DRUID Working Paper No. 09-01
Localised Spillovers and Knowledge Flows:
A Study on the Effects of Proximity and Labour Mobility on Plant
Performance
By
Rikard H. Eriksson
www.druid.dk
Localised Spillovers and Knowledge Flows: A Study on the Effects of
Proximity and Labour Mobility on Plant Performance
Rikard H. Eriksson
Department of Social and Economic Geography
Umeå University
Sweden
E-mail: [email protected]
Abstract:
This paper aims to shed some light on the influence of geographical proximity on both
intra- and inter-industry spillovers by elaborating on the geographical dimension of both
localised spillovers and inter-firm knowledge flows. By means of a unique longitudinal
micro-database with information on all plants and employees in Sweden, both plantspecific agglomeration measurements and labour markets at various distances from each
of the 8,313 plants in the sample were created. OLS-regressions were run to account for
what type of co-located activities that is most beneficial to productivity growth of plants
between 2001 and 2003; how different types of knowledge flows – in and out from the
plant – affect performance, and finally; how geographical proximity influences the effects of
both localised spillovers and knowledge flows. The empirical results indicate that it is not
possible to establish whether either intra- or inter-industry spillovers are most beneficial
unless the geographical dimension is considered. This is because neither too much nor
too little proximity (measured as both geographical and cognitive proximity) between colocated activities is likely to produce significant localised spillovers. This seems also to be
the case when assessing more directly the impacts of inter-plant knowledge flows via
labour mobility – only knowledge flows that are complementary to the existing knowledge
base of plants, and neither characterised by too much nor too little geographical proximity,
affect plant performance positively. Concerning the outflows of skills, the results indicate
that it is less harmful for the dispatching plant if the former employee remains within the
local milieu as compared to leaving for a job in another part of the economy.
Keywords: Agglomerations, Knowledge Spillovers, Labour Mobility, Plant Performance,
Geographical Proximity, Related Variety
Jel codes: R11, R12, O18
ISBN 978- 87-7873-280-4
www.druid.dk
Acknowledgements
Financial support was granted from The Swedish Governmental Agency for Innovation
Systems’ (VINNOVA) Dynamo II-program. The author is grateful for the valuable
comments made by Guido Buenstorf and Bram Timmermans on an earlier draft presented
at the 2009 DRUID-DIME Academy Winter PhD Conference in Aalborg. The usual
disclaimer applies.
www.druid.dk
1 Introduction
In the literature discussing the effects of either regional specialisation (localisation) or
regional diversification (Jacob’s externalities) on knowledge spillovers and regional
growth, the need for also differentiating between related and unrelated variety is
increasingly recognised. This is because Jacob’s externalities are not likely to
automatically
produce
significant
knowledge
spillovers
unless
there
are
complementarities between sectors. Therefore, it is essential to separate between
related and unrelated variety (Boschma and Iammarino, 2009; Frenken et al, 2007).
The empirical support for the importance of intra or inter industry spillovers is
somewhat mixed, which partly is due to a potential miss-specification of the local
dimension of agglomerations in previous studies. Frequently, the analyses rely on
regional aggregates which disregard that (i) knowledge spillovers via face-to-face
contacts are localised to a much larger extent than envisioned, and (ii) plants and
regions are not are isolated entities in the spatial economy which makes it essential
to also consider the combined effect of inward and outward linkages both within and
outside the local milieu. Hence, by adhering to the literature emphasizing the
significance of extra-local linkages (Bathelt et al, 2004), this paper contributes to the
discussion on knowledge spillovers in agglomerations by extending the arguments in
two ways.
Although there is a great consensus about that geographical proximity
facilitates knowledge spillovers and increase the performance of plants in the
agglomeration literature (Glaeser et al, 1992; Henderson et al, 1995; Porter, 2000),
there are various interpretations of the actual spatiality of spillovers resulting in
disparate and quite modest empirical findings (Rosenthal and Strange, 2004). One
explanation may be attributed to the fact that previous empirical studies tend to be
spatially biased towards using pre-defined regionalisations or choosing spatial scales
on the basis of more or less anecdotal assumptions. Thus, proximity is treated as
fixed definitions and not as a specific characteristic of agglomerations (Martin and
Sunley, 2003; Malmberg, 1996; Phelps, 2004; Oerlemans and Meeus, 2005). Even
though recent empirical studies address the spatial bias of agglomeration
externalities (e.g. Burger et al, 2007), these studies remain reliant upon that
1
externalities work within administrative borders which rarely is the case in real life
where externalities can be more or less geographically dispersed (Van Soest et al,
2006; Anselin, et al, 1997). The need for overcoming this conceptual bias and stop
relying on aggregates bounded within administrative borders is particularly crucial
according to the knowledge spillover literature. It is recognized that knowledge is
primarily transferred and utilised within very close surroundings of the plant (e.g.
Jaffe et al 1993), and that learning often requires frequent face-to-face interactions
(Storper and Venables, 2004). The tendency of regular interactions to be highly
localised implies that the daily arena of work is primarily concentrated to close circles
of the workplace. As a consequence, the local dimension of agglomerations cannot
be defined on the basis of regional aggregates, but should rather be perceived as
plant-specific possibilities for labour to regularly interact.
By departing from the literature advocating the importance of related variety,
this paper explores how geographical proximity influences the impact of similar (i.e.
intra-industry spillovers), complementary (i.e. related variety) or very different (i.e.
unrelated
variety)
activities
by
ascribing
each
plant
unique
plant-specific
neighbourhoods and relaxing the influence of administrative borders. We expect that
knowledge spillovers will be most effective if a high degree of complementarity is
present in contrast to very similar or very different activities. Additionally,
geographical proximity is expected to increase the risks of lock-in effects due to high
degrees of similar activities. A spatial context characterised by unrelated variety can
be beneficial if it is combined with geographical proximity since close neighbours are
likely to share the same place-specific corporate cultures that reduce communication
problems.
The notion of the economic effects of related variety is also applied when
addressing the impacts of inter-firm linkages. This is done by drawing on
contributions that advocate the role of labour mobility for local knowledge formation
and the competitiveness of clustered firms (e.g. Pinch and Henry, 1999; Almeida and
Kogut, 1999; Power and Lundmark, 2004). However, since economic activities tend
to cluster in certain parts of the spatial economy, and moreover, processes of job
search largely are dependent on face-to-face contacts and informal networks
constructed via such contacts (Granovetter, 1995), one major drawback of previous
studies is the disregard of different sub-regional knowledge flows that may be driven
by face-to-face contacts. This is despite the indications of empirical studies which
2
show that flows of skilled labour are often highly localised and selective (Power and
Lundmark, 2004; Eriksson and Lindgren, 2009). Previous empirical studies on this
topic tend to discriminate between the potential and the actual local labour market,
whereas the current study provides insights into the spatiality of knowledge flows by
creating plant-specific labour markets. By doing so, the research sets out to explore
whether it is more beneficial to recruit new skills across the street, from another part
of the urban area or from elsewhere in the national economy. Following the notions
put forward by Boschma and Iammarino (2009) and Boschma et al (2009), we argue
that it is not just a matter of having linkages with the outside world, but to what extent
these linkages bring new knowledge that is complementary to the existing knowledge
base. We anticipate that close geographical proximity increases the need for
recruiting related skills since labour previously working for firms in the close vicinity
likely brings similar experiences to the new firm. The new firm can absorb this
knowledge, but the new skills will not add anything to the existing stock of skills
which, in turn, may increase the risk of lock-in if it is not combined with other types of
place-specific knowledge acquired farther away. On the other hand, for inflows of
unrelated skills, geographical proximity is needed since increased distance is
expected to worsen the problems of communication when combining skills originating
from different sectors. Moreover, is it the case that recruited skills merely substitute
skills leaving the plant and therefore not actually add to the existing knowledge base?
In order to control for this potential substitutive effect, outflows of skills to other parts
of the economy are also accounted for.
Hence, the current paper adds to the above theoretical discussion in two
ways. First, it contributes by more thoroughly accounting for the ways geographical
proximity influences knowledge spillovers. This is not carried out not via spillovers at
the regional level, but through the creation of plant-specific agglomeration
measurements that estimate localised spillovers for every single plant in the
economy. Second, the paper contributes to the agglomeration literature by exploring
the combined effects of both inward and outward knowledge flows, which is carried
out by elaborating on inter-plant distances in more detail. These theoretical
propositions will be tested by estimating productivity growth of 8.313 plants in
Sweden between the years 2001 and 2003. By means of a unique micro-database
that connects attributes of individuals (e.g. education and working experience) to
characteristics of plants (e.g. spatial coordinates, sector and productivity) it is
3
possible to create measurements of plant-specific agglomeration externalities based
on the potential range for workers to make face-to-face contacts.
The paper is structured as follows: The main theoretical ideas are introduced
in the next section, which is followed by a presentation of data and research design
in section three. In section four the empirical results are presented and, finally, some
concluding remarks and suggestions for further research are provided in the last
section.
