Who explores further? Evidence on R&D outsourcing from the survey of research and development

Shotaro Yamaguchi, Ryuji Nitta, Yasushi Hara, Hiroshi Shimizu
First published: 27 September 2020


Building off the resource-based view and the knowledge-based view, our study aims to examine determinants of firms’ R&D outsourcing, using annually-conducted firm-level survey data of Japanese R&D companies from 1984–2012. This survey allows us to measure strategic R&D outsourcing, isolated from those more for cost-reducing, such as prototyping, testing and inspecting. The results corroborate the argument of complementarity in scale between internal R&D and R&D outsourcing. We also find that firms employing more doctorate holders and diversifying in knowledge spaces tend to make more use of R&D outsourcing. This study sheds light on firms’ absorptive capacity, associated both with higher-order R&D human capital and diversified knowledge spaces, as determinants of R&D outsourcing.

1 Introduction

Firms’ exploration for external knowledge sources has become an essential ingredient for their innovation. Previous literature highlights that today’s firms increasingly rely on various external agents – universities, research institutions and upstream/downstream firms – as knowledge sources for innovation rather than on their own (Lundvall, 1992; Powell et al., 1996; Chesbrough, 2003).

Accordingly, research and development (R&D) outsourcing has hitherto caught great scholarly attention (Howells, 2008; Huang et al., 2009; Jones, 2000; Lai et al., 2009; Varadarajan, 2009; Stanko and Calantone, 2011). R&D outsourcing is distinguished from outsourcing in general in the sense that R&D investment decisions are highly strategic because not all activities can be outsourced and R&D entails learning, which is highly dependent on individual firms’ heterogeneous resources (Tiwana and Keil, 2007; Huang et al., 2009).

Empirical studies of R&D outsourcing, which usually measure R&D outsourcing as expenditure on contracted R&D or technology acquisition, observe a notable increasing trend (Veugelers, 1997; Lokshin et al., 2008; Hsuan and Mahnke, 2011). Moreover, we are reaching the point where providing capital to start-ups or acquiring technology developed by external organisations is no longer sufficient: not only finding start-ups possessing already established technologies, but they also need to collaborate with partners whose immature technologies have potential but have not yet yielded fabulous outcomes. Hence, for exploring new opportunities, some established firms have started to fund R&D activities by outside organisations, leveraging in-house R&D and accelerating development of new technology.

Firms exploring embryonic technologies that involve high uncertainty may be unwilling to make a capital commitment (Folta, 1998; Van De Vrande et al., 2006). Funding the R&D activities of an external entity can be the first step towards capital investment or acquisition. Established firms can fund outside R&D activities without holding equity or forming joint venture, or as a preliminary to do so. The fiercer competition is between firms striving for external knowledge, the earlier they would try to gain access to outside resources.

Although R&D outsourcing has become increasingly prevalent and caught scholarly attention, we still have the insufficient knowledge of determinants of R&D outsourcing, mainly due to the absence of relevant data sources. Moreover, the previous literature on R&D outsourcing has not distinguished cost-reducing outsourcing such as prototype production, testing and inspecting from more strategic outsourcing that is critical for competitive advantage. We aim at providing evidence on the determinants of such R&D outsourcing by using a unique database: the Survey of Research and Development conducted by the Japanese Ministry of Internal Affairs and Communications (MIC). As explained in detail below, this survey allows us to measure R&D outsourcing, isolated from those more for cost-reducing, such as prototyping, testing and inspecting. Moreover, since this survey has been annually conducted, we can construct a panel data set over around 30 years accounting for time-invariant unobservable firm heterogeneity. To prefigure our main results, we find that firms expend R&D budgets on outside organisations when they have higher internal R&D intensity, employ more doctorate holders and conduct R&D in a broader scope of technological fields. Our results suggest that internal R&D and R&D outsourcing are complementary in scale, and the scope of R&D also plays an important role in firms’ absorptive capacity, which promotes R&D outsourcing.

