Taxonomies of patterns of innovation give a dominant role to large firms, and are often based on empirical studies that exclude micro firms. This paper proposes an empirical taxonomy of the innovative firms at the bottom of the size distribution, based on a new survey of 1,234 small firms and micro firms in the Netherlands, in both manufacturing and services. These firms differ not only in their innovative activities, but also in their business practices and strategies – such as management attitude, planning and external orientation – that they use to achieve innovation. The taxonomy identifies four categories of small innovative firms: science-based, specialised suppliers, supplier-dominated and resource-intensive. It suggests a more diverse pattern of innovation of small firms than in Pavitt’s (1984) taxonomy, a pattern that is shared by both manufacturing and services firms. Finally, the research shows that taxonomies can be effectively used to map differences in the rates, sources and nature of innovation, with the differences in the business strategies of innovative firms.

1. Paper to be presented at the DRUID Tenth Anniversary Summer Conference 2005 on
DYNAMICS OF INDUSTRY AND INNOVATION:
ORGANIZATIONS, NETWORKS AND SYSTEMS
Copenhagen, Denmark, June 27-29, 2005
THE FRUIT FLIES OF INNOVATION:
A TAXONOMY OF INNOVATIVE SMALL FIRMS
Jeroen P.J. de Jong
EIM Small Business Research
E-mail: jjo@eim.nl
Orietta Marsili*
RSM Erasmus University
E-mail: omarsili@fbk.eur.nl
*Corresponding author
Erasmus University Rotterdam, Burg. Oudlaan 50,
3000 DR Rotterdam, The Netherlands
Tel.: +31 (0)10 408 9508
Fax: +31 (0)10 408 9015.
February 2005
Abstract
Taxonomies of patterns of innovation give a dominant role to large firms, and are often based
on empirical studies that exclude micro firms. This paper proposes an empirical taxonomy of
the innovative firms at the bottom of the size distribution, based on a new survey of 1,234
small firms and micro firms in the Netherlands, in both manufacturing and services. These
firms differ not only in their innovative activities, but also in their business practices and
strategies – such as management attitude, planning and external orientation – that they use
to achieve innovation. The taxonomy identifies four categories of small innovative firms:
science-based, specialised suppliers, supplier-dominated and resource-intensive. It suggests
a more diverse pattern of innovation of small firms than in Pavitt’s (1984) taxonomy, a pattern
that is shared by both manufacturing and services firms. Finally, the research shows that
taxonomies can be effectively used to map differences in the rates, sources and nature of
innovation, with the differences in the business strategies of innovative firms.
Keywords: Innovation; Taxonomy; SMEs; Business strategies; Services.

2. 1
1. Introduction
Fruit flies are a favourite object of study in evolutionary biology as things happen fast for
them and their evolution can be observed over short time periods (Maynard Smith, 1996)1
. In
analogy, the firms at the fringe of the business population may take a similar role in the study
of technological change. Innovation creates a lot of “room” at the bottom of the size
distribution, as new small firms continue to enter the market with new ideas for new products
and processes (Audretsch, 1995). Often these firms are short-lived, as they exit the market
within few years after their entry, thus leading to high turnover of firms at the margin (Caves,
1998). However, those that do innovate successfully, can increase considerably their chances
of survival (Cefis and Marsili, 2003; De Jong et al., 2004). Despite the short appearance of
some, the new firms who survive contribute to a large proportion of growth at the economy
wide level (Foster et al., 1998). Yet, their behaviour can vary substantially. Some new firms
survive by competing in a market niche, while others may pursue more radical innovations
and become themselves market leaders. Pavitt (1998) argued that the diversity of innovative
behaviour in reality, for small as for large firms, cannot be reduced to the few invariant laws
of a general model. Instead, in order to understand this diversity, one needs to draw on the
descriptive tools and techniques of biology sciences (more than of physics). In Pavitt’s words:
“…[T]ruths about the real innovating firm will never be elegant, simple or easy to
replicate. Certainly, it is just as wrong to criticise formal models in evolutionary
economics, because they don’t help managers, as to criticise formal biological models,
because they don’t help us to survive in the forest. But in biology, there are also the
empirical and often descriptive disciplines of botany and zoology, which offer insights
that are mostly conditional and contingent, but quite useful in coping with the real
world” (Pavitt, 1998).
Taxonomy, as the science of classification of organisms, has largely diffused to social
sciences, and among these to the study of technological change. As exemplified by Pavitt
taxonomy (1984), empirical classifications are a useful tool that helps understanding the
diversity of innovative patterns in firms and sectors (Archibugi, 2001). Taxonomies are
systems of classification that organize and label many different items into groups that share
common traits. A useful taxonomy is one that reduces the complexity of empirical phenomena
to few and easy to remember categories. Taxonomies provide scholars with a framework that
help to build theories. As well, practitioners and policy makers can use taxonomies to shape
firm strategies and policy decisions (Pavitt, 1984; Archibugi, 2001).
The use of taxonomies of innovation finds a theoretical precedent in the concept of
“technological regime” (Nelson and Winter, 1977; Dosi, 1982). Nelson and Winter (1977)
propose that a “useful theory of innovation” needs to be built on the empirical observation of
the diversity of innovative behaviour of firms; what they call “appreciative theorising”. In
particular, firm behaviour is shaped and constrained by the nature of technology, whose
varieties of forms can be categorised into a few broad categories, or “technological regimes”.
They reflect differences across technologies in (i) the sources and principles of knowledge at
the origin of innovation (Dosi, 1982); (ii) the boundaries that can be achieved in the process
of innovation (Nelson and Winter, 1977); and (iii) the directions, or “natural trajectories”,
along which incremental innovations take place (Nelson and Winter, 1977; Dosi, 1982).
Previous empirical research has shown that the sources, rates and directions of
technological change vary significantly across sectors, both in manufacturing and in services
(Pavitt, 1984; Klevorick et al., 1995; Evangelista, 2000). These patterns however have been
1
SPRU Conference on “Evolutionary Economics”, University of Sussex, UK, March 11 – 12, 1996.

3. 2
explored empirically mainly for relatively large firms, while sectoral variations in innovative
behaviour of small and medium sized firms (SMEs) are yet overlooked. Many studies have
examined the success factors of innovation in SMEs, looking at the characteristics of
products, firms, markets and innovation processes. Yet, most of the studies on SMEs did not
investigate in a systematic way the differences across industries in the patterns of innovation
(Brouwer and Kleinknecht, 1996; Hadjimanolis, 2000; Bougrain and Haudeville, 2002). This
often reflects a lack of feasible data, especially for micro-firms with less than 10 employees.
This paper aims at classifying empirically groups of SMEs with similar innovation
patterns. For this purpose it examines the innovative behaviour of a sample of 1,234 Dutch
firms with less than 100 employees. In doing so, it extends the current literature on innovation
patterns in several directions. First, the classification is built using some new indicators,
which are especially relevant to innovation in SMEs. Second, it captures manufacturing and
service firms at once, thus enabling a direct comparison between the two. Also, it selects the
firm as its unit of classification, while most previous taxonomies are formulated at the sector
level. Finally, it includes micro-enterprises, with less than 10 employees. This adds to
taxonomies based on the data from the Community Innovation Survey, with is carried out for
firms with at least 20 employees in most countries, and for firms with at least 10 employees in
few countries.
The results show that the innovation behaviour of small firms is highly differentiated;
small firms cannot be regarded as a homogenous group in their innovative activities. We find
that a classification into four categories well represents this diversity, with a profile that
resembles Pavitt’s taxonomy. This suggests that small firms display innovative patterns more
diverse than expected with Pavitt’s taxonomy, in which small firms fall within the two
categories of specialised suppliers and supplier dominated firms. In addition, firms in
manufacturing and services sectors share common patterns, as firms from both sectors belong
to the all range of categories.
The paper is organised as follows. In the Section 2 we present an overview of the relevant
taxonomies of innovation in the literature and discuss some methodological issues in the
building of taxonomy. Section 3 describes the sample used in the analysis, the data collection
process, and the variables which we employ to construct and validate the taxonomy. Section
4 illustrates the method of classification of the innovative firms. In section 5 we present the
results of the analysis and describe the profile of the taxonomy obtained into four clusters of
firms. Section 6 is the conclusions.
2. Taxonomies of innovation
In his pioneering article, Pavitt (1984) proposed a taxonomy that distinguishes different
categories of innovative firms based on their structural characteristics and organisation of
innovative activities. The aim of the taxonomy was to provide an empirically based
framework as a basis for developing a theory of innovation as well as guiding S&T policies.
Based on the data from the SPRU innovation survey and review of cases studies, Pavitt
identified four groups of firms: science-based, specialised suppliers, supplier dominated and
scale intensive. The Pavitt taxonomy has become a heavy-cited framework for innovation
researchers to explore deviations across industries. Indeed, the 1984 article is the most cited
of Keith Pavitt’s articles (Meyer et al., 2004). Pavitt taxonomy has been used as a predictive
tool of various dimensions of firms’ behaviour and performance, such as: strategy related
variables (Souitaris, 2002); the sources of international competitiveness (Laursen and
Meliciani, 2000); and innovative performance (Laursen and Foss, 2003).
Although Pavitt taxonomy was in principle formulated at the firm level, the various
categories are often regarded, as also suggested in the 1984 paper, as representative of inter-
industries differences in technological conditions (Archibugi, 2001). Other taxonomies can be

