International inventor teams and technological variety in multinational enterprises

Abstract

This study investigates the relationship between international inventor teams and the technological variety of multinational enterprises. We frame this relationship conceptually by considering two possible attributes of diversity in international inventor teams: cultural differences and heterogeneous knowledge. We employ a dataset for 454 multinational enterprises with 71,126 subsidiaries across 185 countries that applied for 139,066 priority patents during the period 2007–2014. Fixed-effects panel estimations indicate that international inventor teams are positively associated with both related and unrelated technological variety at the level of the MNE. Such relationships display diminishing marginal returns, pointing to management and coordination costs reducing the benefits from international inventor teams. In addition, we find that MNEs with higher technological innovation capability deal with the additional complexity from managing and coordinating international inventor teams by consolidating technological variety.

1 Introduction

Variety in the knowledge base is essential for innovation, since it determines the range of possible reconfigurations of knowledge (Arthur, 2007; Basalla, 1988; Nelson & Winter, 1982; Schumpeter, 1934). Weitzman (1998) concludes that ‘… the ultimate limits to growth may not lie so much in our abilities to generate new ideas, as in our abilities to process to fruition an ever-increasing abundance of potentially fruitful ideas’ (p. 359). Choices regarding variety play an important role in economics and innovation management, but often remain implicit (Van den Bergh, 2008).

Previous research dealt with strategies of multinational enterprises (MNEs) to maintain variety through diversification of technology drawing on evolutionary concepts such as path dependency and technological accumulation (see for example Cantwell, 1995; Piscitello, 2000; Cantwell & Piscitello, 2014) and the capability-based view of the firm (Granstrand, 1998; Teece et al., 1997). This line of thought suggests that firms diversify primarily into ‘related’ technological assets to retain patterns of corporate coherence (Breschi et al., 2003; Piscitello, 2004; Teece & Pisano, 1994; Teece et al., 1994). In terms of cognitive proximity, some degree of relatedness is required to recombine existing pieces of knowledge for incremental innovation by the firm. However, firms also diversify into unrelated technologies, thereby combining previously unconnected knowledge domains. While such recombination is more likely to fail, resulting innovations, when successful, are more likely to be of a radical nature (Fleming, 2001).

This study is one of the first to analyse determinants of the creation of technological variety within firms. While there is evidence that technological variety increases innovation in firms (Fleming, 2002; Garcia-Vega, 2006; Miller et al., 2007; Nesta & Saviotti, 2005; Quintana-Garcia & Benavides-Velasco, 2008) through access to a broader set of recombination paths, we find very little research on the determinants of and mechanisms for the creation of technological variety by firms (see Nerkar & Paruchuri, 2005; Cecere & Ozman, 2014).

Our analysis focuses on international inventor teams as a source of technological variety in MNEs. Recent studies emphasise that intra-firm knowledge transfer and integration ultimately depend upon the actions and interactions of individuals, such as inventors, sharing knowledge within the MNE (Castellani et al., 2022). Structural characteristics of individual inventors affect not only dissemination but also the recombination of knowledge throughout the organisation (Nerkar & Paruchuri, 2005; Paruchuri & Awate, 2017).

This study moves beyond individual inventor characteristics by inquiring whether and, if so, how international inventor teams create technological variety in MNEs. We posit that international inventor teams create novelty through the recombination of knowledge. Drawing from organisation research, we argue that novelty emerges from cultural differences and heterogeneous knowledge embedded in international inventor teams. Cultural differences create value by drawing from diverse perspectives (DiStefano & Maznevski, 2000; Lane et al., 2009), deemed particularly beneficial for conceptual and creative tasks (Hoever et al., 2012; Marino & Quatraro, 2022; McLeod et al., 1996). The second attribute, heterogeneous knowledge, reflects differences in know-how and expertise. We reason that this attribute is more prevalent in international inventor teams compared to domestic ones, since differences in technological specialisations are larger across than within countries (Archibugi & Pianta, 1992; Cantwell & Vertova, 2004; Picci & Savorelli, 2013).

However, diversity is a ‘double-edged sword’ (Milliken & Martins, 1996, p. 403). Excessive levels of diversity may also cause disparate mental models and interpersonal tensions, which might hinder the team’s ability to develop creative outcomes (Bassett-Jones, 2005; Khedhaouria & Jamal, 2015). Cultural differences within teams may inhibit their ability to develop a task strategy, resolve conflicts, and build cohesion (Anderson, 1983; Kirchmeyer & Cohen, 1992; Kirkman & Shapiro, 2005; Watson et al., 1993, 2002). Knowledge heterogeneity can be counterproductive due to a lack of coordination among otherwise homogenous groups (Cegarra-Navarroa et al., 2021; Zhang & Li, 2016). Thus, we propose a non-linear relationship: International inventor teams increase the possibility for knowledge recombination within and across technological domains, which, in turn, would be a source for related and unrelated technological variety at the level of the MNE. However, we expect that coordination and management costs associated with the integration of knowledge from inventor teams spreading over too many countries reduce the benefits from international inventor team diversity upon technological variety. We suggest that MNEs’ technological innovation capability could mitigate potentially negative effects from diversity in international inventor teams related to cultural differences and heterogeneous knowledge.