2 Previous literature and hypotheses

What are the attributes of firms that invest in R&D outsourcing? Following previous literature, we define R&D outsourcing as ‘contractually paid R&D performed by an independent provider that is either a firm or a research organisation’ (Grimpe and Kaiser, 2010; Spithoven and Teirlinck, 2015). While firms are increasingly using R&D outsourcing, there has still been a debate about why firms choose this strategy (Grimpe and Kaiser, 2010; Stanko and Calantone, 2011; Spithoven and Teirlinck, 2015; Un, 2017). The literature on R&D outsourcing has predominantly relied on two perspectives: transaction cost economics (TCE) and the resource-based view (RBV) or its extension, the knowledge-based view (KBV). From TCE perspective, the decision to internalise or outsource depends on transaction costs (Williamson, 1975). It stresses that firms prefer R&D outsourcing to in-house investment, inasmuch as outsourcing would not incur considerable transaction costs due to, for instance, ex ante search and negotiation, and ex post-execution and enforcement of contracts (Veugelers and Cassiman, 1999; Gooroochurn and Hanley, 2007). However, the plausibility of predictions made by TCE literature has been questioned in explaining R&D outsourcing, as several empirical studies show evidence inconsistent with or only weakly supportive for TCE (Love and Roper, 2005; Gooroochurn and Hanley, 2007; Stanko and Calantone, 2011). Another drawback of TCE is that it ignores heterogeneity in resources and firm-specific capabilities and knowledge, which several empirical studies have identified as of high relevance for R&D outsourcing (Holcomb and Hitt, 2007; Spithoven and Teirlinck, 2015).

The RBV or the KBV, on the contrary, regard heterogeneity in resources, capabilities and knowledge as a critical driver of R&D outsourcing. According to those views, firms choose R&D outsourcing when they expect combining internal resources with those outside would yield a substantial impact on their competitive advantage. Some kinds of knowledge and expertise cannot necessarily be developed internally, which can be a strong driver for outsourcing (Howells et al., 2008). The benefits from outsourced R&D can be augmented by absorptive capacity, which has been discussed as the crucial reason why firms simultaneously invest in internal R&D as well as R&D outsourcing. Recent studies provide empirical evidence consistent with the RBV- and the KBV-based constructs rather than TCE (Spithoven and Teirlinck, 2015; Un, 2017).

We primarily build off the KBV (based on which firms’ technological resources are typically examined) to construct three hypotheses, pertaining to the antecedents of R&D outsourcing. The first hypothesis concerns the relationship between internal R&D and R&D outsourcing. Some empirical studies have reported a substitutive (rather than complementary) relationship between internal R&D and R&D outsourcing, although their findings are less robust than those which have observed a complementary relationship (e.g., Laursen and Salter, 2006; Vega-Jurado et al., 2009). For example, a survey of UK manufacturing firms shows that the commitments to internal R&D, operationalised as the ratio of R&D expenditure to sales, was negatively correlated with external searching for new knowledge (Laursen and Salter, 2006). This suggests that the relationship between internal and external R&D activities is substitutive. An anecdote of DSM also corroborates this argument: in the 1970s, DSM cut its basic expenditure on in-house R&D and reduced its R&D personnel (Van Rooij, 2007). As a consequence, some of the R&D personnel transferred to the University of Groningen, where DSM had funded some of their R&D activities. Hagedoorn and Wang (2012) find a more nuanced relationship. Their analysis of pharmaceutical firms indicates that the relationship between internal and external R&D is contingent on the level of internal R&D; Above a threshold level of in-house R&D investment, which is around 1,000 million dollars, internal and external R&D are complementary, whereas below that threshold they are substitutive.

With these exceptions, however, many of the previous work find the complementary relationship between internal R&D and R&D outsourcing (e.g., Veugelers, 1997; Mol, 2005; Cassiman and Veugelers, 2006; Rothaermel and Hess, 2007; Lokshin et al., 2008; Schmiedeberg, 2008; Grimpe and Kaiser, 2010; Howells et al., 2012; Un, 2017). The key proposed mechanism is absorptive capacity (Cohen and Levinthal, 1990); firms can build absorptive capacity by investing in R&D internally, which provides further incentives to explore external knowledge that can potentially be combined (Spithoven and Teirlinck, 2015; Un, 2017).

One of the empirical challenges in exploring the relationship between internal R&D and R&D outsourcing lies in the difficulty in tracking R&D outsourcing behaviours over time. Most of the previous empirical studies rely on one or several-periods observations based on the survey data and so remain to be cross-sectional analyses, or conduct longitudinal observations within a single-industry context. Since our data source is the annually conducted survey over around 30 years across different industries, we can construct a panel data set accounting for time-invariant unobservable firm heterogeneity, and test the hypothesis of complementarity between internal R&D and R&D outsourcing

H1: Internal R&D intensity of the firm is positively associated with its R&D outsourcing intensity.

The second hypothesis pertains to R&D human capital. As explained above, absorptive capacity can be a potential reason why internal R&D and R&D outsourcing are complementary rather than substitutive (Cohen and Levinthal, 1989, 1990; Spithoven and Teirlinck, 2015; Un, 2017). To appraise, assimilate and apply new external knowledge effectively, firms need to invest in R&D on their own and build up and sustain their knowledge bases (Zahra and George, 2002). Previous literature also underscores that investing in-house R&D may not be sufficient; they also need a system which lubricates knowledge transfer beyond the boundaries of the firms (Katz and Allen, 1982; Allen, 1986; Laursen and Salter, 2006; Kathoefer and Leker, 2012). In this sense, gatekeepers can play a significant role in effective knowledge transfer and utilisation of external knowledge (Katz and Tushman, 1981). With the increasing complexity of science and technology, technological breakthroughs demand a wide range of intellectual and scientific skills that far exceed the capabilities of any single organisations. While scientific knowledge is an important frame of reference for firms that seek solutions and road maps in technological development (Fleming and Sorenson, 2004), it may be difficult for them to be well exposed to and digest those scientific knowledge despite their relevance. In such a case, the existence of higher order human capital that possesses sophisticated scientific knowledge mitigates opportunity losses, as they could play a role of gatekeepers of such knowledge.