4. 3
found that focus on the organization of the innovation process in particular industries. A
classification inspired to the views of the innovative firm expressed by Schumpeter (1934;
1942), distinguishes between two categories: “Schumpeter Mark I” (SM-I) and “Schumpeter
Mark II” (SM-II) (Malerba and Orsenigo, 1995). This classification is directly linked to the
concept of “technological regime” (Nelson and Winter, 1977) and the typology suggested by
Winter (1984). This identifies an “entrepreneurial regime” (or SM-I) as one in which the
nature of technology favours innovation by new and small firms. Opposite to that, a
“routinised regime” (or SM-II) is one in which the nature of technology is such that
established firms are the major innovators. The emergence in a sector of one or the other of
the two Schumpeterian patterns of innovation is to be interpreted on the ground of the specific
combination of the level and sources of technological opportunity; appropriability conditions;
cumulativeness of innovation; and the nature of knowledge bases (Dosi, 1988; Malerba and
Orsenigo, 1990; Breschi et al., 2000).
Taxonomies of innovation built on the concept of technological regime have a cognitive
basis; they reflect the diversity in the nature of the technological competencies that shape and
constraint what “firms can and cannot do” (Pavitt, 1998). As pointed out by Nelson and
Winter (1977), “[their] concept is more cognitive, relating to technicians’ believes about what
is feasible or at least worth attempting” (p. 57). Other taxonomies of innovation, which are
based on cognitive mechanisms, focus on the effects that major improvements in technology
have on the firm-specific competencies (Pavitt, 1998). These effects consist of either
enhancing the competencies accumulated by established firms, or of destroying them, and
thus of threatening the position of the established firms in the industry (Tushman and
Anderson, 1986). Based on this distinction, Abernathy and Clark (1985) elaborate a typology
of innovations into four categories: incremental, component, architectural and revolutionary.
Some taxonomies rest on the differences in the nature of technology as new
technologies evolve along the stages of the Product Life Cycle (Utterback and Abernathy,
1975; Klepper, 1997). In the PLC model, industries are classified according to their stage of
evolution, in relation to the changing nature of product and process innovation. Although with
variations in the number of stages among later studies, the basic classification, going back to
the work of Abernathy and Utterback (1978), distinguishes industries that are at a ‘fluid’
stage, a transition stage and a ‘specific’ stage. Recently, McGahan (2004) has proposed a
revisited version of the PLC model, which account for the type of effects that technological
change has on the core activities and core assets of an industry. To the extent that these effects
may threaten industries of obsolescence, four trajectories of industry evolution are identified:
radical, progressive, creative and intermediating.
Classifications based on technology sectors were also developed under the auspices of
the OECD, which contrast high-tech and low-tech industries. This type of classification is
based on the intensity of technology production, as measured by the intensity of R&D
expenditure in a sector, and, in a revised version, on the intensity of technology use and the
process of diffusion across sectors (Hatzichronoglou, 1997). While the OECD classification is
technology-based, recent attempts are made to also include non-technological dimensions
(intangible investments and human capital) among the factors of production (Peneder, 2002).
In sum, the biology analogy in the use of descriptive tools of analysis like taxonomies,
advocated by Pavitt, is widely diffused in the literature on innovation studies. Table 1
provides an overview of the empirical taxonomies of innovation patterns that, in the past
twenty years, have been elaborated in the line of Pavitt’s effort.2
----- Insert Table 1 about here -----
2
A review of industry classifications, which highlights the aim, scope and techniques of major taxonomies that
have been developed in the broader context of applied economic studies, can be found in Peneder (2003).

5. 4
2.1. Methodology issues
Pavitt’s taxonomy is not immune from criticisms and limitations. Archibugi (2001) points out
two main issues: (a) the sectoral composition of the classification and (b) its unit of analysis.
With regard to the sectoral composition, the main issue concerns the treatment of the
services sector in the taxonomy. The 1984 version of the Pavitt taxonomy classifies the whole
services sector within the category of supplier-dominated firms, while the revised version
adds a fifth category of information intensive firms (Tidd et al., 2001). Nevertheless, studies
of innovation in services highlight that there is more diversity of configurations within the
sector than assumed in Pavitt’s only category of information-intensive firms (Miles, 1993;
Evangelista, 2000; Miozzo and Soete, 2001). These studies often use more robust measures of
innovation in services that are now available from the Community Innovation Survey
(Evangelista, 2000; Kleinknecht, 2000). The new classifications display close resemblance to
Pavitt taxonomy, suggesting that the patterns of innovation in the service firms can be as
differentiated as in the manufacturing firms (Evangelista, 2000; Miozzo and Soete, 2001).
Because of this similarity of results, as well as the difficulty to identify the actual activity of
firms, who increasingly bundle manufacturing and services (Miozzo and Soete, 2001),
Archibugi (2001) suggests that taxonomies should be developed jointly for the two sectors.
With regard to the level of observation, the issue is here in the choice between the firm
level and the industry level of analysis (Archibugi, 2001). Pavitt taxonomy, as well as most of
its successors, is based on industry-level indicators of innovative activities (Archibugi et al.,
1991; Hatzichronoglou, 1997; Evangelista, 2000). While the sector level may better respond
to the requirement of policy making (Raymond et al., 2004), it does not account for the
diversity of innovative performance and strategies within a sector (Dosi, 1988; Marsili and
Salter, 2004). Archibugi (2001) argues that to account for this heterogeneity, taxonomies of
innovation need to be developed directly at the firm level, before aggregating the firms into
the standard system of industrial classification. Indeed, Pavitt taxonomy is in principle a
taxonomy of firms, although its empirical validation relies on industry data (Archibugi, 2001).
The few empirical studies that classify firms directly (Cesaratto and Mangano, 1993;
Arvanitis and Hollenstein, 1998) find that firms within the same industry are often dispersed
across several different groups of the constructed taxonomy.
Following the lines of research discussed above, this paper extends previous empirical
work in four respects. First of all it focuses on small and medium sized firms, and especially
on micro enterprises (with less than 10 employees). Pavitt taxonomy is based on a sample of
innovations that under represents the contribution of small firms; these, in turn, are grouped
into two categories: supplier dominated and specialised suppliers. Successive empirical work
has often overlooked the innovative behaviour of small firms, and most often that of micro
firms (see Table 1). Samples are usually limited to firms with a number of employees above
the threshold of twenty (Archibugi et al., 1991), and in few cases above ten (Raymond et al.,
2004) or five (Arvanitis and Hollenstein, 1998). Yet, firm size appears to matter for the
presence and nature of innovative practices (Welsh and White, 1981; Vossen, 1998; Bodewes
and de Jong, 2003). Accordingly, this study aims at a more fine grained representation of
patterns of innovation in small firms, as compared to Pavitt taxonomy.
Second, we use some new variables to build a taxonomy, which have relevance to small
firms, since indicators of R&D and of other innovation costs do not account for the more
informal type of innovative activities that are typical of small firms (Brouwer and
Kleinknecht, 1996; De Jong, 2002). In addition, we extend the set of variable used for the
classification to include variables related to firm strategy. In this respect, the taxonomy is seen
as an “integrative tool” (Souitaris, 2002), which links differences in the patterns of

