Why space matters in technological innovation systems—Mapping global knowledge dynamics of membrane bioreactor technology

Abstract

Studies on technological innovation systems (TISs) often set spatial boundaries at the national level and treat supranational levels as a geographically undifferentiated and freely accessible global technological opportunity set. This article criticizes this conceptualization and proposes instead to analyze relevant actors, networks and processes in TIS from a relational perspective on space. It develops an analytical framework which allows investigating innovation processes (or ‘functions’) of a TIS at and across different spatial scales. Based on social network analysis of a co-publication dataset from membrane bioreactor technology, we illustrate how the spatial characteristics of collaborations in knowledge creation vary greatly over relatively short periods of time. This finding suggests that TIS studies should be more reflexive on system boundary setting both regarding the identification and analysis of core processes as well as in the formulation of policy advice.

Full text

1
Why space matters in technological innovation
systems – Mapping global knowledge dynamics of
membrane bioreactor technology
Christian Binza, Bernhard Trufferb and Lars Coenenc
a Corresponding author. Eawag, Swiss Federal Institute of Aquatic Science and Technology, Überlandstrasse 133, 8600
Dübendorf, Switzerland. christian.binz@eawag.ch. Tel: +41 (0)58 765 5674, Fax: +41 (0)58 765 5028
b Eawag, Swiss Federal Institute of Aquatic Science and Technology, Überlandstrasse 133, 8600 Dübendorf, Switzerland.
benrhard.truffer@eawag.ch
c Circle, Lund University, Sweden. lars.coenen@circle.lu.se
Paper published in Research Policy
Please cite as: Binz, C., Truffer, B., Coenen, L., 2014. Why space matters in technological innovation
systems – The global knowledge dynamics of membrane bioreactor technology. Research Policy 43
(1), 138-155. DOI: 10.1016/j.respol.2013.07.002
© 2016. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
Keywords: Technological innovation system, relational space, system functions, knowledge
creation, social network analysis, membrane bioreactor
Abstract
Studies on technological innovation systems (TIS) often set spatial boundaries at the national
level and treat supranational levels as a geographically undifferentiated and freely accessible
global technological opportunity set. This article criticizes this conceptualization and proposes
instead to analyze relevant actors, networks and processes in TIS from a relational perspective
on space. It develops an analytical framework which allows investigating innovation
processes (or ‘functions’) of a TIS at and across different spatial scales. Based on social
network analysis of a co-publication dataset from membrane bioreactor technology, we

2
illustrate how the spatial characteristics of collaborations in knowledge creation vary greatly
over relatively short periods of time. This finding suggests that TIS studies should be more
reflexive on system boundary setting both regarding the identification and analysis of core
processes as well as in the formulation of policy advice.
1. Introduction
Technological innovation systems (TIS) have become a popular and resourceful approach for
the analysis of innovation processes and early industry emergence, especially in the context of
recently developing clean-tech sectors (Markard et al., 2012). However, this literature’s
narrow geographical focus on industrialized countries and the overriding emphasis on
processes at the national scale are increasingly criticized (Berkhout et al., 2009; Coenen et al.,
2012). Continuing globalization and the fast rise of new industries in emerging economies add
considerable complexity to the spatial extent of innovation processes (Berkhout et al., 2009;
Bunnell and Coe, 2001). It thus becomes increasingly important for innovation scholars and
policy makers to understand how innovative activity is organized globally and how
innovation processes work at and between increasingly interrelated spatial scales.
The technological innovation system (TIS) concept allows in principle for such an
international analysis. Conceptualizing innovation systems without setting a priori territorial
boundaries can be seen as a distinctive feature of the TIS concept. In contrast to other
innovation system approaches that have pre-defined territorial delineations, e.g. at the national
(Lundvall, 1992) or regional (Cooke et al., 1997) scale, TIS proponents argue that by taking
technology as a starting point, the approach cuts across spatial boundaries (Hekkert et al.,
2007). Counter to this original vantage point, most of contemporary TIS literature delineates
empirical studies ex-ante on the basis of territorial (often national) boundaries (Coenen et al.,
2012; Markard et al., 2012). The broader (global) context of the system under study is often
conceptualized as representing a ubiquitous ‘global technological opportunity set’ (Carlsson,
1997a), to which all actors have indiscriminate access.
Mindful of the uneven geographical distribution of innovative activity (Asheim and Gertler,
2005), Coenen et al. (2012) propose a more careful treatment of space in TIS studies which is
pronouncedly relational and multi-scalar, avoiding a priori scalar boundaries and hierarchies.
Also other TIS proponents have started acknowledging the need to better understand

3
relationships between technological and other types of innovation systems (regional, national)
to avoid a reified, decontextualized treatment of technological innovation systems and to
improve policy advice based on the TIS approach (Jacobsson and Bergek, 2011).
Therefore, in applying a relational conceptualization of space, the objective of this paper is to
develop an analytical framework for TIS that is explicitly spatial but at the same time avoids a
fixation on specific territorial units or singular scales. This suggests to start from a network
perspective and ‘follow the network wherever it leads’ throughout its development over time
(Coenen et al., 2012). This means using the relational properties of the actors to identify
relevant places and spatial levels of a TIS, a posteriori. In developing this analytical
framework, the paper elaborates on how to specify whether, why and how space matters in
studies of technological innovation systems, what errors might be incorporated in nationally
delimited case studies and how policy advice could accordingly be improved.
The specific approach presented in this paper aims at explicating the spatial reach of core
processes driving TIS dynamics, the so called TIS functions (Hekkert et al. 2007). By tracking
the activities of core actors over time, processes like knowledge creation, entrepreneurial
experimentation or market formation can be related to specific spatial setups. A relational
view emphasizes that actors contribute to these processes by drawing on resources that they
can access through specific networks. These networks may be confined to specific regions
(e.g. as in the case of industry clusters) but they can as well span over several continents. An
explicit analysis of the geography of these functions thus scrutinizes the differential access of
TIS actors to resources and institutional contexts that are unevenly distributed across space.
The notion of a global opportunity set is therefore replaced by a concept of differential access
to unevenly distributed resources in the spaces of a ‘global TIS’. While the conceptual
argument is explicated for all TIS functions, we are restricting the empirical illustration to one
core function that often plays a dominant role in early formation processes (Bergek et al.,
2008a): knowledge creation. We will measure the spatial structure of actors and their
collaborations by analyzing co-authored ISI publications in the field of membrane bioreactor
(MBR) technology.
In the next section, the problems of limiting TIS studies to a national level will be discussed
and the potential benefits of a relational geographic perspective for assessing the spatial reach
of core functions will be elaborated in more detail. Section 3 introduces social network

4
analysis as a tool for spatial analysis of TIS functions and develops and operationalizes a set
of respective indicators. Section 4 discusses the selected co-publication dataset and applies the
framework to knowledge creation in the TIS of MBR technology. The results suggest that
knowledge creation in this field evolved from a nurturing phase dominated by globally
spanning networks to a Europe-based expansion phase and finally to a multi-scalar, Europe-
and Asia centred consolidation phase. We conclude by discussing the implications of the
observed strong spatial-temporal dynamics in innovation activities for future TIS studies and
policy making.
2. Conceptualizing space in TIS
The TIS concept emerged in the early nineties from a quickly expanding innovation system
literature, which is rooted in evolutionary economics and industrial dynamics (Freeman, 1987;
Lundvall, 1992; Nelson, 1993). TIS are defined as a “network of agents interacting in a
specific economic/industrial area under a particular institutional infrastructure or set of
infrastructures and involved in the generation, diffusion, and utilization of technology”
(Carlsson and Stankiewicz, 1991, p.111). Innovation is conceptualized as an interactive,
recursive process, embedded in a set of co-evolving actors, networks and institutions. TIS
literature thus pronouncedly rejects the idea of linear innovation paths and emphasizes instead
the importance of systemic interplay of complementary actors, interactive and recursive
learning processes and the institutional embeddedness of innovation (Bergek et al., 2008a;
Carlsson and Stankiewicz, 1991). One does typically divide between TIS structure and
processes (or ‘functions’). Structure is defined as the actors, networks and institutions that
conjointly support the generation, diffusion and utilization of a new technology (Bergek et al.,
2008a). A structural analysis is complemented with a dynamic view on innovation system
build-up, by focusing on a set of functions, as defined in two programmatic papers by Bergek
et al. (2008a) and Hekkert et al. (2007). A TIS most successfully creates and diffuses new
technologies if its actors sustain six key system-building processes, namely knowledge
creation, entrepreneurial experimentation, market formation, influence on the direction of the
search, resource mobilization and creation of legitimacy1.
1 We synthesized the two lists of functions slightly: Creation of external economies, which is only mentioned by Bergek et al.
(2008) was not considered here, whereas ‘knowledge development’ (Bergek et al., 2008; Hekkert et al., 2007) and
‘knowledge diffusion through networks’ (Hekkert et al., 2007) are summarized in the shorter term ‘knowledge creation’.

