ABSTRACT
The location quotient (LQ) measures regional industry concentration with the advantages of easy calculation and interpretation. However, it is a weak method for identifying industry clusters that consist of related industries geographically concentrated in contiguous counties. This paper proposes a new spatial input–output location quotient (SI-LQ) accounting for both the co-location of related industries and the spatial spillover of concentration into neighbouring counties. A bootstrap method is used to determine the cut-off values of the new measure. The practical advantages of the SI-LQ over the traditional LQ include attenuation of the extreme values of the LQ in less populous and remote counties and the identification of large substantive clusters. The SI-LQ outperforms the LQ in a regression analysis of the effect of industry concentration on total employment growth.
ACKNOWLEDGEMENTS
The authors thank the W.E. Upjohn Institute for providing the unsuppressed County Business Patterns Data.
DISCLOSURE STATEMENT
No potential conflict of interest was reported by the authors.
Notes
1 See Nakamura and Paul (Citation2009) for a survey.
2 In the following context, we consider these concepts as synonyms: regional industry concentration, industry concentration, spatial concentration, industry agglomeration, industry clusters and spatial clusters.
3 In the literature, the LQ is commonly computed with industry employment, but it is also computed with other variables for the size or importance of an industry, such as establishment counts (Billings & Johnson, Citation2012; Delgado et al., Citation2016) and value-added (Mulligan & Schmidt, Citation2005). Owing to data availability at the county level and industry aggregation as well as common practice in the literature, we chose to use employment data to compute the LQ.
4 Among all 71 industries, 22 industries have a diagonal element that is the maximum value of the column.
5 It is an interesting counterfactual attempt to add to . This can be interpreted as the situation where even though industry does not exist in county , it has the potential to locate there because this region already possesses the localization and urbanization economies for industry .
6 The main reason for choosing such an industry classification is for easy calculation. Mulligan and Schmidt (Citation2012) also point out that it is undesirable to discuss statistical properties of LQs at a very granular level of industry classification and the level of NAICS with three to four digits is appropriate.
7 Rail transportation (NAICS 482) and private households (NAICS 814) are missing in both the CBP and BEA data sources. Therefore, the corresponding rows and columns in the direct requirement matrix of these two sectors are deleted. However, when calculating the direct requirement matrix from the use and make tables, these two sectors are preserved so that part of the IO information of these two sectors is saved in the direct requirement matrix.
8 A five-nearest-neighbour spatial weight is also used, yielding similar results.
9 For the tables comparing descriptive statistics of each industry, see the supplemental data online.
11 The literature on agglomeration generally predicts growth in both employment and productivity (Cohen, Coughlin, & Paul, Citation2019). In the short run, we can expect differences in agglomerative advantage to generate more firm births or, in a recessionary period, fewer plant closings. In contrast, the negative employment effect of enhanced labour productivity is likely to be a long-run phenomenon. Studies that explore the kind of correlations we hypothesize here, and over similar timespans, include Barkley, Henry, & Kim (Citation1999), Rosenthal and Strange (Citation2003) and Gabe (Citation2003). To test the effect of agglomeration on productivity that is often represented by wage rates, we estimated a model with the growth rate of average wages being the dependent variable. The result of this model also shows that the SI-LQ has a higher and more significant effect on the growth rate of average wages than the LQ. For the detailed result, see the supplemental data online.
12 We also estimated spatial panel models with two-way (county and time) fixed effects. The results of the two-way fixed-effects model are similar to those of the county fixed-effects models, but the coefficients on both the LQ and SI-LQ are insignificant. Those results are available from the authors on request.
13 Categorizing LQ and SI-LQ makes them comparable with the same magnitude scale. Using the original values does not appreciably change the regression results. The results are available from the authors on request.
14 To compare the effects of both the LQ and SI-LQ directly, we estimated a model that includes both variables. Although not reported in Table 3 owing to the problem of multicollinearity between the LQ and SI-LQ variables, the nested model yields similar results and the negative coefficient on the LQ is statistically significant.
15 We also compute the Pearson correlation coefficients with similar results.