6. 5
innovation, as expressed by the rates, sources and directions of innovation, with differences in
firm strategies (Malerba and Orsenigo, 1993; Kaniovski and Peneder, 2002; Souitaris, 2002).
Third, as suggested by Archibigi (2001), we consider manufacturing and service firms at
once, thus enabling a comparison in the main patterns of innovative behaviour between the
two sectors. More taxonomic exercises have been carried out for manufacturing than for
services (Table 1), also because of lack of statistical data on innovation in the latter. Most
often taxonomies are developed separately for manufacturing and services industries or
whenever joined, services fall within one or two broad classes. We expect that a similar
distribution of patterns of innovation emerges for manufacturing and services firms.
Finally, while most taxonomies are based on sector-level data, we use firm-level data to
classify firms directly according to their innovative behaviour. This allows testing for the
assumption that firms within an industry share common innovative patterns (Archibugi,
2001).
3. Data and Variables
3.1. Data collection
The data were collected as part of a survey performed by EIM Small business research. This
survey was meant to collect innovation statistics for Dutch policy makers. Its sample was
stratified across 18 industries and two size classes, covering all parts of the Dutch commercial
business society with the exclusion of agriculture. The sample was randomly drawn from the
population of all small and medium-sized firms in the Netherlands, following the Dutch
definition of SMEs as firms with no more than 100 employees (Bangma and Peeters, 2003).
The population was derived from a database of the Chambers of Commerce, containing data
on all Dutch firms. The data were collected in April 2003, over a period of three weeks, by
means of computer assisted telephone interviewing (CATI). All respondents were managers
responsible for day-to-day business processes – usually the owner/entrepreneur, and
otherwise a general manager. Attempts to contact the reference person were made five times
before considering the company as a non-respondent.
3.2. Sample
The initial sample consisted of 2,985 firms. Of these, 1,631 firms were willing to participate
in the enquiry, with a response rate of 55 per cent. To check for non-response bias, the
distribution of respondents and non-respondents across type of industry and size class was
compared. The χ2
-tests contrasting the two groups revealed no significant differences at the
5% level (p = 0.07 for type of industry and p = 0.65 for size class), indicating that non-
response bias was not a serious problem. As the focus of the survey was on innovative firms,
only respondents who had implemented at least one (product or process) innovation in the
previous three years were asked to respond to the full questionnaire. Of the 1,631
respondents, 1,234 innovative firms completed the full questionnaire (76 percent). Another
383 firms did not implement any innovation recently, while 14 firms were discarded because
of incomplete answers. Table 2 shows how the innovative firms are distributed according to
their sector of activity and size class. Of the 1,234 innovative firms, 776 firms are micro-
firms, with less than 10 employees (63 per cent). This percentage under estimates, to a certain
extent, the vast presence of micro firms in the entire population, for which it is 90 per cent of
the total (Bangma and Peeters, 2003). Yet, our sample gives a much larger representation than
other innovation surveys (the CIS excludes firms below the threshold of 10 employees).
--- Insert Table 2 about here ---

7. 6
3.3. Clustering variables
For the construction of the taxonomy, we rely on cluster analysis techniques. These are
sensitive to the selection of the variables used, since the addition of irrelevant variables can
have a serious effect on the results of the clustering (Milligan and Cooper, 1987). Cluster
variables should also be representative for the typology one wants to present (Everitt, 1993).
For this reason, we employ a combination of “core variables” that have been applied
several times before and of new variables, because of their relevance to innovation in SMEs.
Table 3 shows the dimensions and variables that formed the basis of our taxonomy. Some of
them (‘managerial attitude’ and ‘product innovation’) are constructed by summing up the
responses to different statements, for which the Cronbach’s alpha is also reported.
--- Insert Table 3 about here ---
Core variables that have been largely used in building taxonomies of innovation are the
innovativeness of a firm (input and output), also in relation to the different nature of product
and process innovation, and the sources of innovation (Pavitt, 1984; Archibugi et al., 1991;
De Marchi et al., 1996; Arvanitis and Hollenstein, 1998; Evangelista, 2000; Marsili, 2001);
the latter often influence how various patterns of innovation are labelled.
Innovative output. The definition and measurement of product and process innovation in
the EIM survey are similar to those employed in the Community Innovation Survey (OECD,
1997). To account for both the presence and degree of novelty of product innovation,
occurred in the three years previous to the survey, we construct an indicator that combines
two dichotomic variables (Cronbach’s alpha = 0.74). One indicates whether or not a firm has
introduced a product new to the firm, including minor product improvements or mere
imitation. The other indicates whether or not a firm has introduced a product new to the
industry, thus covering those product innovations with a higher degree of novelty. For process
innovation, we use a dichotomic variable of whether the firm has introduced at least one new
work process in the past three years.
Innovative input. It is well known that input measures that rely on indicators based on
R&D expenditure or R&D-personnel do not adequately capture the innovative effort by small
firms (Brouwer and Kleinknecht, 1996; Roper, 1997; Rogers, 2004). For example, Sundbo
(1996) argues that as an alternative to R&D, many small firms empower their workforce to
contribute to the innovation process. Most SMEs are not involved in R&D themselves or, not
being the best bookkeeper, they may fail to record their R&D-related expenses in their
accounting systems. For the above reasons, the EIM survey collected information on
alternative input indicators: the reservation of an annual budget (money) and of capacity
(time) to implement new products or processes, and the presence of innovation specialists.
The first variable reflects the financial constraints that especially within SMEs can act as
bottleneck to realise something new (Acs and Audretsch, 1990; Hyvärinen, 1990). The
investment of time is also regarded as a success factor (Brouwer and Kleinknecht, 1996;
Roper, 1997), not only in terms of the time that one needs for developing innovations, but
also the time that is devoted to the commercialisation stage and customer relationships
(Hyvärinen, 1990). The third variable, presence of innovation specialists, is defined by the
employees occupied with innovation as part of their daily work. In SMEs, this measure has
been identified as one that predicts innovation success (Hoffman et al., 1998).
Sources of innovation. Our dataset provides information on three alternative sources of
innovation: suppliers, customers, and scientific developments. This selection is somehow
limited with respect to the broad range of sources of innovation that are considered in the CIS
questionnaire. However, customers, suppliers and knowledge/education institutes are
generally found to be significant sources for innovation in SMEs (Hyvärinen, 1990; Brouwer

8. 7
and Kleinknecht, 1996; Roper, 1997; Appiah-Adu and Singh, 1998; Oerlemans et al., 1998).
In particular, a strong customer orientation appears to be closely linked to the success of small
firms in developing innovative products and services (Appiah-Adu and Singh, 1998).
A second set of variables that we use in the construction of the taxonomy is composed
of variables that are related to the strategy of the firms (Table 3). Only few studies have used
similar variables in combination with taxonomies of innovation patterns (Evangelista, 2000;
Souitaris, 2002). We consider three dimensions: 1) managerial attitude, 2) innovation
planning and 3) external orientation.
Managerial attitude. In SMEs the owner/entrepreneur has a larger direct influence on
employees compared to managers of large organizations (Bodewes and de Jong, 2003).
Leaders in small firms can successfully instil an ‘entrepreneurial dynamism’ (Davenport and
D., 1999) in the behaviour of others in the organization. A positive attitude towards
innovation correlates with a continuous attention for innovative opportunities and provides
employees with support for innovative behaviour. This, in turn, strongly affects the decision
to innovate and the way innovation is carried out in small firms (Kim et al., 1993; Hoffman et
al., 1998; Hadjimanolis, 2000). Managerial attitude is measured by combination of three
variables (Cronbach’s alpha = 0.67). They reflect a positive attitude of the SME towards
investing resources (time) in innovation and the expected results from innovation, in relation
to customers and competitors.
Innovation planning. Planning is generally recognized as a factor for success especially
in new firms (Delmar and Shane, 2003). This is one of the factors that distinguishes innovate
firms from their less innovative counterparts (Hadjimanolis, 2000). In our dataset, this
variable is measured by the presence of a documented innovation plan, implying that explicit
ambitions, targets and milestones were defined and written down.
External orientation. Empirical evidence supports that SMEs that are aware of and use
external information perform significantly better in terms of innovation success (Hoffman et
al., 1998; Romijn and Albaladejo, 2002; Freel, 2003). To represent this dimension, we use
two variables. One is defined as the number of external parties that a firm consulted for
information and advice, and, although these may not be directly related to the innovation
process (see Table 3), they can be assumed to extend a firm’s knowledge base. The index is
similar to the one introduced by Laursen and Salter (2004), to measure the degree of
“openness” of a firm in the search for innovation ideas (Chesbrough, 2003; Laursen and
Salter, 2004). The second variable refers to the participation in inter-firm cooperation, which
is believed to favour innovation, particularly in SMEs, for a variety of reasons, such as the
ability to overcome a lack of resources, to spread risks, and to acquire complementary assets
(Brouwer and Kleinknecht, 1996; Hadjimanolis, 2000; Hanna and Walsh, 2002).
3.4. Variables used for validation
In order to assess the validity of the taxonomy derived from the above variables we relay on a
new set of variables that are not used to construct the taxonomy but are known or expected to
vary across its clusters (Milligan and Cooper, 1987; Hair et al., 1998). For this purpose we use
the variables indicative of whether the firm is a first mover, whether it has an explicit search
strategy for new knowledge, and whether the firm use innovation subsidies (Table 4).
--- Insert Table 4 about here ---
These variables can be assumed to vary across (groups of) innovative firms. If a cluster is
highly innovative, we expect that its firms are more likely to regard themselves as first-
movers, to implement search strategies for new knowledge, and to use innovation subsidies.

