RURAL VERSUS URBAN INDUSTRIAL DEVELOPMENT:
A SHIFT-SHARE ANALYSIS ON FRENCH DATA
Carl GAIGNÉ, Virginie PIGUET and Bertrand SCHMITT#
(INRA, ESR-Rennes & CESÆR-Dijon, France)
September, 2005
Abstract:
From a shift-share analysis on French data, we examine whether the differential of changes in
industrial employment between urban cores, (peri)urban fringes and rural areas can be due to
differences in their sectoral composition rather than to space-specific factors affecting all sectors
indifferently. A estimable stochastic formulation of shift-share analysis with three composition factors
(sectoral, regional and urban centre size effects) and one geographical factor (urban-rural gradient)
allows us to show the predominance of the sectoral composition and of the urban-rural gradient among
the determinants of local employment growth and the weak role of regional characteristics. We also
find that the increase in the share of industrial employment in rural areas is essentially due to
advantages specific to low-dens ity territories. These ‘rural’ advantages are strong enough to offset the
unfavourable sectoral composition in such areas that attract firms belonging rather to declining sectors.
Conversely, urban centres, and especially the largest ones, combine a positive composition effect (i.e.
preferential urban location of the more dynamic sectors at national level) with a negative geographical
effect (i.e. a tendency for urban industrial employment to fall). The periurban fringes provide
attracting characteristics for location of any industrial sector without the sectoral structure plays a
significant role either positively or negatively.
Keywords: Local Growth – Industrial Employment – Shift-share Analysis – Urban-rural Areas
JEL classification: R12, O4
#
Corresponding author: Bertrand SCHMITT, CESÆR, UMR INRA-ENESAD, 26 Bd Petitjean, BP 87999,
F-21079 Dijon, France, Phone: (+33) 380 77 25 76, Fax: (+33) 380 77 25 71, E-mail: [email protected]
RURAL VERSUS URBAN INDUSTRIAL DEVELOPMENT:
A SHIFT- SHARE ANALYSIS ON FRENCH DATA
1. INTRODUCTION
Whereas the industrialisation of Western economies was accompanied by the urbanisation of
industrial employment (Mokyr, 1995), the crisis of the 1970s seems to have hit urban industrial
employment harder than rural industrial employment. For example, while industrial employment
(including business services) is mainly located in metropolitan areas in France, the growth rate of
these jobs between 1990 and 1999 is negative in metropolitan areas and positive in rural areas. In other
words, a relative decentralization of industrial employment to the benefit of rural areas occurs. This
phenomenon is observed in almost developed countries (Bollman et al., 2005), either in North
America (Holmes and Stevens, 2004) or in Europe (Champion, 1989; Bryden and Bollman, 2000).
However, we do not know whether these contrasting spatial patterns of change in industrial
employment are due to differences in sectoral composition between rural and urban areas or to spacespecific factors affecting all sectors indifferently. The first dimension means that the higher (resp.,
lower) employment growth in a particular type of area is due to the greater (resp., lesser) weight of the
more growing sectors at national level (in terms of employment) in this kind of area. This corresponds
to a sector-based composition effect or to a ‘sectoral effect’. The other dimensions mean that the
employment growth may be locally higher (resp., lower) because each sector is growing faster (resp.,
slower) in this area than in other areas. They correspond to ‘geographical effects’ that can be
supported by differences between regions or/and between categories of areas (urban versus rural, size
of municipalities, and so on).
This distinction between sectoral and geographical effects in the factors determining the local
economic growth suggests lines of research for understanding spatial inequalities in employment
growth. Indeed, if the sectoral effect is predominant, then we need to find out what determines this
particular spatial allocation of industrial sectors. Despite the static nature of these models, the use of
theoretical economic geography frameworks mobilizing both technological and pecuniary static
2
externalities (e.g. Fujita and Thisse, 2002, chs 8 & 9; Duranton and Puga, 2004) is relevant when
accounting for the economic growth of such areas. Conversely, if geographical effects explain the
regional discrepancies in growth, the frameworks that emphasize factors specific to the region or/and
the area seem the more appropriate. One may think of frameworks that consider local dynamic
externalities related to local specialization or to sectoral diversity (Glaeser et al., 1992; Henderson et
al., 1995; Combes, 2000) or of the dynamic effects of local public infrastructure (Gramlich, 1994).
In this paper we used an estimable stochastic formulation of shift-share analysis to determine
whether the different patterns of local change in industrial employment in France are due to sectoral
composition or geographical effects or to both. Contrary to the standard method, this approach
introduced by Berzeg (1978) takes explicitly account of any geographical effect (rather than assuming
a residual geographical effect as in the standard method) and determines the significance of the
composition and geographical effects. We extended the Berzeg’s approach by adding two other
composition factors to the sectoral composition effect: a regional effect and an urban centre size effect.
The first effect lies in the fact that the French regions also have their own engine of growth. Hence,
regional phenomena could bias our results since regions are highly heterogeneous in growth rate and
in degree of urbanization. The second effect considers the urban centres are also heterogeneous what it
is advisable to control for by distinguishing them by their size. We finally defined the geographical
effect as an urban-rural gradient. France is then divided into five major categories often used: urban
centres, (peri)urban fringes and three categories of rural areas (rural under urban influence, rural
centres and remote rural areas). Our study is based on 1990 and 1999 French population censuses data
(wherein the working population in employment is recorded at the place of work).
Our results first highlight the predominance of the sectoral composition and of the urban-rural
gradient among the determinants of local employment growth and the weak role played by the
regional characteristics. We also find that the increase in the share of industrial employment in rural
areas is essentially due to advantages specific to low-density territories. These ‘rural’ advantages are
far from negligible as they offset the fact that such areas attract firms belonging rather to the declining
sectors. This suggests the existence of rural-specific externalities attracting firms whatever their
belonging sector. Conversely, urban centres, and especially the largest ones, benefit to the presence of
3
the more growing sectors at national level but provide disadvantages specific to high-density cities (as
higher land rents or wages) so that the share of urban industrial employment falls. Hence, the urban
employment growth seems to be due to some specific location factors favouring the urban location of
growing sectors. Lastly, the periurban fringes provide favourable characteristics for the location of any
industrial sector and the existing sectoral structure does not play any significant role, neither positive
nor negative. The existence of geographical effect, which is negative for urban centres and positive in
periurban and rural areas, suggests two comments. First, without challenging the existence of
externalities whose intensity increases with the density of firms, lesser density areas are also sources
of externalities for industrial activties. The nature of these externalities remains to be analyzed.
Second, the net effects of industrial agglomeration may, above a certain level, become negative with
urban costs progressively offsetting the technological and pecuniary static externalities related to
agglomeration.
The remainder of the paper is organised as follows. After a brief review of the literature on the
determinants of local growth in Section 2, we develop in Section 3 the shift-share analysis method we
used. Section 4 describes data, various spatial delineations and industrial nomenclature. The main
results are set out in Section 5. Several ways were explored. First, we performed shift-share analysis
with four factors (three composit ion effects and one geographical effect) on the changes in
manufacturing employment and in both manufacturing and business services employment. Business
activities are worked into the analysis alongside manufacturing sectors in an attempt to examine the
role played by outsourcing of these services by manufacturing firms. Next, we tested the hypothesis of
a possible differentiation by size of urban centres of our results. In the final section, we discuss the
implications of our main results for future researches.
