Quantifying entrepreneurship and its impact on local economic performance: A spatial assessment in rural Switzerland

Abstract

Regional and rural development policies in Europe increasingly emphasize entrepreneurship to mobilize the endogenous economic potential of rural territories. This study develops a concept to quantify entrepreneurship as place-dependent local potential to examine its impact on the local economic performance of rural territories in Switzerland. The short-to-medium-term impact of entrepreneurship on the economic performance of 1706 rural municipalities in Switzerland is assessed by applying three spatial random effects models. Results suggest a generally positive relationship between entrepreneurship and local development: rural municipalities with higher entrepreneurial potential generally show higher business tax revenues per capita and a lower share of social welfare cases among the population, although the impact on local employment is less clear. The explanatory power of entrepreneurship in all three models, however, was only moderate. This finding suggests that political expectations of fostering entrepreneurship to boost endogenous rural development in the short-to-medium term should be damped.

Keywords:

• regional development
• rural areas
• entrepreneurship
• regional policy
• Switzerland

Notes

Notes

1. At first glance, this might contrast with the neo-endogenous rural development approach, which Ray (2006, 281) appraises as follows: ‘ … the notion of pure endogenous development in which change is animated solely by local actors […] is useful only as a heuristic device’. However, for Stucksmith (2010), modern neo-endogenous rural development approaches particularly emphasize the role of exogenous interventions to initiate or stimulate local processes.

2. There are no specific studies at the local level in rural areas on these questions. Fritsch (2008, 10), focussing on the regional scale, provides an overview of empirical studies on the impact of start-ups on different performance measures (such as Gross Domestic Product (GDP) or employment rates) over time at the regional level. Most studies find a considerable time lag of 6–14 years between start-up activities and their greatest impact on regional performance measures.

3. According to Eurostat, municipalities correspond to the lowest level of Local Administrative Units (LAU 2, former NUTS 5) in the European Union.

4. Because the variable includes zeros, 0.0001 has been added to every observation to enable the log transformation. Models 1a and 1b in Table 5 were also run using non-transformed values, which produced largely the same results.

5. Spatial clustering was tested by a Moran test (Anselin 2005). To this end, a distance-based, row-normalized spatial weights matrix W was used. This matrix includes a neighbouring municipality if its gravity centre of population is within a distance <25 km, which is a common commuting distance in rural Switzerland (Scherer et al. 2011). The matrix W is also used for estimating the empirical models in section 4.

6. To examine the dimensionality of the set of indicators derived from the Delphi survey, the use of statistical scaling techniques, such as e.g. factor analysis, would have been an alternative. Applying a standard exploratory factor analysis with ‘varimax’ rotation, however, did not result in readily interpretable groupings of variables into factors. Furthermore, the resulting factors did not corroborate the grouping into the four components suggested by the experts in the Delphi study, or follow any logic known to us that would be applicable to local entrepreneurship.

7. Experts had the opportunity to weight all indicators according to their relative importance on a scale from 1 to 10 (see Appendix A). However, only weights between 5 and 8 were applied. Consequently, weighted and unweighted variables led largely to the same results in the empirical analysis. The reported results in Tables 5 and 6 are based on unweighted variables.

8. The modelling approach represents a special case of the ‘Multilevel Model for Change’ (Singer and Willet 2003). The models for SILE and its components focus on inter-community differences in change (not within-community change over time) by considering varying intercepts (but not varying slopes). Both characteristics are motivated by the small sample period: in the short run, the time-invariant differences of local economic performance between communities are very likely to dominate time-varying covariates. Similar panel models (i.e., models that do not include time-varying covariates) are quite common in ‘latent covariates growth-models’ in sociological and medical research (Singer and Willet 2003). Latent growth-models generally include more than two observations over time and consider time as an explanatory variable. Hence, we included time as an explanatory variable, but this did not alter any of the remaining coefficients.

9. Employing this vector is equivalent to estimating a dummy-variable fixed effects model. In such a model a dummy variable would be included for each minus one of the municipalities in the data. This would allow estimation of an intercept for each of the municipalities relative to the reference municipality and thus control for heterogeneity across municipalities. However, a dummy-variable fixed-effects model would not allow inclusion of any non time-varying variables, such as the SILE-index.

10. To grasp the relative explanatory power of SILE or its components, the size of relative to is of particular interest. Computing an intra-class correlation coefficient, such as

indicates the random intercept variance as a share of the variance in the dependent variable that cannot be attributed to the explanatory variables and the spatial correlation. Comparing this measure across models allows us to judge to what extent the inclusion or omission of time-invariant explanatory variables reduces or increases the share of the random-intercept variance. By this comparison, the explanatory power of the unobserved heterogeneity relative to the observed variables can be appraised.

11. The spatial mismatch hypothesis was formulated in the 1960s to explain why low-skilled minorities residing in U.S. inner cities achieve only poor labour market outcomes. The basic hypothesis has been adapted to several other contexts (Gobillon, Selod, and Zenou (2007) for an extended review), such as rural economies, with the assumption that residents in these areas are trapped in low-wage and low-skilled employment opportunities. Empirical assessments of rural spatial mismatch are provided by Blumberg and Shiki (2004) and Partridge and Rickman (2008) in the U.S. and Coombes and Raybould (2004) in the U.K.

12. In accordance to Fritsch (2008) we assumed a time-lag of five years between the foundation of a firm and its maximum impact on local economic performance and operationalized business start-ups as a laged variable that is nevertheless fixed across time. This does not affect the results, however. Alternatively, it is also possible to operationalize business start-ups as a time-varying variable. This, however, would have hampered the comparison with the fixed SILE index.

13. Following Elhorst (2010), the expected bias will decrease as the variance of grows relative to the variance of the remaining error component. This might be assessed through the calculation of θ as follows:

The nearer θ comes to 1, the smaller the bias from unduly applying the random-intercept model.