SEARCH

SEARCH BY CITATION

Keywords:

  • Korea;
  • industry classification;
  • concentration;
  • agglomeration;
  • within-industry distribution pattern;
  • between-industry distribution pattern

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Measuring distribution of industries: a review
  5. Methodologies for measuring and classifying distribution patterns
  6. Data
  7. Correlations between estimated measures of spatial distribution of industries
  8. Industry classification based on spatial distribution patterns
  9. Conclusions
  10. Acknowledgements
  11. References

This study examines the spatial distribution of manufacturing activities in Korea to introduce spatial distributional aspects to industry classifications. One hundred and sixty two administrative units in mainland Korea with 180 manufacturing sectors were analysed for this purpose. In the main analysis, the spatial distribution and association of individual sectors are examined using concentration and agglomeration measures, and 180 sectors are then classified along with the values of these measures. The classification is conducted using factor analysis for identifying groups of industries with a common spatial distribution pattern. The analysis identifies five major industry groups in terms of their spatial distribution patterns.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Measuring distribution of industries: a review
  5. Methodologies for measuring and classifying distribution patterns
  6. Data
  7. Correlations between estimated measures of spatial distribution of industries
  8. Industry classification based on spatial distribution patterns
  9. Conclusions
  10. Acknowledgements
  11. References

The spatial distribution pattern of industry has received a lot of attention. For example, according to Duranton and Overman (2005), clusters have become popular because of the success of Silicon Valley. Many scholars are interested in explaining the benefits of spatial agglomeration, while many practitioners are interested in repeating the successes of elsewhere.

While studies have explored the spatial distribution characteristics of industries themselves, few studies have introduced these spatial perspectives to the industry classification. Classifying industries based on their locational tendencies is very useful for establishing more customised policies for different industry groups. For example, a certain industry's strong concentration implies that firms in this sector prefer locations where others in the same sector are concentrated (for example, because of the benefit of localisation economies; Ohlin 1933; Hoover 1937 1948). Therefore, a cluster of firms in this sector needs to be fostered in order for this industry to be successfully embedded in a target region. On the other hand, a certain industry's strong collocation or spatial association with other industries implies its preference for locations where others are concentrated (for example, because of the benefit of urbanisation economies; Ohlin 1933; Hoover 1937 1948). In this case, a cluster of firms in a diversity of sectors needs to be fostered.

The purpose of this study is to develop the procedure for formally categorising industries based on their spatial distribution patterns in a systematic way. This study first adopts four indices developed for measuring different aspects of spatial distribution (i.e. within-industry versus between-industry and concentration versus agglomeration). The estimated values of these measures are then used for identifying industry groups sharing similar locational characteristics through factor analysis. For this study, 162 cities and counties in Korea with 180 four-digit Korean Standard Industry Classification (KSIC) manufacturing sectors are analysed.

Measuring distribution of industries: a review

  1. Top of page
  2. Abstract
  3. Introduction
  4. Measuring distribution of industries: a review
  5. Methodologies for measuring and classifying distribution patterns
  6. Data
  7. Correlations between estimated measures of spatial distribution of industries
  8. Industry classification based on spatial distribution patterns
  9. Conclusions
  10. Acknowledgements
  11. References

A conventional way to measure the distribution patterns of industries is to calibrate the distribution indices. For example, Gini and Locational Gini (Krugman 1991) coefficients have been widely used for exploring the industry distribution pattern (e.g. Braunerhjelm and Borgman 2004; Sohn 2004a 2004b 2013; Wen 2004; Falcioglu and Akgüngör 2008; Nakamura 2008). Theil and Herfindahl indices are other examples of relatively classic indices often used for analysing distribution patterns (e.g. O'Donnellan 1994; Fan and Scott 2003; Sjöberg and Sjöholm 2004). However, Ellison and Glaeser's (1997) index is considered a more advanced concentration measure in that it measures pure concentration by controlling for the effects of random concentration and the difference in plant size. Subsequently, a number of studies have attempted to analyse the spatial distribution pattern of industries by adopting the Ellison-Glaeser (EG) index (e.g. Maurel and Sédillot 1999; Braunerhjelm and Borgman 2004; Devereux et al. 2004; Sjöberg and Sjöholm 2004; Bertinelli and Decrop 2005; Mori et al. 2005; de Dominicis et al. 2007; Lafourcade and Mion 2007; Holmes and Stevens 2004).