2. Proximity and agglomerations
Economic geographers have for a long time acknowledged that the capability to
innovate and secure continued economic success is not only related to plant-internal
characteristics, but also to attributes external to the plant. This argument can be
summarised by the concept of agglomeration economies, i.e. reduced expenditures
enjoyed by being located nearby other plants. Traditionally (e.g. Weber, 1929;
Hoover, 1937), these cost reductions are tradable outcomes from co-location since
geographical proximity reduces costs of transportation, communication, and
information exchange as well as it sustains a pool of skilled labour available for all
firms in the agglomeration. More recently, scholars have recognised the social and
cultural aspects of co-location. Geographical proximity is in this literature regarded as
an essential feature for the transmission of tacit knowledge between economic
agents via intense face-to-face interactions and shortened social and cognitive
distance. Moreover, proximity is a factor aiding the establishment of trustful relations
between economic agents, increasing the ability to monitor rival firms and facilitating
knowledge flows via labour mobility (Boschma, 2005; Gertler, 2003; Maskell and
Malmberg, 1999; Storper; 1997; Storper and Venables 2004; Porter, 2000; Bathelt et
al, 2004). Such external factors are often referred to as knowledge spillovers and
relate to the economies plants may benefit from when knowledge diffuses between
plants in the local economy. However, the establishment of the relative importance of
these externalities in terms of facilitating knowledge spillovers is strongly contested
(e.g. Glaeser et al, 1992; Henderson et al, 1995). The debate concerns whether
externalities are generated from the benefits of co-located complementary and
competitive plants within the same industry (localisation or MAR-externalities i ) or if
i
This was first mentioned by Marshall (1890) and later on developed by Arrow (1962) and Romer 1896).
Therefore localisation economies are often referred to as MAR-externalities.
4
the externalities are generated through cross-fertilisation of different ideas between
co-located firms in different industries (Jacob’s externalities). Recent contributions,
however, argue that it is not sufficient to only distinguish between intra- and interindustry spillovers.
The empirical section is therefore based upon the literature advocating the
impacts of related variety on economic growth (Frenken et al, 2007; Boschma and
Iammarino, 2009). By doing so we open up for the possibility to both consider the
more qualitative characteristics of agglomeration externalities (in terms of how the
composition of sectors influences spillovers) and the conceptualisation of how
agglomeration externalities may be shaped by geographical proximity. This literature
stresses the importance of distinguishing between intra- and inter-industry spillovers,
but also more thoroughly conceptualising the impacts of diversification. Jacob’s
externalities is likely to cover two different types of agglomeration externalities –
spillover effects stemming from complementarities between sectors and, in times of
demand shocks, risk-reducing portfolio effects stemming from the presence of a
plethora of different activities.
It is in this respect essential to make a distinction between related and
unrelated variety. Whereas unrelated variety is unlikely to produce significant
knowledge spillovers due to problems of communication, plants embedded in much
related variety are more likely to benefit from knowledge spillovers due to a greater
deal of complementarities. Nooteboom et al (2007) empirically support this notion by
showing evidence of an inverted U-shape function connected to the cognitive
distance between technology-based partners and innovative performance. They find
that effective interactive learning and innovation is best facilitated when existing
knowledge is combined with new complementary knowledge that is neither too
similar nor too unrelated. By also considering the effects of intra-industry spillovers,
their study investigates whether high degrees of similarity (i.e. high degrees of
surrounding plants within exactly the same industry), related variety (i.e. surrounding
plants in complementary industries) or unrelated variety (i.e. surrounding plants
within very different industries) enhance plant performance the most.
Nevertheless, irrespective of spillovers are best facilitated within the same
industry or between related/unrelated industries, the main argument of this paper is
to scrutinize how geographical proximity influences the impact of external economies.
In the literature, it is increasingly recognised that face-to-face contacts is a key issue
5
for co-location and close relations, since face-to-face interactions facilitate
socialisation and communication. It also helps solving joint problems, increasing
competition and facilitating the transfer of embodied knowledge (Storper and
Venables, 2004). By following the original contribution by Marshall (1890), the MARliterature stresses that industry-specific knowledge will be as readily available for all
co-located plants within the same industry. The literature frequently treats proximity
synonymously to co-location (within an urban area or an administrative region), and
firms located in such sites are assumed to benefit from ‘a local buzz’ generated by a
concentration of specialised labour interacting on a regular basis. For co-located
firms within similar industries this will give an opportunity to certain surreptitious
activities which may help them to stay in business and benefit from further knowledge
creation. This is also highlighted in the literature on Jacob’s externalities. However,
instead of advocating the importance of knowledge available to all co-located firms
belonging to the same industry, this literature stresses that diversity in general is
more beneficial for knowledge spillovers. A concentration of a diverse set of workers
and human capital will facilitate knowledge spillovers because, for example, a variety
of different – but complementary – industries and occupations have much more open
networks allowing the (re-)combination of different ideas by personal interaction
(Glaeser et al, 1992; Boschma and Iammarino, 2009). Since the probability of interpersonal interaction correlates positively with population density, learning is assumed
to be more intense in larger cities (Glaeser, 1999). Thus, both approaches are
convinced that face-to-face contacts rooted in geographically proximate relations are
crucial for effective knowledge spillovers to take place. However, little attention is
paid to how the degree of proximity may influence the different types of externalities
or how different plants may – or may not – be affected by their co-located
neighbours. On the contrary, previous empirical studies tend to neglect the interplay
between the plant and its geographical situation. They appear to be more interested
in mapping the presence of spillovers, regardless of industrial affiliation and the
specific location of the plant.
Considering
that
many
of
the
notions
on
externalities
derived
by
agglomerations highlight the role played by geographical proximity, one could expect
that the question on how geographical proximity influence localised spillovers is
already thoroughly answered. However, following the discussion put forward by
6
Boschma (2005) and others ii , there is a tendency in previous studies to either be
somewhat vague on the geographical dimension or to take the impacts of
geographical proximity for granted. By focussing on successful regions and
successful industries, quite often based on more or less anecdotal assumptions (c.f.
Malmberg, 1996), geographical proximity is often indirectly specified as being located
on either side of an administrative border neglecting the fact that agglomerations are
more likely to work across or only partially within conventional administrative borders
(Scott, 1982; Anselin et al, 1997; Van Soest et al, 2006).
This critique is also directed towards the numerous studies on localised
clusters presented since the early 1990s (Martin and Sunley, 2003). When defining
‘the local milieu’ in agglomerations or clusters, scholars often use pre-specified
administrative units such as zip-codes, municipalities, counties or functional regions
(e.g. local labour markets). These regionalisations have only moderate connections
to the actual spatial range of externalities and they lack any explicit focus on the local
dimension of such industrial systems. Rosenthal and Strange (2004) emphasize that
this might account for the quite mixed results regarding the localised dimension of
externalities. For instance, Dekle and Eaton (1999) find that agglomeration effects
diffuse nationwide, whereas Ciccone and Hall (1996) – by using US-states as
analytical unit – implicitly assume that agglomeration effects on labour productivity
basically work within regional borders. Rosenthal and Strange (2003) as well as Van
Soest et al (2006) report that spillovers mainly operate within zip-codes areas and
that they show a steep distance decay function. The latter result indicating that
externalities have a much more localised dimension and cannot be assessed
properly by using administrative regions as primary analytical unit is especially
highlighted in the empirical work on knowledge spillovers. For example, Jaffe et al.
(1993) show that the geographical patterns of patent citations very much coincide
with the patents’ place of origin. Zucker and Darby (1996) even argue that the
embodied knowledge of scientists is only available at the workplace level and that
this knowledge is costly for other firms to acquire. In the light of these contributions it
seems reasonable to suggest a shift in focus from administrative regions to the plant
level for the continued analysis of externalities. Regardless of the impacts of similar-,
related or unrelated externalities on plant performance, the spatial scale of such
ii
Also see special issue of Regional Studies (2005, 39:1)
7
externalities should be defined based on where skilled workers are expected to make
most informal face-to-face interactions, that is at their workplace and its vicinity.
If it is assumed that knowledge is spatially sticky and almost certainly requires
face-to-face contacts to be effectively transferred, the local milieu should be defined
as plant-specific milieus corresponding to the potential range of interacting labour.
Moreover, according to findings presented by Essletzbichler and Rigby (2005) and
Rigby and Essletzbichler (2006), routines of firms within a sector tend to be more
similar if they are located nearby as compared to geographically separated locations.
It is implicitly argued that plants are influenced to a relatively larger extent by their
nearby neighbours concerning corporate routines and the like. Hence, since different
types of agglomerations often co-exist within a regional economy, creating ‘sticky
mixes’ (Markusen, 1996), each plant must therefore be attributed its own unique
surrounding For example, if a labour intensive automobile production is located in a
region, the regional aggregate is likely to display high specialisation scores for this
sector. Such an aggregate specialisation probably conceals the presence of other,
unrelated, top-class micro-clusters in in the region.. This bias is probably stronger in
large regions where the combination of many local clusters in different sectors is
likely to produce a relatively diversified whole. This issue will here be addressed by
elaborating on different distances from each plant; from within sight to the maximum
preferred distance of daily trips to work. Hence, by applying the idea suggested by
Boschma (2005) where each plant is ascribed its unique surrounding at various
distances, we expect that geographical proximity will increase the risk of cognitive
lock-in associated with being co-located with very similar industries. Similarly, we
expect geographical proximity to reduce potential problems associated with being
located in a diverse economic setting. For significant spillovers to take place, we
believe that high degrees of related activities must be present in very close relation to
the plant.
These ideas are also applied when accounting for the impact of knowledge
flows between plants. A general assumption put forward in the literature is that
mobility of skilled personnel is crucial for the transfer of embodied knowledge and for
the sustained competiveness of clustered activities (Almeida and Kogut, 1999;
Cooper, 2001; Maskell and Malmberg, 1999; Rodriguez-Pose and Vilalta-Bufi, 2005;
Gertler, 2003; Power and Lundmark, 2004; Pinch and Henry, 1999). Previous studies
nevertheless tend to downplay the impact of more proximate flows than those
8
bounded within functional regions. Yet this is an important aspect since labour is
widely acknowledged to be the most immobile factor of production (Storper and
Walker, 1989), and a majority of all job moves seldom takes place between different
local labour markets (Fischer et al, 1998). Hence, by bundling the majority of all
labour flows into one single intra-regional category, the real features of local
knowledge flows will not be elucidated. This is mainly due to two particular
dimensions of job mobility.
