https://orcid.org/0000-0003-4040-8193
![Sample our Urban Studies journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/110826/TOC-Subject-Sample-2021-Urban-Studies.jpg"})
ABSTRACT
This paper addresses two questions: Which manufacturing industries are associated with the inception of knowledge-intensive business services (KIBS)? Do cross-sectoral interdependencies between the two sectoral macro-domains exhibit geographical heterogeneity? We use a job multiplier approach using decadal employment data of 399 municipalities in Spain over the period 1981–2011. The main result is that only traditional supplier-dominated industries exhibit significant employment effects, both positive and negative, on KIBS. Coherent with the prior literature, this is most prominent in denser areas. Our findings draw attention to the specificities of user industries, and lend support to the notion that KIBS are a heterogeneous group of activities.
ACKNOWLEDGEMENTS
We are indebted to the editor and three anonymous referees for constructive feedback. We are also grateful for comments received at the following events: GeoINNO Conference 2018 (Barcelona, 2018); Spanish Regional Sciences Association (Valencia, 2018); and GeoINNO Conference 2020 (Stavanger, 2020). All errors and omissions are our own.
DISCLOSURE STATEMENT
No potential conflict of interest was reported by the authors.
Notes
1. For exhaustive reviews, see Muller and Doloreux (Citation2009) and Miles et al. (Citation2018).
2. The municipalities of Ceuta y Melilla have been excluded from the analysis due to their peculiarities: these are two provinces home to autonomous cities in North Africa with lower administrative competences compared with other Spanish provinces.
4. Although the original source of information for this survey is the Spanish Statistical Office, the 1981 database is made available by iPUMS (https://www.ipums.org/).
5. We use the weighting factor provided by the Spanish National Office of Statistics. Since census data for 1991 do not include such a factor, we applied a value of 20 for each individual following the iPUMS version of the same database for this year.
6. We only select municipalities for which information is available for at least three out of four years.
7. Figure A1 in Appendix A in the supplemental data online includes a graphical description of the municipalities (lower portion) and provinces (upper portion).
8. NACE codes for KIBS are: 1981 (81–84), where 81–83 are finance; 1991 (23–24), where 23 is finance; 2001 (66–67, 70–71, 73–75), where 66–67 and 70 are finance; and 2011 (64–66, 68–74), where 64–66 and 68 are finance.
9. NACE codes for manufacturing: (1) Metal: 1981 code 31; 1991 codes 11–12; 2001 codes 28–29; 2011 codes 24–25. (2) Food & tobacco: 1981 code 42; 1991 code 5; 2001 codes 16–17; 2011 codes 10–12. (3) Textile & clothing: 1981 code 43; 1991 code 6; 2001 codes 18–19; 2011 codes 13–14. (4) Leather & footwear: 1981 codes 44–45; 1991 code 15; 2001 code 20; 2011 codes 15 and 32. (5) Wood & furniture: 1981 code 43; 1991 code 7; 2001 codes 21 and 37; 2011 codes 16 and 31. (6) Paper & publishing: 1981 code 47; 1991 code 8; 2001 codes 22–23; 2011 codes 17–18. (7) Plastic & chemical: 1981 codes 48–49; 1991 code 10; 2001 codes 25–27; 2011 codes 20–23. (8) Machinery: 1981 codes 32–35; 1991 code 18; 2001 codes 30–32 and 35; 2011 code 29. (9) Transport equipment: 1981 codes 36–38; 1991 code 14; 2001 code 36; 2011 code 30. (10) High-tech manufacturing: 1981 code 39; 1991 code 13; 2001 codes 33–34; 2011 codes 26–28.
10. Following Van Dijk (Citation2016), we computed regressions with a battery of area-specific control variables to account for structural changes, specifically: college–non-college ratio as a proxy of the general level of education; local unemployment rate; share of entrepreneurs; and percentage of immigrant workers. Results of these alternative models are similar to those presented here, but are available from the authors upon request.
11. While the shift–share approach is standard in the literature, it is not exempt from criticism. Jaeger et al. (Citation2018), in particular, call attention to the issue of persistent correlation over time. To illustrate, if it takes time for markets to adjust to shocks, the error term may include ongoing general equilibrium adjustment to past shocks. Their suggested solution is to control for these dynamics by adding a lagged independent variable (lagged increment in manufacture in time t − 10) as well as an additional lagged Bartik instrument (at time t − 10). Results using multiple instruments (see Table A2 in Appendix A in the supplemental data online) confirm our main findings. We also report the first-stage coefficients and the conventional first-stage F-test (Bound et al., Citation1995) and R2 for the joint significance of the instruments in each equation separately. Both instruments are valid, with F-statistics above the standard threshold (respectively, 122.10 and 21.76). The correlation between the current increment and the lagged increment in manufacturing is low (r = −0.37; p = 0.000), thus suggesting that both variables can be part of the regression model. It is important to note that the addition of lagged variables entails a smaller sample size because information is only available for two periods (1991–2001 and 2001–11).
12. See note 9.
13. Following Moretti (Citation2010) we quantify the magnitude of the multipliers by multiplying the elasticity against the relative size of the two sectors. Also, based on prior literature and a concentration on coefficients that are significant and on regressions that have a Kleibergen–Paap Wald F-statistic > 10.
14. The population of the municipalities is calculated as the average population between 1981 and 2011 using census information and taking into account the weights in each particular year. The results are substantially similar when the cut-off is set at the 50th and 75th percentiles of population density.
15. For detailed maps of industry employment by municipality, see Appendix A in the supplemental data online.