5
2.1 The need for a notion of space in TIS
The conceptualization of space in TIS studies is rather simplistic and ignores that system
build-up is an inherently spatial process which might transcend territorial boundaries and
spatial scales. So far, TIS studies have essentially employed national borders as scalar
envelopes (Cooke, 2005) that contain all the relevant processes of an innovation system.
Especially the interaction of various TIS elements with the ‘global technological opportunity
set’ (Carlsson, 1997a) has not been further specified. When outlining the original framework,
Carlsson (1997b, p.776) assumed that “the technological opportunities facing any economic
agent are virtually unlimited; the pool of global possibilities has practically no boundaries”.
This view is increasingly criticized (Coenen et al., 2012). In many sectors, the global
opportunity set is conditioned by differential absorptive capacities at the level of individual
organizations. Actors differ in their ability to tap into external knowledge sources and to make
use of it for innovative activities. This explains why, despite the potential existence of a
ubiquitous global opportunity set, innovation activities are not uniformly or randomly
distributed across the global landscape. Moreover, tacit dimensions of knowledge may be
sticky, which means it does not travel easily beyond the context in which it was generated
(Gertler, 2003). This results in dual knowledge flows for innovation activities that consist not
only of localized learning embedded in local nodes (Maskell and Malmberg, 1999) but also
global knowledge networks in the form of international epistemic communities (Amin and
Roberts, 2008), corporate networks of transnational companies (Chaminade and Vang, 2008)
or temporary proximity and face-to-face interaction at international trade fairs and
conferences (Bathelt and Schuldt, 2008). Scrutinizing these interconnected relational
dynamics has been ignored so far by TIS research, but become one of the hallmarks in the so-
called relational turn in economic geography (Bathelt and Gluckler, 2003; Boggs and Rantisi,
2003).
2.2 Applying a relational perspective on TIS space
In a relational perspective, spaces and places are shaped not only by processes and
interactions happening within a specific territory but also by the impact of wider sets of
structures and processes (Bathelt and Gluckler, 2003; Yeung, 2005), that are fluent and
constantly reorganizing at all scales (Amin, 2002). Actors thus have significant relationships
(through which they seek to access resources to achieve their individual goals) at different
spatial levels that simultaneously influence their behaviour (Amin, 2002; Bunnell and Coe,

6
2001; Coe and Bunnell, 2003; Coenen et al., 2012). Relational economic geography has
therefore put a premium on networks as a conceptual and methodological underpinning to
analyze (uneven) spatial development (Glückler, 2007; Ter Wal and Boschma, 2009).
Networks span space by establishing transversal and topological interlinkages among
geographically dispersed locations or organizational units (Brenner et al., 2011). This does
however not mean that a networked perspective by default presupposes distanced, global
relations. Network spaces may as well be concentrated in a particular locality. Economic
geographers have shown that often a combination of dense local ties and extended extra-
regional connections creates successful long-term innovativeness of actors, places or
innovation systems (Bathelt et al., 2004). We would thus expect that such local ties and extra-
regional connections are equally relevant for the core innovation processes in a TIS. How this
combination plays out empirically is however contingent on a number of factors such as the
type of industry and its knowledge base (Asheim and Coenen, 2006) or the institutional
conditions of a region (Tödtling and Trippl, 2005). A relational perspective on space thus
suggests that relying only on interaction at one scale (e.g. the regional scale in regional
innovation systems or the national scale in technological innovation systems) curtails the
significance of relevant interaction at other scales or treats it as a merely exogenous factor.
This reveals a key challenge for TIS research. While indeed the development of a technology
or technological fields does not stop short of territorial borders, its spatial set-up is neither
randomly spread across the geographical landscape, but contingent on the specific technology
in focus and the resources and relationships of actors involved in driving the relevant
innovation processes.
Analyzing networks therefore potentially allows scrutinizing the spatial extent and structure
of core TIS processes. Networks have held a core position in the TIS approach since the
earliest writings. The actual use of the term has, however, been restricted to a mostly
qualitative and metaphorical level (Coenen and Díaz López, 2010; Kastelle and Steen, 2010).
This is not very surprising as getting a grasp of the plentiful and very diverse types of
networks that define a TIS is a delicate task: they can be formal, informal, short-run, long-
lasting, trans-disciplinary, exclusive, open or strategic and spanning between diverse actor
types (Musiolik and Markard, 2011). Nevertheless, they are of key importance for explaining
how innovation and a supportive institutional context are created by TIS actors. Formal
networks as a recent example have been shown to create system resources that are crucial for
maturation and diffusion of new technologies (Musiolik et al., 2012). The spatiality of these

7
actor networks which enact key system build-up processes (and ultimately structural change)
has however not yet been further specified.
Thus, also when speaking about networks, TIS research so far mostly restricted its analytical
focus to ties within national territories. It would therefore be adequate to explicitly label these
studies as ‘national TIS’ snapshots of a wider ‘global TIS’. Shifting to a relational perspective
on space thus means explicitly analyzing TIS structure and processes from a global, relational
perspective. Whether or not sufficiently coherent systemic interaction may be identified in
specific countries, regions or continents can then be treated as an empirical question.
2.3 A networked perspective on TIS functions
This implies a fundamentally new inroad to the way TIS analysis is approached. Instead of
delimiting system boundaries ex ante we propose to start with a technological boundary and
to then empirically reconstruct whether sufficiently coherent sub-systems overlap with
specific regional or national boundaries. Existing schemes of analysis (Bergek et al., 2008a;
Hekkert et al., 2007) would accordingly have to be adapted. A relational spatial perspective
demands an explicit consideration of spatially structured networks for driving core processes
of TIS development. It thus becomes crucial to discuss where spatially extensive actor
networks become important elements of TIS functions and where, as a consequence, a myopic
focus on nationally bound networks is likely to miss out on important causal factors. 2
Knowledge creation, for instance, is usually defined without reference to the actors or
networks involved in the process, but with a focus on the way it is generated; e.g. Hekkert et
al. (2007) distinguish between knowledge produced through “learning by searching” or
“learning by doing”. In our view, a distinction between codified and tacit knowledge could be
a fruitful extension here: Codified (or ‘explicit’) knowledge can be easily transferred between
creator and recipient; codified knowledge bases of technologies are thus – at least partly –
public goods e.g. created in the science system and “originating from various geographical
areas all over the world” (Bergek et al., 2008a, p.414). Tacit knowledge is in contrast hardly
accessible in conscious thought, only producible in practice and strongly context-dependent
2 Note that at this point it will not be possible to expound an exhaustive ‘theory’ on the relationship between different TIS
structures and functions.

8
(Gertler, 2003). It therefore evolves in much more complex settings: its creation and
dissemination is in many cases still restricted to interaction in densely co-located actor
networks (Gertler, 2003; Maskell and Malmberg, 1999), yet also increasingly mobilized and
effectively shared in international networks and communities (Bathelt and Schuldt, 2008;
Crevoisier and Jeannerat, 2009; Wenger, 1998). As tacit and codified knowledge co-evolve,
one can assume to find considerably complex geographic network structures underlying
knowledge creation: Subnational clusters of dense interaction, combined with increasing
distant connections between actors in international networks and communities. A respective
analytical framework for this function will be proposed in this paper.
Entrepreneurial experimentation depends on new companies entering a field, and especially
the networks forming between them and supportive partners in an experimentation process,
typically in protected market niches (Bergek et al., 2008a, p. 416). This process is inherently
spatial as there are proximity advantages for new firm start-ups: “The social ties of the
potential entrepreneurs are likely to be localized, and induce entrepreneurs to start their firm
in close proximity to their homes and to their current employers” (Stam, 2010, p. 142). At
first sight, one could thus expect entrepreneurial activities to build up mainly in localized
settings. Yet, also entrepreneurial networks can be shaped by more international interrelations.
Transnational entrepreneurship literature shows how e.g. returnee entrepreneurs induce
entrepreneurial experimentation as “new argonauts” (Saxenian, 2007) in places that were
initially unconnected to a TIS emerging in other places and thereby span relevant networks
between at first sight unrelated national subsystems (for a more extensive overview of this
argument see Drori et al., 2009). This function could accordingly be analyzed by
reconstructing the social networks of entrepreneurs and their dynamics over time, e.g. based
on primary survey data, industry association’s member lists, data on actors in R&D projects
or patent data.
Market formation usually develops in different stages with distinctive features of the relevant
user-producer networks (Bergek et al., 2008a). Especially in very early nursing markets,
collocation between users and producers may form important ‘learning spaces’ (Kemp et al.,
1998), which facilitate repeated and trustful feedback loops between companies (or
entrepreneurs) and their customers (Lundvall, 1992). Early markets for wind power and
photovoltaics as an example were strongly shaped by such interactive learning at local to
regional levels (Dewald and Truffer, 2012; Garud and Karnoe, 2003). Yet, especially in later