The hiring practices of Japanese firms allow us to investigate the role played by employees who have advanced knowledge in science. Japanese firms tend to employ fewer people with doctorates to work on R&D than comparable US firms. Instead, they hire more new graduates – with a bachelor’s or a master’s degree – as R&D personnel and provide them with on-the-job training, whereby their working practices align with the firms’ culture. Although an increasing number of PhD-degree holders have recently been employed by Japanese firms, this has still been a dominant path to careers of the R&D personnel in Japanese firms. Thus, PhD-degree holders are scarce human capital, which possesses more advanced and a broader base of scientific knowledge than those with a bachelor’s or a master’s degree (Shimizu and Hara, 2011). We therefore conjecture that R&D personnel with a doctorate play a distinctive role in R&D, for example, by accessing, assessing and absorbing advanced knowledge in academia and hence, enables firms to expand their knowledge sets. They are a critical source of absorptive capacity, which encourages firms to explore outside technological knowledge that is new to them. Hence, we hypothesise that the proportion of R&D personnel with a doctoral degree to the size of R&D divisions positively associated with R&D outsourcing.

H2: Firms where a higher proportion of R&D personnel holds a doctorate have higher R&D outsourcing intensity.

The breadth of firms’ R&D could also affect the extent of R&D outsourcing. Contrary to the first hypothesis pertaining to R&D scale, the third hypothesis sheds light on the scope of R&D. Some literature exploring the level of diversification and external R&D has found a positive relationship (e.g., Veugelers, 1997; Gooroochurn and Hanley, 2007; Lai et al., 2010). Firms diversified broadly in product spaces could also need broad knowledge base and so tend to conduct a variety of R&D projects as well as contract at least some of those projects (Nakamura and Odagiri, 2005).

However, product diversification does not necessarily imply diversification of technological knowledge. Firms can leverage a narrow but deep technological knowledge to target many product lines. Or, it could be said that the addition of product lines is rather a marketing issue at least for some firms, which does not involve the broad expansion of R&D knowledge base. This study directs its attention to the knowledge base of firms rather than product diversification. A broader technological knowledge base can provide firms with capabilities to evaluate, assimilate and utilise a wider range of knowledge outside. In other words, the width of R&D can also promote firms’ absorptive capacity. Thus, we hypothesise that the scope of internal R&D and R&D outsourcing represent a positive association.

However, the opposite might also apply. Firms may be more keen on funding outside R&D to complement internal R&D when their R&D scope is narrower. This could happen when firms lack sufficient resources to invest in complementary knowledge that they have not possessed so cannot help relying on outsourcing. Another reason is that, as TCE predicts, firms selectively internalise only a degree of knowledge that would involve high transaction costs but procure others through market. If this is the case, one would expect to find a negative association between the scope of internal R&D and R&D outsourcing. We therefore construct the conflicting hypotheses, which guide us to examine whether the technological scope of R&D is positively or negatively associated with R&D outsourcing.


H3a: Firms whose R&D activities are broader in scope have higher R&D outsourcing intensity.

H3b: Firms whose R&D activities are narrower in scope have higher R&D outsourcing intensity.

3 Research design

3.1 Data and estimation strategy

Our primary data source is the Survey of Research and Development annually conducted by the Japanese Ministry of Internal Affairs and Communications (MIC) during the period between 1984 and 2012, the survey whose responses are required by law. The average response rate for this survey over the past 16 years is 82%. We further link the survey data to two other data sources: financial information from the Firm Financial Data Bank provided by the Development Bank of Japan (DBJ); patent information from the IIP Patent Database provided by the Institute of Intellectual Property (IIP).1

The respondents of the survey comprises three types of organisations in Japan: all firms with capital stock over 10 million yen (approximately a hundred thousand US dollars); sampled firms with capital stock less than 10 million yen; and non-firm research-oriented organisations.2 The second category (i.e., firms with capital stock less than 10 million yen) is extracted each year from stratified sampling based on research activities, capitalisation and industry categories.