9. 8
Finally, we investigate differences across type of industry and size class to establish
whether some industries are more represented in some clusters of firms than others and to
assess whether different patterns emerge across size classes.
4. Method
Our analysis consists of three steps. We begin with a principal component analysis to reduce
the number of dimensions in our dataset. Next, we apply cluster analysis techniques to build a
taxonomy of innovative firms. Finally, we use analysis of variance and χ2
-test to validate the
taxonomy.
4.1. Principal component analysis
Several studies that perform taxonomic exercises of innovation patterns use Principal
Component Analysis (PCA), as a way to reduce the number of dimensions to be used in the
clustering. However, due to the focus on the sector level of most studies, the component
analysis may suffer from too few observations (Evangelista, 2000; Peneder, 2002), while at
the firm level a greater number of observations can provide more robust results for it. In
general, the PCA reduces the risk that single indicators dominate a cluster solution, and helps
to prevent that irrelevant (non-discriminative) variables are included (Everitt, 1993; Hair et
al., 1998). Another advantage is that the factors obtained from a PCA are uncorrelated and
therefore no variable would implicitly be weighted more heavily in the clustering and thus
dominate the cluster solution (Hair et al., 1998, p. 491).
Since most of our variables have significant correlations, often exceeding absolute
values of 0.20, they seem to be appropriate for a PCA. Furthermore, we test if our data are
suitable for a component analysis, by calculating Measures of Sampling Adequacy (MSA) for
the individual variables (Hair et al., 1998). All the variables, with the exception of process
innovation, have satisfactory MSA values (> 0.60), and they are suitable candidates for a
PPA. In addition, KMO and Bartlett’s test of sphericity met common standards (KMO = 0.80
and p(Bartlett) < 0.001) (Hair et al., 1998). With regard to the process innovation variable, the
value of the MSA test suggests that this is not adequate for a PCA (MSA = 0.46 < 0.50). The
great majority of firms, 92 per cent, carry out process innovations, leaving a small group of
only 8 per cent of the cases to dominate any solution. Therefore we omit this variable from
the taxonomy construction phase, and use it to validate and describe the profile of the clusters.
In performing the principal component analysis, we use the extraction technique with
varimax rotation and, for the selection of the number of factors, we apply the latent root
criterion, requiring that the eigenvalues are greater than one. As result, we obtain a three-
dimensional solution explaining 46% of the variance. Since we use the PCA with the aim of
reducing the number of dimensions for building the taxonomy, whose groups will be
validated and profiled on the basis of the original variables, the output is not presented here.3
4.2. Cluster analysis
Because cluster analysis is sensitive to outliers, we first assess for outlying observations in the
scores of the three principal components. Using Hair et al.’s (1998) method, two extreme
cases are detected: their scores on the first component deviate more than three standard
deviations from the mean. Both cases are then eliminated from the analysis.
In the cluster analysis, we follow a two-step procedure, which combines hierarchical
and non-hierarchical techniques, to obtain more stable and robust taxonomy (Milligan and
Sokol, 1980; Punj and Stewart, 1983). As a first step, we carry out a hierarchical analysis to
group the firms into homogeneous clusters, by using the Ward’s method based on squared
3
This and other results of the cluster analysis that could not be reported for reason of space are available from
the authors on request.

10. 9
Euclidian distances.4
Homogeneous groups are built so as to minimise the distance in scores
of firms within a single cluster and to maximise the distance in scores between companies
from the various clusters. This analysis gives the initial solutions for the taxonomy.
The second step consists of a non-hierarchical cluster analysis to improve on the initial
solutions and to select the number of clusters for the taxonomy. At first, a visual inspection of
the dendogram, plotting the initial solutions of the hierarchical analysis, suggests a taxonomy
with four clusters. For a better assessment of robustness, we then consider the all range of
initial solutions from the hierarchical analysis, going from two up to six clusters. For each
number of clusters (k), we perform a k-means ‘non-hierarchical’ cluster analysis, in which the
firms are iteratively divided into clusters based on their distance to some initial starting points
of dimension k. While some k-means methods use randomly selected starting points, we
employ the centroids of our initial hierarchical solutions for this purpose (Milligan and Sokol,
1980; Punj and Stewart, 1983). Finally, to assess robustness we compute Kappa, the chance
corrected coefficient of agreement (Singh, 1990), between each initial and final solution. The
four-cluster solution appears to have the highest value of Kappa (k = 0.75, while k < 0.72 for
the other solutions) and is thus selected as our final taxonomy.
5. Results
5.1. Descriptive statistics
Table 5 reports the means for all variables used in the analysis. With regard to the output,
process innovation is more widespread than product innovation. While 92 per cent of firms
have implemented process innovations, 45 per cent has not carried out any product
innovation. In input, a larger proportion of firms dedicate time to innovation (69 per cent),
than those who reserve a budget for innovation (50 per cent). Lower is the percentage of those
who write down a formal plan for innovation (37 per cent). As well, the presence of
innovation specialists is fairly low (16 per cent). Customers are considered to be a source of
innovation more important than suppliers and scientific developments. Small firms appear to
be fairly ‘open’ as on average they tend to draw on more than three relevant sources of
knowledge,5
55 per cent of them has an explicit search strategy for new knowledge, and 57
per cent participate in formal collaborations for innovation. In contrast, most firms do not
consider themselves as first mover innovators (76 per cent), and a small proportion uses
innovation subsidies (18 per cent). Finally, respondents display a positive attitude towards
innovation, with a mean score of 4.06 close to the maximum. This is not surprising as the
sample consists of innovative firms only. Yet, we do recognize some variance in the extent to
which respondents agree on this summated scale.6
--- Insert Table 5 about here ---
5.2. Empirical taxonomy
Table 6 shows the descriptive statistics of the four clusters of small innovative firms. On the
basis of the scores of the clustering variables, we label the four groups, to stress the similarity
with Pavitt taxonomy, as supplier-dominated (23 per cent), specialised suppliers (24 per cent),
science-based (26 per cent) and resource-intensive firms (27 per cent).
4
The Ward’s method generally provides good results compared to other clustering methods (Milligan and
Cooper, 1987).
5
In particular, 4 per cent of firms consult with 6 or more external sources; 13 per cent with five sources; 27 per
cent with four; 25 per cent with three; 18 per cent with two; 9 per cent with one source and 4 per cent with none.
6
A closer inspection of the data shows that the mean score falls in the range (4; 5] for 36 per cent of the firms; in
(3; 4] for 58 per cent; in (2; 3] for 5 per cent of the cases, and in [1; 2] for the remaining one percent.

11. 10
-- Insert Table 6 about here --
Supplier-dominated firms. Among the four groups, this displays the lowest score in most
variables and below the average in all but process innovation, the role of suppliers as sources
of innovation (for which it ranks first), and the number of external sources of knowledge (for
which it ranks second). Innovativeness is low in all dimensions: in all forms of input
(financial, time and employment), in formal planning and managerial attitude. Innovation
mainly consists of process innovation, and essentially responds to the proposal of new
applications from suppliers. Firms are relatively “open” as, on average, they consult with
more than three external parties. However, this orientation seems to be more related to the
normal solution of business problems than to formal partnership aimed at innovation.
Specialised suppliers. Innovativeness among these firms is fairly high. It is almost at the
highest level (comparable to science based firms) in product innovation, while it is lowest in
process innovation, implying a distinctive prevalence of product on process innovations. The
innovation process is based on a more diffused use of specialised labour when compared with
financial and time resources. Firms are customer driven, as they heavily rely on understanding
customers’ needs as a source of innovation. Together with the fact that the relevance of the
other sources of innovation, suppliers and scientific development, is the lowest in this group,
it suggests a distinctive customer focus. This is also consistent with the low degree of
“openness” of these firms, given the number of sources they consult is the lowest (about two),
although they frequently participate in formal collaboration (likely with their customers).
Science-based firms. This cluster displays the highest score on most variables with the
exception of the relevance of suppliers as a source of innovations (the only one for which it
ranks third). Innovativeness is high, both in products and processes, and innovation specialists
are most often employed within these firms. Firms are distinguished for using knowledge
from universities and research institutes as a source of innovation, but they also draw heavily
on customers needs (the latter is a common feature with specialised suppliers). Managers have
a strong positive attitude towards innovation, which is most frequently accompanied by a
written-down plan (66 per cent versus the average of 37 per cent). Science-based firms are
also the most “open”; they tend to consult with more than four external parties on average,
and very often participate in innovation collaborations (90 per cent of firms).
Resource-intensive firms. This group shares some characteristics with supplier-
dominated firms, although the degree of innovativeness is relatively higher (not far from the
middle), and the prevalence of process on product innovations, as well as the role of suppliers
as a source of innovation, is less pronounced. The most distinctive feature consists in the high
shares of firms who reserve budgets and time for innovative activities, 92% and 95%
respectively, which are the highest across the four clusters. The external orientation is below
average, in terms of both the consultation of external parties and formal partnership for
innovation. These firms thus appear to allocate financial and time resources to innovation,
although with a limited use of dedicated personnel and external networks.
While the similarities of the first three clusters with Pavitt taxonomy are manifest, and
hence the use of the same labels, the last cluster departs from it. Although it shares some traits
with Pavitt category of scale intensive firms, such as the combination of in-house resources
and reliance on suppliers as sources of (mainly) process innovation, yet, the similarity is the
smallest, because our classification is not based on firm size.
5.3. Validation and industry composition
As a basic requirement for validity, one needs to find significant differences on the variables
that have been used to develop a taxonomy (Milligan and Cooper, 1987). A multivariate
analysis of variance test reveals significant differences on all clustering variables (Pillai’s