9
bridging and mass market phases, producers and users do not necessarily have to be co-
located to form and supply markets: Actors in regions without markets could also sell their
products in other subsystems of the same TIS, e.g. by compensating missing spatial proximity
to foreign market places with other forms of (organizational, institutional, cultural or
cognitive) proximity (Lagendijk and Lorentzen, 2007), or in extended user-producer relations
in global production networks or multinational companies (Coe et al., 2004). Chinese PV
manufacturers as a case in point developed into a market leading position by strongly
exploiting spatially distant foreign markets in Europe and the US (de la Tour et al., 2011).
Empirically, networks of market formation could be mapped based on surveys on relevant
user-producer interactions, market reports, or - in later development phases - trade statistics.
‘Influence on the direction of search’ describes the selection process dealing with variety
emerging from knowledge creation (Hekkert et al., 2007). It works through a combination of
regulations or long term policy goals set by governments and the creation of vision and
collective expectations on a new technology among different TIS actors (Bergek et al., 2008a;
Hekkert et al., 2007). In this context it is often assumed (but seldom verified) that national
institutions constitute the most relevant context for effective policy intervention. However,
supranational political institutions and treaties like the EU, UN, WTO or the clean
development mechanism of the Kyoto protocol can have increasingly strong influence on
innovation processes, especially in clean-tech sectors (Binz et al., 2012; Gosens et al.,
submitted). A similar argument holds for the second main dimension of that function, the
shaping of expectations. Bergek et al. (2008a) explicitly argue that expectations might be
influenced by growth occurring in TISs in other countries or by changes in the socio-technical
landscape, which lies outside the influence sphere even of specific national agents. E.g.
direction of the search in the German wind power TIS was reportedly strongly influenced by
developments in California and Denmark (Bergek and Jacobsson, 2003). Also the spatiality of
this function is therefore far from restricted to a specific spatial scale. Similar arguments hold
for another closely related function, creation of legitimacy. This complex process of
expectation shaping and institutional change is created through e.g. lobbying in political
networks, the global climate change debate or experiences from ‘sister’ TIS (Bergek et al.,
2008b). The performance of both functions is thus closely related to the emergence of
supportive advocacy coalitions, interest groups, networks and intermediaries which jointly opt
for coordinated technological and institutional change (Bergek et al., 2008a), much in the
sense of the work of Musiolik et al. (2012). Also here, whereas some relevant actor networks

10
might be restricted in their spatial reach, others might consciously aim at creating guidance
and legitimacy at a more international scale - as e.g. in the case of membrane technology
policies in the Netherlands and Japan (van Lente and Rip, 1998). Empirical analysis of these
functions should thus focus on the perception of key actors on the potential of new
technologies and the formation of advocacy coalitions. Perception of key actors about the
legitimacy of a technology can be scrutinized with discourse analysis methods (in the context
of TIS studies see for instance (in the context of TIS studies see for instance Konrad et al.,
2012) or some newer forms of discourse network analysis, as done in political sciences (see
e.g. Fisher et al., 2012). Relevant data sources can be newspaper articles or protocols of
parliamentary discussions. Formation of advocacy coalitions and intermediaries could in turn
be analyzed based on affiliation data from the core industry associations or interest networks
in a field or again by conducting surveys.
Resource mobilization, finally, involves the deployment of financial and human capital
(Hekkert et al., 2007). Mobilization of financial capital essentially depends on the investment
decisions of private or public investors. Whereas some of these investments are likely coming
from local sources, venture capital might as likely be mobilized through the global financial
system (Avdeitchikova, 2012). Similarly, human capital could be mobilized in local
specialized labor markets, national education institutes or increasingly also through attracting
foreign talent in the form of entrepreneurs, specialized professionals or academicians
(Saxenian, 2007). Actor networks underlying financial resource mobilization could thus be
reconstructed through data on the investment shares of financial institutes or other investors in
key companies of a field. Scrutinizing the mobilization of human capital could in turn be
followed by e.g. mapping the ego-networks of key actors in a field or through graduation
records of specialized engineers.
In summary, specifying how actor networks at different spatial scales influence functional
dynamics is an important analytical problem that remains to be addressed in TIS research. The
short discussion above reveals that further work is needed in particular to better theorize and
empirically analyze the networked spatialities of TIS functions. Rather than trying to assign
functions to their appropriate spatial level, we suggest to examine in more detail how
processes in networks at different spatial levels interact and thereby shape key processes and
ultimately innovative outcomes of both specific national subsystems and the global TIS as a
whole. Unpacking these high spatial complexities in TIS was avoided for a long time due to

11
problems of data availability - in particular if the focus is extended beyond the borders of
small European countries. Obviously, new methodologies and indicators are needed for
tackling innovation processes in a more global perspective. The following section will
therefore propose a first step in this direction by developing a set of indicators based on social
network analysis that allow for a spatial analysis of the actor networks underlying TIS
functions. To reduce complexity and enable an in-depth study of spatial dynamics, the
analysis will be limited to one function, knowledge creation, whereas the framework’s
potential applicability to the other key processes will be discussed in the concluding sections.
3 Measuring international network topologies of TIS
functions
Based on the discussion above, we can distinguish between three ideal-type network patterns
characterizing the spatial setup of innovative interaction in a specific function or – if the
assessment of different functions are combined – a TIS as a whole. First, as assumed in
existing TIS research - relevant networks might form exclusively in localized setups, at
regional to national scales. In such a setup, innovation would be created based on processes
emerging in largely unrelated subsystems, e.g. in different countries. On the other extreme,
networks might be exclusively global, spanning between actors in distant places, as e.g. in an
innovation network of multinational companies or the networks of open source programming.
In this case, relevant TIS space would hardly be assignable to any fixed place or country, but
rather be completely embedded in internationalized networks. Third and in between these
extreme cases, relevant actor networks might be multi-scalar, incorporating a set of both
spatially proximate and distant ties. This setup essentially represents small-world networks,
which efficiently connect tight clusters of local interaction with occasional nonlocal links to
other clusters (Watts and Strogatz, 1998). Small-world networks are assumed to increase
creative output as they combine spatially dense and trustful collaborative innovation processes
with ties to more distant, complementary ideas (for a critical discussion see Fleming et al.,
2007). Consequently, if actor networks underlying TIS processes show small world properties,
then this has strong implications on a respective TIS study, as it implies that scrutinizing
interrelations between different territorial subsystems gets crucial to understanding the
structural and functional properties of its innovation processes.

12
For analyzing what network setup characterizes a TIS or given function at a specific point in
time, a new methodological approach is needed. Here social network analysis (SNA) enters
the stage as a tool that provides heuristic routines for scrutinizing actor network evolution in
global space (Wasserman and Faust, 1994). A respective analytical framework will be
operationalized based on four types of indicators: First of all, general properties of network
structure can be characterized with conventional SNA indicators like mean distance, network
diameter or centralization index.3 Secondly, a “nationalization index” is developed, which
gives a direct measure for how much of the cooperation in a given function is actually
confined to national borders. Thirdly, areas of dense collaboration in the overall network are
analyzed as ‘coherent subsystems’. Such subsystems are here defined as groups of diverse
actors (companies, academia, government, intermediaries) which show particularly tight
interaction. As TIS research assumes such interaction to be crucial for the innovation process,
coherent subsystems can point to core areas of innovative activity in a given function.
Obviously, such subsystems may be strongly localized, but they may as well develop in
regional agglomeration, form between actors at a national or even international level. Finally,
a measure for the overlap between these subsystems is introduced. Coherent subsystems
might in some cases form in isolation from each other, whereas in other cases they might
strongly overlap, thereby integrating subsystems at different spatial scales to densely
integrated ‘global’ TIS (see Figure 1).
3.1 Measuring the relative relevance of national networks
The ‘nationalization index’ is defined as the average ratio of links among actors inside one
country versus the links with actors outside a country. Its definition is based on the E-I index
by Krackhardt and Stern (1988), but combined with the spatial attributes ‘national’ and
‘international’. This index gives a direct measure for the average importance of nationally
delimited interaction in the actor networks underlying a function. The following formulae
capture this relationship:
1)
3 As these are standard measures in SNA methodology, they will not be introduced here, but directly in the results section.
Detailed descriptions can be found in Appendix A.