We tested these conjectures empirically, using a patent-firm-level dataset for a sample of MNEs headquartered in EU15¹ countries. We analysed data for 454 MNEs with 71,126 subsidiaries across 185 countries with 139,066 priority patent applications during the observation period 2007–2014. Fixed-effects panel estimations on variety measures computed on MNE patent portfolios indicate that international inventor teams are positively associated with both related and unrelated technological variety. However, such relationships display diminishing marginal returns, pointing to management and coordination costs reducing the benefits of cultural diversity and heterogeneous knowledge in international inventor teams. In addition, we find that multinational enterprises with higher technological innovation capability deal with the additional complexity of managing and coordinating international inventor teams by consolidating technological variety in the firm.

Section 2 provides a conceptual framework and develops three hypotheses on the relation between international inventor teams and technological variety in MNEs. Section 3 introduces data and descriptive statistics, as well as the measures and methods. In Sect. 4, we present the findings from the analysis and Sect. 5 concludes with a discussion.

2 Conceptual framework

2.1 Related and unrelated technological variety

Greater diversity in knowledge inputs means a greater potential for creative combinations (Weitzman, 1998; Van den Bergh, 2008). To produce new knowledge, firms use external sources in conjunction with internal R&D, which has been framed as an open innovation model (Almirall & Casadesus-Masanell, 2010; Benkler, 2006; Chesbrough, 2003; Dahlander & Gann, 2010; von Hippel, 2001, 2005). Searching broadly enhances firms’ knowledge through variation and novelty of the knowledge employed by the external source (Laursen & Salter, 2006; Teodoridis et al., 2019). Technological recombination requires language and interface commonality to be able to enter and be diffused within an organization (Forman & van Zeebroeck, 2019; Savino et al., 2017; Trantopoulos et al., 2017; Vaccaro et al., 2009). Today, there is a broad consensus that the probability of a firm, region, or country launching new and specific activities is a function of the number of the related activities they are specialised in (see for example, Frenken et al., 2007; Boschma & Iammarino, 2009; Neffke et al., 2011; Boschma et al., 2015; Hidalgo et al., 2018).

From a technological point of view, relatedness refers to the fact that although no two technologies are identical, they often share commonalities (Nooteboom, 2000). Proximity, commonality, and complementarity of knowledge are factors that might help to explain why firms diversify predominantly into related technologies (Breschi et al., 2003). Knowledge proximity refers to firms’ learning processes via unintended spillovers (Griliches, 1979; Henderson & Cockburn, 1996), as well as intended learning by focusing on technological domains that present similarities in problem-solving and knowledge bases (Dosi, 1997; Nelson & Winter, 1982). Knowledge commonality implies that the same type of knowledge is used in various technologies, whereas knowledge complementarity arises from the need to use technologies jointly (Milgrom & Roberts, 1990; Pavitt, 1998; Scott, 1993). Decision-makers may have limited cognitive capabilities to identify potentially fruitful combinations of pieces of knowledge unrelated to their existing knowledge bases and/or to each other (Nightingale, 1998; Nooteboom, 2000). While recombination of previously unconnected knowledge domains is more likely to fail, resulting innovations, when successful, are often of a radical nature as recombination across unrelated technologies can lead to completely new operational principles, functionalities and applications (Fleming, 2001).

Extant literature offers insights into the extent to which related and unrelated technological variety increases firms’ innovation through the recombination of knowledge (Aarstad et al., 2016; Fleming, 2002; Fornahl et al., 2011; Garcia-Vega, 2006; Miller et al., 2007; Nesta & Saviotti, 2005; Quintana-Garcia & Benavides-Velasco, 2008; Solheim et al., 2020). However, research on the determinants of and mechanisms for changes in related and unrelated technological variety is much more limited (Juhacz et al., 2020; Menzel, 2015).

2.2 Capabilities-based view on technological diversification in the MNE

The capabilities-based approach distinguishes the theory of the MNE from a mere special case of the theory of the firm, based on the specificities of the dynamic and continuous cross-border interaction of MNEs (Cantwell, 2014). To innovate, MNEs build synergistic or complementary portfolios of capabilities drawn from different locational contexts, which involve multidirectional knowledge transfer (Cantwell, 2009). The capabilities-based approach explaines the process of corporate technological diversification as an outcome of more closely integrated MNE networks and their internationalisation (Cantwell & Piscitello, 2000, 2014; Castellani & Zanfei, 2006; Rahko, 2016).

Over the decades, we gained important insights into MNE strategies to maintain variety through the diversification of technology. Firstly, technological competencies are multi-field as they include technological fields outside firms’ distinctive cores (Patel & Pavitt, 1997). Technological diversification is usually greater than product diversification and often anticipates product and market diversification (Pavitt, 1998). Operating within environments of converging or increasingly interrelated technologies, large firms develop and maintain a broad technological base, thereby becoming multi-technology corporations (Granstrand, 1998; Granstrand & Sjölander, 1992; Granstrand et al., 1997; Patel & Pavitt, 1997). Secondly, firms’ technological diversification changes only slowly because of the inertia of specialisation and incremental changes in knowledge production (Cantwell & Andersen, 1996; Fai & von Tunzelmann, 2001), as well as sunk costs of R&D (Narula, 2014). Thirdly, while specialisation results in economies of scale associated with the learning process and knowledge transfer between the core technologies of the firm (Breschi et al., 2003; Garcia-Vega, 2006), diversification generates economies of scale and scope due to cross-fertilisation between core and peripheral technologies (e.g., Granstrand, 1998; Granstrand & Sjölander, 1992; Granstrand et al., 1997; Piscitello, 2000; Suzuki & Kodama, 2004).