12. 11
Trace F-value = 143.0 and p < 0.001). In addition, one-way analyses of variance for each
individual variable confirm this finding (see Table 6). In particular, significant differences are
also found for the indicator of process innovation, which has not been used for cluster
development. These differences are consistent with the clusters profile that emerges from the
variables actually used in the clustering. Besides, validity is supported in the similarity with
the Pavitt categories (1984); with process innovation in supplier dominated firms, a mix of
product and process innovation in science based and product innovation in specialised
suppliers. These results can be considered as a first indication of validity.
A further assessment of the validity of clusters exploits variables not used to form the
clusters, but known or expected to vary across them (Milligan and Cooper, 1987; Hair et al.,
1998, p. 501). Table 6 also reports the descriptive statistics of the three external variables used
for this purpose: whether the firm consider itself as first-mover in innovation, the presence of
a formal policy to collect new knowledge, and the use of innovation subsidies. The analysis of
variance shows significant differences across the four groups on all external variables. The
nature of these differences is consistent with the characterisation of clusters in our taxonomy.
As expected, science-based firms who are highly innovative both in relation to their
internal practices and external networks, rank also highest on all external variables. Indeed
these firms are often the first ones to introduce the innovations, have a well-defined search
strategy for new knowledge (70 per cent of firms), and take the lead in the use of innovation
subsidies. Supplier-dominated firms, with an almost opposite profile, score the lowest also in
all the external variables. In particular, it is here supported that the external orientation that
was observed in terms of consultation of various external parties, is not part of a search
strategy for new knowledge. For the specialised suppliers, who are characterised for product
innovation and close relationship with customers, it is not surprising that speed to the market
is important; indeed, the number of firms that consider themselves as first movers is just
below the maximum value (33% versus 24% average). Finally, the resource intensive firms
again score slightly below or never far from the average of all innovative SMEs.
Some more evidence on the taxonomy validity is found when comparing the distribution
of clusters across industries and size classes (Table 7). The chi-square test shows that within
the industries, the distribution of firms across the clusters is not uniform (χ2
=142.6, df=51, p
< 0.001). In some industries particular patterns appear as more represented than others.
Science-based firms are well present in the chemicals, machinery, office and electrical
equipment, as well as in economic services (e.g. consultancy) and engineering and
architectural services. In the Netherlands, these industries are characterised by a highly
educated workforce and the frequent application of new technologies (Bangma and Peeters,
2003). Specialised suppliers are most often found in wholesale and computer and related
services; while supplier-dominated firms are the prevailing type in the metals industry,
transport and construction, three industries that in the Netherlands have been recognized for
their lack of initiative in innovation. Resources intensive firms are especially represented in
hotels and restaurants, and personal services.
At first glance, the industry composition of clusters suggests that firms from the various
clusters can be found both in manufacturing and services. To establish whether there are
systematic differences between the two sectors, we examine the distribution of firms in the
four clusters within each one broad sector, leaving aside the construction firms (figures
printed in italics in Table 7). A chi-square test shows that overall the differences are
statistically significant at the 1 per cent level (χ2
=14.5, df=3, p < 0.01), with a concentration
of service firms in the resource intensive cluster. However, one should notice that – although
one cannot deny variance across industries – the different patterns are observed in each
individual industry (e.g. in each industry at there is at least 8 percent of the firms that are

13. 12
supplier-dominated). This may reflect the high degree of aggregation of the industrial
classification used in the analysis, which may average the patters of individual industries.
As for size class, we observe significant differences in the distribution of clusters
between micro firms (below 10 employees) and small-medium firms (between 10 and 100
employees). The chi-square test indicates that they are highly significant (χ2
=45.8, df=3, p <
0.001), as science based firms tend to be relatively larger in size than other types of firms.
6. Conclusions
Empirical taxonomies have proven to be a useful tool for understanding the diversity of
innovative behaviour that can be observed across firms. In this paper, we have attempted at
extending this approach to small firms (below 100 employees), which are often excluded or
under-represented in taxonomies of innovation. In particular, within this class of firms, our
research covers a dominant proportion of micro firms, with less than 10 employees. The class
of micro firms finds very little part in earlier empirical taxonomies of innovation.
Drawing upon a sample of 1,234 SMEs in the Netherlands, we derive a taxonomy of
four types of innovative firms, with a profile that supports but also qualifies and extends
Pavitt’s taxonomy to small firms, in both manufacturing and services. This leads to conclude
that, first, the innovation patterns in small firms are more diverse than generally believed, as
for example in Pavitt taxonomy, in which they are represented mainly by two categories. The
similarity between our categories of small and new firms, and the Pavitt taxonomy, suggests
that these firms can in fact play the role of the fruit flies of innovation. They manifest on a
fast changing setting, some underlying properties of technological change that are likely to be
found on the more stable and less selective setting of large and established firms.
Second, our research shows that the diversity of patterns in services firms is as at least
as broad as that observed in manufacturing firms, and the two sectors share, to a large extent,
common patterns. This is consistent with the results of previous studies that have focused on
classifications a la Pavitt of the service sector, and that have revealed highly differentiated
patterns of innovation in the services firms (Evangelista, 2000). Our findings are also in line
with Archibugi (2001) who proposes that the two sectors share some fundamentals in the
process of innovation, such as the interaction with suppliers and the innovation intensity. This
also consistent with the observation that the boundaries between manufacturing and services
have blurred as services and manufacturing activities are often closely bundled within
organizations (Miles, 1993; Schmoch, 2003).
Finally, our results confirm that taxonomies of innovation based on differences in the
patterns of innovation – as defined by the rates, sources and directions of innovation – can
usefully be extended to map differences in business practices and strategies. These variables
are especially important to capture the more informal aspects of the innovation process that
are typical of small firms, and cannot be measured by traditional indicators of input and
output. As in Souitaris (2002), we find that the differences in a number of strategy related
variables – such as innovation budget, business strategy, and management attitude – can be
systematically linked to differences in the rates, sources and nature of innovation. This
supports Souitaris’s view of the Pavitt taxonomy as an “integrative tool” of the management
perspective and the economic perspective in the study of the determinants of innovation. In
this paper, we focused on some elements of a firm’s strategy that are related to its innovative
activities. A question for further research is whether the taxonomic approach can be extended
to link innovation to the broader set of competitive strategies of firms. For example, Laursen
and Salter (2004) conjecture that the strategies firms pursue in their search for innovative
ideas, could find a match with the way firms position themselves along the value chain.
Taxonomies are a valid instruments to organise the diversity of patterns observed in
reality, which can be used for policy making and theory building, to the extent that they are

14. 13
invariant across specific contexts. In this research we have looked at the invariance with
respect to firm size, as our taxonomy of small firms resembles others built on samples of
predominantly large firms, and to the distinction between manufacturing and services.
Another important issue is whether taxonomies are invariant across institutional contexts, in
particular across countries and over time. Across countries, the specific institutional setting
may lead not only to a different distribution of common patterns but also to the emergence of
more idiosyncratic patterns. Over time, new industries may be created per effect of the
introduction of new technologies as well as existing industries may change their profile
(Archibugi, 2001); taxonomies are then to be seen not as static but rather as dynamic tools.