ei
ei
cLL
LL
N

13
Nc := ‘nationalization index’ of all actors in a specific country in the TIS, Li:= internal link, Le:= external link of actors in a
specific country c
2)
:= nationalization index of the TIS as a whole, c:= number of countries
Equation 1) assesses the nationalization of activity in a specific country in the TIS, whereas
equation 2) calculates the average of all nationalization indexes, thus providing a cumulated
measure for the importance of nationally bound cooperation in the whole TIS. If most actors
are cooperating in national or subnational contexts, these ratios will show values above 0 and
tend towards 1. If internal and external links are equally important, the value will be close to
zero. Consequently, if international interaction is dominant, it will take on negative values and
tend towards -1.4
3.2 Identifying coherent subsystems
Coherent subsystems will be assessed by identifying and characterizing network components
and cohesive subgroups. Components depict isolated fractions of a network, whereas cohesive
subgroups are defined as a subset of a network that displays stronger interaction within a
group of actors than with actors outside the group. Subgroup identification will here be based
on n-clan analysis. N-clans are defined as subgraphs in which the largest (geodesic5) distance
between any two nodes is not greater than n and the diameter does not exceed the set n-value
(Wasserman and Faust, 1994, p.258-261). As such, n-clans identify cohesive subgroups based
on reachability. This helps to understand processes that work through an intermediary, like e.g.
the diffusion of knowledge among different actors in a TIS. In addition, it allows specifying
some of the properties of the cohesive subgroups in focus. In the following analysis, an n-
value of 2 was chosen, meaning that every actor in each 2-clan is divided from all other actors
4 Note that this index is partly dependent on the size of countries. Large countries will always have more national cooperation,
simply because there are more potential cooperation partners inside their boundaries. For regression studies this point should
be controlled for, in this contribution it suffices to keep this caveat in mind.
5 The shortest possible path between two connected actors in a network.
c
N
Nc
gTIS 

gTIS
N

14
by no more than one intermediary. In addition, n-clans allow for the definition of a minimum
value of participants, which was set at 9 actors.6
3.3 Analyzing the overlap between coherent subsystems
Finally, the overlap between coherent subsystems has to be assessed. This will be done here
based on exploring 2-clan-overlap matrixes. After identifying all 2-clans in our network, they
can be arranged in a 2-clan overlap matrix which measures overlap between any pair of clans
through the number of shared actors. In some cases, clans might consist of different actors,
whereas in other cases they might almost completely overlap. Analyzing this pattern can
reveal if coherent subsystems of a TIS are isolated from each other or if they are integrated in
an interconnected set of subsystems. Here 2-clan overlap will also be visualized and assessed
from a geographic perspective, in order to identify spatial scales at which cohesive
subsystems are forming and overlapping.
3.4 Analytical framework
Summarizing, the spatial setup of actor networks underlying a given function can thus be
assessed based on the degree of nationalization, the geographic reach of its 2-Clans and the
strength of 2-clan overlap. Based on this selective spatial characterization, the rough typology
of spatial TIS setups developed above can be further differentiated (Figure 1).
6 This value was chosen based on the properties of our co-publication data. One publication in the dataset contains 8 authors,
four of them 6 actors, 96% less than 4 authors. The threshold level was therefore set at nine actors to avoid single
publications from forming one distinct n-clan and therefore biasing the used n-clan measure. If more multi-author
publications appear in a dataset, n-clan measures should be normalized with the number of publications per n-clan.

15
Figure 1: Typology of spatial TIS setups7
Firstly and secondly, innovation in a given function might be based on networks without 2-
clans, but high levels of either national (Figure 1 i) or international interaction (Figure 1 ii). In
both cases cooperation ties are not (yet) integrated into coherent subsystems, thus hinting at
interaction failures in a respective TIS (Jacobsson and Bergek, 2011). Thirdly, networks
might include 2-clans that largely overlap with national boundaries but show weak
interconnectivity (‘localized TIS’ in Figure 1 iii). In this case, subsystems of a TIS would
develop largely independent from each other in different parts of the world. Fourthly,
networks might include 2-clans that do not overlap with national boundaries but at the same
time also not overlap with each other (‘internationalized TIS’ in Figure 1 iv). Such a case
would describe a TIS driven by different international networks (e.g. by multinational
7
This framework strongly profited from inputs of one of the anonymous reviewers.

16
companies) that develop independently from each other. Fourthly, there might be networks
with cohesive subgroups that are mainly confined within national boundaries, but also show
substantial connections among each other (‘multi-scalar TIS’, Figure 1 v). This case
exemplifies a small-world network with plentiful shortcuts between areas of dense local
interaction. Finally, networks might be structured as in Figure 1 vi), with 2-clans transcending
national boundaries and at the same time strongly overlapping each other. In such a case, most
activities in a TIS would get integrated in a complex network of overlapping coherent
subsystems, forming what can be labeled a ‘global TIS’.
3.5 Analyzing knowledge creation in the MBR TIS
For illustrating the benefits of this framework, it will be applied to the illustrative case of
knowledge creation in membrane bioreactor technology. MBR technology represents a case in
point for a recently emerging environmental technology which strongly depends on systemic
innovation (Truffer et al., 2012). MBR plants are based on conventional biological wastewater
treatment, combined with a micro-porous membrane. They produce a directly reusable,
reliably clean effluent and thereby promise to significantly improve the efficiency of
industrial, municipal and particularly on-site wastewater treatment systems (Fane and Fane,
2005). The basic process was invented in 1966 in a lab of Dorr-Oliver Inc. in the USA (Wang
et al., 2008), but innovation in this field remained rather dormant in the following 20 years.
Activities re-gained momentum only after a decisive innovation by a Japanese professor in
1989 and especially in the past ten years (Judd and Judd, 2006; Lesjean and Huisjes, 2008).
The MBR TIS is thus in a late formative phase. Commercial applications are booming
recently (Lesjean and Huisjes, 2008; Zheng et al., 2010), but the technology is still subject to
particular uncertainties, not yet fully standardized and developed by a multifaceted set of
small start-ups, large transnational companies and various research institutes and universities
worldwide (Binz et al., 2012).
3.6 Data sampling
Knowledge creation on MBR technology relies on integrating a mix of synthetic and
analytical knowledge bases from areas as diverse as process engineering, biology and
advanced materials sciences. It is strongly engineering-driven and tightly intertwined with
actors from companies, utilities and government agencies and that foster pilot plant

17
applications. In the MBR field, such pilot plant experimentation is crucial for interactive
learning and the development and diffusion of the technology.
The results of such experimentation are widely published in international academic journals
or presented at specialized conferences. Relatively abundant data about innovative
cooperation is thus included in the MBR publication record, which was chosen as a source of
network data. Publication data could not be complemented with patent data. A respective
search in the global database of the European patent office retrieved 575 patents, among
which more than 87% originated from small Chinese companies and were of questionable
quality, whereas most major commercial players did not file one single patent8. Contextual
knowledge of the sector confirms that most important MBR companies prefer non-disclosure
of their production processes over patenting as a strategy to protect their intellectual property.
This notwithstanding, the co-publication dataset includes a balanced set of actor types (only
53% of actors originate from universities, the rest includes companies, research institutes,
government agencies and associations, see Table 1). We thus maintain that in the specific case
of MBR technology – and despite well documented limitations of publication data (Katz and
Martin, 1997) - a sufficiently indicative part of the knowledge creation network is covered
with this dataset.
Data collection was based on a query in Thomson Reuters web of knowledge.9 A dataset of
1,068 publications covering a timeframe from 1992-2009 was obtained by searching for
TS=(‘membrane bioreactor’ AND water) and filtering for research areas that contribute to
knowledge generation in MBR technology.10 Publications after 2009 were excluded, as the
records did not yet appear to be complete at the time of data sampling. After manually
eliminating thematically completely unrelated entries in the database, 911 publications
8 Search string: “membrane bioreactor” AND “water” in title or abstract. Search performed on October 2, 2012 on the
website of the European Patent Office, http://worldwide.espacenet.com/advancedSearch?locale=en_EP
9 Thomson Reuters Web of Knowledge, http://apps.isiknowledge.com/
10 Search string: TS=("membrane bioreactor" AND water) AND SU=(water resources OR engineering, chemical OR
environmental sciences OR engineering, environmental OR biotechnology & applied microbiology OR polymer science OR
chemistry, multidisciplinary OR biochemistry & molecular biology OR engineering, civil OR energy & fuels OR agricultural
engineering OR food science & technology OR microbiology OR chemistry, analytical OR chemistry, applied OR materials
science, textiles OR multidisciplinary sciences OR ecology OR engineering, aerospace OR engineering, biomedical OR
engineering, electrical & electronic OR engineering, multidisciplinary OR environmental studies), Databases=SCI-
EXPANDED, SSCI, A&HCI, CPCI-S, CPCI-SSH, Timespan=1960-01-01 - 2010-01-01, Lemmatization=On

18
covering a time frame from 1993-2009 were left for a co-authorship analysis. Even though co-
publications are the source of relational data, actors in this study are defined not at the level of
single authors, but at the level of organizations such as companies, universities, research
institutes or government agencies.
11
Network nodes thus represent organizations and ties
between them represent their cooperation in the co-publication process. Nodes that are linked
to themselves indicate cooperation between different departments of the same organization
(e.g. different faculties of the same university). The dataset was evaluated and visualized
using Net Miner 3 software.
4 The spatial evolution of knowledge creation in the
MBR TIS
4.1 General characteristics of the dataset
MBR technology is in a booming period: Publications grew exponentially from 1999 to 2009
(see Figure 2A), in parallel with rapid market growth and increased commercial dissemination
of MBR systems (Lesjean and Huisjes, 2008; Wang et al., 2008).
Figure 2: Publications on MBR technology, 1993-2009
A
number of publications
B
spatial origin of publications
Source: Own design, based on data from web of knowledge
11
Interpretation of the 2-mode data is simplified by operationalizing links between organizations as co-publications and
analyzing them as a one-mode network (organizations interacting in a network with other organizations). We maintain that
this simplification is legitimate as co-authorship in MBR technology often involves pilot-scale experimentation and
prototyping which includes extended cooperation among significant parts of the participating organizations.