More recent analyses on spatial distribution patterns have adopted spatial statistical techniques. Unlike the indices discussed above, the spatial statistical measures consider the relative position in space or real geography (Arbia 2001a 2001b; Sohn 2004b; Lafourcade and Mion 2007; Guillain and Le Gallo 2010). Moran's I and Local Moran have been the representative spatial statistical indices used for the analysis of industrial distribution (Ying 2000; Rigby and Essletzbichler 2002; Sohn 2004a 2004b 2013; Chakravorty et al. 2005; de Dominicis et al. 2007; Lafourcade and Mion 2007; Guillain and Le Gallo 2010). More recently, the use of G-statistics (Getis and Ord 1992; Ord and Getis 1995) has increased (Sohn 2004b; Feser et al. 2005; Carroll et al. 2008).

Methods analysing the spatial association pattern of different industries are paid less attention than those analysing the distribution pattern of a single industry. In this category, a correlation coefficient is often used for analysing the association between industries (O'Donnellan 1994; Sohn 2004a 2013; Chakravorty et al. 2005). In some cases, rank correlation coefficient is used in order to eliminate the effect of outliers in the data (Bergsman et al. 1975). A more sophisticated index of spatial association is the measure of co-agglomeration developed by Ellison and Glaeser (1997). This is a between-industry version of the EG index and many empirical studies have adopted this measure for investigating the locational association patterns of different industries (e.g. Maurel and Sédillot 1999; Devereux et al. 2004; Bertinelli and Decrop 2005). In spatial statistics, Bivariate Moran is the spatial association index derived from Moran's I and, thus far, only a few studies have used it in characterising the spatial pattern of industries (e.g. Sohn 2013). The spatial correlation coefficient (Sohn 2004a) shares the idea with Bivariate Moran in terms of identifying the spillover effect to the spatially adjacent regions.

However, these studies are limited in explaining the spatial distribution pattern of industries for two reasons. First, spatial distribution patterns in many studies have not been tackled in a comprehensive way. Second, most studies have not provided information directly applicable to industrial policies. In an attempt to address these issues, this study establishes a framework that comprehensively incorporates different aspects of spatial distribution patterns of industries, and the systematic procedure for classifying industries based on the characteristics of their spatial distribution patterns.

Methodologies for measuring and classifying distribution patterns

  1. Top of page
  2. Abstract
  3. Introduction
  4. Measuring distribution of industries: a review
  5. Methodologies for measuring and classifying distribution patterns
  6. Data
  7. Correlations between estimated measures of spatial distribution of industries
  8. Industry classification based on spatial distribution patterns
  9. Conclusions
  10. Acknowledgements
  11. References

Spatial distribution pattern needs to be characterised from a range of perspectives. One dimension for characterising the spatial distribution pattern is whether we are interested in the distribution of a single industry or the spatial association between different industries. Within-industry patterns characterise the clustering of economic activities of a single industry. Between-industry patterns characterise the spatial association of economic activities between different industries. The distinction of these patterns or indices is useful in determining more dominant agglomeration forces between localisation and urbanisation economies (Ohlin 1933; Hoover 1937 1948).

Another dimension for characterising spatial distribution is whether or not geography is considered. In this case, concentration and agglomeration can be distinguished (Lafourcade and Mion 2007; Combes et al. 2008). Concentration shows the degree to which industries are clustered based on the values of individual spatial units. However, since this does not consider the geographic locations of the spatial units, it cannot detect the geographic pattern (Arbia 2001b; Sohn 2004b; Guillain and Le Gallo 2010). Agglomeration can determine the degree to which industries are geographically clustered by considering the values of individual spatial units in association with those of adjacent ones. However, this cannot consider the variability of the events (Arbia 2001b). For example, the distribution patterns in Figure 1a and b can be distinguished by agglomeration but not by concentration indices. Meanwhile, the distribution patterns in Figure 1b and c can be distinguished by concentration but not by agglomeration indices.

figure

Figure 1. Hypothetical distribution patterns of industries

Source: Arbia (2001b, 274–5)

Download figure to PowerPoint

The spatial scale of clustering is industry specific (Rosenthal and Strange 2003; Funderburg and Boarnet 2008). Therefore, it is important to use the index appropriate for the spatial scale of the industry of interest. For example, agglomeration indices are especially effective in detecting the spatial clustering when it exists on the spatial scale larger than a single spatial unit, while concentration indices are effective when it occurs at a few scattered spatial units. More recent studies reveal that agglomeration economies have a spillover effect on neighbouring areas (Viladecans-Marsal 2004; Van Oort 2007) and their spatial scales are expanding (Phelps et al. 2001; Phelps and Ozawa 2003; Burger et al. 2008). Although the distance-decay effects still hold in spillover (Rosenthal and Strange 2003; Van Oort 2007), these findings suggest that agglomeration indices become more important in the changing economic environment.