The survey asks respondents for their R&D personnel (e.g., its number, specialty and turnover) and R&D expenditure (e.g., its total amount, by nature (basic/applied) and by product category). It also has an item on ‘expenditure on external R&D’, which we use to capture the amount of R&D outsourcing. The survey, conducted every year since 1984, enables us to construct a panel-structured database, which accounts for time-invariant unobservable factors. We note that our constructed data entail unbalanced panels (see footnote3 in detail). They consist of 1,004 Japanese firms conducting R&D (10,919 observations) for the period between 1984 and 2012, among which approximately 92% comes from firms whose stock capital is over 10 million yen. This means that our samples are largely dominated by large-sized firms.

3.2 Variables

Following the OECD’s guideline and previous literature, this study operationalises R&D outsourcing as an intensity measure: the natural log of the ratio of external R&D expenditure to sales (OECD, 2002; Spithoven and Teirlinck, 2015). The survey defines the external R&D expenditure as ‘money spent for the purpose of providing R&D expenses for outside firms, non-profit organisations, public institutions and universities (expenditure on outsourcing prototype productions, testing, inspecting is not included)’. This item suits our purpose as it excludes more of cost-reducing R&D outsourcing.

We use three explanatory variables to test the hypothesis: internal R&D intensity, R&D doctorate and R&D diversification. Similarly to the dependent variable, we measure internal R&D intensity as the natural log of the ratio of internal R&D expenditure to sales (Spithoven and Teirlinck, 2015). R&D doctorate is a measure of the extent to which firms holds R&D human capital with advanced knowledge, operationalised as the ratio of R&D personnel with PhD degree to the number of R&D-related employees. Since the survey asks the number of PhD holders only since 2002, we can only test its effect (Hypothesis 2) based on the sub-sample from 2002 onwards. R&D diversification assesses the extent to which firms diversify their knowledge scope, which is calculated as the inverse concentration ratio of patenting technological fields. We use primary four-digit IPC subclasses to define each patent’s technological field. Since the concentration measure is highly dependent on the number of patents – the more patents firms obtain, the lower the concentration ratio becomes, we discount it by the number of patents that firms obtain in each year. Thus, the measure of R&D diversification entails following steps: First, urn:x-wiley:00336807:media:radm12437:radm12437-math-0001 where pij denotes i’s number of patent applications in subclass j and ni denotes i’s total number of patent applications in each year. Second, we take the natural log of inverse of the above ratio, then, the higher value of it represents more R&D diversification.4

We include some control variables in our models whose omission could potentially bias our estimates. Firm size is operationalised as the natural log of the number of employees. Firm age, the years lapsed since its legal establishment, is also included in our models. The amount of slack resources can also influence both our predictors and the dependent variable, as firms with excess resources may be more willing to conduct risky or exploratory investments (Singh, 1986). We calculate current ratio for capturing slack resources, which is the natural log of the ratio of current assets to current liabilities. We also account for the extent of market competition, which has been recognised as one of the important determinants for innovation (Malerba and Orsenigo, 1995, 1996; Fontana et al., 2012). It is measured by the natural log of the inverted sum of squared market share in sales within industries. We aggregate sales of all private firms in DBJ’s database sharing the same three-digit Japanese SIC codes. A firm’s primary area of business (i.e., primary SIC) is obtained from the NISTEP name lists. We finally include a set of product class controls: the natural log of the R&D expenditure in 30 product categories provided in the survey. We choose to use continuous expenditure variables rather than binary dummies for sectoral heterogeneity to account for more variations among firms.

4 Empirical analysis

4.1 Descriptive statistics

Figure 1 plots the logged internal R&D intensity and R&D outsourcing across product categories. The variance of R&D outsourcing tends to be larger than that of internal R&D, implying the greater heterogeneity in the extent of R&D outsourcing even within product classes. In terms of the ratio of R&D outsourcing to internal R&D intensity, we see largely the similar pattern across product classes. It is clear that the associations between internal R&D and R&D outsourcing, which examined later, are not driven solely by a couple of industries. Electricity & gas is the only industry in which the median of R&D outsourcing is larger than that of internal R&D, probably due to its public nature. Excluding this industry from our samples does not change our main results described below.

Details are in the caption following the image
Box plots of internal R&D intensity and R&D outsourcing intensity by product classes.

Table 1 reports summary statistics of the variables of interest. The sample means of the expenditures on internal R&D and on R&D outsourcing are 11.6 and 1.9 billion yen – approximately, 116 and 19 million US dollars at the current exchange rate. In other words, 14.0% of the sample firms’ aggregated R&D budget are spent on outsourced R&D. The ratio has gradually increased from 13.0% in 1984 to 18.1% in 2012, implying that R&D outsourcing has become increasingly common and such levels of outsourcing are comparable to those in other countries. For instance, Lokshin, Belderbos and Carree find that Dutch manufacturing firms on average outsourced 9.5%–15.3% of R&D budget during 1996–2001 (Lokshin et al., 2008); Jones reports R&D outsourcing of UK pharmaceutical firms increased from 5% to 16% of internal R&D expenditures from 1989 to 1995 (Jones, 2000).