First, according to the imperfect labour market approach (e.g. Storper and
Walker, 1989; Peck, 1996; Doeringer and Piore, 1971), a local labour market
contains a number of different non-competing submarkets which are separated from
each other by inherent barriers and structural disequilibrium characterised by lack of
substitutability between segments. Martin (2000), for instance, argues that each submarket is having its own geography, with its own employment and wage processes
and its own specific modes of social regulation. Since a majority of all job moves are
local more detailed analyses within regions must be conducted in order to assess the
impacts of intra-regional knowledge flows.
Second, a high concentration of workers is considered to be a strong
contributor to both localisation and urbanisation by giving access to a pool of skilled
labour which is likely to facilitate local circulation of knowledgeable individuals (Pinch
and Henry, 1999). Since job mobility largely is a product of informal networks and
face-to-face contacts (Granovetter, 1995), densely concentrated economic activities
are more likely to benefit from such information flows. These two notions are
confirmed in studies on the role of labour mobility in agglomerations. While Eriksson
et al (2008) show that regional concentrations of similar and related activities
significantly increase the probability of changing jobs within the same region,
Eriksson and Lindgren (2009) demonstrate that only a minor share of all plants within
a local labour market are intertwined by flows of people changing jobs. Power and
Lundmark (2004) find that only some parts of the local labour market are embedded
with dynamism by studying local job mobility of key-personnel within the spatially
concentrated Stockholm ICT-cluster. Their analysis shows that local job mobility
within the cluster is significantly higher than in the rest of the urban economy. This is
confirmed by Bienkowska (2007) who argues that clustered industries within more
local parts of urban economies facilitate the possibility for the firms to both screen
and recruit skilled personnel. According to Waxell (2005) this is because job mobility
9
between firms located within the same metropolitan area does not cause as high
social costs as changing job to another part of the same labour market. Accordingly,
in parallel with the need to account for plant-specific externalities when addressing
plant performance, it is not reasonable to assume that there is only one local labour
market that affects knowledge flows for all plants located in the region. It is rather
more pertinent to assume that each plant has its own local labour market
characterised by factors like industrial affiliation, location and industrial affiliation of
neighbouring plants. Altogether, these patterns are likely to influence the potentials to
acquire new knowledge and to generate plant-specific geographies of knowledge
flows.
However, a too strong reliance upon local knowledge, either via spillovers or
via labour flows, may trigger lock-in effects that are potentially harmful for firms’ longterm performance (Bresnahan et al, 2001; Asheim and Isaksen, 2002; Bathelt et al,
2004; Faggian and McCann, 2006). In order to improve our understanding of the
economic effects of knowledge flows, we therefore argue that the impacts of more
distant inflows as well as the skills leaving the plant must be accounted for. The
ability to absorb and utilise extra-local knowledge may compensate for relative
shortcomings within the local milieu. Similar to the findings in Boschma et al (2009),
we expect that inflows characterised by geographical proximity must be related to
avoid problems of lock-in because such inflows are complementary to the existing
knowledge base of the plant, and thus potentially beneficial to performance. This will
be particularly evident when also accounting for skills leaving the plant. For example,
plants with high turnover rates have to fill the vacancies of skills leaving the plant. For
them it is especially crucial to recruit related skills, since complementary knowledge
can be absorbed and utilised more easily than other types of skills. However, by
accounting for the distance of labour flows and for skills leaving plants we expect our
empirical findings to differ from previous findings. Whereas inflows of geographical
proximate skills similar to the inhouse portfolio are expected to deepen the problems
of lock-in, long-distance recruitment of such skills should be less harmful for firm
performance. Hence, geographical distance is therefore assumed to reduce the need
to recruit employees with related skills because inflow of similar skills from far away is
supposed to bring valuable knowledge. Moreover, the more unrelated recruitment,
the more need for geographical proximity to solve problems of communication and
coordination at the plant. Inflows of unrelated skills might therefore still contribute to
10
firm performance, as long as they are recruited locally. This effect fades away as
distance increases. Before empirically validating these propositions, the next section
describes data and statistical models used in the analysis.
3. Research design
The above theoretical propositions are tested by analysing changes in productivity
among 8.313 plants within the entire Swedish economy. We use data retrieved from
a unique longitudinal micro-database, ASTRID, which is a compilation of several
administrative registers at Statistics Sweden and contains annual information of all
people, firms and workplaces. In ASTRID, attributes of individuals (working
experience) is connected to features of plants (spatial coordinates at hectare squares
and sector). The high resolution of the socio-economic data makes it possible to both
create plant-specific neighbourhoods as well as analysing flows of employees at
various distances. Before turning to the variables used in the analyses, some notes
on the sampling procedure need to discussed.
At the initial stage, all workplaces with information about industrial affiliation
and performance indicators were selected (256,985). From this population only
workplaces with inflows of skilled labour were selected, which resulted in a sample of
17,098 workplaces.
The database contains no variable directly indicating job mobility. Therefore
this variable had to be created based on a number of conditions. In order to confirm
that the selection only measures the impacts of job movers established on the labour
market and working full-time, individuals had to meet the following income and age
criteria. Job movers had to (i) earn more than USD 20,500 annually (200,000
Swedish crowns in 2001 monetary values), (ii) aged 25 to 64 , and (iii) must have a
registered change in both workplace identity and workplace coordinates (hectare
grid) between the years 2000 and 2001. The first and second conditions are set to
exclude part-time workers (e.g. students) or people not yet established on the labour
market. The third condition is set so as to check for that an actual job move has
taken place. Due to the widespread idea that knowledge transfers between
workplaces mainly is the result of mobility of key persons (e.g. Power and Lundmark,
2004), a fourth criterion is added: (iv) individuals must have obtained at least a
bachelor degree or belong to the top 20 percent income earners. Two indicators for
key persons are used since these persons do not automatically have a higher
11
academic training. A final selection was made to include only manufacturing units
and knowledge-intensive service sectors (e.g. financial services) because knowledge
spillovers are assumed to be strongest in these sectors (e.g. Frenken et al, 2007). By
only modelling the performance of workplaces within manufacturing and knowledgeintensive service sectors with skilled inflows, we end up with a final sample of 8,313
workplaces.
After having selected workplaces and job moves to be included in the
analysis, the next step is to define plant-specific neighbourhoods. Although there are
no obvious ways of defining geographical proximity because knowledge spillovers
and externalities derived from market-size are likely to work at different spatial scales
(Martin, 1999), the analysis applies three different distances similar to the ones used
by Li et al. (2009). The areas include from every single plant economic activities
within radii of 0.5, 5 and 50 kilometres, respectively. The closest range is defined to
cover the location of economic activities within the same business park or urban
district. Each plant and its employees are likely to be aware of the whereabouts of
other economic activities, since it is practically within arms-length and within sight
from the workplace. Employees are therefore likely to cross each others paths at
several occasions during a working week and firms in related industries are able to
monitor each other. Moreover, according to Fischer et al. (1998) about 50 percent of
all mobility in Sweden is shorter than 5 kilometres, whereas 83 percent is shorter
than 50 kilometres. This insight makes it reasonable to define the two other plantspecific surroundings accordingly, as they are likely to reflect both the local labour
market of each plant and the potential for regular face-to-face interactions. For
instance, there is no major problem to regularly travel 5 kilometres to another urban
district for business meetings and this distance is also likely to cover many of the
small- and medium-sized urban regions in Sweden. Hence, this distance should not
cause any major hindrance for effective collaboration and socialisation. Finally, a
plant is less likely to actually have close relations to other plants up to 50 kilometres
away since daily interactions are more difficult to maintain. In addition, in order to
asses the effects of relaxing administrative borders more accurately, agglomeration
measurements were also calculated for each municipality (n=290, average
size=1.420 km2). Most economic activities are institutionalised in line with this
administrative division.
12
3.1 Dependent variable: Labour productivity
The effects of labour mobility will most likely not materialize instantly at the plant,
which justifies a dynamic measure of firm performance. Therefore we use changes in
labour productivity between 2001 and 2003 as a measure of performance. Since the
database does not carry information on employees’ hours of work, labour productivity
has been defined as value added per employee iii . However, value added in our
dataset is reported for firms and not workplaces. For circa 25 percent of the firms in
this sample, with more than one plant, value added was distributed to the workplaces
in the same proportion as the distribution of the sum of wages across workplaces
(Wictorin, 2007). Thereafter, the calculated sum of value added was divided by the
number of employees of the workplace. This procedure potentially takes both
education and experience into account when measuring labour productivity at the
plant level. This aspect would be neglected if only distributing value added according
to the workplace’s share of firm employees. Finally, by using the unique identification
number associated to each workplace making it possible to follow workplaces over
time, the level of productivity in 2001 was subtracted from the level in 2003. This
calculation renders it possible to measure growth. All numbers were adjusted to the
2001 price levels. In the model, log values are used to reduce the impact of skewed
distributions.
3.2 Independent variables
In order to asses the impacts of being co-located nearby similar, related or very
different industries, all independent variables are measured at the beginning of the
period (i.e. 2001) and based on the entire population of plants within all sectors.
When estimating the effects agglomeration externalities on firm performance, entropy
measurements similar to Frenken et al (2007) and Boschma and Iammarino (2009)
have been used. Briefly, this calculus compares the industrial affiliation (SNI-codes)
of all plants with the sectoral belonging of all other plants within their respective
neighbourhoods. The SNI-nomenclature consists of 753 different five-digit categories
iii
To control for part-time work and increased efficiency which would have been possible with information on
hours of work, a proxy controlling for this was created. It held information on the per capita social benefits
received of all employees at each workplace (including parental leave, unemployment insurances and sickleave), which implicitly account for the relative share of absence from work during 2001 (Eriksson and
Lindgren, 2009). This variable did not affect the estimates and was omitted from the final model.