In this study, EG indices and Moran's coefficients are used for measuring concentration and agglomeration, respectively. These two indices are chosen because they provide between- as well as within-industry measures. We can maintain consistency between within- and between-industry measurements by using these indices. Table 1 summarises the two dimensions of spatial distribution and the specific indices used in the analysis. Although individual indices are not new, especially amongst recent work, a combination of these indices in this study enables us to understand a complex spatial distribution pattern of industries in a more systematic way.

Table 1. Dimensions and indices of spatial distribution used in the analysis
 Within-industryBetween-industry
ConcentrationEllison-Glaeser's γkEllison-Glaeser's γkl
AgglomerationMoran's IkBivariate Moran BMkl

Equation (1) presents γ (Ellison and Glaeser 1997) of industry k. This index measures the degree of geographic concentration (Gk) while controlling for the effects of random concentration and the difference in plant size distribution (Hk). If the industry has no concentration, E(γk) becomes 0. As this value rises, we expect stronger concentration. It is noted that plant Herfindahl Index of industry k, Hk, in equation (1) is estimated by the method proposed by Ellison and Glaeser (1997).

  • display math(1)

where

Gk =

measure of industry k's geographic concentration

Hk =

Herfindahl index of industry k's plant size distribution

xi =

the share of total employment in spatial unit i (i = 1, 2, … , M)

ski =

the share of industry k's employment in spatial unit i

zpk =

plant p's share of industry k's employment (p = 1, 2, … , N)

γkl is an inter-industry concentration index extended from γk (equation (2)). This measures the degree to which industries k and l are jointly concentrated (Gkl) while controlling for each industry's concentration (γk and γl) and plant size distribution (Hkl). The interpretation of the value is basically the same as γk. E(γkl) = 0 means that there is no joint concentration of plants between industries k and l than that attributable to the general tendencies of plants to locate near other plants (Ellison and Glaeser 1997). A positive γkl value indicates evidence of joint concentration.

  • display math(2)

where

Gkl =

measure of a geographic concentration of industries k and l as a whole

inline image =

weighted average of plant Herfindahl indices of industries k and 1

ωk =

employment share of industry k (ωk + ω1 = 1)

Moran's Ik in equation (3) (Anselin 1995) is used to complement the within-industry concentration index γk and to check the within-industry agglomeration pattern of industry k.1 This index measures the degree to which plants of industry k are spatially agglomerated by measuring the degree of spatial autocorrelation between plants of industry k (ski and skj) in terms of their spatial distributions across the spatial units (wij). With a large sample, we expect an agglomerated pattern with positive Ik and a dispersed pattern with negative Ik.

  • display math(3)

where

  • display math
wij =

element in the spatial weight matrix corresponding to the observation pair of spatial units i and j

Note that the measure of economic activity used for examining agglomeration pattern in equation (3) is ski–xi (Lafourcade and Mion 2007) to make the comparison of the outcomes between concentration and agglomeration indices more straightforward. First- and second-order contiguity-based spatial weight matrices are used for estimating Ik. The average distances of first- and second-order neighbours are 32.2 and 56.1 kilometres, which would be the approximate spatial scales of agglomeration to be examined.

Between-industry agglomeration is examined using Bivariate Moran in equation (4) (Anselin 2005). This index measures the degree of joint agglomeration of two industries by examining the spatial correlation between the plants in one industry in one spatial unit (ski) and those in another industry in another spatial unit (slj). Similar to Ik, positive coefficients suggest an evidence of joint agglomeration of two industries.

