How Do Universities Contribute to Employment Growth? The Role of Human Capital and Knowledge Bases

Abstract

The aim of this paper is to analyse whether employment growth is faster in regions housing a university compared to non-university regions. We argue that universities per se are less likely to trigger externalities that facilitate employment growth. Instead, we propose that it depends on the concentration of different skills in that particular region. This is analysed by running a number of ordinary least squares regressions, based on official data on a municipal level from Statistics Sweden, on how concentrations of human capital, analytic, synthetic and symbolic knowledge bases in Swedish university regions influence employment growth in 2002–2008. The results indicate that the presence of universities per se does not influence employment growth. However, the findings suggest that university regions with high concentrations of human capital and, in particular, with employees characterized by the synthetic knowledge base, show higher growth rates. This implies that the influence of universities on employment is greatest in regions with high concentrations of skills able to apply the knowledge created in universities. Consequently, the regional composition of skills needs to match the knowledge produced by universities for significant university-induced spillovers to occur.

Acknowledgements

This paper was originally an unpublished magister thesis that has been adapted to an article. We would like to thank Aina Tollefsen and fellow students for their comments on the early version of this paper. We are also grateful to Magnus Strömgren for providing the inter-municipal distance matrix used for the accessibility measures, as well as to the two anonymous referees whose comments significantly improved the paper. The usual disclaimers apply.

Notes

1. Due to a change in the administrative structure of Sweden the municipality of Knivsta has been excluded from the analysis as it was broken out from the Uppsala municipality in 2003. To cope with the sometimes drastic changes in the time series data between the years when the municipality was broken out it has been added back to the Uppsala municipality again and will therefore not constitute a problem in the analysis. This means that the total number of cases in the analysis is 289.

2. An alternative way to handle the spatial autocorrelation would be to estimate the models as hierarchical two-level multilevel models (Semyonov & Lewin-Epstein, 2009). However, a subsequent likelihood ratio-test revealed that the multilevel model was not superior to the OLS and was consequently excluded from the analysis.

3. It would have been desirable to also distinguish between different universities depending on when they where established to see if there were differences between new an old universities. Unfortunately the universities that could be considered as newly established (after the year of 1990) but has not been established after the starting point of the time span for this analysis (2002) were quite few in numbers (4) which made it not possible to include in the analysis as an OLS regression then would have a hard time finding significant relations because of the shortage of samples.

4. The general histogram of the residuals of the different models did not show any particular signs of severe skewness. The normal probability plots, indicated a very slight sinus wave immediately adjacent to the normal distribution line indicating that the models residuals are not 100% normally distributed. This pattern was however assessed to be of marginal importance to the performance of the models. A residuals vs. fits plot was also conducted for each model with similar results showing the residuals plotted without any special patterns visible.