2. THEORY: STATIC VERSUS DYNAMIC SPATIAL EXT ERNALITIES AS SOURCES OF LOCAL GROWTH
The literature exploring the main sources of local growth is based on ‘localised’ externalities
as a key concept. More precisely, two categories of externalities have been put forward to explain
local development. The first one includes static externalities while the second one includes dynamic
4
externalities. We aim in this section to precise the sources of differences in sectoral composition
between rural and urban areas and some components of space-specific advantages.
According to urban economists, metropolitan areas favour the emergence of static externalities
or urban agglomeration economies. As initially suggested by Marshall (1890) and developed by recent
urban economics literature (Fujita and Thisse, 2002; Duranton and Puga, 2004), there are three sources
of urban agglomeration economies: (i) interactions on local labour markets, (ii) linkages between
buyers and suppliers on intermediate- and final-goods markets and (iii) knowledge spillovers. Cities
provide productive advantages due to wide variety of differentiated inputs (intermediate goods or
workers) and large size of final markets. These externalities can imply increasing returns at the region
level despite constant returns to scale in production (see Abdel-Rahman and Fujita, 1990; Helsley and
Strange, 1990). In addition, geographical proximity to individuals with high skills or knowledge
facilitates the acquisition of skills and the diffusion of knowledge. This latter type of externalities is
thus based on immediate information spillovers about current market conditions. At the opposite, rural
areas could be viewed as low-density areas. In this case, they enable firms to reduce land rents and
labour costs because of weaker competition on local markets of land (Fujita, 1989) or of labour
(Krugman and Venables, 1995). In addition, due to transportation costs of natural resources, some
firms have incentive to locate in rural areas in order to produce some of intermediate goods. Therefore,
firms intensive in R&D or using high technology favour urban location while rural areas attract firms
using low technology and intensive in labour (especially in low skilled workers) or producing less
differentiated good. This framework is consistent to Duranton and Puga (2001), who showed that new
industries evolve in large cities and then they move to smaller cities ones after achieving maturity.
In parallel, recent empirical works have shown that the local economic structure is an
important source of dynamic externalities. This literature gives a major role of prior information
accumulations in the area. According to Glaeser et al. (1992), Henderson et al. (1995) and Combes
(2000), sectoral specialization and diversity of regions could foster local growth. Dynamic
externalities arising from specialization corresponds to benefits captured by firms from clustering with
other firms in the same sector. Additionally, large agglomerations accommodating several industries
imply positive inter-sectoral linkages. Indeed, the innovations of a given sector can also diffuse from
5
this sector to the others. High diversity raises the probability of such a cross-fertilization of ideas. For
example, Audrescht and Feldman (2004) showed the positive impact of local diversity on local rate of
innovation. Notice also that the quantity of exchanged information raises with the number of firms
since the potential complementarities increases too. But, in these empirical works on local growth, the
sector-specific dynamic is not always controlled for. They focus mainly on the impact of local
environment on local growth. Sectors also have their proper dynamic because of foreign competition,
change in consumers’ preference or in market potential.
A shift-share analysis could be useful to control for these sectoral effects in order to evaluate
the respective role of local features and of sectoral location factors in local growth. In addition, it can
allow us to determine if dynamic externalities are also at work in low density (i.e. rural) areas while
foregoing works implicitly suggest that externalities are only related to medium and large cities.
3. AN ESTIMABLE STOCHASTIC FORMULATION OF SHIFT-SHARE ANALYSIS
The traditional formulation of shift-share analysis consists to partition the growth rate rij (over
the time interval from 0 to 1) of the i-th industry in the j-th category of area – urban versus rural areas
or region – in three 3 components as follows:
(
rij = r + (ri − r ) + rij − ri
)
(1)
where r is the national growth rate of the all industrial sectors and ri , the national growth rate of the ith
industry. Expressions of these three terms are given by: 1
(
)
rij = X ij 1 − X ij 0 X ij 0
X ij 0
r = ( X1 − X 0 ) X 0 = ∑ ∑
rij
i j X0
X ij 0
ri = ( X i1 − X i 0 ) X i 0 = ∑
rij
j X i0
with Xijt , the level of employment in sector i and spatial category j at period t (t = 0,1), X it = Σ j X ijt ,
the total employment of sector i at period t and X t = Σi Σ j X ijt , the national number of industrial jobs
at time t. Then, by calculating the regional mean of the equation (1), we obtain the following
composition-geographical equality:
1
These equalities hold when Xij0 ≠ 0 and Xij1 ≠ 0, ∀ i and ∀ j.
6
r j = r + s j + g j where r j = ∑
i
X ij 0
X j0
rij , s j = ∑
i
X ij 0
X j0
(ri − r ) and
gj =∑
i
X ij 0
X j0
(rij − ri )
Hence, the employment growth rate in the jth region, rj , is equal to the sum of the three
following components: (i) r, the national growth rate; (ii) sj , the composition effect (or sectoral effect);
and (iii) gj , the geographical effect.
However, we cannot apply a significance test of both composition and geographical effects
because this conventional shift-share approach is formulated as a tautology. An alternative approach
was suggested by Berzeg (1978) and developed by Jayet (1993), Esteban (2000) or Gaschet (2002).2
This method involves an estimable stochastic formulation and consists to rewrite the equation (1) as a
linear model given by:
rij = α + β i + γ j + ε ij
(2)
where α is a constant, βi (resp., γj ) is the fixed effect of the i-th sector (resp., the j-th spatial category)
and ε ij is a random error term. Furthermore, the model is weighted by wij =Xij0 /X0 . Without constraints
on the parameters, this model is not identified. Therefore, the two following constraints were
introduced in model (2):


X
∑  ∑ wij  β i = ∑ i 0 β i = 0
i  j
i X0

X j0


γ j =0
∑  ∑ wij  γ j = ∑
j  i
j X0

so that expression (1) and model (2) are identical because, in this case, α̂ = r . Hence, the equation to
be estimated is the following:
I −1  X
 J −1  X
X
X
rij = α + ∑ β i'  0 1i − 0 1I  + ∑ γ 'j  0 1 j − 0 1J
X I 0  j =1  X j 0
XJ 0
i =1  X i 0

 + ε ij


(3)
Determining the significance of both composition and geographical effects only needs the use
of the Fisher statistics to test the null hypotheses β1′ = β 2′ = L = β I′ −1 = 0 and γ 1′ = γ 2′ = L = γ ′J −1 = 0 ,
respectively. In addition, we can also calculate the composition and geographical effects for each type
of spatial category and examine their respective significance, what is our main objective. Composition
2
Knudsen and Barff (1991) also suggested three alternative shift-share linear models. All give qualitatively
identical results.