  • display math(4)

Four maps in Figure 2 show how these indices are complementary in explaining the spatial structure of the Korean economy. In the distribution maps of ski–xi with the same class intervals, sectors 1079 and 1419 show highly distinctive patterns. The former has low concentration, but high agglomeration, while the latter has high concentration, but low agglomeration. The former has more cities in the fifth class (above 0.0015) and many of them are spatially contiguous. The latter only has a small number of cities in the same class scattered in the country and a few of them have extremely high values (e.g. 0.551 of Seoul located in the northwest).

figure

Figure 2. Spatial distribution patterns of selected industries

Download figure to PowerPoint

Although fewer in number, there are sectors with both high concentration and agglomeration or low concentration and agglomeration. For example, sector 2911 (both high) has a relatively smaller number of cities (17 cities) above zero, but they are spatially contiguous. Some cities have extremely high values (e.g. 0.301 for Changwon, located in the southeast). On the other hand, sector 2811 (both low) has a relatively higher number of cities (30 cities) above zero and they are scattered. In summary, Ik can effectively detect the spatial pattern in which cities with high values are spatially clustered, whereas γk can effectively detect the spatial pattern in which a few cities have extremely high values.

After obtaining indices, factor analysis is used for grouping 180 industries into a few more generalised industry groups using distribution measures as variables. Conventional principal component analysis based on a correlation matrix is used as the extraction method. The rotation of the factors is made using the varimax method for maintaining the independence between factors. The cutoff eigenvalue for extracting meaningful industry groups is set to one. For the industry groups identified, group names are assigned based on the groups' average index values compared with those of the entire industries.

Data

  1. Top of page
  2. Abstract
  3. Introduction
  4. Measuring distribution of industries: a review
  5. Methodologies for measuring and classifying distribution patterns
  6. Data
  7. Correlations between estimated measures of spatial distribution of industries
  8. Industry classification based on spatial distribution patterns
  9. Conclusions
  10. Acknowledgements
  11. References

For the empirical analysis, 162 cities and counties in mainland Korea are used as the spatial units. The GIS administrative boundary file is obtained from Statistical Geographic Information Service provided by Statistics Korea. Use is also made of the manufacturing data on the employment profile of individual cities and counties and the number of establishments by employment-size class for individual industries from the 2009 Census on Establishments (Statistics Korea 2010). This analysis considers 180 manufacturing sectors at the four-digit KSIC level. The size of the data matrix is 180 industries × 162 cities and counties or 29 610 cells. If this value is too large, problems related to missing data due to non-disclosure of confidential information may arise. On the other hand, if it is too small, little information can be extracted.2

Correlations between estimated measures of spatial distribution of industries

  1. Top of page
  2. Abstract
  3. Introduction
  4. Measuring distribution of industries: a review
  5. Methodologies for measuring and classifying distribution patterns
  6. Data
  7. Correlations between estimated measures of spatial distribution of industries
  8. Industry classification based on spatial distribution patterns
  9. Conclusions
  10. Acknowledgements
  11. References

Table 2 reports the correlation coefficients between within- and between-industry concentration and agglomeration measures estimated for 180 industries. Note that in case of γkl and BMkl a vector of average values of 179 estimates (measuring association between industry k and industries 1, 2, 3, … , k − 1, k + 1, … , 180 respectively) is obtained and used for conducting the correlation analysis.

Table 2. Correlations between measures
 γkγklIk1stBMk1stIk2nd
  1. ** significant at 99%

  2. 1st (2nd): first (second)-order contiguity spatial weight matrix

γkl0.674 (0.000)**    
Ik1st−0.249 (0.001)**−0.359 (0.000)**   
BMk1st−0.139 (0.062)−0.217 (0.003)**−0.004 (0.959)  
Ik2nd−0.236 (0.001)**−0.307 (0.000)**0.875 (0.000)**0.034 (0.649) 
BMk2nd−0.126 (0.092)−0.211 (0.004)**0.046 (0.540)0.861 (0.000)**0.039 (0.604)

There are several notable relationships between different measures in Table 2. First and most notably, concentration and agglomeration (γk versus Ik and γkl versus BMkl) are negatively correlated, which confirms an earlier study on US manufacturing industries (Sohn 2004a). This suggests that few industries are both concentrated and agglomerated. Instead, a large number of industries are either more concentrated, but less agglomerated or more agglomerated, but less concentrated. These empiric results confirm the complementarity between the concentration and the agglomeration measures mentioned in the methodology section. Second, γk and γkl are positively correlated, implying that industries with strong within-industry concentration also show strong between-industry concentration. Third, Ik and BMkl are not correlated at statistical significance, which is different from the relationship between γk and γkl. Finally, two sets of Ik and BMkl derived using different spatial weight matrices (first- and second-order contiguity matrices) show a very strong positive correlation, as expected.

