Concentration, agglomeration and the size of plants

https://doi.org/10.1016/j.regsciurbeco.2006.04.004Get rights and content

Abstract

This paper investigates whether the geographic distribution of manufacturing activities depends on the size of plants. Using Italian data, we find, as in Kim [Kim, S., 1995. Expansion of markets and the geographic concentration of economic activities: the trends in U.S. regional manufacturing structure, 1860–1987, Quarterly Journal of Economics 110 (4), 881–908.], Holmes and Stevens [Holmes, T.J., and Stevens, J.J., 2002. Geographic concentration and establishment scale, Review of Economics and Statistics 84, 682–690.], and Holmes and Stevens [Holmes, T.J. and Stevens, J.J., 2004. Spatial distribution of economic activities in North America, in: J.V. Henderson and J.F. Thisse, eds., Handbook of Regional and Urban Economics, Vol.4, (Elsevier-North Holland, Amsterdam).], that large plants are more concentrated than small plants. However, considering distance-based patterns via spatial auto-correlation, we find that small establishments actually exhibit a greater tendency to be located in adjacent areas. These apparently contradictory findings raise a measurement issue regarding co-location externalities and suggest that large plants are more likely to cluster within narrow geographical units (concentration), while small establishments would rather co-locate within wider distance-based clusters (agglomeration). This picture is consistent with different size plants engaging in different transport-intensive activities.

Introduction

Economists, geographers and historians share a considerable interest in analyzing the causes of regional specialization. Among the myriad of determinants which have been explored, particular attention has been paid to regional endowments or raw-material intensity, comparative advantage, localization externalities, or, more recently, transport costs and market potential. In this paper, we focus on one particular aspect of this complex set of mechanisms which has remained relatively unexplored: the size of plants. Our main contribution is to investigate whether the geographic distribution of manufacturing activities is related to plant size, and in particular to look for a separate role for physical distance in explaining location.

We address this question using a number of years of Italian census data on manufacturing industries, at varying geographic and industrial scales. By extending the empirical focus to Europe, we complement the seminal empirical studies of Kim (1995), Holmes and Stevens (2002), and Holmes and Stevens (2004), which focused on North America. Kim (1995) reports a positive correlation between concentration and both average plant size and the intensity of raw materials, across U.S. manufacturing industries. Holmes and Stevens (2002) find strong evidence of the same phenomenon within industries: plants located in areas with higher industry concentration are larger, on average, than those located outside such areas, and this holds especially for the manufacturing sector. Holmes and Stevens (2002) extend Kim's findings by emphasizing that this positive relationship is robust to controlling for the establishment's own size effect on concentration.

We extend the analysis of the relation between plant size and spatial location patterns to include spatial dependence among geographic units. A number of recent papers (for example, Arbia, 2001a, Duranton and Overman, 2005, Marcon and Puech, 2003) have underlined significant differences in the concentration patterns obtained from continuous distance-based measures compared to more standard indicators like the index proposed by Ellison and Glaeser (1997). Since labor productivity is positively related to employment density (Ciccone and Hall, 1996, Ciccone, 2002), a clearer picture of how establishment density varies within industries and of how this depends on the number of employees remains high on the regional policy agenda.

Although our approach builds on this recent revival of distance, we continue to work with a discrete vision of space as a finite number of spatial units (which can be aggregated to various degrees). We can therefore complement the Ellison and Glaeser concentration index, which measures the strength of co-location spillovers within each geographic unit, with an indicator of spatial auto-correlation (the Moran index) that accounts for possible distance-based co-location patterns across geographical units. Spatial auto-correlation exists if, for a particular industry, data on the location of plants in region i is ‘linearly’ informative about the location of other plants within the same industry in regions ‘close’ to i.1 Spatial auto-correlation as a feature of location decisions has received relatively little attention compared to concentration and can be related to the minimization of transport costs, as in the New Economic Geography (henceforth, NEG) literature. In this paper, we use the term “agglomeration”, which is the term commonly used in the NEG to refer to distance-based location patterns, as a synonym for positive spatial auto-correlation.

Using Italian manufacturing data, we find strong evidence of a significant positive relationship between plant size and concentration, as in Kim (1995) and Holmes and Stevens (2002). We go further and examine the sensitivity of this result to distance and find that, on average, small plants exhibit more pronounced agglomeration patterns. Large establishments are therefore more concentrated but less agglomerated (in the sense of spatial auto-correlation), while small establishments are by contrast more agglomerated and less concentrated. These apparently contradictory results raise a measurement issue linked to the spatial magnitude of co-location spillovers by plant size. This issue is likely related to the properties of the indices used.