7
and geographical effects for each spatial category can be calculated as the following linear
combinations of the estimates βˆi and γˆ j from the model (2), respectively:
X ij 0 ˆ
βi
i X j0
sj = ∑
g j = γˆ j
and
Thus, we can calculate their variance and determine whether these effects are significant or not.
For avoiding some possible biases in examining the respective role of sectoral composition and
urban-rural gradient on local growth, we introduced two other factors in the model: the regional effect
and the effect of the urban centre size (called by ‘size effect’ in what follows) in order to control for
heterogeneity of urban centres and regions. Hence, model (2) is extended as follows:
rijkl = α + β i + γ j + δ k + φl + ε ijkl
(4)
where rijkl is the employment of sector i located in the category of area j (j=urban centre, urban fringe,
rural areas) with a size l and in the region k whereas δ k and φ l are the regional effect and the ‘size
effect’, respectively. Applying the same approach, this model is weighted by wijkl = X ijkl0/ X 0 with
the following constraints:
Xi0
∑X
i
βi = 0 ,
0
By setting βi =
∑
j
X j0
γ =0,
X0 j
∑
k
Xk0
δk = 0,
X0
X
X0
X
X
β i′ , γ j = 0 γ ′j , δ k = 0 δ k′ , φl = 0 φl′
X
X i0
Xk0
X l0
j0
X l0
∑X
l
φl = 0 .
0
and introducing these
constraints in (4), the equation to be estimated becomes:
I −1  X

 J −1  X
X
X
rijkl = α + ∑ β i'  0 1i − 0 1I  + ∑ γ 'j  0 1 j − 0 1J 
X I 0  j =1  X j 0
X J 0 
i =1  X i 0
K −1
 X
 L −1  X

X
X
+ ∑ δ k'  0 1k − 0 1K  + ∑ φl'  0 1l − 0 1 L  + ε ijkl
XK0
X L0 
k =1
 Xk0
 l =1  X l 0
(5)
As previously, we can test the level and the significance of four effects at overall sample level by
implementing Fisher tests and for each spatial category (urban/rural, for instance) with the following
expression:
8
X ji 0 ˆ
X jk 0 ˆ
X jl 0 ˆ
r j = αˆ + ∑
βi + ∑
δk + ∑
φl +
γˆ j
{
i X j0
k X j0
l X j0
14243 14243 14243 geographic al
sectoral effect
regional effect
size effect
(6)
effect
where the geographical effects correspond to the urban-rural gradient effect.
4. DATA AND NOMECLATURES
USED
The data used are taken from the 1990 and 1999 population censuses (quarter sample)
conducted by the French national statistics institute (INSEE). This database can be used to evaluate the
local employment by sector by counting the working population in employment at their work place. In
contrast to other data sources often used for examining sectoral location or local growth in France, the
population census is comprehensive both for the sectors covered by the analysis and for the types of
employment taken into account. The census data are put together and yield the 1990-99 changes in
number of jobs by sector and by spatial category. We now examine the nomenclatures and the
divisions used.
4.1.
Spatial delineations
For geographical divisions, we used the 1990 delineation proposed by the INSEE and the
INRA called Urban and Rural Area Zoning (Schmitt and Perrier-Cornet, 1998). It divides France into
five categories of area:
(i)
Urban centres: Urban units with more than 5,000 jobs (361 centres in all), corresponding to
almost all urban units of more than 20,000 inhabitants and to a fifth of those numbering 5,000 to
20,000 inhabitants,
(ii) Periurban fringes: Communes (i.e. French municipalities) from where at least 40% of the resident
working population commute to urban centres,
(iii) Rural areas under urban influence: communes from where between 20% and 40% of the resident
working population commute to urban areas (i.e. urban centres and their periurban fringes),
(iv) Rural centres and their fringes: communes or urban units providing at least 2,000 jobs communes
where more than 20% of the resident population work in rural centres,
(v) Remote rural areas: all the other communes or urban units that are communes free from influence
9
urban or rural centres.
The two first spatial categories together make up the predominantly urban areas, which are
contrasted with the predominantly rural areas, corresponding to the three last categories. Main
characteristics of these five spatial categories are provided in Table 1.
[Table 1 about here]
To allow for possible ‘regional bias’ due to the specialization of French regions and to their
specific dynamics, we used a spatial delineation introducing a ‘regional’ dimension into the analysis.
Indeed, between 1990 and 1999, 11 out of the 22 French administrative regions (European Union
NUTS 2) had a positive employment growth rate (the maximum, +17%; is obtained by Brittany) while
the others regions had a negative one (the minimum is -6% for Ile -de-France and Champagne).
The third delineation we used aims to allow for the heterogeneity of urban centres and its
consequences on the four other spatial categories. Six types of urban centres were distinguished in
relation to their employment size: centres smaller than 10,000 jobs; centres with 10,000–20,000 jobs;
centres with 20,000–50,000 jobs; centres with 50,000–100,000 jobs; centres greater than 100,000 jobs
(except Paris); and, finally, Paris. Further, the periurban fringe communes were linked to the urban
centre to which most of their working population commutes while the communes belonging to the
three categories of rural areas were related to the nearest urban centre (evaluated through the shortest
distance as the crow flies between the commune and the boundary of an urban area).3 Then, all
periurban fringes and rural areas were subdivided in the six foregoing categories in relation to the
employment size of their corresponding urban centre.
4.2.
Industrial nomenclatures
The changes made to the sectoral nomenclatures in France in 1993 make it an intricate matter
to match the 1999 and 1990 census data at a fine scale. The INSEE built such a connection but the
sectoral data derived can only be used for aggregated levels of sectoral and geographical divisions.
The usable sector-based aggregation (NES36) distinguishes 17 manufacturing sectors4 and 4 business
3
Bessy et al. (2001) constructed this subdivision in order to identify ‘rural hinterlands of urban centres’.
The manufacturing sectors are: Food product, beverage and tobacco; Dressing and leather; Publishing, printing
and reproduction of recorded media; Pharmaceutical industry; Household equipment; Automotive; Other
4
10
service sectors.5 But these data are reliable only for areas greater than 10,000 jobs. The exclusion of
areas smaller than 10,000 jobs in 1990 removes a non negligible number of observations: 2,054 out of
7,440 observations when looking at manufacturing sectors alone and 2,566 out of 9,224 observations
for manufacturing sectors and business services. However, this restriction does not affect the results.
Finally we performed two shift-share analysis with two sets of sectors. The first set covers the
17 manufacturing sectors while the second set includes also four business service sectors. We included
the employment change in business services in the analysis to take into account the outsourcing by
manufacturing firms of some of their service jobs. Although accounted for in the manufacturing
sectors at the beginning of the period (in 1990), these jobs are assigned to the business services at the
end of the period (in 1999). In other words, a non-negligible share of jobs that were classified in 1990
as belonging to industry were therefore reclassified in 1999 as being in the business services sector.
Comparing results obtained without the business services with those obtained with them allows us to
analyse the consequences of the industrial outsourcing on the spatial difference in employment
growth.
5. EMPIRICAL RESULTS
Prior to give the results of the shift-share analysis, we will present some descriptive statistical
features to examine the main tendencies of urban-rural employment growth in France.