Industry classification based on spatial distribution patterns

  1. Top of page
  2. Abstract
  3. Introduction
  4. Measuring distribution of industries: a review
  5. Methodologies for measuring and classifying distribution patterns
  6. Data
  7. Correlations between estimated measures of spatial distribution of industries
  8. Industry classification based on spatial distribution patterns
  9. Conclusions
  10. Acknowledgements
  11. References

Using a total of 360 estimated indices of both concentration and agglomeration (γk (one value), Ik (one value), γkl (179 values) and BMkl (179 values)), 180 industries are summarised into several factors using factor analysis. Between first- and second-order contiguities, since factors extracted using the latter explained more variance with a fewer number of factors (95.3% with 12 factors versus 96.1% with 10 factors), the results of the latter are reported hereafter. Table 3 reports the factor analysis results.

Table 3. Factor analysis: total variance explained
ComponentExtraction sums of squared loadingsRotation sums of squared loadings
Eigenvalue% of varianceCumulative %Eigenvalue% of varianceCumulative %
  1. Note: Only the components with an eigenvalue larger than one are listed

 169.07838.37738.37745.79825.44325.443
 246.25035.69464.07134.35519.08644.530
 327.90315.50279.57329.22416.23660.765
 410.4455.80385.37626.00714.44875.214
 55.1362.85388.22920.77111.53986.753
 64.4092.44990.6784.8182.67689.429
 73.5111.95192.6294.0042.22591.654
 82.7061.50394.1323.0061.67093.324
 91.9011.05695.1882.9611.64594.970
101.6530.91896.1072.0471.13796.107

For the ten factors, factor loading values are reviewed from the component matrix in order to examine the correlation between factors and 180 industries. For simplification purposes, factor loading values of 0.6 or higher are identified among 1800 values (180 industries × 10 factors) in this study. Among the ten factors, factor 6 only has one value higher than 0.6 and factors 7 through 10 have none. This suggests that these factors do not represent the distribution pattern of any industry. Therefore, they are excluded from the analysis that attempts to group industries based on the distribution pattern.

Factors 1 through 5, on the other hand, show a stronger association with industries in terms of the number of factor loading values higher than 0.6: factor 1 (48 industries), factor 2 (38 industries), factor 3 (27 industries), factor 4 (26 industries) and factor 5 (20 industries). Among them, only three industries show strong correlation (factor loading values higher than 0.6) with more than one factor at the same time: printed circuit boards and loaded electronic components onto printed circuit boards (KSIC 2622) with factors 1 and 2, internal combustion piston engines and turbines (KSIC 2911) with factors 2 and 4, and motorcycles (KSIC 3192) with factors 2 and 5. Overall, 156 out of 180 industries are successfully classified into five factors and 153 of them are exclusively classified into a single factor, suggesting factors or classified groups of industries are sufficiently independent from each other.

In order to characterise and name the five factors or groups of industries based on the distributional characteristics, a group average value of the concentration and the agglomeration indices in each industry group (factor) relative to the overall average value is examined. Table 4 summarises this result. In case of within-industry pattern, positive (negative) sign means that the average value of distribution indices of an industry group is higher (lower) than the overall average of 180 industries. For example, factor 1's within-industry concentration is negative, which means that the average value of γk of 48 industries in factor 1 is lower than that of 180 industries.

Table 4. Distributional characteristics of factors
 Within-industryBetween-industry
ConcentrationAgglomerationConcentrationAgglomeration
Factor 1+088
Factor 2+17881
Factor 3+881
Factor 4065
Factor 596101

While the same procedure is applied in handling the between-industry pattern, it is repeated 180 times since each of 180 industries has different sets of γkl and BMkl values respectively (e.g. 179 values of γkl measuring association between industry k and industries 1, 2, 3, … , k − 1, k + 1, … , 180 respectively). Values in the table show the number of cases in which the average index value of an industry group is higher than the overall average (see equation (5)). For example, factor 3's between-industry concentration is 8, which means that in eight cases the average γkl value (an average of 27 industries' γkl values in factor 3) is higher than the overall average of γkl.