19
Table 1 further reveals that the publication record of MBR technology contains a mixed set of
academic, commercial and public actors and abundant data on cooperation between them.
Actors from 46 countries are involved in the network. Seen from this aggregate perspective,
knowledge creation is thus forming around three key blocks of innovative activity in Europe,
Asia and – to a lesser extent – North America (Figure 2B).
Table 1: Actors and form of cooperation in publications on MBR technology
Actor type Number % Actors per publication
University 273 53.2 1 44.8% 3 15.4%
Company 109 21.2 2 35.8% ≥4 4%
Research Institute 84 16.4
Government Agency 39 7.6 Form of cooperation
Research Institute of Company 5 1.0 international cooperation 22.1%
Association 2 0.4 national cooperation 24.1%
Government Research Institute 1 0.2 internal cooperation 9.0%
Total 513 100 single authored publication 44.8%
To discuss temporal dynamics, the evolution of the MBR TIS will be divided into distinct
development phases, based on the dynamics observable in the evolution of the co-publication
network.12 As publications are relatively sparse in the first ten years of development, the
aggregated network data between 1993 and 2001 is taken as a starting point for a more
detailed analysis between 2001 and 2009. Appendix D and network measures in Table 2 show
how co-variation of key network measures allows distinguishing three stylized development
phases. Between 2001 and 2003, relatively short paths span between most actors in a dense
network. After 2003, the network expands quickly; new actors enter the field and mean
distance between actors grows longer. This trend reverses only after 2007, when average
12 Note that the focus of this contribution is introducing our analytical frame based on an illustrative example. We had to
refrain from a thorough control for many of the problems typically arising when comparing networks over time and across
different contexts. E.g. most network measures are very sensitive to network size as in growing networks the number of
existing ties increases linearly when the number of maximum possible ties increases in quadratic terms. Future studies
applying this framework and comparing properties of networks of various sizes, across contexts or over time should address
these caveats, by e.g. comparing properties of each observed network against values expected in equal-size (i.e. same density,
same number of nodes or similar degree distribution) random networks.

20
connecting paths get shorter again. In parallel, the network oscillates from a centralized to a
more equally connected and back to a more centralized setup (see Appendix D).
Table 2: Three phases of network evolution (non-cumulative except for ‘number of actors’)
Number of actors Number of Links Mean Distance Network
diameter
Centralization
index
Components
> 9 actors
93-03 104 201 2.597 7 19.174 2
03-07 291 553 5.540 15 4.456 6
07-09 513 945 4.963 13 6.953 1
Explanations of the indicators used in this table are summarized in Appendix A, a threshold value of 9 actors was chosen for
the component analysis to avoid publications with many co-authors from being interpreted as a distinct component.
These three phases can be characterized as follows: First a nurturing phase between 1993 and
2003 in which activity is growing and first cooperative ties form around a few central actors
in a dense and centralized network. Subsequently a rapid expansion phase (2003-2007) in
which knowledge creation grows exponentially and many new actors enter an increasingly
broad and decentralized network. Third and finally a consolidation phase (2007-2009) in
which growth slows down and knowledge creation gets intensified among existing actors.
A comparison with secondary sources shows that the first 20 years of TIS development are
not covered by this dataset. Publication records only start after a decisive invention at the end
of the 80ies (Judd and Judd, 2006). Our dataset thus misses the very early invention phase
between 1960 and 1990, but covers the later nurturing phase between 1990 and early 2000
when activities start growing and first commercial MBR plants emerge (Lesjean and Huisjes,
2008; Wang et al., 2008). The subsequent phase matches an expansion phase in the TIS when
commercial applications start booming and many new actors enter the field in different parts
of the world (Judd and Judd, 2006). The last phase finally corresponds with a consolidation in
the MBR industry where dominant designs emerge and some companies leave the field or are
bought by large transnational companies (Binz et al., 2012; De Wilde et al., 2008). This very
general characterization of our data can now be complemented with the spatial analytical
framework and indicators outlined in section 3.

21
Nationalization index
Figure 3 shows that knowledge creation on MBR technology is most internationalized at the
beginning of the nurturing phase. In the consecutive expansion phase the trend is reversing
and cooperation at a national level gets slightly more important, whereas the consolidation
phase is characterized by another dip towards more internationalized values. This pattern
interestingly suggests that knowledge creation in the MBR TIS started in a rather globalized
network structure and turned into more differentiated multi-scalar spatial setups only in the
later expansion and consolidation phases.
Figure 3: Nationalization index of knowledge creation in the whole TIS and 4 national subsystems
Source: Own design, based on data from ISI web of knowledge. Values depict shifting (3 years) averages.
The dominant form of interaction in specific countries shows strong temporal variation, too.
E.g. Chinese actors’ nationalization index values are exclusively international in the first two
years and then increasingly switch to nationalized index values until 2006. This shift happens
at a time when many new Chinese actors enter the TIS and MBR technology gets increasingly
integrated into strategic national R&D programs (Wang et al., 2008; Zheng et al., 2010). This
pattern thus reveals a catching-up process in which Chinese actors first tapped into global
knowledge sources before domestic technological capabilities and policy incentives were built
up. South Korean actors, in contrast exemplify a geographically stable cooperation strategy
which was in all periods mainly confined to a national level.

22
Coherent subsystems identification and overlap analysis
The results of the 2-Clan analysis in 3 further substantiate the precedent insights. In the
nurturing phase, all three identified 2-Clans are of global outreach. The expansion phase is
dominated by six 2-Clans at a ‘continental’ level (mainly in the EU). Coherent subsystems get
spatially more differentiated only in the consolidation phase when continental 2-clans are the
dominant level of interaction, but global and national 2-clans emerge, too.
Table 3: Spatial reach of 2-clans
Type of 2-clan Nurturing 1993-2003 Expansion 2003-2007 Consolidation 2007-2009
National 2-clan 0 0 9
Continental 2-clan 0 6 45
Global 2-clan 3 0 3
Source: own design. National 2-clans: 2-clans with more than ½ of the actors from one specific country; Continental: 2-
clans with more than ½ of the actors from different countries of the same continent; Global: 2-clans containing actors from
at least three different continents, without a dominant region
As the analysis of 2-clans and especially of 2-clan overlap needs careful interpretation, the
next section will discuss these results in more detail and with contextual information.
4.2 1993-2003: Global, company-based knowledge creation
Figure 4 illustrates that in the nurturing phase, knowledge creation is split into two main
components and three strongly overlapping 2-clans, containing actors from eight countries.
The core coherent subsystem is centered on CIRSEE (Centre International de Recherche Sur
l'Eau et l'Environnement), a French company owned research institute, and its subsidiaries in
Malaysia (ASTRAN Malaysia) and Australia (ASTRAN Sydney). Another subsystem is
forming around an isolated network component comprising Cranfield University, the National
University of Seoul and other institutes in South Korea, the USA and Malaysia. However,
cooperation in this component is less tight than in the main component around CIRSEE and
no 2-clans can be identified in this part of the network.
Network measures in Table 2, the nationalization index and coherent subsystem analysis thus
assert that international interaction is most relevant in the nurturing phase (also see network
visualization in Appendix B). These results thus suggest that knowledge in the MBR TIS
originated from a globalized coherent subsystem initiated by French water companies. As
public funding for research and development on MBR technology was very limited at this
early point of development (Lesjean and Huisjes, 2008), first innovative activities were

23
pushed by private actors that mobilized financial resources and their extended international
innovation network to developed the first commercial applications of the technology.
Figure 4: Coherent subsystem in MBR knowledge creation, 1993-2003
Source: Data from web of knowledge, visualized with NetMiner 3 software. Node size depends on sum of publications.
4.3 2003-2007: Multi-scalar, Europe-centered knowledge creation
The subsequent expansion phase was so far characterized as a multi-scalar setup with six 2-
clans at a continental level and a sharp increase of involved actors. Also in this second period,
most identified 2-clans are strongly overlapping. Figure 5 identifies a core coherent subsystem
spanning between actors in 6 overlapping 2-clans in the European Union, connected mainly
by German actors. Dense cooperation in the networks of French water companies is still
relevant in that subsystem,
13
but the network around CIRSEE has lost its central position.
13
Anjou Recherche and Berlin Competence Centre for Water are closely related to Veolia, a large French water corporation

