- 23 May 2007
- Working Paper Summaries
What Causes Industry Agglomeration? Evidence from Coagglomeration Patterns
Executive Summary — Most industries exhibit some degree of geographic concentration. Although many theories attempt to explain this agglomeration, empirical tests of these theories are difficult as they all predict similar outcomes within individual industries. This study considers how industries coagglomerate—that is, which industry pairs locate together—to form a tractable analysis. The authors specifically study the relative importance of proximity to suppliers and customers, to firms using similar labor, and the sharing of ideas for explaining agglomeration. Key concepts include:
- In manufacturing, transport costs for goods, people, and ideas all still seem to matter. Given the remarkable decline of transportation costs over the 20th century, it is striking that transport costs remain so important.
- It is unclear how these results would extend to non-manufacturing industries. Services are costly to transport since they involve face-to-face interaction, and idea flows might take precedence in non-manufacturing industries.
Many industries are geographically concentrated. Many mechanisms that could account for such agglomeration have been proposed. We note that these theories make different predictions about which pairs of industries should be coagglomerated. We discuss the measurement of coagglomeration and use data from the Census Bureau's Longitudinal Research Database from 1972 to 1997 to compute pairwise coagglomeration measurements for U.S. manufacturing industries. Industry attributes are used to construct measures of the relevance of each of Marshall's three theories of industry agglomeration to each industry pair: (1) agglomeration saves transport costs by proximity to input suppliers or final consumers, (2) agglomeration allows for labor market pooling, and (3) agglomeration facilitates intellectual spillovers. We assess the importance of the theories via regressions of coagglomeration indices on these measures. Data on characteristics of corresponding industries in the United Kingdom are used as instruments. We find evidence to support each mechanism. Our results suggest that input-output dependencies are the most important factor, followed by labor pooling.