  • display math(5)

where

  • display math
inline image =

average γkl of industry group G (kG)

inline image =

average γkl of 179 industries (k = 1, 2, … , 180, kl)

Table 4 reveals that factor 2 shows a stronger (than average) within-industry concentration, while factors 1 and 3 show a stronger within-industry agglomeration. Meanwhile, factor 2 shows a very strong (178 values are above the average) and factor 5 shows a moderate between-industry concentration. Between-industry agglomeration is not as distinctive in terms of the values reported in the table as in the case of between-industry concentration, although factor 5 shows a relatively stronger agglomeration. Based on these observations, five factors can be characterised in terms of their prominent spatial distributional characteristics:

  • Factor 1: industries showing strong within-industry agglomeration.
  • Factor 2: industries showing strong within- and between-industry concentrations.
  • Factor 3: industries showing strong within-industry agglomeration.
  • Factor 4: industries showing weak concentration and agglomeration.
  • Factor 5: industries showing moderate between-industry concentration.

Table 5 lists the main industries included in individual factors and the ones with which the factors are spatially associated at the two-digit KSIC level. The list of main industries in the table are compiled based on the factor loading values in the component matrix of the factor analysis, while the list of spatially associated industries are compiled during the process explained in equation (5).3 It should be noted that the classification made in Table 5 is substantially different from the conventional industry classification. This suggests that even the industries similar in the type of products or production processes can be substantially different in their spatial distribution patterns. From a policy perspective, industrial clustering strategies may not be successful if they do not take account of industries' spatial distributional features, as shown in Tables 4 and 5.

Table 5. Factors' main industries and their associated industries in terms of spatial distribution
FactorMain industryBetween-industry
ConcentrationAgglomeration
1Wood, paper, chemical, electronic, electric, machinery, equipment Clothes, leather, paper, printing, plastic, electronic, electric, other
2Clothes, leather, printing, electronic, other transport equipmentAll industriesPlastic, metal, equipment, other transport equipment
3Food, beverage, chemical, non-metallic Food, beverage, tobacco, chemical, plastic, non-metallic
4Textile, metal, machinery, equipment Motor vehicle
5MiscellaneousTobacco, printing, other transport equipmentTobacco, textile, wood, petroleum, metal, machinery, other transport equipment, furniture

Finally, for the 156 industries that belong to the five factors, cities with the highest value of ski–xi are identified in order to check if these factors are associated with specific cities and regions. Some major findings are summarised as follows (also refer to Figure 3):

  • Factor 1: 38 of 48 industries that belong to this group show the highest value at the cities in the Seoul Metropolitan Region including the capital city Seoul (10 industries).
  • Factor 2: 22 of 38 industries show the highest value at the cities in the southeastern region including Changwon (7 industries) with large industrial complexes and 15 industries also show the highest value at the cities in the Seoul Metropolitan Region including Seoul (11 industries).
  • Factor 3: 15 of 27 industries show the highest value at the cities in the central region including Cheonan (2 industries) located right outside of the Seoul Metropolitan Region.
  • Factor 4: 21 of 26 industries show the highest value at the cities in the southeastern region including Busan (6 industries), the second largest city in Korea, and Changwon (6 industries).
  • Factor 5: although the Seoul Metropolitan (9 industries) and the southeastern regions (6 industries) have higher shares, cities with the highest values are observed in other regions as well.
figure

Figure 3. Factors with their dominant regions

Download figure to PowerPoint

These results show an interesting spatial pattern in that individual factors are spatially associated with different cities and regions. Along with the characteristics of the factors from Table 4, these suggest that the spatial distribution pattern of industry is location- as well as industry group-specific. Therefore, policies should be sensitive to the local context and to the context of particular industries (Funderburg and Boarnet 2008).

Conclusions

  1. Top of page
  2. Abstract
  3. Introduction
  4. Measuring distribution of industries: a review
  5. Methodologies for measuring and classifying distribution patterns
  6. Data
  7. Correlations between estimated measures of spatial distribution of industries
  8. Industry classification based on spatial distribution patterns
  9. Conclusions
  10. Acknowledgements
  11. References

This study has the potential to have significant implications for industrial policy. For example, the method adopted here can provide knowledge on the spatial extent and type of clustering that economic development policy targeted to this sector must consider. Furthermore, if groups of industries are classified based on the spatial distribution pattern as in this study, we can introduce more customised industrial policies to individual groups of industries. For example, a few policy implications in the Korean spatial economy can be derived based on the findings of this study.

First, industries in factors 1 and 3 are considered to be influenced by the localisation economies with which industries of a kind are spatially clustered. The clusters to foster these industries should be established in a way that facilitates networking between firms in these industries. Considering that their spillover effects reach spatially adjacent cities and counties, the clusters do not need to be confined to a small area such as industrial complexes. In this case, coordination across multiple municipalities is also important (Funderburg and Boarnet 2008).