13
which are nested within 224 different three-digit categories and 10 different one-digit
categories. To begin with, three measurements on the average regional composition
of sectors within municipalities were calculated in order to address the impact of
regional externalities before calculating the plant-specific externalities surrounding
each plant.
First, the effect of externalities at the regional level was calculated by
employing the entropy measurements in Frenken et al (2007) and Boschma and
Iammarino (2009) to capture the degree of similarity, relatedness and unrelatedness
within the Swedish municipalities. The degree of similarity (i.e. the degree of
specialisation within each region) was measured by the inverted entropy at the fivedigit industrial level. Let piV be the share of plants within five-digit industry i and let N V
be the number of five-digit classes. The similarity measurement is then calculated as:
NV
SIMILARITY = 1 /
∑p
V
i
i =1
⎛ 1 ⎞
⎟
log 2 ⎜
⎜ V ⎟
p
⎝ i ⎠
(1)
where the higher the score, the higher the concentration of similar industries within
regions.
After constructing a measure for intra-industry spillovers, the effect of being
surrounded by complementary activities is addressed. As noted by Frenken et al
(2007),
performance
is
assumed
to
be
supported
by
concentrations
in
complementary sectors since this facilitates knowledge to spill over between plants
within the regional economy. This is measured by calculating the weighted sum of
entropy at the five-digit level within each three-digit industrial category iv . Thus, the
more variety within the three-digit level, the better for firm performance due to a
higher degree of complementarities between plants. The degree of related variety is
calculated as follows: All five-digit sectors piV belong to a three-digit category S III
j ,
where j=1,…,NIII. Therefore, we can derive the three-digit shares p III
by summing the
j
shares of all five-digit sectors nested within S III
j :
p
III
j
=
∑
i∈ S
p iV
(2)
III
j
Related variety is then defined as the weighted sum of entropy within each three-digit
industry category, given by:
iv
It should be noted that also the entropy within each two-digit category has been estimated but that did not
change the effect of related variety in neither of the models.
14
N
RELATED
VARIETY
=
III
∑
p
III
j
H
j
(3)
j =1
where:
H
j
=
∑
i∈ S
III
j
p iV
p III
j
⎛
1
⎜
log 2 ⎜
V
⎜ p i / p III
j
⎝
⎞
⎟
⎟
⎟
⎠
(4)
Finally, a variable capturing the degree of unrelated activities within each
region is calculated in order to assess how relative regional diversification affects the
performance of plants. This variable is measured as the entropy at the one-digit level
where a high variety at the one-digit level symbolises that a region is characterised
by very different activities v .
plI
Let
be the share of one-digit industrial
sector S lI ∈ S1I ,..., S NI I . We now get:
NI
UNRELATED VARIETY =
∑
l =1
⎛ 1 ⎞
⎟
p lI log 2 ⎜
⎜ I ⎟
⎝ pl ⎠
(5)
where, the higher the score, the more diversified the region.
However, as put forward in section two, the use of regional aggregates for
focussing on the geography of knowledge spillovers may create biased estimates
since regional aggregates probably conceal the effects of different micro clusters
within regions and do not consider the plant-specific geography. In order to address
the potential impact of geographical proximity, each plant is ascribed a unique plantspecific measurement of similarity, related variety and unrelated variety, repeatedly
calculated for each of the distances. This is done for capturing the different types of
externalities surrounding each plant. By adopting modified entropy measurements
considering both the plant-specific industry and all the other co-located activities, it is
possible to determine how similar, related or unrelated the local environment is in
relation to every single plant. The degree of similarity (i.e. the degree of surrounding
plants with exactly the same SNI-code) was calculated as follows (where p ipV is the
share of neighbouring plants belonging to exactly the same five-digit industry I
and p uI the total share of plants belonging to the same 1-digit category):
v
Because of the decomposable nature of the entropy measure differentiating variety at various digit levels, this
variable should not be interpreted as the inverse of the Similarity variable (see Frenken, 2007 for more details).
15
⎞
⎛
⎜ 1 ⎟
V
SIMILARITY = ( p uI * p ip
) * log 2 ⎜
V ⎟
⎟
⎜ p ip
⎠
⎝
(6)
Plants scoring high on this measurement are located in an environment with many
identical industries and are, according to the MAR-literature, assumed to benefit from
localised spillovers due to a high level of available-for-all intra-industry knowledge. As
hypothesised in section two we, however, expect that this effect will only be prevalent
within a certain distance from the plant due to the risk of cognitive lock-in associated
with too much proximity (both geographical and cognitive).
In the following stage, the effect on plant performance of being surrounded by
complementary activities is assessed. Knowledge is assumed to spill over more
easily between co-located plants and generate enhanced firm performance as the
complementarity of the local milieu increases. As noted in section two, we expect that
a plant in automobile production will be able to absorb and utilise knowledge spilling
over from co-located plants specialised in supporting or complementary activities but
not from a manufacturer specialised in, for instance, pulp-production because the
cognitive distance between them is too far. This is measured for each plant by
calculating the entropy at the five-digit level within its three-digit industrial category.
The higher the variety within each three-digit category and the more complementarity
in the surrounding milieu, the more beneficial it is assumed to be for the plant
because it expected that plants with such surroundings more easily can benefit from
localised
spillovers.
This
measurement
does
not
only
reveal
that
more
complementary sectors as compared to similar sectors are located nearby, it also
relates to the industrial affiliation of all other co-located activities. The degree of
related variety is defined as the entropy within each three-digit industry category p ipIII ,
except for those plants belonging to exactly the same industry (i.e. similar industries:
V
),
p ip
by also considering the total share of industries within both the same 1-digit
category ( p uI ) and the same 3-digit category, which is given by:
RELATED
VARIETY
= ( p uI * p ipIII ) * H
ip
(7)
where:
H ip =
∑
i∈ S III
j
V
p ip
p ipIII
⎛
1
⎜
log 2 ⎜
V
⎜ p ip / p ipIII
⎝
⎞
⎟
⎟
⎟
⎠
(8)
16
We expect that the impacts of related variety will be occurring particularly within short
distance from each plant. This is because proximity is likely to facilitate the recombination of complementary knowledge without the risks of lock-in effects due to
cognitive proximity and similar activities.
Finally, if a plant-specific surrounding is characterised by very different types
of industries it is assumed that the plant will not be able to benefit from knowledge
spillovers due to communication problems. Geographical proximity is assumed to
reduce such problems and therefore we anticipate that increasing distance from the
plant will make it more difficult to sustain effective communications (e.g. greater
dissimilarities in place-specific routines). This variable is measured as the entropy at
the one-digit level where a high value represents a situation of being surrounded by
very different activities. Let p Ijp be the share of co-located plants belonging to the
same one-digit industry code but another three-digit sector than the specific plant.
The measurement for unrelated variety is then defined in relation to the share of colocated plants in all other one-digit categories p uI . The higher the score, the more
difference to the other co-located activities. This is in comparison to both the variation
within the same 1-digit category as well as the activities in other 1-digit categories:
UNRELATED VARIETY
= 1 /( p uI
*
p Ijp
⎛
⎞
⎜ 1 ⎟
) * log 2 ⎜
⎟
⎜ p Ijp ⎟
⎝
⎠
(10)
As mentioned in section two, different local- and extra local linkages may
compensate for relative shortcomings in the local milieu. Old knowledge may become
obsolete which calls for new knowledge, either via local or extra-local sources. Since
labour mobility is assumed to facilitate the transfer of knowledge between economic
agents, variables measuring similarity, relatedness and unrelatedness of labour
inflows were also created to analyse the impacts of inter-plant linkages. However,
ASTRID only provides information concerning the main output for each workplace,
which implies that only one single five-digit sector code is available. This means that
it is not possible to use entropy measures when estimating inflows at the workplace
level. Nevertheless, by comparing the background of new employees and
summarising the total number of different types of inflows, it is possible to obtain
information on how different extra-firm linkages affect plant performance. The degree
of similar inflows is measured as the total number of inflows originating from the
same five-digit sector code, while the related inflows are defined as the number of
17
new employees from the same three-digit code, excluding the inflows from the same
five-digit code (i.e. similar inflows). Finally, unrelated inflows are defined as the
number of employees with a background in all other five-digit industries. Similar to
the findings in Boschma et al (2009), we expect that inflows similar to the plant
specialisation are absorbable, but will not add new knowledge. It is therefore not
expected to increase performance unless such knowledge is combined with other
place-specific routines obtained farther away from the plant. Moreover, for high levels
of unrelated inflows, the cognitive distance between the existing knowledge base and
the new knowledge is expected to be too great and will therefore not improve plant
performance due to problems of communication. Because of the assumed cognitive
similarity between co-located plants we, however, expect that the communication
problems associated with inflows of unrelated skills will be reduced if recruited from
the next-door neighbour – by being co-located, plants are likely to share common
values despite being specialised in different sectors. On the other hand, high levels
of related inflows will complement the existing knowledge base, which increases
learning opportunities and potentially contributes to increased performance. This is
an effect we expect to be particularly strong in combination with close proximity, but it
will slowly decrease as more place-specific knowledge is added from further
distances.
However, it is not reasonable to assume that plants will constantly increase
their numbers of employees by continually recruiting new skills which adds to their
existing knowledge. In order to determine the set of skills brought to the plant and its
potential economic effects, it is also necessary to control for skills leaving the plant.
Due to the mono-structure of the SNI-codes, skills leaving the plant do not show any
variation since the plant is only ascribed one single sector-code. The constitution of
data renders it impossible to draw any conclusions about the outflows of similar,
related or unrelated skills leaving the plant (data can indicate to which sector the
former employee leaves for, but it cannot tell anything about how the departing skill
matches the knowledge base of the old plant). It is however possible to take into
account the number of employees leaving the plants and determine whether they go
elsewhere in the local milieu or to jobs further away.