Table 1. Summary statistics
Mean SD p25 Median p75
Expenditure on R&D outsourcing (million yen) 1,904.539 19,193.480 5.200 28.000 242.720
Internal R&D expenditures (million yen) 11,696.450 41,424.360 584.620 1,713.910 6,568.000
R&D outsourcing intensity 0.003 0.010 0.000 0.000 0.002
Internal R&D intensity 0.037 0.042 0.010 0.027 0.049
R&D diversification 3,119.648 13,667.870 34.091 171.329 1,120.039
Firm size (number of employees) 4,554.934 11,848.450 742.000 1,575.000 3,783.000
Firm age 60.226 18.768 48 59 71
Market competition 7.546 5.285 3.794 6.331 9.260
Current ratio 1.827 1.395 1.093 1.444 2.058
R&D doctorate 0.048 0.065 0.000 0.026 0.069
    • Descriptive statistics are based on absolute values (without log-transformation).

Tables 2 and 3 show the correlation matrices of the variables of interest during the period between 1984 and 2012 and between 2002 and 2012, respectively. Overall, the pairwise correlations represent consistent signs with our hypotheses. Maybe a tricky thing is the negative relationship between internal R&D intensity and PhD doctorate in Table 3. This is probably because that higher internal R&D intensity is associated with larger size of firm laboratories (i.e., the denominator of R&D doctorate) and given the scarcity of PhD degree holders in Japan, this may drive down the density of the doctorates within laboratories.

Table 2. Correlation matrix (1984–2012)
1 2 3 4 5 6 7 8
1. R&D outsourcing intensity 1.000
2. Internal R&D intensity 0.389* 1.000
3. R&D diversification 0.165* 0.337* 1.000
4. Firm size 0.094* −0.011 0.699* 1.000
5. Firm age −0.091* 0.031* 0.161* 0.071* 1.000
6. Market competition 0.072* 0.025 −0.091* −0.048* 0.017 1.000
7. Product diversification −0.033* 0.064* 0.285* 0.133* 0.196* −0.089* 1.000
8. Current ratio 0.096* 0.335* −0.148* −0.285* −0.110* 0.099* −0.044* 1.000
    • * P < 0.01.
Table 3. Correlation matrix (2002–2012)
1 2 3 4 5 6 7 8 9
1. R&D outsourcing intensity 1.000
2. Internal R&D intensity 0.414* 1.000
3. R&D diversification 0.183* 0.320* 1.000
4. Firm size 0.111* −0.039* 0.668* 1.000
5. Firm age −0.089* 0.027 0.194* 0.105* 1.000
6. Market competition 0.074* 0.043* −0.074* −0.043* −0.004 1.000
7. Product diversification −0.020 0.046* 0.284* 0.120* 0.263* −0.097* 1.000
8. Current ratio 0.144* 0.300* −0.181* −0.283* −0.123* 0.118* −0.072* 1.000
9. R&D Doctorate 0.055* −0.152* −0.006 0.103* 0.110* 0.208* −0.011 −0.103* 1.000
    • * P < 0.01.

5 Main results

Our main results are shown in Table 4. All the estimations are based on firm fixed effect models (the P-value of the Hausman test: P < 0.001), including year fixed effects and a set of product class controls to account for time and industry heterogeneity. The standard errors are clustered at a firm level. The signs of coefficients on internal R&D intensity (H1) remain consistently positive and significant across Model 2 and 3 for the whole period and Model 5 for the sub-period, estimating that an increase of 100% in internal R&D intensity is associated with 47.3% to 59.2% increase in outsourcing intensity.

Table 4. Estimation results
Model1 Model2 Model3 Model4 Model5
1984–2012 1984–2012 1984–2012 2002–2012 2002–2012
Internal R&D intensity 0.488*** 0.473*** 0.592***
[0.07] [0.07] [0.07]
R&D diversification 0.077*** 0.055*
[0.03] [0.03]
R&D doctorate 1.358* 1.816**
[0.76] [0.73]
Firm size −0.300** −0.224** −0.263*** −0.242* −0.136
[0.12] [0.10] [0.10] [0.14] [0.13]
Firm age 0.079 0.214 0.21 −0.005 −0.001
[0.21] [0.21] [0.21] [0.01] [0.01]
Market competition 0.071 0.031 0.026 −0.334 −0.189
[0.21] [0.19] [0.19] [0.29] [0.28]
Current ratio −0.217** −0.193** −0.197** −0.266** −0.191*
[0.10] [0.09] [0.09] [0.11] [0.11]
Product diversification −0.303*** −0.194* −0.548 −2.579* −0.201
[0.11] [0.11] [1.41] [1.38] [0.15]
Constant −4.897 −11.415 −11.289 −0.548 −2.579*
[9.23] [9.26] [9.22] [1.41] [1.38]
Year FE Yes Yes Yes Yes Yes
Product class controls Yes Yes Yes Yes Yes
R-square 0.024 0.052 0.055 0.033 0.069
N of obs. 10,919 10,919 10,919 5,610 5,610
N of firms 1,004 1,004 1,004 903 903
    • SEs in parentheses are clustered at firm level.
    • *** P < 0.01,
    • ** P < .05,
    • * P < 0.1.