5.1 Some preliminary results: beyond the urban agglomeration, a recent industrial employment
spread to periurban fringes and rural areas
Table 2 shows some features of descriptive statistics. They confirm that in France as in the
other major industrialised countries manufacturing employment is concentrated in urban centres: 63%
of manufacturing jobs are located there in 1999. However, this category of jobs is less concentrated
than the total industrial employment (i.e. manufacturing and business service sectors): 71% of total
industrial employment is located in urban centres in 1999. Furthermore, their degree of concentration
transport equipment; Mechanical equipment; Electrical machinery and apparatus; Mining and quarrier; Textiles;
Wood and papers; Chemical, rubber and plastics products; Basic metals; Electrical and electric components;
Fuel; Water, gas and electricity.
11
tends to decline between 1990 and 1999. This relative decentralization is the result of the growth gap
between the urban decline of manufactured employment (-16% in 9 years), its growth in periurban
fringes (+5%) and its moderate decline in rural areas (-6%). The share of rural manufactured
employment then rose from 23 % to 25% and that of periurban employment from 10% to 12%
between 1990 and 1999. In addition, when we include business services that are very expanding in all
spatial categories, the growth rate of industrial employment remains negative in urban centres and it
becomes positive in rural areas.
As also shown in Table 2, each industry is affected by this shift in relative balance of
industrial employment between urban areas and rural/periurban ones. However, this shift in balance
occurs at different rates and affects industries with unequal initial degrees of concentration.
[Table 2 about here]
Indeed, the agro-food industries, sectors where the fall in employment is the lowest at national
level, are the least spatially concentrated industrial sectors. Additionally, over the period 1990-1999,
employment of these sectors becomes less agglomerated. The strong employment growth in these
sectors in rural and periurban fringes, +7% and +14% respectively, contrasts with its 10% fall in urban
areas. As a result, in 1999, the distribution of agro-food industries between urban areas and rural and
periurban fringes is close to 50–50%. Somewhat less concentrated at the beginning of the period,
intermediate goods industries (which account for 35% of manufactured employment nationwide) also
become deconcentrated, although a little slower than the agro-food industries. The national decline of
employment observed in these sectors is faster in urban areas (-18%) than in the rural ones, where
employment still fell by 7% over the period. The growth in periurban fringes is limited to 5%.
Accordingly, the weight of urban centres in the spatial employment distribution of these industries
declined from 60% to 57% between 1990 and 1999.
Historically very concentrated industrial sectors (equipment goods industries, automobile
manufacturing, and energy industries) follow a comparable decentralization but it does not bring into
question their high degree of agglomeration. These sectors account respectively for 20, 6.5, and 6% of
5
The business service sectors are: Post and telecommunications; Consultancy and others; Renting without
operator; Research and development
12
national manufactured employment. Behind a national decline of the same magnitude, decentralization
in these sectors is not based on exactly the same factors. The equipment goods industries and, to a
lesser extent, the energy sector, follow the ‘classical’ scheme of decentralisation: the fall in
employment is more intense in urban areas (-16% for equipment goods industries) than in rural ones
(-1%) together with a spectacular periurban growth (+11%). By contrast, the automobile industry
recorded a 6.5% fall in periurban employment and a 6% rise in rural employment.
Further, the consumption goods industries, another important industrial sector that accounts
for 18% of manufactured employment in France, are a little more concentrated than the average.
However, they have undergone a stronger geographical decentralisation of their employment than the
other sectors. The decline, which has been large everywhere, reached 20% in urban and rural areas
alike, this decline in the periurban fringes is just 4% in 9 years.
Finally, employment in business services (around 40% of the total industrial employment in
France) is geographically very concentrated. Their growth rate is positive in each spatial categories.
However, the growth rate achie ves its higher values in periurban fringes (+72%) and in rural areas
(+48%) and its lowest value in urban centres. Despite this, the employment in business services keeps
agglomerated in this last spatial category. In 1999, around 83% of jobs in this sector are located in
urban centres. However, their inclusion within industrial sectors does not challenge the observation
made earlier of employment decentralization in these sectors. The 3% fall in such employment in
urban centres goes along with fast growth in periurban fringes (+18%) and with stability or moderate
growth in the different rural categories. This can be explained by the fact that the two largest business
activities (consultancy and others and renting without operator that make up 80% of business service
jobs) have clearly higher growth in employment outside the urban centres as opposed to within urban
centres. The post & telecommunications and research & development sectors behave in a different
way: the fall in jobs in the former affects both urban and rural centres and the remote rural areas while
the latter continues to have a strong tendency toward concentration.
So, overall, industrial employment in France has undergone relative dispersion with a
combination of a spread of industrial employment from urban areas to their periurban fringes and a
relative redistribution of employment between predominantly urban and predominantly rural areas.
13
The latter phenomenon is the result of differences in the rates of decline in employment. An analysis
by sectors confirms this trend and also brings out a spatial differentiation in the rates of sectoral
change, which follow identical trends within each spatial category. This may suggest the existence of
positive geographical effect in periurban fringes and rural areas and a systematically negative one in
urban centres. Nevertheless, results are much less clear when it comes to the existence of possible
sectoral effect. The shift-share analysis whose results are set out below should allow us to analyse this
latter effect and to measure the true significance of the former.
5.2 Shift-share analysis allowing for three composition effects (sectoral, regional & urban centre
size) and for the urban-rural gradient as geographical effect
Table 3 gives the Fisher statistic s obtained for the first shift-share analysis. They allow us to
examine the overall level of significance of each factor included in the analysis. Although each factor
acts on the local employment growth, a hierarchy appears between them. The sectoral effects and the
geographical effects (i.e. the urban-rural gradient) largely dominate both regional and urban centre
size effects. Further, this hierarchy holds for both sets of sectors (only manufacturing sectors and with
the business services). The overall significance of the first two factors is even greatly intensified when
business services are included, and the sectoral effect becomes massively predominant. While we
might expect the strong impact of the sectoral effects on the local growth, the predominance of the
urban-rural gradient on the regional effects is an important result: the local economic growth depends
mainly on the urban, periurban or rural characteristics of the areas and slightly on the other features of
the region at which the area belongs. This result is consistent to Bessy et al. (2001), who showed that
the population growth in French rural areas mainly depends on the distance to and growth of the
nearest urban centre rather than the other regional characteristics.
[Table 3 about here]
When examining the break-down of local employment growth by the four factors (Table 4),
the sectoral effects appear to be significantly positive in urban areas and significantly negative in rural
areas. However they are offset by geographical effects that are significantly negative in urban areas
and significantly positive in the periurban fringes and in rural areas. In other words, the increase in the
14
share of industrial employment in the three types of rural areas then appears to be due to a
significantly positive geographical effect (whatever the rural category) that partly offsets
systematically significant and negative sectoral effects. Hence, rural areas are characterised by a
specialisation in industrial sectors having experienced more intense decline in terms of jobs at national
level. But they provide specific advantages thereby offsetting the drawback of the preferentially rural
location of declining sectors. This means that rural industrial employment has fallen at a slower rate
than the national average. An opposite phenomenon seems to be at work in the urban centres: the fall
in their industrial employment is due to a negative geographical effect that is not offset by the slightly
positive sectoral effect. Thus, despite the urban concentration of the faster growing industrial sectors at
national level (in terms of changes in employment), urban centres seem to offer substantial
disadvantages that are sufficiently intense to offset the positive sectoral effect.