2.3 Role of inventors for variety in MNEs

Knowledge transfer and integration are highly dependent on the actions and interactions of individuals, such as inventors, sharing knowledge within the MNE (Castellani et al., 2022). When it comes to the circulation of technological knowledge, inventors are the key characters to observe (Fleming, 2001; Miguelez & Morrison, 2023), since they carry out the actual knowledge-creation processes (Allen & Cohen, 1969). Yet not all inventors are equally effective knowledge creators. For example, cross-border mobility of inventors within an MNE enables the temporary co-location of mobile inventors and, in turn, wider dissemination of their knowledge within the organisation (Castellani et al., 2022). Organisation research also hints at a relation between individual inventors’ structural characteristics, especially the centrality of inventors and the spanning of structural holes, and the processes of recombination of knowledge (Nerkar & Paruchuri, 2005). Inventors who span many structural holes are aware of and have access to organizational knowledge that is distributed in diverse pockets of the organization (Paruchuri & Awate, 2017). When such inventors generate innovations by recombining organisational knowledge, they gain a wider reach in the intra-organizational inventor network. Similarly, there is evidence of a positive relationship between the strength of intra-firm inventors’ ties and technological variety (Cecere & Ozman, 2014). However, if inventors are excessively embedded in networks, they become more like each other, which reduces opportunities for novel combinations of knowledge (ibid).

2.4 Hypothesis development

Moving beyond individual inventor characteristics, this paper looks at international inventor teams as a possible source for technological variety in MNEs. Previous research argued that multinational breadth, in terms of the dispersion of subsidiaries across countries, correlates positively with investments in R&D (Castellani et al., 2017) and also diversifies an MNE’s knowledge base, which increases the likelihood of discovering new and valuable combinations of ideas (Kafouros et al., 2012). MNEs undertake international R&D via strategic alliances (Almeida et al., 2002), acquisition of external R&D facilities (Awate et al., 2015), knowledge acquisition from local suppliers (Li et al., 2010), collaboration with foreign universities (Belderbos et al., 2020, 2021; Ivarsson et al., 2017), or ‘listening posts’ (Gassmann & von Zedtwitz, 1999; Monteiro & Birkinshaw, 2017). All these modes can involve a team of international inventors that appropriates the returns from R&D in the form of a cross-country patent application with inventors residing in different countries (Alkemade et al., 2015; Frost & Zhou, 2005; Laurens et al., 2015). Thus, ‘international inventor teams’ reflect not only the outcome of joint R&D (collaboration in a strict sense) but also the results of R&D contracts (acquisition) and R&D advice (services) (Bergek & Bruzelius, 2010). We already know that MNEs use international teams to coordinate dispersed but interdependent R&D activities (Ambos & Schlegelmilch, 2004). Furthermore, the value of innovative work by international inventor teams can exceed the returns from purely domestic inventor teams of the same MNE (Kerr & Kerr, 2018). This study investigates whether and, if so, how diversity associated with international inventor teams shapes technological variety in MNEs.

2.4.1 Diversity as a driver of variety within international inventor teams

Team diversity favours the search for innovative solutions by introducing more heterogeneous sources of knowledge (March, 1991). Cognitive diversity is often associated with the generation of ideas and creativity (Farr et al., 2003). Diversity as a team characteristic denotes the extent to which members differ with regard to a given attribute (Joshi & Roh, 2009; van Knippenberg & Schippers, 2007). This view focuses upon underlying differences in perspectives on a task as the more proximal indicator of a team’s increased cognitive resources (Harrison & Klein, 2007; van Knippenberg et al., 2004). Regarding international inventor teams, our argument focuses on two possible attributes: cultural differences and heterogeneous knowledge.

International inventor teams with members representing two or more cultures leverage creativity for product and service innovation more effectively than single-culture teams (e.g. Bouncken et al., 2015; DiStefano & Maznevski, 2000; Hoever et al., 2012; Jang, 2017; Lane et al., 2009). Diversity in cultural background is particularly salient (Stahl et al., 2010) because it affects the team members' beliefs, attitudes, and mindsets (van Knippenberg et al., 2013). Although ‘domestic’ teams can also be multi-cultural, international inventor teams are so by definition.

The second attribute, heterogeneous knowledge, reflects differences in terms of technological knowledge and expertise rather than differences in perspectives. We also find heterogeneous knowledge in exclusively domestic inventor teams. However, we expect this attribute to be more extensive in international inventor teams, since differences in technological specialisations are larger across than within countries (Archibugi & Pianta, 1992; Cantwell & Vertova, 2004; Picci & Savorelli, 2013). This provides a higher potential for knowledge complementarities in international inventor teams compared to purely domestic ones. Team members can acquire knowledge from one another to make further abstractions and analogies between problems and, thus, use experience-based skills to solve new problems (Zhuge et al., 1997).

Given that MNEs locate foreign R&D predominantly in host countries with a technological specialisation, to exploit an already existing technological advantage at home (Laurens et al., 2015; Le Bas & Sierra, 2002), international inventor teams in such R&D projects are likely to focus on technological domains that present high similarities in problem-solving and knowledge bases. In this case, collaboration within international inventor teams facilitates the effective exchange of proximate knowledge to work on incremental innovations. This collaboration often requires face-to-face communication and personal feedback to overcome the ambiguities of tacit knowledge (Teece, 1981). Knowledge proximity lowers the cost of combining knowledge and enables systematic and creative work, which is likely to foster related technological variety within the MNE.