The effect of R&D doctorates is tested by the sub-sample models from 2002 to 12 (Model 4 and 5). The coefficients are significant and positive, and the increase of one standard deviation (6.5%) in PhD doctorate leads to the 8.8% (Model 4) or 11.8% (Model 5) increase in R&D outsourcing intensity.

The coefficient on R&D diversification (H3) is also positive (Model 3 and 5), which indicates that more diversified firms in R&D tend to rely on outside R&D by outsourcing. We note that the Model 3 accounts for product diversification in R&D, and the coefficients and the significance levels of R&D diversification remain unchanged. This could suggest that even accounting for the diversification in the application side, the diversity in the invention side matters for the extent of using R&D outsourcing.

Another notable result is that the coefficients on firm size represent consistently negative, suggesting that smaller firms tend to have higher R&D outsourcing intensity. This is consistent with the previous study (O’Regan and Kling, 2011).

While the estimated results in Table 4 are consistent with the hypotheses, the punctual conditional mean estimators give less information about their sensitivity, in particular for the complementary relationship between internal R&D and R&D outsourcing. Hence, we further employ a couple of additional analyses. First, we conduct quantile regressions with the same control variables in Model 3 and 5 in Table 4. Quantile regressions estimate conditional values at particular quantiles (e.g., median) rather than conditional means of the dependent variable, allowing for varying coefficients on the independent variables across the positions in the distribution of the dependent variable.

Figure 2 plots the coefficients of the variables of our primary interest, in which Panel A and B are based on Model 3 in Table 4 and Panel C based on Model 5. The coefficients on internal R&D intensity are consistently significant and increase slightly in quantiles, which implies the association between internal R&D and outsourcing becomes larger at the higher level of outsourcing intensity, although the differences in coefficients across quantiles are not statistically significant. On the contrary, the coefficients on R&D diversification are decreasing (0.125 at 0.1-st quantile and 0.030 at 0.9-th quantile) and not significant over 0.8-th quantile. One interpretation is that while diversification in R&D can drive further exploration of outside technological knowledge that is potentially combined, such outsourcing is decreasing returns in its scale so that the association becomes small at the low level of outsourcing intensity. Finally, the coefficients on PhD doctorate are also decreasing and only significant at lower quantiles. This may imply that the existence of gatekeepers with advanced skills matters particularly at the outset of R&D outsourcing, and such effects become smaller once firms accumulate experiences of outsourcing (at higher level of intensity).

Details are in the caption following the image
Quantile regression estimates.

Results in Panel A and B are based on Model 3 in Table 4 and results in Panel C are based on Model 5 in Table 4. The shaded areas are 90% confidence intervals.

Next, we conduct sub-sample regressions under the same specifications of Model 3 and 5, with the presumption that the effects of the explanatory variables could also vary across the different positions in the distributions of key independent variables. We divide our pooled samples based on the quartiles of internal R&D intensity and firm size. The results are shown in Table 5.

Table 5. Sub-sample regressions
Panel A: Sub-sample regressions by internal R&D intensity (Model 3)
below p25 p25 – median median – p75 over p75
Internal R&D intensity 0.141 0.563*** 0.670*** 0.511***
[0.10] [0.18] [0.25] [0.14]
R&D diversification 0.066** 0.05 0.076 0.021
[0.03] [0.05] [0.05] [0.06]
N of obs. 2,730 2,729 2,731 2,729
N of firms 427 505 448 364
Panel B: Sub-sample regressions by firm size (Model 3)
below p25 p25 – median median – p75 over p75
Internal R&D intensity 0.533*** 0.445*** 0.346** 0.245**
[0.12] [0.11] [0.17] [0.11]
R&D diversification 0.077* 0.080* 0.059 0.069
[0.04] [0.05] [0.06] [0.06]
N of obs. 2,725 2,738 2,723 2,733
N of firms 460 406 310 208
Panel C: Sub-sample by internal R&D intensity (Model 5)
below p25 p25 – median median – p75 over p75
R&D Doctorate 1.064 0.838 5.919*** 2.85
[1.04] [1.82] [2.06] [2.48]
N of obs. 1,446 1,368 1,352 1,444
N of firms 333 376 339 302
Panel D: Sub-sample by firm size (Model 5)
below p25 p25 – median median – p75 over p75
R&D Doctorate 1.810* 2.532 −1.833 2.756
[1.09] [1.81] [1.78] [1.73]
N of obs. 1,681 1,385 1,357 1,187
N of firms 396 295 241 170
    • * P < 0.1.
    • ** P < 0.5.
    • *** P < 0.01.