[Table 4 about here]
The case of the periurban fringes is a little different: the change in their industrial employment
is only affected by an intense positive geographical effect, while the sectoral effect is non-significant
(when focusing on manufacturing sectors) and negative (when extending industrial employment to
business services). These areas therefore seem quite indifferent to the type of sectors they
accommodate and to their composition changes. On the other hand, they seem to provide intense
advantages able to attract firms whatever their belonging sector.
These results confirm the urban/rural differences in the two main dimensions studied here.
First, rural areas enjoy intrinsic advantages able to attract any industry whereas the urban centres
suffer from symmetrical disadvantages. Se cond, the more dynamic industrial sectors at national level
are located mostly in urban centres, whereas the industries that are in the worst decline are overrepresented in the predominantly rural areas. Third, the periurban fringes are fundamentally opposed
to the urban centres. They look more like the rural areas in the size of the geographical advantages
they procure, but less like them in terms of the low specificity of the sectors locating there.
Extending our industry definition in including business services generally intensifies these
effects (more so for sectoral effects than for geographical effects). This reinforces the positive urban
sectoral effect without changing the negative urban geographical one and the negative rural sectoral
15
effect as well as the positive geographical effects in rural areas and periurban fringes. Furthermore the
non-significant periurban sectoral effect becomes significantly negative. The reinforcement of both
sectoral and geographical effects could be interpreted in different ways. On the one hand, outsourcing
gives firms that take on such activities a new autonomy in terms of location. Their generally
innovative character associated with the search for new outlets and for scale economies may reinforce
their tendency to agglomerate. This should intensify the sectoral effects (positive in the urban centres
and negative in rural areas). On the other hand, it may be thought that outsourcing is differentiated not
only by sector but also within a sector depending on the firm location. Indeed, within a given sector, a
firm located in an urban centre can outsource some of its activities more easily as it finds a denser and
more diversified supply of services locally. For opposite reasons, rural firms will tend to maintain a
larger number of such functions within the firm. Obviously, they could buy services from providers
located in urban centres, but all else being equal, the costs related to distance imply a less incentive to
outsourcing. This may explain the increase of the posit ive geographical effect observed in rural areas.
Finally, observe that the regional effects have a negative impact on urban centres (that’s, the
regions experiencing the least favourable change in industrial employment seem to be urban regions)
and a positive impact on the three categories of rural areas (that’s, regions experiencing the least
unfavourable change in industrial employment seem to be rural ones). The same type of phenomenon
is found for the urban centre size effect.
All these results first suggest that among the factors explaining local employment growth,
local sectoral structure and the urban versus rural nature of places play stronger role than
characteristics of the belonging region. They also reveal an opposition between rural areas and urban
centres. The former tends to attract preferentially the more declining sectors in terms of employment,
such as agro-food or intermediate goods industries, sectors which, in France, are known to be labourintensive and/or competitive. It may be that these sectors find advantageous production conditions in
these areas, such as lower labour costs or lower land rents. Alongside this, the specific environment of
rural areas seems favourable to the relative development of industrial employment. This might be
interpreted as the existence of positive dynamic externalities as a result of low densities, as shown by
Kim et al. (2000) from US data. At the opposite and consistently with the theoretical results obtained
16
by Duranton and Puga (2001), the faster growing sectors in terms of employment set up historically in
urban centres so as to benefit from the advantages of agglomeration, such as the presence of
diversified inputs and workers, the possibility of sharing information that can only be exchanged by
face-to-face communication, and so on. However, the recent conditions of change in the firm’s set up
in urban centres have become highly unfavourable. This may be the effect of land rents that have
become prohibitive for some firms. This is an incentive to relocate part of their production. They seem
to do this, largely, toward the periurban fringes, as suggested by the very large positive geographical
effect in the periurban areas. As shown by Delisle and Lainé (1998), this decentralisation of industrial
employment around urban centres largely occurred in France by relocation of plants. This can be
explained, on the one hand, by lower land rents and by lower labour costs in the periurban fringes (as
shown by Ota and Fujita, 1993; or Fujita et al., 1997) and, on the other hand, by the indirect effects of
population spread (Schmitt and Henry, 2000).
5.3 Shift-share analysis with geographical effects combining the urban-rural gradient and the size
of the influencing urban centre
Despite the low level of overall signific ance of the urban centre size effect, we tested a
geographical effect constructed as the combination between our urban-rural categories and the size of
the urban centre to which they are related. Urban centre size was combined on the one hand with each
of the five spatial categories and, on the other hand, with the only urban centre and periurban fringe
one.
Table 5 gives the Fisher test results for both shift-share analyses. It shows that allowing for a
geographical effect combining urban-rural gradient and size categories of the urban centre of influence
has little effect on the overall level of significance of the remaining composition effects, the sectoral
and regional ones. On the other hand, such allowance does not significantly reduce the overall
significance of the geographical effect. However, this effect remains significant and stronger than the
regional effect.
[Table 5 about here]
The detailed results of the shift-share analysis are reported in appendix (Table A1 and A2).
17
For rural areas, results are qualitatively similar: the sectoral effects remains significantly negative
whereas the geographical effects remain positive despite a fall in the significance (especially in rural
centres and remote rural areas).
In the case of urban centres, previous results evolve .The sectoral and geographical effects
obtained for the urban centres (and their periurban fringes) according to the urban centre size exhibits
a contrast between large and small urban centres (see Figure 1). Indeed, the sectoral effect remains
positive for urban centres of more than 100 000 jobs while it becomes negative or non-significant for
the smallest ones. The geographical effect remains negative for the largest urban centres but it
becomes significantly positive or non-significant for the smallest ones. In other words, the smallest
urban centres exhibit an employment growth pattern close to this observed in rural areas and only
largest cities combine positive sectoral effect and negative geographical one.
[Figure 1 about here]
A quite similar opposition is observed for the sectoral effect in periurban fringes: periurban
fringes related to smallest urban centres have a strong negative sectoral effect while those related to
largest cities exhibit a positive one. This means that sectoral structure in the periurban fringe is close
to this one of the related urban centre. But this opposition between periurban fringes according to the
size of their related urban centre does not affect the level of employment growth because the
geographical effects in periurban fringe remain highly positive whatever the size of the related urban
centres and largely dominate the sectoral effects.
These results confirm our previous analysis. In addition to the opposition between urban and
rural areas, there exists an opposition between large cities and small cities in terms of sources of local
growth. High densities seem to involve static spatial externalities that historically attracted firms
belonging to faster growing sectors at national level while low dens ities involve dynamic spatial
externalities acting as an engine of employment growth.