15. 14
References
Abernathy, W. J., Clark, K., 1985. Innovation: Mapping the winds of creative destruction,
Research Policy 14, 3-22.
Abernathy, W. J., Utterback, J. M., 1978. Patterns of industrial innovation, Technology
Review 80 (41-47).
Acs, Z. J., Audretsch, D. B., 1990, Innovation and Small Firms. MIT Press, Cambridge, MA.
Appiah-Adu, K., Singh, S., 1998. Customer orientation and performance: a study of SMEs.,
Management Decision 26 (6), 385-394.
Archibugi, D., 2001. Pavitt's taxonomy sixteen years on: a review article, Economic
Innovation and New Technology 10, 415-425.
Archibugi, D., Cesaratto, S., Sirilli, G., 1991. Sources of innovative activities and industrial
organization in Italy, Research Policy 20, 299-313.
Arvanitis, S., Hollenstein, H., 1998, 'Innovative activity and firm characteristics - A cluster
analysis with firm-level data of Swiss manufacturing', EARIE 25th Annual
Conference, Copenhagen, 27-30 August 1998.
Audretsch, D. B., 1995. Innovation, growth and survival, International Journal of Industrial
Organization 13 (4), 441-457.
Bangma, K. L., Peeters, H. H. M. (2003). Kleinschalig ondernemen 2003: structuur en
ontwikkeling van het Nederlandse MKB (Small-scale entrepreneurship: structure and
developments of Dutch SMEs). EIM, 2003. Zoetermeer.
Bodewes, W., de Jong, J. P. J., 2003, Innovatie in het midden- en kleinbedrijf" (Innovation in
small- and medium sized enterprises). In: Risseeuw, P.Thurik, R. (Eds.), Handboek
ondernemers en adviseurs: management en economie van het midden- en kleinbedrijf.
Kluwer, Deventer, pp. 323-338.
Bougrain, F., Haudeville, B., 2002. Innovation, collaboration and SMEs internal research
capacities, Research Policy 31 (5), 735-747.
Breschi, S., Malerba, F., Orsenigo, L., 2000. Technological regimes and Schumpeterian
patterns of innovation, Economic Journal 110 (463), 388-410.
Brouwer, E., Kleinknecht, A., 1996. Firm size, small business presence and sales of
innovative products: a micro-econometric analysis, Small Business Economics 8,
1898-1201.
Caves, R. E., 1998. Industrial organization and new findings on the turnover and mobility of
firms, Journal of Economic Literature 36 (4), 1947-1982.
Cefis, E., Marsili, O. (2003). Survivor: The role of innovation in firm's survival. WP T.
Koopmans Institute, USE, Utrecht University. No. 03-18, 2003.
Cesaratto, S., Mangano, S., 1993. Technological profiles and economic performance in the
Italian manufacturing sector, Economic Innovation and New Technology 2, 237-256.

16. 15
Chesbrough, H., 2003, Open Innovation. Harvard Business School Press, Boston,
Massachusetts.
Davenport, S., D., B., 1999. Rethinking a national innovation system: the small country as
'SME', Technology analysis and strategic management 11 (3), 431-462.
De Jong, J. P. J. (2002). Midden- en Kleinbedrijf terughoudend met innovatie (Small and
medium-sized firms hesitate to invest in innovation). EIM, 2002. Zoetermeer.
De Jong, J. P. J., Vermeulen, P. A. M., O'Shaughnessy, K. C., 2004. Effecten van innovatie in
kleine bedrijven (Effects of innovation in small firms), M&O 58 (1), 21-38.
De Marchi, M., Napolitano, G., Taccini, P., 1996. Testing a model of technological
trajectories, Research Policy 25 (1), 13-23.
Delmar, F., Shane, S., 2003. Does business planning facilitate the development of new
ventures?, Strategic Management Journal 24 (12), 1165-.
Dosi, G., 1982. Technological paradigms and technological trajectories : A suggested
interpretation of the determinants and directions of technical change, Research Policy
11 (3), 147-162.
Dosi, G., 1988. Sources, procedures and microeconomic effects of innovation, Journal of
Economic Literature 26 (3), 1120-1171.
Evangelista, R., 2000. Sectoral patterns of technological change in services, Economics of
Innovation and New Technology 9, 183-221.
Everitt, B. S., 1993, Cluster Analysis. Oxford University Press, London.
Foster, L., Haltiwanger, J., Krizan, C. J. (1998). Aggregate productivity growth: lessons from
microeconomic evidence. NBER Working Papers6803, National Bureau of Economic
Research, 1998.
Freel, M., 2003. Sectoral patterns of small firm innovation, networking and proximity,
Research Policy 32, 751-770.
Hadjimanolis, A., 2000. An investigation of innovation antecedents in small firms in the
context of a small developing country, R&D Management 30 (3), 235-245.
Hair, J. F., Anderson, R. E., Tatham, R. L., Black, W. C., 1998, Multivariate Data Analysis,
5th ed. Prentice Hall, Englewood Cliffs, NJ.
Hanna, V., Walsh, K., 2002. Small firm networks: a successful approach to innovation?, R&D
Management 32 (3), 201-207.
Hatzichronoglou, T. (1997). Revision of the High-Technology Sector and Product
Classification. OECD STI Working Paper Series No. 1997/2, 1997. Paris.
Hoffman, K., Parejo, M., Bessant, J., L., P., 1998. Small firms, R&D, technology and
innovation in the UK: a literature review, Technovation 18 (1), 39-55.

17. 16
Hyvärinen, L., 1990. Innovativeness and its indicators in small- and medium-sized industrial
enterprises, International Small Business Journal 9 (1), 64-79.
Kaniovski, S., Peneder, M., 2002. On the structural dimension of competitive strategy,
Industrial and Corporate Change, 11 (3), 257 – 279.
Kim, Y., Song, K., Lee, J., 1993. Determinants of technological innovation in the small firms
of Korea, R&D management 23 (3), 215-225.
Kleinknecht, A., 2000, Indicators of Manufacturing and Service Innovation: Their Strengths
and Weaknesses. In: Metcalfe, S.Miles, I. (Eds.), Innovation systems in the service
economy: Measurement and case study analysis. Kluwer academic, Boston, Dordrecht
and London.
Klepper, S., 1997. Industry life cycle, Industrial and Corporate Change 6 (1), 145-181.
Klevorick, A. K., Levin, R. C., Nelson, R. R., Winter, S. G., 1995. On the sources and
significance of inter-industry differences in technological opportunities, Research
Policy 24 (2), 185-205.
Laursen, K., Foss, N. J., 2003. New human resource management practices,
complementarities and the impact on innovation performance, Cambridge Journal of
Economics 27, 243-263.
Laursen, K., Meliciani, V., 2000. The importance of technology-based intersectoral linkages
for market share dynamics, Weltwirtschaftliches Archiv 136 (4), 702-723.
Laursen, K., Salter, A., 2004, 'Open for Innovation: The role of openness in explaining
innovation performance among UK manufacturing firms', DRUID Academy Winter
2004 PhD Conference, Aalborg, Denmark, January 22-24 2004.
Malerba, F., Orsenigo, L., 1990, Technological Regimes and Patterns of Innovation: A
Theoretical and Empirical Investigation of the Italian Case. In: Heertje, A.Perlman, M.
(Eds.), Evolving Technology and Market Structure. Michigan University Press, Ann
Arbor, pp. 283-306.
Malerba, F., Orsenigo, L., 1993. Technological regimes and firm behavior, Industrial and
Corporate Change 2, 45-74.
Malerba, F., Orsenigo, L., 1995. Schumpeterian Patterns of Innovation, Cambridge Journal of
Economics 19 (1), 47-65.
Marsili, O., 2001, The Anatomy and Evolution of Industries: Technological Change and
Industrial Dynamics. Edward Elgar, Cheltenham, UK and Northampton, MA, USA.
Marsili, O., Salter, A., 2004. The 'inequality' of innovation: skewed distributions and the
returns to innovation in Dutch manufacturing, Economics of Innovation and New
Technology forthcoming.
McGahan, A. M., 2004, How Industries Evolve: Principles for Achieving and Sustaining
Superior Performance. Harvard Business School Press.