Since competence-exploitation and competence-creation strategies are often employed simultaneously, MNEs also use foreign R&D to explore and search beyond existing technological specialisations at home (Cantwell & Mudambi, 2005, 2011; De Beule & Van Beveren, 2019). In this case, we would expect the diversity of knowledge within international inventor teams to be relevant, since it offers the opportunity to develop a broader set of recombination paths. These are essential for more radical innovations (Fleming, 2002), which are rare but usually come from a recombination of already existing knowledge (Hargadon, 2003), often based on the combination of mature and emerging technologies (Schoenmakers & Duysters, 2010). International inventor teams could be a way to access emerging technologies developing abroad for integration with existing technologies of the MNE. Thus, international teams could be a mechanism to attain greater unrelated technological variety within an MNE. Against this background, we hypothesise:

H1

Other things equal, the presence of international inventor teams is associated with greater technological variety of the MNE.

2.4.2 Coordination and management costs of international inventor teams

Diversity favours the opportunity for creativity, while at the same time increasing the likelihood that some team members will feel dissatisfied and detached from the group (Bassett-Jones, 2005; Milliken & Martins, 1996). Organisation research confirms that an excessive level of diversity may cause disparate mental models and interpersonal tensions, which hinder a team's ability to develop creative outcomes (Khedhaouria & Jamal, 2015). Cultural differences within a team may inhibit the team’s ability to develop a task strategy, resolve conflicts, build cohesion, and foster effective interaction among team members (Anderson, 1983; DiStefano & Maznevski, 2000; Kirchmeyer & Cohen, 1992; Kirkman & Shapiro, 2005; Watson et al., 1993, 2002). Negative social interaction and group incohesiveness are also likely to reduce the benefits from heterogeneous knowledge in international inventor teams in the effective creation of technological variety (Huang, 2009; Nissen et al., 2014). Maintaining regular international knowledge exchange, in turn, is costly, and spreading MNE networks across a larger number of countries increases organisational complexity (Castellani et al., 2017).

Knowledge heterogeneity of international inventor teams itself could be counterproductive if coordination is lacking among otherwise homogenous groups (Cegarra-Navarroa et al., 2021; Zhang & Li, 2016). MNE research team geographic diversity has a curvilinear relationship with the team’s innovation performance (Seo et al., 2020), which implies that some degree of homogeneity in collaborative research teams is beneficial (Coad et al., 2017). We would, therefore, expect that coordination and management costs associated with integrating knowledge from multinational inventor teams reduces the benefits from diversity upon technological variety at the level of the MNE. Thus, we hypothesise:

H2

Other things equal, there are diminishing marginal returns from the breadth of knowledge sourcing via international inventor teams upon an MNE’s technological variety.

2.4.3 Technological innovation capability as a moderator

Arguably, some MNEs are better than others at managing international knowledge flows to create technological variety. Many studies in innovation performance focus on R&D investments (e.g., Ahuja & Katila, 2001; Castellani et al., 2017; Mairesse & Mohnen, 2005). Others emphasize non-R&D inputs. For example, Bell (2009) refers more broadly to firms’ ‘capability to create new configurations of product and process technology and to implement changes and improvements to technologies already in use’ (p. 11). Such innovation capability consists of various assets related to physical capital, knowledge capital, human capital, and organisational capital (ibid.). The latter has been recognised as an important element in firms’ searches for innovative inputs from external sources (e.g., Katila & Ahuja, 2002; Laursen & Salter, 2006).

We expect that MNEs with higher innovation capability are better positioned to identify and integrate knowledge flows from international inventor teams, so that stronger positive effects upon technological variety can materialise. This, for example, could be facilitated by access to superior R&D equipment, staff experienced in international R&D collaborations, or a larger stock of previous knowledge within the MNE. Such technological innovation capability would then mitigate the potentially negative effects from diversity in international inventor teams related to cultural differences and heterogeneous knowledge. Therefore, we hypothesise:

H3

Other things equal, technological innovation capability positively moderates the relation between the presence of international inventor teams and an MNE’s technological variety.

3 Data, measurement and method

3.1 Data

Extant research used firm-level databases with information on patent applicants to measure R&D internationalisation and/or to approximate the incidence of foreign R&D (Alkemade et al., 2015; Dernis et al., 2015; Laurens et al., 2015). We extend this approach by accounting for affiliates’ entries into and exits from MNEs groups over time to relax the restrictive assumption of static MNE ownership structures (see for example Dernis et al., 2015 for a discussion of this limitation). This increases precision in the allocation of patents to the actual patent owners over time. We employ a dataset for 454 MNEs headquartered in the EU15 for the observation period 2007–2014.

Our data source is the BvD ORBIS database. We started by extracting data on active enterprises headquartered in an EU15 country, with at least 100 employees, being active and not dissolved, classified as an industrial company, having manufacturing as a primary activity (NACE Rev. 2 Section code C) and having at least one foreign subsidiary in 2014. Based on individual ownership links, we derived an historical ownership hierarchy for each MNE for each year of the observation period.² We excluded individuals or financial entities as global ultimate owners (GUOs) and instead selected the next industrial entity in the hierarchy as GUO. Then, we extracted all patent documents available in the patent section of ORBIS for all MNEs and their subsidiaries. The identified patent documents were matched with PATSTAT (Version 2019a) through the application number and the corresponding identification number of the entity.³ This enabled us to identify priority patent applications, their associated International Patent Classification (IPC) codes, and the countries of residence of their inventors. Using priority applications⁴ reduces country bias and enables a more accurate representation of inventor locations (De Rassenfosse et al., 2013). In line with existing research (see for example Picci & Savorelli, 2013; Laurens et al., 2015; Alkemade et al., 2015), we inferred the inventors’ location from the addresses of residence of the inventors, as indicated in the priority patent application.