6. Summary and implications for future research
By performing an estimable stochastic formulation of shift-share analysis on French
employment data (1990–1999), we have shown that the increase in the share of rural industrial
18
employment in national industrial employment may be due to geographical advantages specific to
low-density areas and liable to attract any industrial sector. These ‘rural’ advantages are far from
negligible because they offset preferentially rural location of firms belonging to the declining sectors
(at national level). These geographical advantages also concern periurban fringes. It is worth stressing
that geographical advantages expressed in periurban fringes and in rural areas differ. The former are
characterized by a geographical position allowing them to benefit from advantages of urban
agglomeration without supporting its costs; the latter are too remote from cities to be able to benefit
directly from such advantages. Conversely, urban centres, especially the largest ones, are affected by
important geographical disadvantages (to be related to the urban costs they incur), while the
favourable character of the sectoral structure fails to offset such local disadvantages. In addition, we
highlight the other regional characteristics (other than their urban-rural pattern) have a weak influence
on the local growth.
Further, in order to understand the local growth factors, we first need to focus on the basis of
the sectoral distribution of industry between small urban centres and large cities as well as between
urban and rural areas. It has been shown that the growing industrial sectors are located preferentially
in urban areas, whereas the more fragile industrial sectors are located rather in rural areas. This is
consistent with the idea that the most innovative sectors, the ones that create most jobs and have more
intensive demand for a highly skilled workforce, begin by locating in cities and then relocate in
periurban fringes or in more peripheral areas as the production processes become standardised. To
account for this phenomenon, we can drawn on the theoretical frameworks of economic geography as
developed by Krugman (1991) or by Ottaviano et al. (2002) as did Kilkenny (1998) or Duranton and
Puga (2001).
A second important line of inquiry suggested by our results is to work on a more precise
picture of the specific rural features favouring the dynamics of these low-density areas. It is thought
that a rural location allows firms to exploit ‘economies of small agglomeration’ which are not found in
urban settings. Without calling into question the existence of urban externalities that grow in intensity
with the firm density, this result tends to show that low-density areas may also be affected by such
dynamic externalities or may be the site of other dynamic externalities. There is always scope to
19
investigate what these ‘specifically rural advantages’ are. Our findings also show that the net effects of
agglomeration economies may become negative over a certain threshold as increased urban costs
progressively offset the agglomeration economies.
While periurban areas are distinctive too because of the growth of their industrial employment
related to specific advantages, these advantages are probably not the same kind as those found in rural
areas. Naturally, the role of land rent comes to mind here. Industrial sectors attracted by urban areas
may find an incentive to set up on the fringes of urban centres because such areas offer non-negligible
advantages in terms of land rents and their proximity to urban centres allows them to benefit from the
advantages of agglomeration while avoiding its costs to some extent. This calls for specific analyses of
the decentralization around urban centres of industrial employment with a special focus on the role of
the land market. Note that little work firms have been done in this field (Fujita et al., 1997; Cavailhes
et al., 2004).
Acknowledgements : The authors are grateful to thank Hubert Jayet and Maureen Kilkenny for very
helpful comments and suggestions. We also thank participants of the North-American RSAI meetings
at Philadelphia and of research seminars at LEA-Paris, at GREMAQ-Toulouse and at GdR ASPEParis.
20
References
Abdel-Rahman H, Fujita M, 1990, “Product Variety, Marshallian Externalities and City Sizes”
Journal of Regional Science 30 165 - 183
Audrescht D B, Feldman M P, 2004, “Knowledge Spillovers and the Geography of Innovation”, in
Handbook of Regional and Urban Economics (vol. 4) Eds V. Henderson, J. Thisse, pp 2713 2739.
Berzeg K, 1978, “The Empirical Content of Shift-share Analysis” Journal of Regional Science 18 463
- 469
Bessy-Pietry P, Hilal M, Schmitt B, 2001, “Des évolutions démographiques rurales contrastées liées à
la proximité et aux dynamiques urbaines : le cas de la France” Espace, Populations, Sociétés
1-2 19 - 36
Bryden J, Bollman R, 2000, “Rural Employment in Industrialised Countries” Agricultural Economics
22 185 – 197
Bollman R, Terluin I, Godeschalk F, Post J, 2005, “Comparative Analysis of Leading and Lagging
Rural Regions in OECD Countries in the 1980s and 1990s”, paper presented at the 45th
Congress of the European Regional Science Association (ERSA), Vrije Universiteit
Amsterdam, 23-27 August.
Cavailhes J, Gaigné C, Thisse J F, 2004, Trade Costs versus Urban Costs. Do Jobs Move to the
Suburbs or to the Sticks? CEPR Discussion Paper.
Champion A, 1989, Counterurbanization, (Ed Arnold, London).
Combes P P, 2000, “Economic Structure and Local Growth: France, 1984-1993” Journal of Urban
Economics 47 329 - 355
Delisle P, Lainé F, 1998, “Les transferts d’établissements contribuent au desserrement urbain”
Economie et Statistique 311 91 - 106
Duranton G, Puga D, 2001, “Nursery Cities: Urban Diversity, Process Innovation and the Life Cycle
of Products” American Economic Review 91(5) 1454 - 1477
Duranton G, Puga D, 2004, “Micro-foundations of Urban Agglomeration Economies”, in Handbook of
Regional and Urban Economics (vol. 4), Eds Henderson V., Thisse J.F., pp 2063 – 2117.
Esteban J, 2000, “Regional Convergence in Europe and the Industry Mix: A Shift-share Analysis”
Regional Science and Urban Economics 30 353 - 364
Fujita M, 1989 Urban economic theory (Cambridge University Press, Cambridge MA)
Fujita M, Thisse J F 2002, Economics of Agglomeration (Cambridge University Press, Cambridge
MA).
Fujita M, Thisse J F, Zénou Y, 1997, “On the Endogenous Formation of Secondary Employment
Centres in a City” Journal of Urban Economics 41 337 - 357
Gaschet F, 2002, The New Intra-urban Dynamics: Suburbanisation and Functional Specialisation in
French Cities Papers in Regional Science, 81(1) 63 - 81
21
Glaeser E, Kallal H, Scheinkman J, Shleifer A, 1992,” Growth in Cities” Journal of Political Economy
100 1126 - 1152
Gramlich E, 1994, “Infrastructure Investment: A Review Essay” Journal of Economic Literature 32
1176 - 1196
Helsley R W, Strange W C, 1990, “Matching and Agglomeration Economies in a System of Cities”
Regional Science and Urban Economics 20 189 - 212
Henderson V, Kuncoro A, Turner M, 1995, “Industrial Development in Cities” Journal of Political
Economy 103 1067 - 1090
Holmes T, Stevens J, 2004, “Spatial Distribution of Economic Activities in North America” in
Handbook of Regional and Urban Economics (vol. 4), Eds Henderson V., Thisse J.F., pp 2797
- 2843.