Irrespective of these data issues it is nonetheless possible to draw some
general conclusions about the impacts of outflows. For instance, by following the
notion that job mobility will create linkages of occupational weak ties between the old
18
and the new workplace (Granovetter, 1995), Bienkowska (2007) argues that former
employees may be regarded as ‘ambassadors’ for the previous firm due to their role
as mediating connections to new customers and arranging recruitment of new staff.
This insight leads to the conclusion that the impacts of job mobility do not necessarily
involve a relative gain for the receiving firm as compared to the loss for the
dispatching firm. On the contrary, such social ties between former colleagues are
likely to further facilitate inter-firm knowledge flows (Breschi and Lissoni, 2003). Due
to the localised dimension of job mobility, such networks are formed locally and will
therefore enhance knowledge accumulation within the local area (Dahl and
Pedersen, 2003). Hence, we therefore expect that it will be less damaging if
employees change job to another employer within the local milieu since their
embedded knowledge will remain within close proximity and stay available for the old
plant, either indirectly via localised spillovers or via the social link produced by the job
move. If the former employee leaves for a new position far away it is likely that more
efforts are needed to keep the connections and benefit from his/her role as
‘ambassador’ vi .
In total 18 different variables on similar, related and unrelated inflows from
within and outside the defined neighbourhoods are constructed. Similarly, six
variables measuring the distance of outflows within and outside the neighbourhoods
are calculated (see further details in Table A1 in the Appendix).
3.3 Control variables
A number of other factors, such as sector, plant size and educational level are likely
to co-determine labour productivity at the plant level. These factors also need to be
incorporated in the analysis. In order to control for industry-specific effects, a dummy
variable separating manufacturing units from service sectors are included. Two
additional controllers are plant size and the ratio of workers with an education
equivalent to a bachelor degree or beyond. While large plants are expected to show
higher levels of productivity, they are not expected to show as high levels of relative
productivity growth as smaller plants. We also expect that a greater share of formal
human capital measured as educational level will affect performance positively. The
vi
Agrawal et al (2006), however, point out that knowledge is admittedly highly localised, but social networks
maintained via more geographically distant job mobility may overcome the problems associated with greater
spatial distances. This ensures knowledge transfers between socially interlinked individuals working for firms in
different localities.
19
more people are being clustered to each plant, the higher the possibility of
employees to interact and generate knowledge spillovers (Glaeser, 1999). But the
impact of this interaction is dependent on the type of economic activity each worker is
affiliated to. Therefore, a general measure of population density was added in order
to more accurately control for this effect. The uneven population distribution justifies
logarithmic transformation of the number of workers per square kilometre within each
municipality and neighbourhood. This transformation makes the measurements
comparable over different spatial units. In order to control for nonlinear relationships,
a quadratic term of population density is also included. It should be noticed that a
variable controlling for the number of other co-located workplaces within the same
corporate group was also included in the analyses, since this is likely to influence the
impacts of agglomeration measurements as well as different types of labour flows.
This variable did not however have any effect on the estimation scores and therefore
it was excluded in the final analysis. Definitions of variables and descriptive statistics
of all included variables are provided in the Appendix (Table A1). Despite the
potential risks of extensive multicollinearity, no such problems are identified vii . For the
empirical analyses, ordinary least-squares (OLS) models were applied.
4. Empirical results
In this section the empirical estimates are presented (Table 1 and Table 2). The
tables display the coefficients and the z-values (within brackets) showing the sign
and the relative effect of each covariate. Table 1 presents the effects of
agglomeration externalities on plant performance within administrative regions
(Models B), within plant-specific surroundings of 0.5 kilometres (Models C), 5
kilometres (Models D) and 50 kilometres (Models E), respectively. In Model A,
estimating only control-variables, all variables show expected signs. A high share of
employees with a university degree will increase performance, whereas plant size
has the greatest effect on performance. In general, small plants show higher levels of
productivity growth. In comparison to the knowledge-intensive service sectors,
manufacturing units have a negative effect on productivity change during the period
2001 to 2003. The base model fits the data reasonably well. Despite the high degree
vii
For example, the only correlation higher than 60% and significant at the 5% level was between the similarity
and unrelated variety variables. The highest correlation was between the similarity and unrelated variety
variables calculated for the municipalities (correlation = 0.74).
20
of unexplained heterogeneity involved when modelling micro data, the model
explains about 58% of the total variation.
In Model B1 and B2 (the municipal level) the impacts of external factors like
density and composition of economic activities are presented by separating between
pure intra-industry spillovers (Model B1) and inter-industry spillovers (Model B2) viii .
The results show that high concentrations of plants within a municipality generally
have a positive effect on performance (Model B1) but this effect weakens when
adding variables on diversity in Model B2. When differentiating between the effects of
intra- and inter-industry spillovers, the scores show expected signs. Both similar and
related activities contribute to plant performance but when comparing the z-values
and the r2-values between the models it is possible to conclude that the impact of
related variety is larger. However, high degrees of very different activities do not have
any effect on performance, which indicates that the cognitive distance between
plants is too far for significant knowledge spillovers to occur. In sum, these two
models confirm previous findings (e.g. Frenken et al, 2007; Boschma and Iammarino,
2009) indicating that plants are more likely to benefit from spillovers and real learning
opportunities if the region is characterised by complementary activities as compared
to very similar or very different activities.
viii
These effects have been separated mainly for two reasons: First, it is collinearity between the measurements
for similarity and unrelatedness and, second, we wanted to explicitly analyse how geographical proximity
influences both intra- and inter-industry spillovers. It should be noted that models estimating only related or
unrelated variety also have been calculated because these variables could affect each other. Since neither the sign
nor the levels of significance of covariates were affected by this procedure, the results indicate robustness and
could be interpreted with confidence.
21
Table 1: OLS estimates on the effects of agglomeration externalities on productivity growth (2001-2003) for workplaces with skilled inflows 2001. Coefficients
and t-values (within brackets) are reported. Significant at the *** 0.01 level, ** 0.05 level and * 0.10 level
Base
Region
Region
0.5 km
0.5 km
5 km
5 km
50 km
50 km
Labour
model
(B1)
(B2)
(C1)
(D1)
(D2)
(E2)
(C2)
(E1)
Productivity
(A)
0.033***
0.007
-0.022***
10.018***
Similarity
(4.510)
(-2.850)
(4.680)
(1.230)
7.310***
-0.002
0.008**
0.006**
RelVar
(6.120)
(-0.250)
(2.050)
(1.780)
-0.103
-0.006
-0.058***
-0.127***
UnrelVar
(-0.450)
(-0.440)
(-3.620)
(-7.320)
0.044**
0.005
-0.094***
-0.060**
-0.050*
-0.035
-0.043
-0.074
2
PopDensKm
(2.210)
(0.210)
(-3.410)
(-2.050)
(-1.650)
(-1.200)
(-0.620)
(-1.070)
0.001
0.003
0.010***
0.006**
0.009**
0.007**
0.002**
0.002**
PopDensKm2^2
(0.200)
(0.890)
(2.710)
(1.660)
(2.570)
(2.070)
(1.990)
(2.240)
0.495***
0.464***
0.416***
0.510***
0.512***
0.476***
0.484***
0.449***
0.467
HEducRatio
(11.080)
(10.270)
(9.120)
(11.220)
(10.660)
(10.020)
(11.260)
(10.490)
(10.440)
-0.765***
-0.765***
-0.762***
-0.765***
-0.764***
-0.768***
-0.769***
-0.761***
-0.762***
PlantSize
(-87.560)
(-87.520)
(-78.100)
(-87.270)
(-86.970)
(-87.620)
(-87.750)
(-87.370)
(-87.710)
-0.310***
-0.283***
-0.204***
-0.338***
-0.328***
-0.271***
-0.098
-0.139***
0.119*
Manufacturing
(-9.960)
(-8.770)
(-6.110)
(-10.140)
(-7.160)
(-7.620)
(-1.640)
(-3.400)
(2.000)
-1.036***
-2.587***
-0.880***
-0.925***
-0.928***
-0.991***
-0.912***
-1.040***
-0.816***
Intercept
(-29.170)
(-8.220)
(-7.280)
(-17.920)
(-17.910)
(-16.070)
(-13.990)
(-9.410)
(-7.010)
R2
0.583
0.586
0.587
0.583
0.583
0.584
0.584
0.588
0.590
N
8 313
8 313
8 313
8 313
8 313
8 313
8 313
8 313
8 313
22
By shifting focus to the models C1 to E2 where the administrative borders are
relaxed and the effects of the unique plant-specific agglomerations at 0.5 kilometres
(Models C1 and C2), 5 kilometres (Models D1 and D2) and 50 kilometres (Models E1
and E2) are measured, the findings in Models B on regional aggregates are
somewhat altered. Within all three radii, the scores on plant density indicate a nonlinear relationship with increased productivity. When assessing the influence of
geographical proximity on agglomeration externalities by comparing the estimate
scores and z-values through Models C1 to E2, the models indicate that in close
proximity (within 0.5 kilometres) the composition of activities is subordinated the
effect of relative concentration. A high concentration of similar activities is strongly
negatively correlated with productivity growth. Neither related nor unrelated variety
show any significant effects. However, by increasing the geographical distance the
influence of intra-industry spillovers increases and become significantly positive when
measured within 50 kilometres from each plant (Model E1), whereas the effect of
being located nearby very different activities is more detrimental to performance in
combination with increased distance. Except for the non-significant effect when
measured within 0.5 kilometres from the plant, concentrations of complementary
activities have positive effects on plant performance. Thus, these results are in line
with our expectations – neither too much proximity nor too little proximity
(geographical or cognitive) is beneficial for plants (Boschma, 2005). However, if
cognitive proximity is combined with plant-specific knowledge originating from places
further away it may reduce the risks of cognitive lock-in associated with many similar
industries. Moreover, the effect of unrelated activities seems to be less harmful in
combination with close proximity, since geographical proximity may reduce the
communication problems associated with too much cognitive distance. Finally,
although related variety was expected to be particularly beneficial in combination with
close proximity, the results indicate that complementary spillovers are effective as
well; some distance is necessary but the results give at hand that increased
geographical distance reduces the positive effects of related variety. By comparing
the original entropy measurements estimated in Models B (on municipalities) and the
modified entropy measurements estimated in Models E (which are set to cover most
of the Swedish municipalities), it is possible to argue that the modified measurements
serve their purpose very well as the estimation scores of the modified measurements
in Models E correspond to the estimated score on the original entropy-measurements
23
in Models B. These results remain stable when adding covariates on labour flows in
the following models.