This process yielded our final sample, which includes 454 MNEs with 71,126 subsidiaries across 185 countries and 139,066 priority patent applications. The average MNE was observed in the sample for 4.5 years and accounted for about 68 priority applications per year (see Appendix Table 3). Nearly all patent applications were made by inventor teams (only 0.2% report a single inventor). Inventors whose listed address is not in their MNE’s headquarters country constitute a relevant phenomenon: about 1 of every 5 inventors are foreign (i.e. 19.2%, 12.1% residing in the EU15 and 7.1% outside it) and about 1 of every 4 (i.e. 26.1%) patent teams are international, i.e. include a foreign inventor (see Appendix Tables 4, 5). Aggregated across the parent and subsidiaries, we found that 78% of MNE-year observations had patent applications with more than one IPC code and about 48% had inventors from more than one country (see Appendix Table 6).

3.2 Measurements and methods

Earlier studies also used patents as proxies for codified technological knowledge and used patent classifications to measure various aspects of technological relatedness and coherence at the level of the firm (e.g., Breschi et al., 2003; Colombelli et al., 2014; Nesta & Saviotti, 2005). Technological proximity can be estimated in several ways (Juhacz et al., 2020). We applied a hierarchical technological classification scheme as an ex-ante imposed structure that defines relatedness and unrelatedness. We computed entropy measures using the hierarchical structure of 610 IPC subclasses that fall exclusively under one of the IPC’s eight sections.⁵ Formally, in the equations that follow, k denotes IPC subclasses, j IPC sections and ${S}_{j}$ the set of IPC subclasses belonging to section j. We first computed the IPC subclass shares ${p}_{i,t,k}$ as the sum of the full count of patent applications of MNE group i falling in IPC subclass k in year t.⁶ Then, we computed the IPC section share ${P}_{i, t,j}$ by summing the IPC subclass shares ${p}_{i,t,k}$ of MNE group i falling in IPC section j in year t. Unrelated technological variety ${UV}_{i,t}$ of MNE i in year t is then given by:

$${UV}_{i,t}= \sum_{j=1}^{8}{P}_{i,t,j}\mathit{ln}\left(\frac{1}{{P}_{i, t, j}}\right)$$

(1)

and related technological variety ${RV}_{i,t}$ by:

$${RV}_{i,t}= \sum_{j=1}^{8}{P}_{i,t,j}{H}_{i,t, j},\mathrm{ where} {H}_{i,t,j}=\sum_{k\in {S}_{j}}\frac{{p}_{i,t,k}}{{P}_{i, t,j}}\mathit{ln}\left(\frac{1}{({p}_{i,t,k}/{P}_{i,t,j})}\right)$$

(2)

Thus, to account for cognitive proximity between technologies in an MNE’s patent portfolio, unrelated variety measures the extent of technological diversity across the eight broad IPC sections. MNE groups that patent in more sections have a higher index value for unrelated technological variety. Related variety, in turn, measures the extent of technological diversity within each IPC section across IPC subclasses and is larger for MNEs whose patents are distributed across more IPC subclasses within a single IPC section.

To test our first hypothesis (H1), we regressed our technological variety measures on variables based on the location of the corresponding inventors and other MNE characteristics. We applied a fixed-effect panel regression specified by:

$${TV}_{i, t}={\beta }_{1}{MNInv}_{i, t}+{\beta }_{2}{NCInv}_{i,t}+{\delta }^{\mathrm{^{\prime}}} {Z}_{i, t-1}+{d}_{t}+{u}_{i}+ {\varepsilon }_{i, t}$$

(3)

${TV}_{i,t}$ indicates the respective technological variety measure for MNE group i (i = 1, …, 454) at year t (t = 2008, …, 2014); ${\beta }_{1}$ is the parameter of ${MNInv}_{i, t}$, a dummy that indicates whether one or more patents applied by MNE i in year t includes a team of inventors with different countries of residence⁷; ${\beta }_{2}$ is the parameter of ${NCInv}_{i,t}$, which is the number of the countries of residence of the inventors in the pool of patent applications made by MNE i in year t (‘breadth of international knowledge sourcing’); ${\delta }^{\mathrm{^{\prime}}}$ is a vector of parameters associated to a vector of time-varying MNE-specific controls ${Z}_{i, t-1}$; ${d}_{t}$ denotes year fixed effects; ${u}_{i}$ is an MNE-specific time-invariant fixed effect; and ${\varepsilon }_{r,t}$ is the error term.⁸ The vector ${Z}_{i, t-1}$ includes the number of patent applications (${{\text{NPat}}}_{i, t-1}$, in log) made by MNE i, the number of subsidiaries (in log), the share of foreign subsidiaries to take into account MNE’s internationalisation of production, the number of employees and the value of intangible assets (both in logs) as well as measures for MNE related and unrelated variety across industries. All control variables are lagged by one year in order to minimise the impact of simultaneity between MNE’s attributes and technological variety.