Panel A and B in Table 5 display the estimation results of Model 3 in the sub-samples on internal R&D intensity and firm size. The coefficients notably vary across the quartiles. In Panel A, the coefficient on internal R&D intensity in the first quartile (0.141) is far smaller than the punctual estimator (0.473) and statistically nonsignificant, whereas those in the other three quartiles are greater than that and statistically significant. The pairwise statistical tests yield significant differences in coefficients between the first quartile and each of the second-fourth quartiles, but not among the second-fourth quartiles. The strong complementarity at the high level of internal R&D is also seen in the previous study (Hagedoorn and Wang, 2012). On the contrary, Panel B indicates that the coefficient on internal R&D intensity in the first quartile is the greatest (0.533), and it decreases in quartiles. Again, the coefficient on internal R&D intensity in the first quartile is significantly different from the counterparts in the other quartiles. Thus, the strong positive associations between internal R&D intensity and R&D outsourcing is observed at higher levels of internal R&D as well as among smaller firms. These results corroborate the complementarity between internal R&D intensity and R&D outsourcing, while such complementary becomes stronger among smaller firms.

Both Panel A and B also show the coefficients on R&D diversification, which require more caution to interpret. Though the coefficients remain relatively stable across sub-samples except for the fourth quartile in Panel A, lying between 0.05 and 0.08, only the first quartile in Panel A and the first and the second quartiles in Panel B indicate statistical significance. The possible reason could be that in diversifying knowledge spaces, the necessity to rely on R&D outsourcing is greater for firms with scant resources and R&D experiences. To diversify their own knowledge spaces, simply combining knowledge in the existing sets that firms possess may not be sufficient: they need to reach out to the area that they had not targeted before. Such necessity to outsource R&D may be larger for firms whose knowledge bases are not well established yet.

Panel C and D provide the sub-sample estimations in the period 2002–2012 to examine differential effects of R&D doctorate across internal R&D intensity and firm size. The estimated coefficients are quite unstable across quartiles and only significant in the third quantile of Panel C and the first quantile of Panel D. The variability and the non-significance of the coefficients could be due to less statistical power, given that the samples in those estimations are already partial in terms of the observation period. It should be emphasised that the effect of possessing higher order human capital (i.e., PhD holders) on R&D outsourcing is, if any, contingent on firms’ characteristics such as internal R&D intensity and firm size.

Finally, we employ a couple of additional tests to see the robustness of our findings. First, we utilise two alternative measures of internal and outsourcing R&D scales instead of the ratios of those expenditures to sales. One is the log of the absolute amount of internal R&D and R&D outsourcing and the other is the log of the ratio of these two expenditures to operating income, instead of sales. Using these alternative measures generates similar results of complementarity, both in terms of the punctual estimators and increasing coefficients in the level of the internal R&D. Second, we use one-year lagged independent variables to account for simultaneity. This again does not change the results related to hypotheses 1 and 3, but the coefficient on PhD doctorate turns nonsignificant, which implies the higher sensitivity of its effect. Finally, we exclude from our samples firms operating in the electricity & gas industry due to its public nature and so the higher extent of R&D outsourcing, and obtain the consistent results in all the hypotheses.

6 Conclusion

This study aims to explore the determinants of R&D outsourcing from the perspective of RBV and KBV. Having access to firm-level information on R&D outsourcing over around 30 years from the Survey of Research and Development allows us to construct a panel and estimate firm fixed effect models, which account for heterogeneity in time-invariant unobservable factors such as organisational culture.

The findings substantiate the argument that internal R&D and R&D outsourcing are complementary. It should be noted, however, that in our panel there are indeed six firms whose outsourcing ratio within R&D budgets is over 90% at least for one year during the observation period. They include firms operating in electricity, telecommunication, shipping and transportation equipment industries. It seems that in these cases, internal and R&D outsourcing are substitutive activities. However, all of those firms had separated their R&D division from their business entity; they had set up their R&D laboratory as an independent organisation and they channelled their R&D expenditure to this laboratory. Thus, in this sense they still relied on internal R&D. Only twenty firms in our sample had used this approach, however, and excluding them as outliers does not alter the results of the analysis.