Jayet H, 1993 Analyse spatiale quantitative (Economica, Paris)
Kilkenny M, 1998, “Transport Costs and Rural Development” Journal of Regional Science 38 293 312
Kim Y, Barkley D, Henry M S, 2000, “Industry Characteristics Linked to Establishment
Concentrations in Nonmetropolitan Areas” Journal of Regional Science 40(2) 231 - 259
Knudsen D, Barff R, 1991, “Shift-Share Analysis as a Linear Model” Environment and Planning A
23 421 - 431
Krugman P, 1991, “Increasing Returns and Economic Geography” Journal of Political Economy
99 483 - 499
Krugman P, Venables T, 1995, “Globalization and the Inequality of Nations” Quarterly Journal of
Economics 110(4) 857 - 880
Marshall A, 1890 Principles of Economics (Macmillan, London)
Mokyr J, 1995, “Urbanization, Technological Progress and Economic History” in Urban
Agglomeration and Economic Growth Ed H Giersch (Springer Verlag, Berlin) pp 3 - 37
Ota M, Fujita M, 1993, “Communication Technologies and Spatial Organization of Multi-unit Firms
in Metropolitan Areas” Regional Science and Urban Economics 23 695 - 729
Ottaviano G, Tabuchi T, Thisse J F, 2002, “Agglomeration and Trade Revisited” International
Economic Review 43 101 - 127
Schmitt B, Henry M S, 2000, “Size and Growth of Employment Centres in French Labor Market
Areas: Consequences for Rural Population and Employment” Regional Science and Urban
Economics 30 1 - 21
Schmitt B, Perrier-Cornet P (eds), 1998 Les campagnes et leurs villes (INSEE, Paris)
22
Table 1 - Some characteristics on urban centres, periurban fringes and rural areas
Number of
communes
Land area
(%)
Population
(N/1000)
(%)
Employment
(N/1000)
(%)
Urban centres
Periurban fringes
Rural under urban influence
Rural centres
Remote rural areas
2,793
10,431
8,887
3,528
10,926
7.4
22.1
23.8
9.6
37.1
35,217
9,674
5,307
3,284
5,037
60.2
16.5
9.1
5.6
8.6
16,164
2,249
1,401
1,310
1,676
Total
36,565
100.0
58,518
100.0
22,800 100.0
70.9
9.9
6.1
5.7
7.4
Sources: 1990 and 1999 Population Censuses
Table 2 - Industrial employment: break-down and recent changes by sector and category of areas
Agro
food
Consumption Automobile Equipment Intermediate Energy Manufactured Business
Total
goods
goods
goods
sectors
services industry
Urban
centres
Jobs a
Share b
Growthc
309
49.7
-10.0
512
68.1
-21.3
199
74.0
-14.7
605
73.7
-15.5
816
56.5
-17.6
186
76.3
-11.1
2,628
63.3
-16.4
2,321
82.9
+18.9
4,949
71.2
-2.9
Urban
fringes
Jobs a
Share b
Growthc
83
13.3
+13.6
68
9.1
-3.8
39
14.5
-6.5
92
11.2
+10.7
197
13.7
+4.9
14
5.7
+4.8
493
11.9
+5.0
193
6.9
+72.2
686
9.9
+17.9
Rural areas Jobs a
b
under urban Share
influence Growthc
71
11.4
+9.3
48
6.4
-11.7
12
4.5
+14.0
40
4.8
+2.9
134
9.3
-7.7
12
4.8
+5.2
317
7.6
-2.6
90
3.2
+61.9
407
5.9
+6.9
Rural
centres
Jobs a
Share b
Growthc
64
10.2
+1.8
63
8.3
-22.2
12
4.5
+7.0
46
5.6
-3.9
147
10.2
-8.0
20
8.0
-4.1
350
8.4
-8.2
101
3.6
+43.0
451
6.5
-0.3
Remote
rural areas
Jobs a
Share b
Growthc
96
15.4
+9.3
61
8.1
-22.9
7
2.5
-8.3
39
4.7
-2.1
150
10.4
-6.4
13
5.2
-6.7
364
8.8
-5.8
94
3.3
+42.2
458
6.6
+1.2
Total
Jobs a
Share b
Growthc
623
100.0
-1.5
752
100.0
-19.6
268
100.0
-11.6
821
100.0
-11.2
1,443
100.0
-12.1
244
100.0
-8.9
4,152
100.0
-11.8
2,800
100.0
+24.0
6,952
100.0
-0.2
a: 1999 number of jobs in thousands ; b: 1999 share in sectoral employment in % ; c: 1990-99 growth rate.
Sources: 1990 and 1999 Population Censuses
23
Table 3 - Fisher Statistics for shift-share analysis with four factors
Manufacturing
sectors only
Manufacturing
sectors & Business
services
43.81
5.27
4.32
53.58
168.78
6.56
7.33
86.76
Sectoral effect
Regional effect
Urban centre size effect
Geographical effect (Urban-rural gradient)
Table 4 - Composition effects and geographical (urban-rural gradient) effect
Local
Sectoral
growth rate effect
Regional
effect
Manufacturing sectors (national growth rate = -0.121)
Urban centres
-0.164
0.008 *** -0.007
Periurban fringes
0.051
-0.002
-0.001
Rural under urban influence -0.017
-0.021 *** 0.027
Rural centres
-0.085
-0.030 *** 0.015
Remote rural areas
-0.057
-0.029 *** 0.028
**
***
***
Size
effect
-0.005
-0.007
0.014
0.022
0.025
Geographi
-cal effect
**
-0.039 ***
0.183 ***
0.084 ***
0.029 *
0.040 **
***
**
***
Manufacturing sectors & Business services (national growth rate = -0.004)
Urban centres
-0.029
0.028 *** -0.007 ** -0.007 ***
Periurban fringes
0.181
-0.053 *** 0.003
-0.004 *
***
***
Rural under urban influence
0.079
-0.101
0.032
0.024 **
***
**
Rural centres
-0.005
-0.102
0.023
0.035 ***
***
***
Remote rural areas
0.012
-0.111
0.035
0.039 ***
-0.040 ***
0.240 ***
0.128 ***
0.043 **
0.053 ***
*, ** and ***: significant respectively at 1%, 5% and 10%
Table 5 - Fisher Statistics for shift-share analysis combining urban-rural gradient and urban centre size
Manufacturing
sectors only
Urban-rural gradient * urban centre size (including for rural areas)
Sectoral effect
44.23
Regional effect
5.32
Geographical effect (i.e. Urban-rural
9.79
gradient * Size of urban centre)
Manufacturing sectors
& Business services
169.72
6.58
15.99
Urban-rural gradient * urban centre size (for urban centres & periurban fringes only)
Sectoral effect
38.26
140.47
Regional effect
4.65
5.55
Geographical effect (i.e. Urban-rural
15.36
24.13
gradient * Size of urban centre)