24
Figure 5: Core coherent subsystem in MBR knowledge creation, 2003-2007
Source: Data from ISI web of knowledge, visualized with NetMiner 3 software. Node size depends on the number of
publications.
Dense interaction now gets dominant especially inside the European Union, whereas the USA and Canada
become the most disconnected region with a high number of single authored papers and correspondingly
isolated actors (
Appendix B). The actor base in Asia in contrast is expanding quickly and relevant cooperation
forms especially among and between South Korean, Chinese and Japanese actors. In addition,
many small components now appear, mainly connecting European and/or Asian actors.
Knowledge creation as a whole is thus fragmenting into a main coherent subsystem and
several isolated components in different regions of the world.
Comparing the results of the nurturing and the expansion phase reveals that the overall spatial
setup and the composition of the most central actors in knowledge creation have switched
considerably over a short period of time. Actors from Germany as an example occupied a
rather peripheral position in the network until 2003 but quickly moved to a central position

25
between 2003 and 2007. The core coherent subsystem furthermore changed qualitatively from
a global, company-dominated mode to a more trans-disciplinary mode, now connecting seven
universities, five companies, five research institutes, three government organizations and one
company research institute mainly inside Europe.
This major spatial shift very likely reflects activities induced by MBR research programs of
the European Union (Lesjean and Huisjes, 2008). Because European actors were increasingly
lagging behind in the MBR field, a new relevant level of interaction was constructed in four
large research initiatives of the 6th European framework program. These comprehensive
projects were not only aimed at creating scientific knowledge, but also inducing
entrepreneurial experimentation, guidance on the search and connecting different actors in a
series of international conferences. The relative decline of the activities of transnational
companies in knowledge creation might accordingly be explainable with the fact that they
increasingly focused on internal optimization of their MBR technology and left more basic
R&D activities to smaller actors in an increasingly vibrant surrounding technological
innovation system in that second phase.
4.4 2007-2009: Multi-scalar knowledge creation between Europe
and Asia
In the last phase of development, cooperation intensifies in an increasingly consolidating
environment. Most actors are now included in a giant network component, connecting 340
nodes. Section 4.1 described this phase as a multi-scalar to globalized setup with 57 2-clans.
The high number of frequently overlapping 2-clans (see Figure 6) now allows differentiating
different coherent subsystems. First of all, a coherent subsystem with strongly overlapping 2-
clans exists in central Europe, dominated by actors from German speaking countries.
Secondly, a new subsystem now evolves in Asia, dominated by South Korean and Japanese
actors in connection to international partners. Thirdly, a relevant subsystem is forming in an
international network between European, Asian and North American actors, largely
disconnected from the other coherent subsystems. Finally, also the national scale now
contains a significant number of isolated subsystems, e.g. in Israel and Italy.

26
Figure 6: 2-Clan overlap in MBR technology knowledge creation, 2007-2009
Source: data from ISI web of knowledge, visualized with NetMiner 3 software. Node size depends on number of actors in 2-
clans, line thickness on number of overlapping actors, threshold value of links: 6.
Overall, the network takes on increasing small world properties with most 2-clans showing
considerable overlap. Especially the subsystems in Europe and Asia are strongly coupled to
each other through a major 2-clan containing 38 actors centred on TU Berlin (Figure 7).

27
Figure 7: Central hub in MBR technology knowledge creation, 2007-2009
Source: data from ISI web of knowledge, visualized with NetMiner 3 software. The most central actor in the core of the 2-
clan is the Technical University Berlin.
This ‘hub’ perfectly exemplifies the importance of multi-scalar interaction in knowledge
creation of MBR technology in that phase: On the one hand, innovation in this hub exhibits a
global dimension, connecting actors from 16 countries and 5 continents. On the other hand,
cooperation inside the European Union is the core level of activity (more than half of the
actors are located in EU member states). Finally, cooperation among 8 actors at a national
level in Germany (and dominantly in Berlin) is present in the structure, too. Innovative
activity of German actors like the TU Berlin can accordingly not be solely attributed to the
specific context constituted at a national scale. It rather has to be interpreted as the outcome of
multiple relations established concomitantly at different scales and at the intersection between
two coherent subsystems of a wider global TIS.

28
Summarizing, in the consolidation phase, the spatial setup of knowledge creation again differs
considerably from the precedent phase and gets increasingly complex: International,
continental and national scales now all contain relevant coherent subsystems, with the core of
activity still in Europe, but increasingly shifting towards Asia and getting more integrated at a
global level. This last switch likely reflects the increasing maturation of the TIS and the
formation of an increasingly well-structured research and development community around
MBR technology, which turns the underlying knowledge networks more and more into a
globalized small world setup. Still, spatial imbalances in the global distribution of activity are
much accentuated: Concentrated innovation efforts of Asian actors, especially in South Korea,
Japan and China (Zheng et al., 2010), increasingly establish a relevant scale of interaction also
in this part of the world. North American actors, in contrast, are still underrepresented in the
most relevant subsystems of knowledge creation. This finding corresponds with empirical
studies claiming that North American actors are partly decoupled from mainstream research
activities and following a distinct technology development path focussing on side-stream
MBR systems (Wang et al., 2008).
4.5 Discussion
Two main findings stand out from the observed strong spatial dynamics in knowledge
creation of MBR technology. First, our results indicate that TIS function’s underlying actor
networks can shift considerably in space and that innovation processes in national (sub-
)systems might be more strongly interconnected and influenced by a ‘global TIS’ level than
could be assumed from existing studies. We thus support arguments from economic
geographers and innovation system scholars that innovation (system) research should explore
multi-scalar processes and especially the global scale in much more detail (Bunnell and Coe,
2001; Carlsson, 2006).
Secondly, the presented case study illustrates how assessing the spatial setup of functions can
improve the understanding of innovation processes in a TIS. Knowledge created in networks
spanning transnational companies and their research partners (as in the nursing phase of MBR
technology) is clearly of a different quality than knowledge created in small world networks
connecting different trans-disciplinary subsystems in a multi-scalar setup (as in the
consolidation phase of MBR technology). Also the dominant level of the core coherent
subsystems may shift in space. Policy interventions to sustain system buildup in specific

29
countries should thus be responsive to (and try to anticipate) the shifts in the spatial
configuration of core subsystems of a TIS.
A further direct added value of this framework is that it allows identifying spatial errors that
might be incorporated in nationally delimited TIS studies (see Binz and Truffer, 2012). Firstly,
in a TIS with functions dominated by localized interaction (setup i and iii in Figure 1),
‘isolation errors’ might occur: A study in a single country would only inform about
innovation in one specific subsystem of the overall TIS. Decisive technological advances
might however develop independently in another subsystem without the TIS analyst taking
note. In the case of MBR technology, focusing on the USA, whose actors were in most phases
of TIS evolution relatively decoupled from a dynamic international knowledge network,
would likely have produced such isolation errors. Secondly, in multi-scalar or globalized
setups (Figure 1 v and vi), errors of ‘omitted context’ might be conducted; a national case
study would likely overestimate the importance of processes working at national to
subnational scales. Developments stemming from outside could falsely be attributed to
developments inside the focal subsystem and thereby again lead to inefficient policy advice.
Internationalized TIS setups, finally (Figure 1 iv), could induce ‘system misinterpretation
errors’. Here, innovation predominantly stems from activities embedded in international
networks. National delimitations would accordingly lead to a complete misinterpretation of
the most relevant level of innovative activity. In the case of MBR technology, doing
nationally delimited TIS studies in the nurturing phase would arguably have produced this
type of errors: As the central knowledge creating subsystem was dominated by globally
operating companies at that time, nationally delimited studies would arguably not have
identified the core actors and spatial level of this technology’s development.
Some limitations of the presented results also have to be mentioned here: First, we could only
scrutinize one function in more detail and, secondly, left institutional contexts underexplored.
To address the first issue, one could analyze the other functions of the MBR TIS with the
same framework (following the suggestions of section 2.3) and try to identify overlaps
between coherent subsystems in different functions. Places and scales where subsystems of
different functions overlap could then be interpreted as the innovative core of a TIS at a given
point in time and theories could be developed on how and why this core moves in space. In
contrast, if only few overlaps between coherent subsystems in different functions of the same
TIS exist, then the TIS in focus would have to be understood as a conglomerate of spatially