Second, industries in factor 2 consider proximity to other industries (for seeking urbanisation economies) as well as to themselves to be critical. Clusters to foster these industries need to be established in cities (preferably large cities) where the spillover effects of other industries as well as their own are abundant. Considering that their spillover effects do not go beyond the boundary of a city, the clusters need to be spatially more compact.

Third, although we observe a mild between-industry concentration from industries in factor 5, industries in factors 4 and 5 are generally insensitive to spatial proximity and as a consequence to the spillover effects of the agglomeration economies. Therefore, for these industries, attempts to establish clusters may not have anticipated effects.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Introduction
  4. Measuring distribution of industries: a review
  5. Methodologies for measuring and classifying distribution patterns
  6. Data
  7. Correlations between estimated measures of spatial distribution of industries
  8. Industry classification based on spatial distribution patterns
  9. Conclusions
  10. Acknowledgements
  11. References

This study was supported by Seoul National University Research Grant for Humanities and Social Sciences. The author thanks the editor and the referees for their helpful comments and suggestions on earlier drafts of the paper. The author also thanks Kyusang Kwon, a graduate student in the Department of Geography, SNU, for compiling the database.

Notes
  1. 1

    Ik is appropriate for measuring the within-industry agglomeration pattern because it is a spatial autocorrelation measure. That is, it measures the correlation between individual observations (cities and counties) of a single variable (industry) rather than between individual variables (industries) as in the case of the correlation coefficient.

  2. 2

    The median number of cells in the 17 studies in the literature review section with explicit information of the data matrix is 25 935, which suggests that 29 160 cells are not extraordinary in either direction.