Appendix Table 8 shows descriptive statistics of the variables entering the regression analysis.⁹ MNEs in the sample are large (about 20,000 employees on average) and highly innovative (with intangible assets and yearly R&D expenditures of about 1,750 and 289 million Euro, respectively, on average). They have a broad geographical presence, with 105 foreign subsidiaries spanning 23.6 countries, on average. They file an average of about 69 patents every year, about half of which involve international inventors spanning five countries of residence.

To test for decreasing marginal returns from the breadth of knowledge sourcing via international inventor teams on technological variety (H2), Eq. (4) introduces an additive quadratic term of the number of countries of residence of inventors (${NCInv}_{i,t}$):

$${TV}_{i,t}={\beta }_{1}{MNInv}_{i,t}+{\beta }_{2}{NCInv}_{i,t}+{\beta }_{3}{NCInv}_{i,t}^{2}+{\delta }^{\mathrm{^{\prime}}} {Z}_{i, t-1}+{d}_{t}+{u}_{i}+ {\varepsilon }_{i, t}$$

(4)

Finally, we tested for the role of innovation capability in moderating the effects of international knowledge sourcing on technological variety (H3) by augmenting Eq. (5) with an interaction term: the product of the number of countries of residence of inventors (${NCInv}_{it}$) and the number of patent applications¹⁰$({{\text{NPat}}}_{i, t-1}$):

$${TV}_{i,t}={\beta }_{1}{MNInv}_{i,t}+{\beta }_{2}{NCInv}_{i,t}+{\beta }_{4}{{\text{NPat}}}_{i, t-1}+{\beta }_{5}{NCInv}_{i,t}{*{\text{NPat}}}_{i, t-1}+{\delta }^{\mathrm{^{\prime}}} {Z}_{i, t-1}+{d}_{t}+{u}_{i}+ {\varepsilon }_{i,t}$$

(5)

Regression parameters’ of models (3)–(5) may suffer from endogeneity issues that our fixed effects panel estimators can not address, and should therefore be interpreted as correlates of technological variety. While sources of potential endogeneity with an inventor team’s composition and diversity look hard to identify, the relationship between technological variety and technological capacity at the core of H3 seems trickier. In fact, time-varying omitted shocks that are specific to certain countries, industries or technological fields, induced for instance by national policies encouraging innovation in specific domains, the upsurge of technological breakthroughs, and market dynamics entailing changes in the demand for products, might influence both the number of patents and the variation of MNEs patenting across fields. In order to attenuate biases coming from such potential confounders, we also considered models’ specifications that include a time-varying measure of the technological specialization¹¹ of the MNE’s headquarters country, as well as year-specific controls for MNE industry and headquarters country. The inclusion of these additional controls left the parameters of interest unaltered, so we opted to exclude them from our baseline specification for the sake of parsimony.

4 Findings

Table 1 shows the estimation results of fixed effects regression models for related and unrelated variety. It reports the results of four specifications for each variety measure. All models include a dummy for the presence of international inventors and the number of inventor countries of residence,¹² which we expected to be positively associated with the variety measures. Models (2) and (6) introduce a quadratic term of the number of inventor countries of residence to test potential diminishing marginal returns. Models (3) and (7) introduce an interaction term between the number of inventor countries and the number of patent applications.

Table 1 Fixed effects models on MNEs’ technological variety

Models (4) and (8) include all variables, i.e. the quadratic specification of the number of inventor countries and its interaction with the number of patents.

In line with H1, the dummy indicating the presence of international inventor teams is positive and significant across all specifications. Thus, the presence of international inventor teams is positively associated with both related and unrelated technological variety. The same applies to the number of inventor countries of residence. However, the quadratic term introduced in models (2), (4), (6) and (8) is negative, though not significantly different from zero when the interaction between the number of inventors’ countries and the number of patents is also introduced in the model (models (4) and (8)). The result indicates the presence of positive correlations between variety measures and the number of inventor countries (breadth of knowledge sourcing). However, these positive correlations fall slightly as the number of inventor countries increases.

Models (3) and (4), which include the interaction term between the number of inventor countries and the number of patents, show negative and significant estimates for related variety. The sign is negative also for unrelated variety, but the estimate becomes statistically non-significant in model (8), which also includes the quadratic term of the number of inventors’ countries.

In order to better assess the relationships between international inventor teams and technological variety in the models that include the quadratic and interaction terms, Fig. 1 plots the average marginal effects (AMEs) of the number of inventor countries on technological variety. Figure 1 indicates that the AMEs fall for both related and unrelated variety as the number of inventor countries itself and the number of patents increase. Statistically significant positive effects are present for low numbers of inventor countries and patent applications, but are not statistically different from zero at high values. Figure 1 also indicates that point estimates become increasingly imprecise as the number of inventor countries increases, with the confidence interval getting larger as the number of data points decreases. Yet, this feature of the specifications does not affect the qualitative pattern of the focal relationships. In line with H2, the results provide evidence for decreasing marginal returns as the breadth of international inventor teams across countries becomes increasingly broad.

