Firm growth in industrial clusters of the United Kingdom

More complex measures of diversity exist in the literature (see Henderson et al. 1995; Glaeser et al. 1992; or Combes 2000). In our regressions, we tried these different measures, but the simplest Herfindahl worked best and will be used in this article.

There are two ways to treat the consolidated account problem. First, holding companies can be treated separately from their subsidiaries. A problem then arises when we encounter subsidiaries of subsidiaries. Second, dummy variables for holding companies and consolidated accounts can be introduced. This second method is preferred because it does not reduce the sample size more than is necessary. Another issue is how we should treat entry by and growth of multi-sector firms. In this study, such companies were only counted in the principal sector where the firm was active in the sample year. That may be an unsatisfactory assumption when the firm has diversified from one original sector into others, but in the absence of detailed information on such diversification in each firm, this seems a reasonable working assumption.

We have checked for the robustness of the regressions with respect to outliers that may arise from firms filing consolidated accounts, and found no such problem: coefficients and standard errors showed only slight variations when such observations were removed.

NUTS: Nomenclature des Unités Territoriales Spatiales. In the UK, NUTS level 1 corresponds to 11 regions, while NUTS level 3 represents the 65 counties.

The use of NUTS level 3 data is an improvement on some previous studies (which used NUTS levels 1 and 2), but because of this regional disaggregation, we had to compromise on the industrial disaggregation. In an ideal world, we would have NUTS level 3 data using three- or four-digit industrial codes. However, for the present study the 56 two-digit industrial codes (1992 rev.) give a sufficient disaggregation for our immediate interest.

A description of these codes is provided in Appendix A.

This is somewhat unsatisfactory, since the cluster is generally smaller than a NUTS 2 area—really more in line with a NUTS 3 area. However, studies such as Glaeser et al. (1992) and Jaffe et al. (1993) have shown that external effects of the kind that are explored here seem to grow stronger as the regional unit becomes smaller. Any bias introduced here should be to underestimate the strength of clustering effects. Nevertheless, the use of regions as a spatial unit has some administrative sense. In some countries, for example, government policies and incentives towards new industries are to some extent defined at a regional level.

For England and Wales, this data was obtained from the British office, while for Scotland, population estimates were provided by the Scottish office.

One of the problems encountered with data published by the Office of National Statistics, especially within Regional Trends, is the change in the classification of the regions. For a number of years now, and probably since the introduction of the Scottish Parliament and the Welsh Assembly, the boundaries of various ‘official’ regions in Scotland and Wales have changed and no longer correspond strictly to the NUTS level 3 regions. It was however possible to obtain proxy measures for the NUTS level 3 regions from the various government offices responsible to provide this information to Eurostat, the European office of statistics.