30
dispersed functional dynamics, a finding that would strongly contradict existing TIS studies.
Finally, as system functions are inherently interrelated, identifying the core actors and
coherent subsystems of one function could be used for predicting the probability of activities
in other functions emerging in the same place. Knowledge spillover theory of
entrepreneurship as an example suggests that entrepreneurial activities emerge in close spatial
proximity to the core knowledge creating subsystem of a TIS (Audretsch and Lehmann, 2005).
Considering the missing focus on institutional contexts, empirically identifying innovative
cores of a TIS could also allow reconstructing to some degree which institutional settings in
which places have been crucial for system development at specific points in time. In the case
of MBR technology, institutional contexts of the EU were identified as being of key
importance to foster knowledge creation also in other parts of the global TIS. However, the
fundamental question on whether actor networks shape institutional contexts or vice versa
could not be addressed here and clearly needs focused elaboration in future work.
Finally, our approach also leaves ample room for methodological improvements and the
exploitation of new data sources. The observed high importance of international linkage in all
development phases of knowledge creation in MBR technology might be partially attributable
to the bias of publications from ISI web of knowledge towards research in international
projects and published in international journals (Nelson, 2009). For a more balanced view,
other case studies would have to integrate additional data types like patents or licenses,
publications from non-ISI journals or other relational data from industry associations or
conferences. SNA methodology also offers plentiful additional heuristic routines that might
be fruitfully exploited for assessing network evolution over time and identifying cohesive
subsystems.
5 Conclusions
This paper aimed at discussing the implications of the spatially implicit system boundary
setting in current TIS studies and at illustrating how a spatialized TIS framework could
contribute to empirically identifying meaningful system boundaries and analyzing linkages
and relationships between its (territorial) subsystems. As shown in the literature review,
adding relational space to TIS and functional TIS analysis is a promising way forward for
improving conceptual rigor, empirical application and the policy advice derived from this

31
conceptual approach. Mapping the global (yet uneven) TIS helps clarifying how national sub-
TIS are related to each other and how specific spaces in the TIS might generate comparative
advantage. The empirical case study indicates that knowledge creation in MBR technology
happened in a global company-based, a science-driven Europe-centred, as well as in a multi-
scalar Europe- and Asia based spatial setup. As national subsystems are embedded differently
in each of these setups, nationally delimited studies would have to be adapted accordingly.
TIS space is thus fluent and innovation processes can change quickly both in spatial reach and
nature.
Apart from showing where and when specific innovations develop and diffuse, a more
explicit spatial perspective also sheds light on how innovation processes might interrelate
between seemingly unrelated places. The MBR example shows that networks transcending
national borders might be more relevant for innovation processes in TIS than has been
acknowledged in previous studies. The ‘global technological opportunity set’ should
accordingly not be understood as a ubiquitous resource for TIS actors. It rather has to be
characterized as an uneven and dynamically evolving network structure to which actors with
different relational positions and capabilities have differential access at different points in
time. We thus argue in line with Carlsson (2006) that this scale needs more attention in future
conceptual, empirical and especially methodological work.
The presented results also imply a central lesson for policy making: Innovation or industrial
policy, for instance in the form of subsidies for specific technologies, have to consider the
global spatial setup of a technological field (see Truffer, 2012). National support of specific
technologies may otherwise lead to unintended effects like supporting industry growth in
other countries (as exemplified by the impact of feed-in tariffs for photovoltaics in Germany,
which strongly supported the growth of Chinese at the expense of German companies). Also
in the specific case of knowledge creation, policy interventions are often predominantly
targeting processes at a national level even though knowledge production increasingly takes
place in complex international networks. Supporting couplings between national actors and
their international TIS environment has accordingly been underrated as a policy option.
Future TIS research could be inspired by this contribution in two ways: Firstly, our
framework could be used for spatially sensitive studies of other TIS functions, which could in
turn improve the generalizability and explanatory power of the approach. Secondly, respective

32
studies could feed into a spatialized TIS lifecycle theory. Understanding which spatial scales
are relevant in what fields of technology and at what phase of system development could
generate important input for improving TIS-based theory development and policy advice.
Finally, our study just covers one illustrative case in water recycling technology. Similar
studies in other technological fields are needed to further validate and improve the proposed
framework.
Acknowledgements
The authors would like to thank the Sino-Swiss Science and Technology Cooperation (SSSTC)
for the funding of this project. Part of the work was conducted while one of the authors (B.
Truffer) was a fellow in the sustainability Science Program at Harvard University. This paper
profited from constructive input at the AAG annual meeting 2010, a DIME workshop 2010,
the GLOBELICS conference 2010 and a TIS summer school in 2011. We would especially
like to thank Karin Ingold, Frans Berkhout, Staffan Jacobsson, Philip Leifeld and Koen
Frenken for very helpful inputs on earlier drafts. Finally, we would like to thank two
anonymous reviewers for their extraordinarily constructive inputs to this paper.
Appendix
Appendix A: SNA indicators for network characterization
Indicator Definition
Mean distance Mean distance measures the average geodesic distance (shortest paths) between any pair of
nodes in a network
Network diameter Diameter describes the largest geodesic distance between any pair of nodes in a network.
This indicator thus measures how many intermediaries a piece of information has to pass
in order to travel on the shortest possible path between the two most distant actors in the
network.
Centralization index Index of variability of individual centrality scores. The most centralized network is a star
network, where one actor has direct access to every other actor, the least centralized a
circle network, where every actor has only access to two neighbors and thus all actors
possess identical centrality

33
Appendix B: Knowledge creation of MBR technology 1993-2003

34
Appendix B: Knowledge creation of MBR technology 2003-2007
Appendix C: Knowledge creation of MBR technology 2003-2007

35
Appendix D: Three phases of network evolution
Note that the data point in 2001 comprises the cumulated network data from 1993-2001
References
Amin, A., 2002. Spatialities of globalisation. Environment and Planning A 34, 385-399.
Amin, A., Roberts, J., 2008. Knowing in action: Beyond communities of practice. Research
Policy 37 (2), 353-369.
Asheim, B., Coenen, L., 2006. Contextualising Regional Innovation Systems in a Globalising
Learning Economy: On Knowledge Bases and Institutional Frameworks. Journal of
Technology Transfer 31, 163-173.
Asheim, B.T., Gertler, M.S., 2005. The Geography of Innovation: Regional Innovation
Systems. In: Fagerberg, J., Mowery, D., Nelson, R. (Eds.), The Oxford Handbook of
Innovation. Oxford University Press, Oxford, pp. 291-317.

36
Audretsch, D.B., Lehmann, E.E., 2005. Does the Knowledge Spillover Theory of
Entrepreneurship hold for regions? Research Policy 34 (8), 1191-1202.
Avdeitchikova, S., 2012. The geographic organization of ‚venture capital‘ and ‚business
angles‘. In: Landström, H., Mason, C. (Eds.), Handbook of Research on Venture Capital:
Volume 2. Edward Elgar, Cheltenham, UK, pp. 175-208.
Bathelt, H., Malmberg, A., Maskell, P., 2004. Clusters and knowledge: local buzz, global
pipelines and the process of knowledge creation. Progress in Human Geography 28 (1), 31-56.
Bathelt, H., Schuldt, N., 2008. Between luminaires and meat grinders: International trade fairs
as temporary clusters. Regional Studies 42 (6), 853-868.
Bathelt, H., Gluckler, J., 2003. Toward a relational economic geography. Journal of Economic
Geography 3 (2), 117-144.
Bergek, A., Jacobsson, S., Carlsson, B., Lindmark, S., Rickne, A., 2008a. Analyzing the
functional dynamics of technological innovation systems: A scheme of analysis. Research
Policy 37 (3), 407-429.
Bergek, A., Jacobsson, S., Sandén, B.A., 2008b. 'Legitimation' and 'development of positive
externalities': Two key processes in the formation phase of technological innovation systems.
Technology Analysis and Strategic Management 20 (5), 575-592.
Bergek, A., Jacobsson, S., 2003. The Emergence of a Growth Industry: A Comparative
Analysis of the German, Dutch and Swedish Wind Turbine Industries. In: Metcalfe, J.S.,
Cantner, U. (Eds.), Change, Transformation and Development. Physica-Verlag (Springer),
Heidelberg, pp. 197-228.
Berkhout, F., Angel, D., Wieczorek, A., 2009. Sustainability transitions in developing Asia:
Are alternative development pathways likely? Technological Forecasting & Social Change 76
(2), 215-217.
Binz, C., Truffer, B., 2012. Technological innovation systems in multiscalar space. Analyzing
an emerging water recycling industry with social network analysis. Geographica Helvetica 66
(4), 254-260.

37
Binz, C., Truffer, B., Li, L., Shi, Y., Lu, Y., 2012. Prospective analysis of leapfrogging
trajectories – An analytical framework and a case study in the Chinese wastewater sector.
Technological Forecasting and Social Change 79 (1), 155-171.
Boggs, J.S., Rantisi, N.M., 2003. The relational turn in economic geography. Journal of
Economic Geography 3, 109-116.
Brenner, T., Cantner, U., Fornahl, D., Fromhold-Eisebith, M., Werker, C., 2011. Regional
innovation systems, clusters, and knowledge networking. Papers in Regional Science 90 (2),
243-249.
Bunnell, T.G., Coe, N.M., 2001. Spaces and scales of innovation. Progress in Human
Geography 25 (4), 569-589.
Carlsson, B., 1997a. Technological systems and industrial dynamics. Springer, Boston.
Carlsson, B., 1997b. On and off the beaten path: the evolution of four technological systems
in Sweden. International Journal of Industrial Organization 15 (6), 775-799.
Carlsson, B., 2006. Internationalization of innovation systems: A survey of the literature.
Research Policy 35 (1), 56-67.
Carlsson, B., Stankiewicz, R., 1991. On the nature, function and composition of technological
systems. Journal of Evolutionary Economics 1 (2), 93-118.
Chaminade, C., Vang, J., 2008. Globalisation of knowledge production and regional
innovation policy: Supporting specialized hubs in the Bangalore software industry. Research
Policy 37 (10), 1684-1696.
Coe, N.M., Hess, M., Yeung, H.W., Dicken, P., Henderson, J., 2004. Globalizing regional
development: a global production networks perspective. Transactions of the Institute of
British Geographers 29 (4), 468-484.
Coe, N.M., Bunnell, T.G., 2003. 'Spatializing' knowledge communities: towards a
conceptualization of transnational innovation networks. Global Networks 3 (4), 437-456.