18. 17
Meyer, M., Santos Pereira, T., Persson, O., Granstrand, O., 2004. The scientometric world of
Keith Pavitt: A tribute to his contributions to research policy and patent analysis,
Research Policy 33 (9), 1405-1417.
Miles, I., 1993. Services in the new industrial economy, Futures 25 (6), 653-672.
Milligan, G. W., Cooper, M. C., 1987. Methodology Review: clustering methods, Applied
Psychological Measurement, 11 (4), 329-354.
Milligan, G. W., Sokol, L. M., 1980. A two-stage clustering algorithm with robust recovery
characteristics, Educational and Psychological Measurement. 40, 755-759.
Miozzo, M., Soete, L., 2001. Internationalization of Services: A Technological Perspective,
Technological Forecasting and Social Change 67 (2-3), 159-185.
Nelson, R. R., Winter, S. G., 1977. In search of a useful theory of innovation, Research Policy
6 (1), 36-76.
OECD (1997). Oslo Manual - Proposed Guidelines for collecting and interpreting
Technological Innovation data. OECD, 1997. Paris.
Oerlemans, L. A. G., Meeus, M. T. H., Boekema, F. W. M., 1998. Do networks matter for
innovation? The usefulness of the Economics Network Approach in Analysing
Innovation, Tijdschrift voor economische en sociale geografie 89 (3), 298-309.
Pavitt, K., 1984. Sectoral patterns of technical change: towards a taxonomy and a theory,
Research Policy 13 (6), 343-373.
Pavitt, K., 1998. Technologies, products and organization in the innovating firm: what Adam
Smith tells us and Joseph Schumpeter doesn't, Industrial and Corporate Change 7 (3),
433-452.
Peneder, M., 2002. Intangible investment and human resources, Journal of Evolutionary
Economics 12, 107-134.
Peneder, M., 2003. Industry classifications. Aim, scope and techniques, Journal of Industry,
Competition and Trade 3 (1-2), 109-129.
Punj, G., Stewart, D. W., 1983. Cluster analysis in marketing research: review and
suggestions for application, . Journal of Marketing Research 20, 134-148.
Raymond, W., Mohnen, P., Palm, F., Schim van der Loeff, S. (2004). An Empirically-Based
Taxonomy of Dutch Manufacturing: Innovation Policy Implications. University of
Maastricht, mimeo, 2004. Maastricht.
Rogers, M., 2004. Networks, firm size and innovation, Small Business Economics 22, 141-
153.
Romijn, H., Albaladejo, M., 2002. Determinants of innovation capability in small electronics
and software firms in southeast England, Research Policy 31 (7), 1053-1067.

19. 18
Roper, S., 1997. Product innovation and small business growth: a comparison of the strategies
of German, U.K. and Irish companies, Small Business Economics 9, 523-537.
Schmoch, U., 2003. Service marks as novel innovation indicator, Research Evaluation 12 (2),
149-156.
Schumpeter, J. A., 1934, The Theory of Economic Development. Harvard Economic Studies,
Cambridge, Mass.
Schumpeter, J. A., 1942, Capitalism, Socialism and Democracy. Harper & Row, New York.
Singh, J., 1990. A typology of consumer dissatisfaction response styles, Journal of Retailing
66 (1), 57-99.
Souitaris, V., 2002. Technological trajectories as moderators of firm-level determinants of
innovations, Research Policy 31, 877-898.
Sundbo, J., 1996. The balancing of empowerment: a strategic resource based model of
organizing innovation activities in service and low-tech firms, Technovation 16 (8),
397-409.
Tidd, J., Bessant, J., Pavitt, K., 2001, Managing Innovation. John Wiley and Sons.
Tushman, M., Anderson, P., 1986. Technological discontinuities and organisational
environments, Administrative science quarterly 31, 439-465.
Utterback, J. M., Abernathy, W. J., 1975. A dynamic model of process and product
innovation, OMEGA 3 (6), 639-656.
Vossen, R. W., 1998. Relative strengths and weaknesses of small firms in innovation,
International small business journal 16 (3), 88-94.
Welsh, J., White, J., 1981. A small business is not a little big business, Harvard Business
Review (july-august), 18-32.
Winter, S. G., 1984. Schumpeterian competition in alternative technological regimes, Journal
of Economic Behavior and Organization 5 (3-4), 287-320.

20. 19
Table 1 Overview of empirical taxonomies of patterns of innovation
Author Relevant dimensions and variables Data source and
sample
Industry classification Method
Paviit (1984),
extended in
Tidd et al.
(2001)
• Sources of technology: R&D,
design, suppliers, users, public
science
• Type of user: price or quality
sensitive
• Means of appropriation: patents,
IPR, secrecy, etc.
• Objective: cost-cutting or product
design
• Nature of innovation: ratio of
product on process innovation
• Firm size
• Rate and direction of technological
diversification
• SPRU Innovation
survey
• 2,000 significant
innovations in
Great Britain
(1945-1983)
• Dominance of
large firms (53%
with more than
10,000
employees, 25%
with less than
1000)
Manufacturing and services: (1)
science-based, (2) scale
intensive, (3) specialised
suppliers, (4) supplier
dominated.
Extended in Tidd et al (2001)
with a fifth category: (5)
information intensive
• Sector-level
• Quantitative and
qualitative
analysis
Archibugi et
al (1991)
• Innovation intensity: share of
innovators; share innovation sales;
ratio of internal on external
sources of knowledge
• Nature of innovation: ratio product
on process innovation
• Sources of knowledge: design;
R&D; patents; capital embodied.
• Firm size: average size and
concentration index of innovators
• CNR-ISTAT
innovation survey
1987
• 16,700 Italian
firms, with more
than 20
employees
Manufacturing: (1) traditional
consumer goods; (2) traditional
intermediate goods; (3)
specialised intermediate goods;
(4) assembled mass-
production; (5) R&D based
• Sector-level
• Cut-off points in
ratios of industry
level indicators to
the mean across
industries
De Marchi et
al (1996)
• Innovation intensity: R&D, design,
patents
• Nature of innovation: ratio of
product on process innovation
• CNR-ISTAT
innovation survey
1987
• 16,700 Italian
firms, with more
than 20
employees
Manufacturing: Pavitt’s (1984)
taxonomy
• Sector-level
• Test of Pavitt’s
taxonomy based
on predicted
rankings across
pre-assigned
groups and
ANOVA
Malerba &
Orsenigo
(1996)
• Firm size of patenting firms
• Concentration
• Persistence of innovation
• Technological entry and exit (firms
patenting for the first or last time)
• Patent activities in
7 industrialised
countries
• Institutions and
firms excluding
individual
inventors
Manufacturing: ‘Schumpeter
Mark I’ (entrepreneurial) and
‘Schumpeter Mark II’
(routinised)
• Technology-level
• Factor analysis
and cut-off points
on factor scores
Hatzichronog
lou (1997)
• Technology intensity: intensity of
direct and indirect (embodied)
R&D
• ANBERD STAN
dataset
• Sampling of small
firms, varying
across countries
Manufacturing: (1) high tech, (2)
medium-high tech, (3) medium-
low tech, (4) low tech
• Sector-level
• Cut-off points of
technology
indicators
Arvanitis &
Hollenstein
(1998)
• Innovation intensity: inputs (R&D,
design) and outputs (innovations’
value and shares of innovative
sales)
• Sources of knowledge: other firms,
institutions, universally accessible
information and other inputs
(machinery, licences, personnel)
• Swiss KOF-ETH
innovation survey
1996
• 516 firms with
more than 5
employees
Manufacturing: 5 clusters • Firm level
• Factor analysis
and clustering
Evangelista
(2000)
• Innovation intensity: innovation
costs per employees, % innovators
• Nature of innovation: ratio of
product on process innovation
• Type of innovation inputs: R&D,
design, software, training,
• ISTAT-CNR
innovation survey
1997
• 19,000 firms with
more than 20
employees
Services: (1) technology users,
(2) S&T based, (3) interactive
and IT based; (4) technical
consultancy.
• Sector-level
• Factor analysis
and clustering

21. 20
Author Relevant dimensions and variables Data source and
sample
Industry classification Method
machinery, marketing
• Sources of information: internal
(R&D lab) and external (other
firms, institutions, etc)
• Innovation strategies: objectives of
innovation (market driven,
efficiency, etc)
Marsili
(2001)
• Technological intensity
• Technological entry barriers (share
of innovative activity in large firms)
• Persistence of innovation
• Inter-firm diversity
• Technological diversification
• Sources of knowledge
• SPRU databases
on innovative
activities of large
firms
Manufacturing: (1) science
based, (2) fundamental
processes, (3) complex
systems, (4) product
engineering, (5) continuous
processes
• Sector-level
• Qualitative and
quantitative
analysis
OECD
(2001)
• Knowledge intensity: direct and
indirect R&D expenditure; skill
levels.
• ANBERD STAN
dataset
• Sampling of small
firms, varying
across countries
Manufacturing and services: (1)
high-tech manufacturing, (2)
low-tech manufacturing, (3)
knowledge intensive services,
(4) traditional services
• Sector-level
• Cut-off points of
indicators
Peneder
(2002)
• Input intensity: labour; capital;
advertising sales ratio; R&D sales
ratio.
• Expenditure by
investment
category in US
firms
•
Manufacturing: (1) technology-
driven, (2) capital intensive, (3)
marketing driven, (4) labour
intensive, (5) mainstream
manufacturing.
• Sector-level (3
digit)
• Factor analysis
and clustering
Raymond et
al (2004)
• Model of innovative behaviour:
Estimated effects of firm-level and
industry-level characteristics on
the decision to innovate and the
returns to innovation.
• Manufacturing
firms with more
than 10
employees in the
Netherlands
• CIS-2, CIS-2.5
and CIS-3
Manufacturing: (1) high-tech,
(2) low-tech, (3) wood industry.
• Econometric
model at the firm
level with industry-
specific
coefficients