  3. 3

    More specifically, for each factor all the four-digit industries with factor loading value larger than 0.6 are counted by each two-digit industry. Next, for each two-digit industry, the factor with the largest number of four-digit industries (with the value >0.6) is identified. This two-digit industry is then acknowledged as the main industry of this factor. In a similar fashion, for each factor all the four-digit industries that satisfy inline imagein equation (5) are counted by each two-digit industry. The spatially associated industry of the factor can then be obtained by following the same steps as above.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Measuring distribution of industries: a review
  5. Methodologies for measuring and classifying distribution patterns
  6. Data
  7. Correlations between estimated measures of spatial distribution of industries
  8. Industry classification based on spatial distribution patterns
  9. Conclusions
  10. Acknowledgements
  11. References
  • Anselin L 1995 Spacestat version 1.80 user's guide Regional Research Institute, West Virginia University, Morgantown
  • Anselin L 2005 Exploring spatial data with GeoDaTM: a workbook Center for Spatially Integrated Social Science, University of California, Santa Barbara CA
  • Arbia G 2001a Modelling the geography of economic activities on a continuous space Papers in Regional Science 80 411424
  • Arbia G 2001b The role of spatial effects in the empirical analysis of regional concentration Journal of Geographical Systems 3 271281
  • Bergsman J, Greenston P and Healy R 1975 A classification of economic activities based on location patterns Journal of Urban Economics 2 128
  • Bertinelli L and Decrop J 2005 Geographical agglomeration: Ellison and Glaeser's index applied to the case of Belgian manufacturing industry Regional Studies 39 567583
  • Braunerhjelm P and Borgman B 2004 Geographical concentration, entrepreneurship and regional growth: evidence from regional data in Sweden, 1975–99 Regional Studies 38 929947
  • Burger M J, van Oort F G and van der Knaap B 2008 A treatise on the geographical scale of agglomeration externalities and the modifiable areal unit problem ERIM Report Series Reference No. ERS-2008-076-ORG
  • Carroll M C, Reid N and Smith B W 2008 Location quotients versus spatial autocorrelation in identifying potential cluster regions Annals of Regional Science 42 449463
  • Chakravorty S, Koo J and Lall S V 2005 Do localization economies matter in cluster formation? Questioning the conventional wisdom with data from Indian metropolises Environment and Planning A 37 331353
  • Combes P-P, Mayer T and Thisse J-F 2008 Economic geography: the integration of regions and nations Princeton University Press, Princeton NJ
  • Devereux M P, Griffith R and Simpson H 2004 The geographic distribution of production activity in the UK Regional Science and Urban Economics 34 533564
  • de Dominicis L, Arbia G and de Groot H L F 2007 The spatial distribution of economic activities in Italy Tinbergen Institute Discussion Paper TI2007-094/3
  • Duranton G and Overman H G 2005 Testing for localization using micro-geographic data Review of Economic Studies 72 10771106
    Direct Link:
  • Ellison G and Glaeser E L 1997 Geographic concentration in U.S. manufacturing industries: a dartboard approach Journal of Political Economy 105 889927
  • Falcioglu P and Akgüngör S 2008 Regional specialization and industrial concentration patterns in Turkish manufacturing industry: an assessment for the 1980–2000 period European Planning Studies 16 303323
  • Fan C C and Scott A J 2003 Industrial agglomeration and development: a survey of spatial economic issues in East Asia and a statistical analysis of Chinese regions Economic Geography 79 295319
  • Feser E, Sweeney S and Renski H 2005 A descriptive analysis of discrete U.S. industrial complexes Journal of Regional Science 45 395419
  • Funderburg R G and Boarnet M G 2008 Agglomeration potential: the spatial scale of industry linkages in the southern California economy Growth and Change 39 2457
  • Getis A and Ord J K 1992 The analysis of spatial association by use of distance statistics Geographical Analysis 24 189206
  • Guillain R and Le Gallo J 2010 Agglomeration and dispersion of economic activities in and around Paris: an exploratory spatial data analysis Environment and Planning B 37 961981
  • Holmes T J and Stevens J J 2004 Spatial distribution of economic activities in North America in Henderson J V and Thisse J-F eds Handbook of regional and urban economics (volume 4): cities and geography Elsevier, Amsterdam 27972843
  • Hoover E M 1937 Location theory and the shoe and leather industries Harvard University Press, Cambridge MA
  • Hoover E M 1948 The location of economic activity McGraw-Hill, New York
  • Krugman P 1991 Geography and trade The MIT Press, Cambridge
  • Lafourcade M and Mion G 2007 Concentration, agglomeration and the size of plants Regional Science and Urban Economics 37 4668
  • Maurel F and Sédillot B 1999 A measure of the geographic concentration in French manufacturing industries Regional Science and Urban Economics 29 575604
  • Mori T, Nishikimi K and Smith T E 2005 A divergence statistic for industrial localization Review of Economics and Statistics 87 635651
  • Nakamura R 2008 Agglomeration effects on regional economic disparities: a comparison between the UK and Japan Urban Studies 45 19471971
  • O'Donnellan N 1994 The presence of Porter's sectoral clustering in Irish manufacturing Economic and Social Review 25 221232
  • Ohlin B 1933 Interregional and international trade Harvard University Press, Cambridge MA
  • Ord J K and Getis A 1995 Local spatial autocorrelation statistics: distributional issues and an application Geographical Analysis 27 286306
  • Phelps N A and Ozawa T 2003 Contrasts in agglomeration: proto-industrial, industrial and post-industrial forms compared Progress in Human Geography 27 583604
  • Phelps N, Fallon R and Williams C 2001 Small firms, borrowed size and the rural-urban shift Regional Studies 35 613624
  • Rigby D L and Essletzbichler J 2002 Agglomeration economies and productivity differences in US cities Journal of Economic Geography 2 407432
  • Rosenthal S S and Strange W C 2003 Geography, industrial organization, and agglomeration The Review of Economics and Statistics 85 377393
  • Sjöberg Ö and Sjöholm F 2004 Trade liberalization and the geography of production: agglomeration, concentration, and dispersal in Indonesia's manufacturing industry Economic Geography 80 287310
  • Sohn J 2004a Do birds of a feather flock together? Economic linkage and geographic proximity Annals of Regional Science 38 4773
  • Sohn J 2004b Information technology in the 1990s: more footloose or more location-bound? Papers in Regional Science 83 467485
  • Sohn J 2013 A quantitative analysis of the spatial agglomeration pattern among the Korean cities Journal of the Korean Geographical Society 48 5671
  • Statistics Korea 2010 The 2009 census on establishments Statistics Korea, Daejeon
  • Van Oort F G 2007 Spatial and sectoral composition effects of agglomeration economies in the Netherlands Papers in Regional Science 86 530
  • Viladecans-Marsal E 2004 Agglomeration economies and industrial location: city-level evidence Journal of Economic Geography 4 565582
  • Wen M 2004 Relocation and agglomeration of Chinese industry Journal of Development Economics 73 329347
  • Ying L G 2000 Measuring the spillover effects: some Chinese evidence Papers in Regional Science 79 7589