In Table 2, the effects of skilled labour mobility are presented. Since
municipalities were included to reveal the effects of relaxed administrative borders
and plant-unique externality measures, only flows concerning plant-specific
neighbourhoods of 0.5 kilometres (Models C3 and C4), 5 kilometres (Models D3 and
D4) and 50 kilometres (Models E3 and E4) are presented here. For each
neighbourhood two models have been estimated. First, the gross flows of skilled
labour to and from each plant both within and outside the defined neighbourhoods
were calculated. In the next stage all the inflows were separated into similar, related
or unrelated local and extra-local flows, since we argue that it is not the linkages per
se that affect performance but rather the types of skills that enter the plant and the
matching with the in-house skill portfolio (c.f. Boschma et al, 2009). Concerning the
estimates in Table 2, four characteristics should be highlighted in particular.
First, by comparing Models C3, D3 and E3, it is apparent that inflows per se
do not reveal much of the effects of either local or extra-local inflows, regardless of
the geographical distance involved. Thus, these findings confirm previous studies
addressing this issue. Similar to a study on Finnish high-technology industries where
McCann and Simonen (2005) found a negative correlation between innovative
performance and local labour mobility, these results indicate that local inflows of
skilled labour do not show any substantial effects on the economic performance of
plants. However, by estimating the impact of more distant inflows (Models E3 and
E4), the effects of manufacturing units increase implying that such units benefit
relatively more from inflows than service units.
24
Table 2: OLS estimates on the effects of different local and extra-local labour flows on productivity
growth (2001-2003) for workplaces with skilled inflows. Coefficients and t-values (within brackets) are
reported. Significant at the *** 0.01 level, ** 0.05 level and * 0.10 level
0.5 km
0.5 km
5 km
5 km
50 km
50 km
Labour Productivity
(C3)
(C4)
(D3)
(D4)
(E3)
(E4)
0.025
0.030
0.054
LocalInSima
(0.820)
(0.820)
(0.910)
0.347
0.276**
0.193**
LocalInRelVar
(1.550)
(2.500)
(2.340)
0.034
0.020
0.003
LocalInUnrelVar
(0.600)
(0.590)
(0.130)
-0.062**
-0.116***
-0.151***
ExtraLocalInSima
(-2.320)
(-3.480)
(-3.610)
0.072
0.114
0.163**
ExtraLocalInRelVar
(0.683)
(1.725)
(2.280)
-0.040**
-0.083***
-0.127***
ExtraLocalInUnrelVar
(-1.790)
(-3.190)
(-3.580)
0.057
0.032
0.010
TotalLocalIn
(1.410)
(1.260)
(0.470)
-0.054***
-0.107***
-0.140***
TotalExtraLocalIn
(-2.760)
(-4.860)
(-5.090)
0.096
0.095
0.111***
0.114***
0.057**
0.054***
TotalLocalOut
(1.520)
(3.290)
(3.330)
(2.260)
(2.100)
(1.550)
0.057***
0.056**
0.036
0.036
0.053
0.055
TotalExtraLocalOut
(2.610)
(2.510)
(1.380)
(1.350)
(1.550)
(1.590)
-0.001
-0.001
0.009**
0.009**
0.006*
0.006**
RelVar
(-0.210)
(-0.024)
(2.230)
(2.400)
(1.870)
(2.030)
-0.007
-0.007
-0.057*** -0.057***
-0.124***
-0.124***
UnrelVar
(-0.530)
(-0.550)
(-3.570)
(-3.540)
(-7.160)
(-7.150)
-0.059**
-0.058**
-0.026
-0.027
-0.054
-0.052
PopDensKm2
(-2.010)
(-1.970)
(-0.930)
(-0.940)
(-0.780)
(-0.760)
0.006
0.006
0.005
0.006
0.018*
0.018*
PopDensKm2^2
(1.640)
(1.600)
(1.620)
(1.630)
(1.810)
(1.790)
0.515***
0.512***
0.496***
0.491***
0.484***
0.480***
HEducRatio
(10.870)
(10.630)
(10.560)
(10.490)
(10.450)
(10.910)
-0.768*** -0.772*** -0.766*** -0.771***
-0.763***
-0.767***
PlantSize
(-62.240) (-64.050) (-65.170) (-67.280)
(-65.650)
(-67.600)
-0.327*** -0.321***
-0.099*
-0.094
0.111*
0.120*
Manufacturing
(-7.070)
(-6.940)
(-1.650)
(-1.560)
(1.860)
(2.010)
-0.914***
0.914***
-0.912*** -0.908***
-0.834***
-0.833***
Intercept
(-16.740) (-16.750) (-13.320) (-13.300)
(-7.030)
(-7.030)
R2
N
0.584
8 313
0.584
8 313
0.586
8 313
0.587
8 313
0.591
8 313
0.592
8 313
Second, in Models C4, D4 and E4 where both local and extra-local inflows at
various distances are separated into similar, related and unrelated skills, the results
indicate that inflows of related skills is the only type of inflow displaying substantial
positive
effects
on
plant
performance.
This
outcome
is
expected
since
complementary knowledge flows are possible for the plant to absorb. Real learning
opportunities are induced because the cognitive distance between the existing
knowledge base of the plant and the inflows is neither too short nor too long. Thus,
25
the estimates confirm the propositions made in previous sections indicating that
related inter-plant knowledge flows are essential for acquiring and utilising new
embodied knowledge. However, even if we differentiate between various types of
skills, labour recruited from other parts of the national economy (beyond the defined
radii) do not show any positive effects on plant performance.
Third, the geographical dimension highly influences the impact of the different
types of skills. However, the estimates on externalities in Table 1 indicate that
increased distance also enhance the positive effects of intra-industry spillovers, but
such an effect is not evident concerning inflows of skills from exactly the same
sectors. In line with expectations, similar inflows characterised by geographical
proximity is possible for the plant to absorb, but it will not add to the existing
knowledge base that could trigger a positive effect on performance. In contrast
increased geographical distance does not seem to contribute to increased
performance. This result holds even if increased distance is combined with other
types of place-specific knowledge. The results of this study differ from the findings
reported by Boschma et al (2009) who found no positive effects of unrelated inflows,
even if they were combined with geographical proximity. However, the results
correspond to expectations, since geographically distant inflows of unrelated skills
appear to act as a check on productivity growth as compared to inflows from nearby.
This is due to that this type of labour mobility entails both cognitive and geographical
distance, which may give rise to communication problems and inertia at the plant. In
addition, similar to the findings presented in Table 1, the results on related inflows
indicate that a certain degree of geographical distance (i.e. between 5 and 50
kilometres) is needed for such inflows to produce significant positive effects on
performance, but too long distances will reduce the effect as indicated by the
estimation scores on extra local inflows of related skills in Model D2 and Model E2.
A fourth observation concerns the effects of skills leaving the plant. In previous
sections it was theorized that skills leaving the plant not necessarily imply negative
outcomes, since linkages of weak occupational ties between old and the new
workplaces would be established via physical mobility of people. This would in turn
facilitate further knowledge flows between the firms (Breschi and Lissoni, 2003).
Although the economic effect of outflows is likely to be more instantaneous than
those generated by inflows, the empirical results confirm this notion but they also
reveal that the geographical dimension is clearly related to this issue. It is less
26
detrimental to the dispatching plant if their former employees remain within the local
economy as compared to if they depart for jobs in other parts of the national
economy. Hence, knowledge is likely to disseminate both forwards and backwards
via social linkages established between the old and new workplace (c.f Dahl and
Pedersen, 2003). However, in contrast to previous empirical findings a certain
distance between the plants is needed for efficient backward linkages to occur. The
results indicate that outflows of skills into the immediate neighbourhood (closer than
0.5 kilometres) have no significant effects on the dispatching firm. It is more
beneficial to the firm if the former employee moves across greater distances.
The final observation is related to the overall explanatory power shown in the
two output tables. Despite using micro data the R2 values reach reasonably high
levels (58 percent and higher). What should be noted is the relative change of R2
between the different models. As compared to the base model, which only included
characteristics internal to the plant, the subsequent models in Table 1 show relative
moderate increases in explanatory power. Small amounts of explanatory power are
gained by increasing the hinterlands in the models, but relatively more explanatory
power is gained in comparison to models merely assessing the impact of externalities
(see Table 1). This outcome clarifies that different types of inter-firm knowledge flows
via labour mobility tell more about differences in productivity growth than information
gained from externality variables. Nevertheless, the reported overall moderate effect
is in line with previous studies on the effects of agglomeration externalities and
labour mobility in Sweden (e.g. Eriksson and Lindgren, 2009). Agglomerative effects
internal to the workplace affect productivity the most, but the relative effects of labour
mobility tend to be stronger than the effects derived from localised externalities.
5 Summary and conclusions
The analyses carried out in this paper have hopefully contributed to the perennial
discussions on the benefits of intra- and inter-industry spillovers. By creating both
plant-specific agglomeration measures and labour markets at 0.5 kilometres, 5
kilometres and 50 kilometres from each of the 8,313 plants in the sample, it was
possible to account for (i) what type of co-located activities that are most beneficial
the productivity growth during a three year period; (ii) how different types of
knowledge flows, in and out from the plant, affect performance, and; (iii) how
geographical proximity interacts with both localised spillovers and knowledge flows.