22. 21
Table 2 Distribution of innovative SMEs by industry and size class
Size class
Industry
1-9
employees
10-99
employees
Total %
Manufacturing:
Food, beverages and tobacco 51 29 80 6
Textiles, leather and paper 39 25 64 5
Wood, construction materials and furniture 45 29 74 6
Metals 48 23 71 6
Chemicals, rubber and plastic products 49 31 80 6
Machinery, motor vehicles and transport equipment 40 24 64 5
Office, electrical, communication and medical instruments 42 29 71 6
Services:
Retail and repairs 35 25 60 5
Hotels and restaurants 41 18 59 5
Personal services 35 26 61 5
Transport 35 23 58 5
Financial services 48 26 74 6
Business services (cleaning, surveillance, etc) 42 23 65 5
Wholesale 48 25 73 6
Computer and related services 51 29 80 7
Economic services (accountancy, consultancy, etc) 45 26 71 6
Engineering and architecture 48 28 76 6
Other:
Construction 34 19 53 4
Total 776 458 1,234 100
% 63 37 100

23. 22
Table 3 Variables used to develop the taxonomy of firms
Dimension Variable Description and response code±
(1) Innovative
output
Product
innovation
Mean score of two items (Cronbach’s alpha = 0.74):
1. Firm introduced any product new to the firm in the past three
years (yes/no)
2. Firm introduced any product new to the industry in the past three
years (yes/no)
Process
innovation
Firm implemented at least one new work process in the past three
years (yes/no)
(2) Innovative
input
Innovation
budget
Firm reserved an annual budget (money) to implement new products
or processes (yes/no)
Innovation
capacity
Firm reserved capacity (time) to implement new products or
processes (yes/no)
Innovation
specialists
Firm employed people who were occupied with innovation in their
daily work - e.g., specialised staff members, new product developers,
etc. (yes/no)
(3) Sources of
innovation
Suppliers Firm innovates when suppliers propose new applications (5-point
Likert scale)
Customers Firm innovates when customers express new desires/needs (5-point
Likert scale)
Scientific
development
Firm innovates to commercialise universities/knowledge institutes’
new technologies or findings (5-point Likert scale)
(4) Managerial
attitude
Innovative
orientation
Mean score of three items (Cronbach’s alpha = 0.67)
1. It is worth to spend my time on innovation (5-point Likert scale).
2. Innovation enables my firm to better serve its customers (5-point
Likert scale).
3. Innovation is needed to keep up with our competitors (5-point
Likert scale).
(5) Innovation
planning
Documented
plans
Firm had a documented plan describing renewal ambitions, targets
and milestones (yes/no)
(6) External
orientation
Consultation
of external
sources
Number of sources consulted for information or advice on any
business problem in the past three years (e.g., suppliers, colleague
firms, commercial consultants, sector organisations). Measure based
on respondents’ indication of consulted parties (min = 0, max= 6
sources or more)
Inter-firm
cooperation
Firm formally co-operated with other firms or institutes to initiate or
develop renewal activities - based on a formal agreement (yes/no).
Note:
±
Dichotomic responses are coded as ‘yes’ = 1 and ‘no’ = 0. Responses on a 5-point Likert scale are coded as
‘totally agree’ = 5, ‘agree’ = 4, ‘neither agree nor disagree’ = 3, ‘disagree’ = 2, ‘totally disagree’ =1.

24. 23
Table 4 Variables used to validate the taxonomy of firms
Variable Description and response code±
First-mover Firm is among the first to introduce new products, services or techniques
(self-rated) (yes/no)
Policy to collect new
knowledge
Firm has explicit policy to collect new knowledge (yes/no)
Use of innovation
subsidies
Firm used innovation subsidies in the past three years (yes/no)
Type of industry 18 industries (see Table 2)
Size class ‘1-9 employees’ and ‘10-99 employees’
Note:
±
Dichotomic responses are coded as ‘yes’ = 1 and ‘no’ = 0.

25. 24
Table 5 Descriptive statistics of innovative SMEs (n=1,234)
Dimension Variable Mean score
Clustering variables
Innovative output Product innovation 0.42
Process innovation 0.92
Innovative input Innovation budget 0.50
Innovation capacity 0.69
Innovation specialists 0.16
Suppliers 2.64
Customers 3.19
Sources of innovation
Scientific development 2.19
Managerial attitude Innovative orientation 4.06
Innovation planning Documented plans 0.37
Consultation of external sources
(no. of sources)
3.15External orientation
Inter-firm cooperation 0.57
Variables for validation
First-mover in innovation 0.24
Policy to collect new knowledge 0.55
Use of innovation subsidies 0.18
Note:
For the frequencies of firms by industry and size class, also used for the validation
and profiling of the taxonomy, see Table 2.

26. 25
Table 6 Profile of clusters of firms and analysis of variance tests
Cluster
Dimension Variable
Supplier
dominated
Specialised
suppliers
Science
based
Resource
intensive
F-value
Clustering variables
Product innovation 0.13 0.59 0.60 0.36 103.5**Innovative
output Process innovation±
0.93 0.84 0.96 0.94 10.3**
Presence of budgets 0.14 0.18 0.67 0.92 323.2**
Presence of capacity 0.25 0.57 0.91 0.95 238.9**
Innovative input
Innovation specialists 0.01 0.19 0.31 0.11 41.3**
Suppliers 3.33 1.91 2.47 2.83 110.7**
Customers 2.56 3.56 3.58 3.06 58.9**
Sources of
innovation
Scientific development 1.65 1.14 4.61 1.27 668.2**
Managerial
attitude
Innovative orientation 3.72 4.08 4.33 4.10 69.2**
Innovation
planning
Documented plans 0.10 0.36 0.66 0.36 81.3**
Number of sources 3.22 2.24 4.47 2.64 223.3**External
orientation Inter-firm cooperation 0.39 0.57 0.90 0.43 81.1**
Variables for validation
First-mover in
innovation (self-rated)
0.06 0.33 0.34 0.21 30.5**
Policy to collect new
knowledge
0.44 0.48 0.70 0.57 16.2**
Use of innovation
subsidies
0.07 0.15 0.39 0.11 48.6**
N obs 291 293 317 331
Notes:
** p < 0.001, * p < 0.01, ^ p < 0.05,
±
This variable was not used to construct the typology.

27. 26
Table 7 Distribution of firms across clusters within industries and size classes
Cluster
Variable N
Supplier
dominated
Specialised
suppliers
Science
based
Resource
intensive
Industry
Manufacturing:
Food, beverages and tobacco 80 28% 24% 26% 23%
Textiles, leather and paper 64 31% 22% 19% 28%
Wood, construction materials and furniture 74 24% 32% 18% 26%
Metals 70 39% 24% 19% 19%
Chemicals, rubber and plastic products 80 16% 31% 35% 18%
Machinery, motor vehicles and transport
equipment
64 22% 27% 36% 16%
Office, electrical, communication and medical
instruments
71 21% 21% 38% 20%
Services:
Retail and repairs 60 32% 17% 22% 30%
Hotels and restaurants 59 32% 17% 10% 41%
Personal services 61 25% 10% 15% 51%
Transport 58 41% 17% 16% 26%
Financial services 74 20% 28% 27% 24%
Business services (cleaning, surveillance etc) 65 14% 28% 29% 29%
Wholesale 73 18% 33% 22% 27%
Computer and related services 79 8% 34% 24% 34%
Economic services (accountancy, consultancy) 71 15% 24% 41% 20%
Engineering and architecture 76 13% 17% 41% 29%
Other:
Construction 53 40% 11% 17% 32%
Totals:
Manufacturing 503 26% 26% 27% 21%
Services 676 21% 23% 25% 31%
Size class:
1-9 employees 774 26% 26% 19% 29%
10-99 employees 458 20% 20% 37% 23%
Total 1232 24% 24% 26% 27%