Fig. 1

Average marginal effects of the number of inventor countries on technological variety. Notes: Average marginal effects’ (AMEs) are derived using the sample mean of all explanatory variables other than the number of inventors’ countries and the number of patents. The vertical lines indicate the means of the number of inventors’ countries and the number of patents. The graphs for related variety use model (4)’s estimates (Table 1), while the graphs for unrelated variety use model (8)’s estimates. The grey area indicates 95% confidence intervals

As for the moderating role of innovation capability, we find, in contrast with H3’s expectation, that higher innovation capability is associated with a lower correlation between multinational inventor teams and MNE’s technological variety. As for the other explanatory variables, the direct effect of innovation capability is positive and significant for related variety, and not statistically different from zero at standard confidence levels for unrelated variety. Thus, diversification into related technologies increases with higher innovation capability; diversification into unrelated technologies does not. Most of the coefficients of other explanatory variables, shown in Appendix Table 9, are not statistically significant.¹³

The results of the analysis are robust to a series of robustness checks shown in Table 2, which reports estimates of baseline (models (4) and (8) of Table 1) and alternative fixed effects models.¹⁴ First, we use fractional rather than full counting for the patents with multiple applicants in the construction of variety measures.¹⁵ Estimates of models (2) and (8) in Table 2 are nearly unchanged with respect to the relevant baseline.

Table 2 Robustness checks

Second, we replicate the analysis using only MNEs-years observations with more than 5 patents.¹⁶ Variety measures show quite large variation, also for MNEs with one or few patents because most patents (78%, see Appendix Table 6) are associated to multiple IPC classes. This enables us to test that results are not driven by MNEs with few patents. Also, in this case, estimates of models (3) and (9) in Table 2 are very similar to the relevant baseline as the estimated coefficients display the same sign and significance. With respect to the baseline, we notice a larger positive association between innovation capabilities and related variety.

Third, we estimated a model that also includes R&D expenditure (in log and lagged by one year). R&D expenditures are an important input for an MNE’s innovative activities and could influence variety measures. While we do not use R&D expenditures in the baseline models due to a high number of missing values in the sample (25%), models (4) and (10) in Table 2 indicate that, once innovation capability and international knowledge sourcing are controlled for, R&D expenditures are not associated with an MNE’s technological variety. In brief, the results of the baseline models do not change when R&D expenditures are included.

In addition, we tested the robustness of our baseline models by excluding countries that could bias our findings. Models (5) and (11) in Table 2 exclude Germany, which is the most-represented country in the sample, accounting for about 31% of MNEs (see Appendix Table 3). While the the dummy variable indicating the presence of international inventor teams becomes statistically non-significant due to an increase of the associated standard error leading the p-value to 0.103, at the edge of statistical significance at the 90% level in the case of related variety, the findings of the baseline analysis are substantially confirmed for both related and unrelated variety. We also report estimates of models excluding MNEs from the UK, which account for about 17% of MNEs in the sample but show comparatively low patent intensity (on average, 11 patents per year versus 68 by the average MNE).¹⁷ All results are very similar to those of baseline models (see models (6) and (12) in Table 2).

Results are also confirmed by further robustness checks that we do not report for the sake of brevity.¹⁸ In order to test how the financial crisis may have influenced our estimates, we augment the baseline specification, which simply controls for common year effects, with country- and industry-specific business-cycle variables. In particular, results do not change when we include unemployment rates in MNE headquarter-countries and industry-specific (at the NACE Rev. 2 Section level) GDP levels and growth rates. Models’ results are also robust to the inclusion of a time-varying measure of the technological specialization of the MNE headquarters country (computed over alternative time windows to smooth potential volatility in patenting activities across technological fields, especially for smaller countries), as well as of year-specific dummies for the industry and MNE headquarter-countries. Finally, in order to test the extent to which our results change when using lag structures that allow for more time for a patent to be generated, we lagged control variables by two and three years, re-estimated all models, and find no substantial difference with the findings obtained in the baseline using a one-year lag.

5 Discussion

5.1 Main findings and contributions

This research builds upon the capability-based view that international integration of MNE networks facilitates the diversification of corporate technology (Cantwell, 2009; Cantwell & Piscitello, 2000, 2014; Castellani & Zanfei, 2006; Rahko, 2016). We advance the state-of-the-art in two ways. Firstly, this is one of the first studies to focus on the determinants rather than on the consequences of variety by using the related and unrelated technological variety of MNEs as dependent variables. Secondly, we focus on international inventor teams as creators of technological variety in firms. Conceptually, we frame this relationship by considering two possible attributes of diversity in international inventor teams: cultural difference and heterogeneous knowledge.

Our econometric results indicate that international inventor teams might enhance both related and unrelated technological variety of MNEs. This suggests that, in principle, MNEs can overcome the need for spatial proximity in the creation of related variety, which has been postulated in the literature on evolutionary economic geography (Boschma & Iammarino, 2009; Frenken et al., 2007). Our results also suggest that international inventor teams enhance an MNE’s ability to create unrelated variety, which might reflect their quest to combine previously unconnected knowledge domains, which in turn are key to rare radical or breakthrough innovations (Castaldi et al., 2015; Fleming, 2001; Solheim et al., 2020). This finding challenges the conclusion from previous research that ‘…firms go abroad for augmenting or exploiting their home base, not for acquiring new bits of knowledge outside the technologies they master at home’ (Laurens et al., 2015, p. 773).

