Working Paper: CEPR ID: DP3379
Authors: Gilles Duranton; Henry Overman
Abstract: To study the detailed location patterns of industries, and particularly the tendency for industries to cluster relative to overall manufacturing, we develop distance-based tests of localization. In contrast to previous studies, our approach allows us to assess the statistical significance of departures from randomness. In addition, we treat space as continuous instead of using an arbitrary collection of geographical units. This avoids problems relating to scale and borders. We apply these tests to an exhaustive UK data set. For four-digit industries, we find that (i) only 51% of them are localized at a 5% confidence level, (ii) localization takes place mostly at small scales below 50 kilometres, (iii) the degree of localization is very skewed, and (iv) industries follow broad sectoral patterns with respect to localization. Depending on the industry, smaller establishments can be the main drivers of both localization and dispersion. Three-digit sectors show similar patterns of localization at small scales as well as a tendency to localize at medium scales.
Keywords: clusters; kdensity; localisation; spatial statistics
JEL Codes: C19; L70; R12
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| proximity (R32) | localisation (R32) |
| degree of localisation skewed across industries (L89) | localisation (R32) |
| industries within the same branch (L89) | similar localisation patterns (C59) |
| localisation patterns (R32) | clustering of establishments (R32) |
| localisation occurs at small scales (R12) | clustering (C38) |