Business centre or bedroom community? The development of employment in small and medium-sized towns

ABSTRACT

This study scrutinizes the development of employment towards business centres or bedroom communities in Swiss small and medium-sized towns (SMSTs). It analyzes an original data set of all 152 SMSTs in Switzerland. Bayesian multilevel and spatial regression techniques examine regional and local explanatory factors affecting the development of employment. Evidence is found for employment growth if an SMST is embedded in a dynamic network, meaning that employment growth in neighbouring cities and towns creates spillover effects for SMSTs. Thus, if local administrations want to influence their employment structure, they are well advised to engage in regional economic policy-making.

KEYWORDS:

• small and medium-sized towns
• employment
• regional networks
• local autonomy
• spatial regression
• spillover effects

JEL:

• F63
• H7

ACKNOWLEDGEMENTS

The authors contributed equally to this paper. The authors thank Fritz Sager, Heike Mayer and Rahel Meili, as well as the editors and anonymous referees, for their useful comments. They also thank Rahel Meili and Arev Shahinian for help with data collection and the creation of the map.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

ORCID

David Kaufmann http://orcid.org/0000-0002-4185-4550

Stefan Wittwer http://orcid.org/0000-0002-6843-2526

Notes

1. We use the categorization of economic sectors provided by Meili and Mayer (2017), which are based on the Nomenclature generale des activites economiques (NOGA) (EUROSTAT, 2016).

2. For its distribution, see Figure A1 in Appendix A in the supplemental data online.

3. Using the online timetable of www.sbb.ch and GoogleMaps. Twenty minutes is considered a good proxy for a commuting distance that indicates closeness (Kloosterman & Musterd, 2001), given that reaching distances within Swiss cities also easily require more than 20 min of driving time.

4. On average, Swiss SMSTs have 3.5 neighbours. For the distribution of the number of neighbours, see Figure A4 in Appendix A in the supplemental data online.

5. Cities with more than 50,000 inhabitants, such as Zurich or Geneva.

6. The centres of metropolitan regions are Zurich, Lausanne, Geneva, Basel and Milan (Italy) (Raumkonzept Schweiz, 2012).

7. We used diffuse priors ρ(β_k) ∼ 1 for the parameters and inverse Wishart priors ρ(1/σ²_μ0) ∼ Gamma(0.01, 0.01) for the variance component. All models are estimated using the Markov chain Monte Carlo (MCMC) estimation in Stata15. We let the models run for 209,999 iterations (50,000 for models 4–6, respectively), with a respective burn-in of 10,000 (5000 for models 4–6, respectively) and a thinning of 2 (1 for models 4–6, respectively). Extensive diagnostics based on the graphical inspection of the trajectories and the Raftery–Lewis and Brooks–Draper diagnostics lead to the conclusion that the chains have mixed well and converged.

8. See Figure A2 in Appendix A in the supplemental data online for cantonal means of the dependent variable.

9. The models and variables were tested for multicollinearity, heteroskedasticity, non-linearity and influential outliers (e.g., Figure A3 in Appendix A in the supplemental data online). Variables with outliers were logarithmized. For convergence diagnostics, see Appendix B in the supplemental data online.

10. The lower the DIC, the better the model fits the data (Gelman & Hill, 2007, pp. 525–526). Since only one variable is significant, and the DIC penalizes for additional parameters, the first model, unsurprisingly, fits best.

11. The coefficients regarding the political variables were also allowed to vary across cantons in order to control for the fact that SMSTs in cantons that are very active in promoting their SMSTs may only have a small political impact. The results, however, show that the cantonal variance of the coefficient is negligibly low, which is why the random slope is not presented here.

12. See Table A3 and further explanations in Appendix A in the supplemental data online.

13. This can be seen when comparing the predicted and observed values in the dependent variable.

Additional information

Funding

This work was supported by the Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung [grant number 159324].