Our results suggest that large plants cluster within narrow geographical areas such as local labor markets. By contrast, small plants would rather co-locate within wider areas in which a distance-based pattern emerges. One interpretation is that large manufacturing plants are more export-intensive and thus more oriented towards international markets, as shown, for instance, by Eaton et al. (2004). As they are less sensitive to domestic distances, they co-locate within few dense clusters of activities where they can benefit from particular features, such as Marshallian labor market externalities. By contrast, since small establishments mainly serve local demand, the need to save on domestic transport costs produces a spatial distribution of small plants which corresponds more closely to that of the Italian population (as shown in our data), which is itself spatially auto-correlated. Some exceptions arise, however, for most of the so-called “Italian Districts” industries, which are highly concentrated, but only weakly agglomerated, despite small plant size.2

Finally, we show that these results are robust to different definitions of space, of plant size, of industries, and of distance, and that they are also robust to industry characteristics. Furthermore, we find that concentration (agglomeration) has slightly decreased (increased) over the period 1981–1996, small plants exhibiting more movement.

The paper proceeds as follows. Section 2 presents the analytical framework relating plant scale, industry concentration and industry agglomeration. Section 3 describes the Italian data we use to investigate the geographic distribution of manufacturing and its relation to plant size. We also discuss briefly how we deal with some well-known spatial issues such as the Modifiable Areal Unit Problem. Section 4 provides the results of the cross-section analysis of Italian Local Labor Systems and 3-digit manufacturing industries in 1996. Section 5 checks the robustness of the results and explores long-run trends. Finally, Section 6 concludes and suggests some new lines of research.

Section snippets

Analytical framework

Various indices can be used to investigate the location patterns of economic activity.3 A large group of indices used by economists, which we refer to as the “concentration” family, splits space into a certain number of geographic units and looks for relative differences in the

Data and methodological issues

We use data from the Italian Census of economic activities, provided by the Italian National Statistic Institute (ISTAT), which includes information on the location and employment of the whole population of Italian plants. The data are very detailed in its geographic coverage of manufacturing industries. The geographic scale of observation can be disaggregated up to the 8192 Italian communes and the industrial scale up to the 3-digit NACE nomenclature (revision 1) for the years 1981 and 1991,

Basic results for Italy: LLS, 3-digit industries, 1996

Section 4.1 assesses the impact of plant size on the geographic distribution of activities by comparing the concentration and agglomeration indices computed on both employment and number of plants bases. As there are significant differences in the results, Section 4.2 considers separate samples of large and small plants. Section 4.3 considers extreme cases of concentration and agglomeration in order to gain a richer understanding of different location patterns.

Robustness checks and long-run trends

This section considers the robustness of the positive (negative) correlation found between concentration (agglomeration) and plant size, and suggests some explanations of these apparently contradictory findings in the light of previous work (Section 5.1). It further explores the time dimension of our panel data, investigating the evolution of concentration and agglomeration over the 1981–1996 period. Last, it shows that changes have been triggered by small rather than large plants and that

Conclusions

This paper has analyzed the spatial distribution of manufacturing in Italy with respect to two different features. The first, concentration, can be defined as the degree of variability across data for a given partition of space. The second, agglomeration, explicitly considers distances between observations and thus their spatial dependence. Although much work has focused on concentration, agglomeration has received far less attention.

Examining the influence of plant size on both concentration

Acknowledgments

We thank Giuseppe De Arcangelis, Jérôme Carreau, Bart Jourquin, Renato Santelia and Stefano Usai for their critical help in collecting the data, and Antonio Ciccone, Andrew Clark, Esther Lamouroux, Thierry Mayer, Henry Overman, Frédéric Robert-Nicoud, Fabiano Schivardi, Eric Strobl, Stefano Usai and three anonymous referees for insightful comments. Mion gratefully acknowledges financial support from the CEPR Research Network on “The Economic Geography of Europe: Measurement, Testing and Policy

References (35)

  • R.E. Baldwin et al.

    Heterogeneous firms, agglomeration and economic geography: Spatial selection and sorting

    Journal of Economic Geography

    (2006)
  • Barrios, S., Bertinelli, L., Strobl, E., in press. Geographic concentration and establishment scale: can panel data...
  • M. Brülhart

    Evolving geographical specialisation of European manufacturing industry

    Weltwirtschaftliches Archiv

    (2001)
  • A. Ciccone et al.

    Productivity and the density of economic activities

    American Economic Review

    (1996)
  • A.D. Cliff et al.

    Spatial Processes: Models and Applications

    (1981)
  • P.P. Combes et al.

    Transport Costs: Measures, determinants and regional policy implications for France

    Journal of Economic Geography

    (2005)
  • P.P. Combes et al.

    The Spatial Distribution of Economic Activities in the European Union

  • Cited by (49)

    View all citing articles on Scopus
    View full text