24
Figure 1 - Sectoral and geographical effects for urban centres and periurban fringes
by urban centre size (in thousand of jobs)
1a. Urban centres
Manufacturing sectors only
Sectoral effect
Manufacturing sectors & Business services
Sectoral effect
Geographical effect
0,1
0,1
0,05
0,05
0
0
-0,05
-0,05
-0,1
Geographical effect
-0,1
<10
[10;20)
[20;50)
[50;100)
>100
Paris
<10
[10;20)
[20;50)
[50;100)
>100
Paris
1b. Periurban fringes
Manufacturing sectors only
Manufacturing sectors & Business services
Sectoral effect
Sectoral effect
Geographical effect
Geographical effect
0,4
0,3
0,3
0,2
0,2
0,1
0,1
0
0
-0,1
<10
[10;20)
[20;50)
[50;100)
>100
Paris
-0,1
<10
25
[10;20)
[20;50)
[50;100)
>100
Paris
Table A1. Composition effects and geographical effect
(urban-rural gradient combined with urban centre size for all categories of areas)
Manufacturing sectors only
Manufacturing sectors & Business services
(national growth rate = -0.121)
(national growth rate = -0.004)
Local
Sectoral Regional GeographiLocal
Sectoral Regional Geographigrowth rate
effect
effect
cal effect growth rate
effect
effect
cal effect
UC * S<10
-0.080
-0.013 *** 0.018 **
0.037 *
0.005
-0.064 *** 0.021 **
0.052 ***
**
***
UC * S:10-20
-0.121
-0.002
0.018
-0.016
-0.021
-0.035
0.012
0.006
UC * S:20-50
-0.151
-0.000
0.012
-0.041 *** -0.027
-0.025 *** 0.018 *
-0.016
UC * S:50-100
-0.130
-0.001
0.025 ** -0.032 *
-0.015
-0.006 *** 0.033 *** -0.038 **
***
***
***
UC * S>100
-0.136
0.017
0.038
-0.070
0.001
0.046 *** 0.031 ** -0.072 ***
***
***
*
UC * Paris
-0.269
0.023
-0.106
-0.065
-0.072
0.088 *** -0.079 ** -0.077 **
***
***
***
PF * S<10
0.181
-0.013
0.038
0.278
0.299
-0.104 *** 0.032 **
0.375 ***
***
**
***
***
PF * S:10-20
0.130
-0.028
0.026
0.253
0.215
-0.102
0.016
0.304 ***
***
*
***
***
**
PF * S:20-50
0.041
-0.014
0.017
0.159
0.148
-0.077
0.024
0.204 ***
***
***
PF * S:50-100
0.059
-0.002
0.007
0.175
0.161
-0.059
0.010
0.214 ***
***
***
***
***
**
PF * S>100
0.107
0.006
0.031
0.191
0.248
-0.041
0.029
0.263 ***
PF * Paris
-0.083
0.010 *** -0.094 *** 0.122 *** 0.105
-0.018 *** -0.069 *** 0.196 ***
RUI * S<10
0.029
-0.021 *** 0.041 *** 0.130 *** 0.146
-0.104 *** 0.044 *** 0.209 ***
***
*
RUI * S:10-20
-0.056
-0.020
0.021
0.064
0.037
-0.098 *** 0.010
0.128 ***
***
***
*
***
***
RUI * S:20-50
-0.045
-0.022
0.027
0.071
0.034
-0.107
0.038
0.106 ***
***
*
***
***
***
RUI * S:50-100 -0.000
-0.028
0.020
0.129
0.089
-0.105
0.036
0.162 ***
***
***
**
***
***
RUI * S>100
0.003
-0.020
0.045
0.100
0.115
-0.092
0.044
0.168 ***
***
**
***
**
RC * S<10
-0.071
-0.040
0.015
0.075
0.028
-0.115
0.025
0.121 ***
***
*
***
RC * S:10-20
-0.112
-0.021
0.021
0.009
-0.044
-0.086
0.021
0.025
RC * S:20-50
-0.090
-0.031 *** 0.026 **
0.036
-0.017
-0.103 *** 0.033 *** 0.057
RC * S:50-100
-0.074
-0.027 *** 0.023 **
0.052
-0.004
-0.103 *** 0.032 **
0.071
***
RC * S>100
-0.062
-0.029
-0.005
0.093
0.026
-0.100 *** 0.002
0.127 **
***
***
*
***
***
RR * S<10
-0.066
-0.028
0.033
0.050
0.002
-0.115
0.039
0.082 ***
***
**
***
*
RR * S:10-20
-0.089
-0.028
0.023
0.037
-0.017
-0.108
0.022
0.073 *
***
***
**
***
***
RR * S:20-50
-0.045
-0.032
0.029
0.080
0.027
-0.112
0.038
0.105 ***
***
***
**
***
***
RR * S:50-100
-0.006
-0.050
0.038
0.127
0.054
-0.132
0.051
0.139 ***
***
***
***
RR * S>100
-0.039
0.001
0.052
0.030
0.034
-0.080
0.054
0.064
Table A2. Composition effects and geographical effect
(urban-rural gradient combined with urban centre size for urban centres and periurban fringes only)
Manufacturing sectors only
Manufacturing sectors & Business services
(national growth rate = -0.121)
(national growth rate = -0.004)
Local
Sectoral Regional GeographiLocal
Sectoral Regional Geographigrowth rate effect
effect
cal effect growth rate effect
effect
cal effect
UC < 10
-0.080
-0.013 *** 0.018 **
0.036 *
0.005
-0.064 *** 0.021 *
0.052 **
*
***
UC = 10-20
-0.121
-0.001
0.018
-0.016
-0.021
-0.035
0.011
0.006
UC = 20-50
-0.151
-0.000
0.012
-0.041 ** -0.027
-0.025 *** 0.018
-0.016
UC = 50-100
-0.130
-0.001
0.026 ** -0.034 *
-0.015
-0.006 *** 0.034 ** -0.039 **
***
***
***
UC > 100
-0.136
0.017
0.038
-0.070
0.001
0.046 *** 0.030 ** -0.072 ***
***
***
UC * Paris
-0.269
0.023
-0.106
-0.065
-0.072
0.088 *** -0.080 ** -0.077 **
***
***
***
PF of UC<10
0.181
-0.013
0.038
0.277
0.299
-0.104 *** 0.033 **
0.374 ***
***
*
***
***
PF of UC=10-20
0.130
-0.028
0.024
0.255
0.215
-0.102
0.014
0.306 ***
***
***
***
*
PF of UC=20-50
0.041
-0.014
0.017
0.159
0.148
-0.077
0.024
0.205 ***
***
***
PF of UC=50-100
0.059
-0.002
0.008
0.174
0.161
-0.059
0.011
0.212 ***
***
***
***
***
**
PF of UC >100
0.107
0.006
0.031
0.191
0.248
-0.041
0.029
0.263 ***
***
***
***
***
***
PF of Paris
-0.083
0.010
-0.094
0.122
0.105
-0.018
-0.069
0.196 ***
***
***
***
***
***
RUI
-0.016
-0.022
0.031
0.096
0.080
-0.102
0.035
0.151 ***
***
*
**
***
*
RC
-0.084
-0.031
0.019
0.049
-0.004
-0.102
0.025
0.077 ***
RR
-0.056
-0.030 *** 0.032 *** 0.063 *** 0.014
-0.112 *** 0.038 *** 0.092 ***
*, ** and ***: significant respectively at 1%, 5% and 10% ;
UC= Urban centres; PF= Periurban fringe; RUI= Rural under urban influence; RC= Rural centres; RR= Remote rural areas;
UC<10: (UC<10,000 jobs); UC=10-20: (10,000 < UC < 20,000 jobs); etc.
26