. Thus, as with the case for the manufacturing results, we take this as evidence that this variable impacts on whether an establishment exports (or not) via its (highly significant) impact on R&D, rather than through any direct impact of its own.
When R&D is treated as exogenous, the impact of R&D spending was to increase the probability that an establishment also exported by more than 6%.
However, when we take account of the simultaneous relationship between exporting and R&D, the impact of R&D on exporting was (as expected ) much smaller at around 2.5%. 17 Compared with the results for manufacturing, the export-R&D relationship was much weaker in the nonmanufacturing sector.
As with manufacturing, we concentrate on the parameter estimates obtained for the second model where R&D is taken to be endogenous. The size of the establishment had a positive impact on whether any exporting took place; vis-a`-vis the baseline group (establishments employing less than 20), ceteris paribus, moving to 20-49 employees increased the probability of exporting by 1.8%; 50-199 workers increased the probability by 1.9%. These values are considerably smaller than those reported for manufacturing (Table 3). Those establishments with higher labour productivity were also more likely to enter export markets; a doubling of this variable (from its mean value of just over £62k turnover per worker) increased the probability of exporting by nearly 6%. Again, this is a smaller impact than that obtained for manufacturing. Establishments that belonged to enterprises that operated plants in more than one region were (ceteris paribus) over 3% more likely to export, which is in contrast with the manufacturing model where the impact of this variable was negative. However, and in line with manufacturing, being under foreign control increased exporting, with US-owned establishments nearly 18% more likely to export, while other foreign-owned had 10.8% higher probabilities of selling overseas.
Selling to national markets had a significant impact on increasing the likelihood that the establishment was also engaged in exporting abroad (i.e. the probability of exporting was nearly 19% greater for such establishments, although this is significantly lower when compared with the results presented in Table 3 for manufacturing). Establishments engaged in international co-operation on innovative activities were around 11% more likely to export. Establishments that received government support for innovation activities were also some 6% more likely to sell overseas. In line with the results obtained for manufacturing, none of the variables capturing innovation-related market failures had any adverse impact on the export-market entry. Contrary to the case in manufacturing, the external localization of establishments in the services sector does not seem to play a significant role in determining their decision to export. Specifically, the results in Table 5 show that industry agglomeration was only marginally linked to a greater probability of exporting in the nonmanufacturing sector; a five-fold increase in the percentage of industry output (at the five-digit SIC level) in the local authority district in which the establishment was located would have resulted in a merely 4% increase in the probability of exporting. This could, in part, be explained by lower level of industrial agglomeration in the services compared with the manufacturing sector. Moreover, vis-a`-vis the results from the manufacturing sector, increasing the Herfindalh index of market concentration did not boost the probability of an establishment going international.
Regional impacts exhibited a different pattern to those in manufacturing. For instance, all the regions of southern England plus Scotland had higher propensities to export (between 3% and 6% more likely to export). Sector also mattered; the sectors with the highest propensities to export were (ceteris paribus) the R&D sector, transport support, wholesale trade and computing. Again, none of the other variables entered in the model (Table 1) proved to impact on the establishment's propensity to export.

VI. Conclusion
There is a strong expectation in the literature that exporting and innovation activities (particularly R&D) are strongly related, and that the need to be innovative is increasing over time due to globalization. In this study, we find that R&D is endogenous in a model that determines which British establishments enter export markets, and when such simultaneity is taken into account the strength of the export-innovation relationship is generally quite weak (especially in the nonmanufacturing sector).
Rather, we find that establishment and firm size, foreign ownership, the extent of international co-operation, and most importantly the industry sector to which the establishment belongs, are more important in explaining which establishments are able to overcome entry barriers into overseas markets. There are some important differences between the manufacturing and nonmanufacturing sectors; for example, agglomeration and market structure have a stronger role in manufacturing, and there are some interesting differences when looking at the impact of location (in terms of region) on who exports. From a policy perspective, given the differing needs of (potential) exporters (i.e. the internal resources available to them), our results suggest that government assistance needs to be flexible so as to reflect the heterogeneous nature of firms. In particular, we find little evidence (given the data available) to suggest that market failure is a key barrier to exporting, and this suggests that policies might instead need to concentrate on helping firms in particular sectors to acquire certain characteristics (e.g. larger size, higher productivity, greater AC and learning capabilities) so as to confer the ability to overcome sunk costs that act as barriers to entry into international markets.
information has been aggregated to reporting unit level. We have been able to match 14 299 of these establishments in CIS4 with the 2003 ARD. The nonmerged CIS4 data mostly include Northern Ireland (accounting for 63% of the nonmatched observations) and financial services (accounting for 31%), both of which are not included in the 2003 ARD. The remaining 5% of nonmatched observations comprise mostly those that started operations in 2004.
The matching procedure seems to give rise to two potential problems in our merged dataset. First, strictly speaking, CIS4 should be merged with the 2004 ARD (rather than the 2003 version), since the CIS4 sample were based on the population of establishments as existing in the 2004 IDBR, but 2004 data are not yet available. Therefore using only 2003 data for the ARD means that CIS4 establishments that started up in 2004 are excluded from any matching of CIS4 and the 2003 ARD. Nevertheless, this is unlikely to be problematic as those start-ups in 2004 only comprise less than 5% of the nonmatched observations. In addition, CIS4 covers only those with 10þ employees; nevertheless, in practice this results in some 8% (weighted ) of the establishments covered having employment of less than 10, for the IDBR information includes a level of inaccuracy due to the timing of the information obtained on employment, and some units when sampled may have downsized by the time they received the CIS questionnaire. Table 1 sets out the list of variables used in our empirical modelling.