27
As summarized in Figure 1, the empirical results indicate that neither too much nor
too little proximity (measured as both geographical and cognitive proximity) between
co-located activities is likely to produce significant localised spillovers. This also
seems to be the case when assessing more directly the impacts of inter-plant
knowledge flows via labour mobility – only knowledge flows that are complementary
to the existing knowledge base of plants and characterised by just the right
geographical proximity have a positive effect on performance.
Similar knowledge
Dissimilar
knowledge
Close proximity
Long distance
Redundant
Useful
Useful
Incomprehensible
Figure 1. Spatial dimensions on the economic impact of external knowledge.
Concerning the estimates on localised spillovers, the results indicate that the
spillover effect related to Jacob’s externalities is more important than the effect
derived from intra-industry spillovers. However, by relaxing the influence of
administrative borders the shortcomings of using fixed spatial entities have been
shown. The results indicate that the geographical dimension of spillovers is of great
significance for the understanding of labour market dynamism. At close range
externalities derived from complementary sectors are relatively more gainful,
whereas intra-industry spillovers are relatively more beneficial at further distances.
By comparing traded aspects of agglomerations, as measured by the composition of
activities, with untraded aspects related to labour mobility (Wolfe and Gertler, 2004;
Storper, 1997; Phelps, 2004), the findings presented in this paper is coherent with
previous findings (c.f. Boschma et al, 2009). Regarding the impacts of untraded interfirm linkages it appears that high degrees of complementarities are more beneficial
than very similar or very different linkages. Thus, the relative importance of
complementary knowledge is affirmed, irrespective of if it is measured as relative
concentrations or as labour flows. The results also stress that the impacts of
knowledge flows does not to any great extent compensate for shortcomings in the
local milieu, the agglomeration variable remain unaffected when adding the mobility
28
variables. However, the direct inter-plant linkages derived via mobility seems to
enhance more effective knowledge transfer since the mobility variables explain
relatively more of the variations in plant productivity than the composition of colocated activities. Moreover, the findings indicate the relative importance of the local
milieu as compared to relationships to more distant places. Although the estimates
on localised spillovers by definition confirm the impact of co-location (the variables
are defined to cover the local milieu), this notion is also corroborated by the untraded
linkages of labour flows; it is only within distances mostly defined as the maximum
functional hinterland that such linkages produce significant positive effects. This
insight makes it possible to conclude that other types of extra-local relationships, not
only inflows of skilled labour, are crucial for retrieving new knowledge and achieve
sustained competitiveness.
In the light of these general findings there are several challenges for future
research. In order to find out more about the causality of relationships identified here
and the impact of different types proximities, these quantitative findings could be
supplemented with case-studies. Adopting a more dynamic approach to the impact of
both externalities and labour mobility by following a certain cluster or a particular
industry over time would, for instance, make it possible to investigate the importance
of different stages in the product life-cycle and its connections to spatial
dependences of externalities (c.f. Gordon and McCann, 2000; Neffke, 2007). A caseorientated approach would also make it possible to assess whether it is more
beneficial to a particular plant if skills enter the cluster milieu as compared to if it goes
directly to the plant. A major delimitation of the current study is the use of static
industry codes as a measure of externalities and labour flows. In order to determine
how different combinations of individual skills (e.g. different types of education,
occupation and accumulated work experience) within a plant match the surrounding
milieu of individual skills and how this changes over time, future studies would benefit
from constructing refined measures of relatedness (c.f. Neffke and SvenssonHenning, 2008). Another delimitation of this study refers to the definition of spatial
proximity. The use of Euclidian distance provides only moderate increases in r2values, which indicates that other types of spatial relationships may be more
influential on firm performance. For example, other types of temporary clustering
(e.g. trade fairs and conferences) may play a crucial role for retrieving new
knowledge (Bathelt and Schuldt, 2008) implying that co-location is only one of many
29
possible aspects of proximity. It could therefore be fruitful to include more social
aspects of proximity and co-location including face to face interactions. In addition,
the moderate effects of distant linkages may in fact reveal more about the labour
market of the plants than about the impact of extra-local pipelines. Future studies
should therefore go beyond labour flows and more explicitly address the effects of
other types of extra-local linkages.
30
References
Agrawal A., Cockburn I. and McHale J. (2006) Gone but not forgotten: knowledge flows, labor mobility,
and enduring social relationships, Journal of Economic Geography, 6: 571-91
Almeida P. and Kogut B., (1999) Localization of knowledge and the mobility of engineers in regional
networks, Management science, 45: 905-17
Andersson L. F. (2006) Företagsdynamik och tillväxt: En kartläggning och analys av företagsdynamik
och arbetsproduktivitet i Sverige [Business dynamics and economic growth: A mapping and
analysis of business dynamics and labour productivity in Sweden], Report A2006:16, Swedish
Institute for Growth Policy Studies (ITPS)
Anselin, L., A. Varga & Z. Acs (1997): Local Geographic Spillovers between University Research and
High Technology Innovations, Journal of Urban Economics, 42: 422-448.
Asheim, B.T. and Isaksen, A. (2002) Regional innovation systems. The integration of local ‘sticky’ and
global ‘unbiquitous’ knowledge, Journal of Technology Transfer, 27: 77-86.
Bathelt, H., Malmberg A. and Maskell P. (2004) Clusters and knowledge: local buzz, global pipelines and
the process of knowledge creation, Progress in Human Geography, 28: 31-56
Bathelt, H. and Schuldt, N.A. (2008): Between Luminaires and Meat Grinders: International Trade
Fairs as Temporary Clusters, Regional Studies, 42: 853-868.
Bienkowskaya, D. (2007) Arbetskraftens rörlighet och klusterdynamik. En studie av IT- och
telekomklustren i Kista och Mjärdevi [Labour mobility and cluster dynamics. A study of ICT
clusters in Kista and Mjärdevi, Sweden]. PhD-thesis, Geografiska regionstudier 75, Department
of Social and Economic Geography, Uppsala University, Uppsala
Boschma R. A. (2005) Proximity and Innovation: A Critical Assessment, Regional Studies, 39: 61–74
Boschma R. A. and Iammarino S. (2009) Related variety, trade linkages and regional growth in Italy,
Economic Geography, forthcoming.
Boschma, R., Eriksson, R. and Lindgren, U. (2009): How does labour mobility affect the performance
of plants? The importance of relatedness and geographical proximity. Journal of Economic
Geography, 9(2): 169-190
Burger, M., van Oort, F. and van der Knaap, B. (2007): A Treatise on the Scale-Dependency of
Agglomeration Externalities and the Modifiable Areal Unit Problem. Paper presented at the
Workshop on “Agglomeration and Growth in Knowledge-Based Societies”, Kiel, Germany, April
20-21
Breschi S. and F. Lissoni (2003) Mobility and social networks: Localised knowledge spillovers
revisited. CESPRI Working Paper 142. (http://www.cespri.unibocconi.it/)
Bresnahan T., Gambardella A. and Saxenian A. (2001) ‘Old economy’ inputs for ‘new economy’
outcomes: cluster formation in the new Silicon Valleys. Industrial and Corporate Change, 10:
835–60
Ciccone, A. and Hall, R.E. (1996). Productivity and the Density of Economic Activity, American
Economic Review, 86: 54–70.
Cooper D. (2001) Innovation and reciprocal externalities: information transmission via job mobility,
Journal of Economic Behavior and Organization, 45(4): 403–25.
31
Dahl M.S. and Pedersen C.O.R. (2003) Knowledge flows through informal contacts in industrial
clusters. Myth or reality? Research Policy, 33: 1673-1686.
Dekle, R. and Eaton, J. (1999): Agglomeration and Land Rents. Evidence from the Prefectures,
Journal of Urban Economics, 46: 200–214.
Doeringer, P, Piore, M. (1970) Internal Labour Markets and Manpower Analysis. Lexington, Mass.
Eriksson, R., Lindgren, U. and Malmberg, G. (2008): Agglomeration mobility: effects of localisation,
urbanisation and scale on job changes. Environment and Planning A, 40 (10): 2419-2434
Eriksson, R. and Lindgren, U. (2009): Localized mobility clusters: Impacts of labour market
externalities on firm performance. Journal of Economic Geography, 9(1): 33-53
Essletzbichler, J. and Rigby D.L. (2005) Technological evolution as a creative destruction of process
heterogeneity. Evidence from the US plant level data, Economic Systems Research, 17: 25-45.
Faggian, A. and McCann, P. (2006) Human capital flows and regional knowledge assets. A
simultaneous equation approach, Oxford Economic Papers, 58(3): 475-500.
Fischer P. A., Holm E., Malmberg, G. and Straubhaar T. (1998) Why do people stay? The Insideradvantages Approach: Empirical Evidence from Swedish Labour Markets, Discussion Paper 1952,
Centre for Economic Policy Research (CEPR), London.
Frenken K., Van Oort F. and Verburg T. (2007) Related Variety, Unrelated Variety and Regional
Economic Growth, Regional Studies, 41(5): 685-97
Gertler M. S. (2003) Tacit knowledge and the economic geography of context, or the undefinable tacitness
of being (there), Journal of Economic Geography, 3: 75-99.
Glaeser, E.L., Kallal, H.D., Schinkmann, J.A., and Shleifer, A., 1992. Growth in cities. Journal of
Political Economy 100: 1126-1152.
Glaeser, E.L. (1999): Learning in Cities. Journal of Urban Economics, 46:. 254-277
Grabher G. (2002) Cool projects, boring institutions: temporary collaboration in social context.
Regional Studies 36: 205-14.