38
Coenen, L., Benneworth, P., Truffer, B., 2012. Toward a spatial perspective on sustainability
transitions. Research Policy 41 (6), 968-979.
Coenen, L., Díaz López, F.J., 2010. Comparing systems approaches to innovation and
technological change for sustainable and competitive economies: an explorative study into
conceptual commonalities, differences and complementarities. Journal of Cleaner Production
18 (12), 1149-1160.
Cooke, P., 2005. Regionally asymmetric knowledge capabilities and open innovation:
Exploring ‘Globalisation 2’—A new model of industry organisation. Research Policy, 34 (8),
1128-1149.
Cooke, P., Gomez Uranga, M., Etxebarria, G., 1997. Regional innovation systems:
Institutional and organisational dimensions. Research Policy 26 (4-5), 475-491.
Crevoisier, O., Jeannerat, H., 2009. Territorial knowledge dynamics: From the proximity
paradigm to multi-location milieus. European Planning Studies 17 (8), 1223-1241.
de la Tour, A., Glachant, M., Ménière, Y., 2011. Innovation and international technology
transfer: The case of the Chinese photovoltaic industry. Energy Policy 39 (2), 761-770.
De Wilde, W., Richard, M., Lesjean, B., Tazi-Pain, A., 2008. Towards standardisation of the
MBR technology? Desalination 231 (1-3), 156-165.
Dewald, U., Truffer, B., 2012. The Local Sources of Market Formation: explaining regional
growth differentials in German photovoltaic markets. European Planning Studies (3), 397-420.
Drori, I., Honig, B., Wright, M., 2009. Transnational entrepreneurship: An emergent field of
study. Entrepreneurship: Theory and Practice 33 (5), 1001-1022.
Fane, A.G., Fane, S.A., 2005. The role of membrane technology in sustainable decentralized
wastewater systems. Water Science and Technology 51 (10), 317-325.
Fisher, D.R., Leifeld, P., Iwaki, Y., 2012. Mapping the ideological networks of American
climate politics. Climatic Change, 1-23.

39
Fleming, L., King, C., Juda, A.I., 2007. Small worlds and regional innovation. Organization
Science 18 (6), 938-954.
Freeman, C., 1987. Technology and economic performance: lessons from Japan. Pinter,
London.
Garud, R., Karnoe, P., 2003. Bricolage versus breakthrough: distributed and embedded
agency in technology entrepreneurship. Research Policy 32 (2), 277-300.
Gertler, M.S., 2003. Tacit knowledge and the economic geography of context, or The
undefinable tacitness of being (there). Journal of Economic Geography 3 (1), 75-99.
Glückler, J., 2007. Economic geography and the evolution of networks. Journal of Economic
Geography 7 (5), 619-634.
Gosens, J., Lu, Y., Coenen, L., submitted. Clean-tech Innovation in Emerging Economies:
Transnational Dimensions in Technological Innovation System Formation. Technological
Forecasting & Social Change.
Hekkert, M., Suurs, R., Negro, S., Kuhlmann, S., Smits, R., 2007. Functions of innovation
systems: A new approach for analysing technological change. Technological Forecasting and
Social Change 74 (4), 413-432.
Jacobsson, S., Bergek, A., 2011. Innovation system analyses and sustainability transitions:
Contributions and suggestions for research. Environmental Innovation and Societal
Transitions 1 (1), 41-57.
Judd, S., Judd, C., 2006. The MBR book. Elsevier, Oxford.
Kastelle, T., Steen, J., 2010. Using network analysis to understand innovation. Innovation:
Management, Policy and Practice 12, 2-4.
Katz, J.S., Martin, B.R., 1997. What is research collaboration? Research Policy 26 (1), 1-18.
Kemp, R., Schot, J., Hoogma, R., 1998. Regime Shifts to Sustainability Through Processes of
Niche Formation: The Approach of Strategic Niche Management. Technology Analysis &
Strategic Management 10 (2), 175-195.

40
Konrad, K., Markard, J., Ruef, A., Truffer, B., 2012. Strategic response to hype and
disappointment: the case of stationary fuel cells. Technological Forecasting & Social Change
79, 1084-1098.
Krackhardt, D., Stern, R.N., 1988. Informal Networks and Organizational Crises: An
experimental simulation. Social Psychology Quarterly 51 (2), 123-140.
Lagendijk, A., Lorentzen, A., 2007. Proximity, knowledge and innovation in peripheral
regions. On the intersection between geographical and organizational proximity. European
Planning Studies 15 (4), 457-466.
Lesjean, B., Huisjes, E.H., 2008. Survey of the European MBR market: trends and
perspectives. Desalination 231 (1-3), 71-81.
Lundvall, B., 1992. National Systems of Innovation - toward a Theory of Innovation and
Interactive Learning. Pinter, London.
Markard, J., Raven, R., Truffer, B., 2012. Sustainability transitions: An emerging field of
research and its prospects. Research Policy 41 (6), 955-967.
Markard, J., Truffer, B., 2008. Technological innovation systems and the multi-level
perspective: Towards an integrated framework. Research Policy 37, 596-615.
Maskell, P., Malmberg, A., 1999. The competitiveness of firms and regions. 'Ubiquitification'
and the importance of localized learning. European Urban and Regional Studies 6 (1), 9-25.
Musiolik, J., Markard, J., 2011. Creating and shaping innovation systems: Formal networks in
the innovation system for stationary fuel cells in Germany. Energy Policy 39, 1909-1922.
Musiolik, J., Markard, J., Hekkert, M., 2012. Networks and network resources in
technological innovation systems: Towards a conceptual framework for system building.
Technological Forecasting and Social Change 79 (6), 1032-1048.
Nelson, A.J., 2009. Measuring knowledge spillovers: What patents, licenses and publications
reveal about innovation diffusion. Research Policy 38 (6), 994-1005.

41
Nelson, R., 1993. National Innovation Systems: A Comparative Analysis. Oxford University
Press, New York.
Saxenian, A., 2007. The new argonauts: Regional advantage in a Global Economy. Harvard
University Press, Boston, MA.
Stam, E., 2010. Entrepreneurship, evolution and geography. In: Boschma, R., Martin, R.
(Eds.), The handbook of evolutionary economic geography. Edward Elgar, Cheltenham, UK,
pp. 139-161.
Ter Wal, A.L.J., Boschma, R.A., 2009. Applying social network analysis in economic
geography: Framing some key analytic issues. Annals of Regional Science 43 (3), 739-756.
Tödtling, F., Trippl, M., 2005. One size fits all? Towards a differentiated regional innovation
policy approach. Research Policy 34 (8), 1203-1219.
Truffer, B., 2012. The need for a global perspective on sustainability transitions.
Environmental Development 3, 182-183.
Truffer, B., Binz, C., Gebauer, H., Störmer, E., 2012. Market success of on-site treatment: a
systemic innovation problem. In: Larsen, T., Udert, K., Lienert, J. (Eds.), Wastewater
Treatment: Source Separation and Decentralisation. IWA Publishing, London, pp. 209-223.
van Lente, H., Rip, A., 1998. The Rise of Membrane Technology: From Rhetorics to Social
Reality. Social Studies of Science 28 (2), 221-254.
Wang, Z., Wu, Z., Mai, S., Yang, C., Wang, X., An, Y., Zhou, Z., 2008. Research and
applications of membrane bioreactors in China: Progress and prospect. Separation and
Purification Technology 62 (2), 249-263.
Wasserman, S., Faust, K., 1994. Social Network Analysis: Methods and Applications.
Cambridge University Press, Cambridge, UK.
Watts, D.J., Strogatz, S.H., 1998. Collective dynamics of 'small-world' networks. Nature 393
(6684), 440-442.

42
Wenger, E., 1998. Communities of practice - Learning, meaning, and identity. Cambridge
University Press, Cambridge, UK.
Yeung, H.W., 2005. Rethinking relational economic geography. Transactions of the Institute
of British Geographers 30 (1), 37-51.
Zheng, X., Zhou, Y., Chen, S., Zheng, H., Zhou, C., 2010. Survey of MBR market: Trends
and perspectives in China. Desalination 250, 609-612.