Our results also imply that doctoral scientists play an important role in the absorption of knowledge, as firms having a higher fraction of PhD holders within R&D personnel tend to outsource their R&D. It is consistent with the suggestion that corporate scientists who hold a doctoral degree play a bridging role in external and internal R&D activities (Shimizu and Hara, 2011). However, the coefficients on PhD doctorate are highly sensitive and contingent across sub-samples. The effect of higher order human capital in R&D outsourcing would need more exploration in future studies.

This study sheds light on R&D scope, which is another aspect of firms’ absorptive capacity, in addition to internal R&D and R&D human capital. It suggests that a broader technological knowledge base allows firms to increase their absorptive capacity, even controlling for the amount of investment in different sectors. Thus, not only the scale, but the scope of internal R&D is also related to absorptive capacity, which is another contribution of this study.

As with any previous literature, there are several limitations in this study. First, the survey used in our analyses does not provide detailed information about the recipients of outsourced R&D budgets. It only provides the type of organisation (either firms, universities or other research institutions) and around 72% of the outsourced R&D in our sample are devoted to firms. While our results do not change much even when we limit the samples to those whose recipients of outsourcing are universities, it is still possible that our baseline results could have been distorted if most of the outsourced R&D go to firms’ subsidiaries. As the survey does not distinguish between expenditure on R&D carried out by subsidiaries and by other firms, further research is needed to determine how R&D outsourcing behaviour varies across recipient organisational types.

A further limitation concerns with causality. Our findings are correlational rather than causal. It must be noted that although this study provides evidence on what kinds of firms outsource their R&D, more studies are necessary to identify the causality as to firms’ R&D outsourcing decisions.


    1. The Survey of Research and Development per se does not provide a standardized identifier such as a ticker symbol. We follow three steps to link the respondents with their financial and patenting information. First, based on their names, we connect the survey respondents to the corporate name list, the NISTEP Corporate Name Dictionary provided by the National Institute of Science and Technology Policy (NISTEP), to find ticker symbols, by which we gain the respondents’ financial information in DBJ’s database. Second, for unmatched respondents in the previous step, we manually match them according to their demographics and business characteristics (e.g., address, deliverable and industry). Third, to obtain patent records, we use unique identifiers in the NISTEP name list called comp_id to link them with patent applicants in the IIP database.
    2. Non-firm research-oriented organizations include public/private research universities, public/private research institutions, and NPOs. We note here that the threshold of 10-million-yen separating the sampling way only changed one time in 1995, increased from five million to 10 million yen. While this could distort the selection of the sample, our regression results shown later do not change even if we limit our samples to those after 1995.
    3. There are three reasons for our panel to be unbalanced. First, some sample attrition exists in our observation period: respondents drop out from the survey due to bankruptcy or acquisition. The average sample attrition rate by year is 4.5%. Second, since respondents whose stock capital is below the threshold (10 million yen) are sampled, they may not persistently appear throughout the period. The third reason pertains to the respondent identification method used in the survey. Before 2002, the Statistics Bureau of Japan had not revealed respondent names, but assigned 10-digit identifiers to each respondent. In 2002, the Bureau changed the method: It replaced new seven-digit identifiers with the previous 10-digit, and started revealing respondent names. It provided a correspondence table between the old 10-digit and the new seven-digit identifiers, but we can only match the respondents being surveyed both before and after 2002. This is only the case for the respondents under the threshold, and those respondents above it have been surveyed throughout the periods.
    4. We do not distinguish solo-invented patents from co-invented patents. While co-inventing patents are not highly frequent in Japan – just over 10% of Japanese inventions in 2007 involved an external co-inventor (Walsh and Nagaoka, 2009), this could be directly related to the dependent variable, which is one of the limitations of our study.


    • Shotaro Yamaguchi is a PhD student in Robert H. Smith School of Business, University of Maryland, College Park. His research interests lie in inventors’ mobility, industrial evolution and innovation patterns.
    • Ryuji Nitta is a PhD student in Innovation at Graduate School of Business and Administration, Hitotsubashi University. He recently researches National Innovation System and Knowledge Transfer.
    • Yasushi Hara is a faculty of Graduate School of Economics, Hitotsubashi University. Using his IT literacy as ex-ICT infrastructure engineers with economic and management academic discipline in his research activities, he has been conducting empirical studies relevant to policy making in the field of science, technology and innovation.
    • Hiroshi Shimizu is Professor of Waseda University, Faculty of Commerce. He received his Ph.D degree in 2007 from London School of Economics. His research includes Innovation, Entrepreneurship, Technological Change and Competitive Strategy. His recent book is General Purpose Technology, Spin-out, and Innovation: Technological Development of Laser Diodes in US and Japan published from Springer. He has published in Research Policy, Business History Review, Business History, and Journal of Evolutionary Economics.