In line with H1, our econometric results document that the breadth of MNEs inventor countries increases technological variety. However, in line with H2, increases in variety flatten as the number of inventor countries increases. These diminishing marginal returns could be explained by cognitive limits to knowledge identification (Noteboom, 2000), disparity between originating ideas (Olsson & Frey, 2002), inhibited social interaction and group cohesion due to cultural differences within international teams (DiStefano & Maznevski, 2000; Kirkman & Shapiro, 2005; Lane et al., 2009) – all resulting in increasing costs for coordination and management of knowledge transfer and integration from international R&D (Castellani et al., 2017; Narula, 2014). Our finding of diminishing marginal returns supports the view that team diversity might both provide opportunities for creativity and hinder a team's ability to develop creative outcomes (Bassett-Jones, 2005; Khedhaouria & Jamal, 2015; Milliken & Martins, 1996).

We also find that an MNE’s technological innovation capability, approximated by the number of previous patents, is an important direct predictor of related variety. Unrelated variety, by contrast, shows similar values for MNEs with patent portfolios of varying sizes. Thus, diversification into related technologies gains in importance as an MNE’s innovation capability increases. This finding supports the view that technological diversification is characterised by expansion into related assets to establish patterns of corporate coherence (Breschi et al., 2003; Piscitello, 2004; Teece & Pisano, 1994; Teece et al., 1994). Our results also indicate that international knowledge sourcing via multinational inventor teams for MNEs with lower innovation capability might complement strategies that aim to create unrelated rather than related technological variety. This can be associated with the argument that it is possible for MNEs to ‘jump’ stages of technological evolution via international knowledge sourcing (see, for example, Enderwick & Buckley, 2021).

Finally, in contrast to H3, our results show that the positive contribution of international inventor teams upon technological variety declines as an MNE’s innovation capability increases. This finding could suggest that MNEs with higher technological innovation capability deal with the higher complexity of managing and coordinating heterogeneous inventors from different countries by reducing technological diversity more than MNEs with a lower technological innovation capability. In other words, it is likely that MNEs with higher technological innovation capability engage in international knowledge sourcing in ways that reflect consolidation towards their technological core, rather than diversification into related or unrelated variety—given that they already operate at a higher level of variety than MNEs with lower technological innovation capability.

5.2 Limitations

This study exploits a firm-patent level dataset that is not exempt from limitations. First, even though control variables are introduced in the empirical analysis with a one-year lag to minimise the impact of the potential simultaneity between MNEs' attributes and technological variety, the results should be interpreted as descriptive of the variety of an MNE’s patent portfolio, without any presumption of causality.

Second, the observation period is restricted to eight years.¹⁹ This is relatively short in comparison to studies of long-term patterns of firm-level technological change (e.g., Cantwell & Vertova, 2004), though it compares well to most studies on technological variety at the firm level (e.g., Aarstad et al., 2016; Capozza et al., 2020; Solheim et al., 2020). In addition, our observation period includes the global financial crisis, which may distort the relationships of interest due to firms having scaled back their R&D (OECD, 2009). Yet, due to sunk and opportunity costs, firms may have reduced R&D less than other expenditures for other functions such as sales, marketing and post-sales activities during the crisis (Rafferty, 2003). Moreover, during the crisis, the existence of an internal R&D department, as well as strategies aimed at exploring new markets and new product developments, proved to be important determinants of the persistence of innovation (Archibugi et al., 2013).

Finally, our approach exploits patents to measure firms’ technological knowledge, which have limitations well known and extensively discussed in previous studies (e.g. Griliches, 1990; Hall et al., 2014; Hussinger, 2006; Pavitt, 1988). However, previous studies highlighted the usefulness of patents as measures of the discovery of new knowledge. In particular, a firm’s patent portfolio is a reliable measure of technological knowledge and innovation usable for production, especially as compared to innovation measures based on surveys (Acs et al., 2002).

5.3 Future research

In this study, we do not differentiate intra-organisational vs. inter-organisational MNE networks in the production of technological variety. MNEs might rely primarily on internal networks between international inventors to create related variety but might shift to external networks to diversify towards unrelated variety. This aspect might also be important, since organisational research has argued that institutional diversity hampers effective knowledge sharing and negatively affects innovation outcomes of international R&D teams (Brunetta et al., 2020). Differentiating between MNE’s inventor teams involving internal and external networks was beyond the scope of this study but could be accomplished empirically by analysing patent citations (Frost, 2001). Future research could also investigate whether the mode of entry helps to explain the diminishing marginal returns from international inventor teams on technological variety, since the cost of coordination and organisational complexity could be higher in case of acquired MNE units vs. greenfield projects.

Finally, future research could question how the presence of international inventor teams moderates the relationship between technological variety and the innovative performance of MNEs. This would integrate previous research in evolutionary economic geography that considers the effect of unrelated and related variety on innovation performance depending on firm location (see for example, Solheim et al., 2020) with research that looks at the effect of foreign knowledge networks on MNEs’ radical vs. incremental innovation (see for example, Berry, 2018). While we focused on the ‘international’ dimension of knowledge sourcing, future research could employ a research design that allows for within-country variation in the international search for technological variety.

Appendix

See. Tables 3, 4, 5, 6, 7, 8, 9 and 10

Table 3 Number of patents and MNEs by country in the sample

Table 4 Inventors team size by country

Table 5 Frequent combinations of MNEs' HQ countries and inventors' countries of residence

Table 6 Number of inventors, patents and IPC classes by country in the sample

Table 7 Definition of variables

Table 8 Descriptive statistics

Table 9 Fixed effects full models on MNEs’ technological variety

Table 10 Robustness checks—full models on MNEs’ technological variety