Granovetter, M. (1995): Getting a job. A study on contacts and careers (2nd edition). Chicago,
University of Chicago Press
Gordon, I.R. & P. McCann (2000): Industrial Clusters: Complexes, Agglomeration and/or Social
Networks?, Urban Studies 37: 513-532.
Henderson, J.V., Kuncoro, A., and Turner, M., 1995. Industrial development in cities. Journal of
Political Economy 103: 1067-1085.
Hoover, E.M. (1937) Location Theory and the Shoe and Leather industries. Cambridge, MA: Harvard
University Press.
Jaffe, A., Trajtenberg, M., Henderson, R. (1993), Geographic Localization of Knowledge Spillovers as
Evidenced by Patent Citations. Quarterly Journal of Economics, 108: 577-598.
Li, W., Holm, E. and Lindgren, U. (2009): Attractive Vicinities. Population, Space and Place, 15(1): 118
Malmberg, A. (1996), Industrial geography: agglomeration and local milieu. Progress in Human
Geography 20: 392-403.
32
Markusen, A. (1996): Sticky places in slippery spaces: a typology of industrial districts. Economic
Geography, 72: 293-313
Martin, R. (1999): The New “Geographical Turn” in Economics: Some Critical Reflections, Cambridge
Journal of Economics, 23: 65-91.
Martin, R. (2000) Local Labour Markets: their Nature, Performance, and Regulation. In: Clark, G. L.,
Feldman, M. P. & Gertler, M. S. (eds.): The Oxford Handbook of Economic Geography. Oxford
University Press, Oxford.
Martin, R., Sunley, P. (2003), Deconstructing Clusters: Chaotic Concept or Policy Panacea? Journal of
Economic Geography 3: 5-35.
Maskell P. and Malmberg A. (1999) Localised learning and industrial competitiveness. Cambridge Journal
of Economics, 23: 167-85.
McCann, P. and Simonen, J. (2005), Innovation, knowledge spillovers and local labour markets, Papers in
Regional Science, 84(3): 465-485.
Neffke, F.M.H. (2007): Time-Varying Agglomeration Externalities in UK Counties, Mimeo, Department
of Economic Geography, Utrecht University.
Neffke, F.M.H. and Svensson-Henning, M. (2008) Revealed relatedness, Department of Economic
Geography, Utrecht.
Nooteboom, B., Van Haverbeke W., Duysters G., Gilsing V., Van Den Ooord A. (2007) Optimal cognitive
distance and absorptive capacity Research Policy, 36: 1016-1034.
Oerlemans, L. and Meeus, M. (2005): Do Organizational and Spatial Proximity Impact on Firm
Performance? Regional Studies, 39(1): 89-104
Peck, J. (1996): Workplace: The social regulation of labour markets. New York, Guilford
Pettersson, Ö. (2001), ‘Microregional Fragmentation in a Swedish country’, Papers in Regional
Science, 80: 389-409.
Pinch S. and Henry N. (1999) Paul Krugman’s Geographical Economics, Industrial Clustering and the
British Motor Sport Industry, Regional Studies, 33: 815-27.
Phelps, N.A. (2004), Clusters, dispersion and the spaces in between: for an economic geography of
the banal, Urban Studies, 41: 971 - 989
Porter, M. E. (2000): Locations, Clusters and Company Strategy. In: Clark, G. L., Feldman, M. P. &
Gertler, M. S. (eds.): The Oxford Handbook of Economic Geography. Oxford University Press,
Oxford
Power D. and Lundmark M. (2004) Working Through Knowledge Pools: Labour Market Dynamics, the
Transference of Knowledge and Ideas, and Industrial Clusters. Urban Studies, 41: 1025-44.
Rigby D.L. and Essletzbichler J. (2006) Technological variety, technological change and a geography
of production techniques, Journal of Economic Geography, 6: 45-70.
Rodriguez-Pose A. and Vilalta-Bufi (2005) Education, migration, and job satisfaction. The regional
returns of human capital in the EU, Journal of Economic Geography, 5: 545-566.
Rosenthal, S.S. & W.C. Strange (2003): Geography, Industrial Organization, and Agglomeration, The
Review of Economics and Statistics, 85(2): 377-393.
33
Rosenthal, S.S. & Strange, W.C. (2004): Evidence on the Nature and Sources of Agglomeration
Economies, in J.V. Henderson & J.F. Thisse, Handbook of Regional and Urban Economics:
Cities and Geography.
Scott, A.J. (1982): Locational Patterns and Dynamics of Industrial Activity in the Modern Metropolis,
Urban Studies, 19: 111-142.
Storper, M., Walker, R. (1989) The Capitalist Imperative: Territory, Technology and Industrial Growth.
Oxford: Blackwell.
Storper, M. (1997) The Regional World: Territorial Development in a Global Economy. New York: Guilford
Press
Storper, M., Venables, A.J. (2004), Buzz: face-to-face contact and the urban economy. Journal of
Economic Geography, 4: 351-370.
Van Soest, D.P., S. Gerking & F.G. van Oort (2006): Spatial Impacts of Agglomeration Externalities,
Journal of Regional Science, 46(5): 881-899.
Waxell, A. (2005) Uppsalas biotekniska industriella system: en ekonomisk-geografisk studie av
interaktion, kunskapsspridning och arbetsmarknadsrörlighet [The Uppsala Biotech Cluster –
Economic-Geographical Studies of Interaction, Knowledge Spillover and Labor Mobility]. PhDThesis, Department of Social and Economic Geography, Uppsala University, Uppsala.
Weber, A. (1929) Theory of the Location of Industries. Chicago: University of Chicago Press.
Wictorin, B. (2007) Är kluster lönsamma? En undersökning av platsens betydelse för företags
produktivitet [Are clusters profitable? The role of place for firm productivity]. PhD-Thesis,
Department of Business Studies, Uppsala University, Uppsala
Zucker. L.G., Darby, M. (1996), Star Scientists and Institutional Transformation. Patterns of Invention
and Innovation in the Formation of the Biotechnology Industry. Proceedings of the National
Academy of Sciences, 93: 12709-12716.
34
APPENDIX. Table A1: Variable description (N=8.313). Note that the statistics for the dependent variable and the plant-specific variables are the same
throughout the models and are therefore only displayed once.
Variable
Municipalities
Description
Mean
S.D
-2.88
1.74
Min
0.5 km
Max
Mean
S.D
Min
5 km
Max
Mean
50 km
S.D
Min
Max
Mean
S.D
Min
Max
Productivity growth
Change in labour
productivity 2001-2003
(log)
LocalInSima
Local inflows from
similar workplaces (log)
0.02
0.21
0.00
4.76
0.07
0.37
0.00
4.99
0.11
0.45
0.00
5.25
LocalInRelVar
Local inflows from
related workplaces (log)
Local inflows from
unrelated workplaces
(log)
0.00
0.06
0.00
3.56
0.01
0.12
0.00
3.56
0.02
0.15
0.00
3.58
0.02
0.23
0.00
5.59
0.12
0.44
0.00
5.62
0.27
0.64
0.00
5.68
ExtraLocalInSima
Extra-local inflows from
similar workplaces (log)
0.16
0.51
0.00
5.29
0.11
0.42
0.00
4.54
0.07
0.32
0.00
4.19
ExtraLocalInRelVar
Extra-local inflows from
related workplaces (log)
Extra-local inflows from
unrelated workplaces
(log)
0.03
0.18
0.00
3.53
0.01
0.14
0.00
3.53
0.01
0.10
0.00
3.53
0.39
0.73
0.00
5.61
0.29
0.63
0.00
5.49
0.13
0.42
0.00
4.67
Total local inflows (log)
Total extra-local inflows
(log)
0.04
0.31
0.00
5.95
0.21
0.59
0.00
6.05
0.40
0.77
0.00
6.10
0.57
0.86
0.00
5.85
0.42
0.75
0.00
5.73
0.21
0.54
0.00
4.73
Total local outflows (log)
Total extra-local
outflows (log)
Degree of similar
activities (log)
Degree of related
activities (log)
Degree of unrelated
activities (log)
Number of employees
per km2 (log)
Number of employees
per km2 (quadratic term)
Share of employees with
a bachelor degree or
higher
Number of employees
within plant (log)
0.02
0.21
0.00
5.80
0.12
0.46
0.00
6.21
0.25
0.65
0.00
6.71
0.36
0.76
0.00
6.76
0.28
0.66
0.00
5.92
0.14
0.46
0.00
4.86
LocalInUnrelVar
ExtraLocalInUnrelVar
TotalLocalIn
TotalExtraLocalIn
TotalLocalOut
TotalExtraLocalOut
Similarity
RelVar
UnrelVar
PopDensKm
2
2
PopDensKm ^2
HEducRatio
PlantSize
-11.61
5.90
0.14
0.01
0.13
0.20
0.04
0.05
0.00
0.53
0.03
0.03
0.00
0.13
0.02
0.02
0.00
0.08
0.01
0.01
0.00
0.07
-0.01
0.03
-0.38
0.09
-0.01
0.02
-0.12
0.02
-0.01
0.01
-0.05
0.01
2.80
0.16
2.50
3.21
2.43
1.38
0.00
9.63
2.97
1.66
1.33
9.28
3.16
1.57
1.66
8.09
2.80
2.17
-4.08
5.57
3.12
0.86
-0.73
4.24
3.51
0.75
-4.70
4.25
3.47
0.42
-4.35
3.89
12.55
12.46
0.00
30.98
10.47
4.40
0.04
17.97
12.87
4.32
0.01
22.10
12.23
2.65
0.13
18.97
0.40
0.32
0.00
1.00
2.53
1.56
0.00
8.52
35
Manufacturing
Dummy =1 if plant is
defined as
manufacturing unit
0.35
0.48
0.00
1.00
36
37