Clusters of stem jobs across Europe

Abstract

Using geographic information system techniques, this article explores the formation of STEM jobs from a regional perspective. It uses data on European regions, specifically from the Nomenclature of Territorial Units, to first show (by means of Getis-Ord general G statistic and the global Moran’s I statistic) the existence of regional clusters of STEM jobs, which then it explores by means of Getis-Ord Gi*, a local indicator of spatial association. Finally, it estimates, both globally (fitting an ordinary least squares, regression model) and locally (fitting a geographically weighted regression), the extent to which different institutional and firm-based factors contribute to their formation. Findings reveal North–South and West–East differences in how factors contribute to the formation of STEM clusters in Europe. In particular, whereas STEM employment in the North seems to depend more on tertiary education, secondary education is more important in the South. On the other hand, whereas in Western Europe, a favorable learning environment seems to trigger the concentration of STEM jobs, extrinsic factors such as pay seem to be more important in Eastern Europe.

Appendix

The general G statistic of overall spatial association is given as

$$G = \frac{{\mathop \sum \nolimits_{i = 1}^{n} \mathop \sum \nolimits_{j = 1}^{n} w_{ij} x_{i} x_{j} }}{{\mathop \sum \nolimits_{i = 1}^{n} \mathop \sum \nolimits_{j = 1}^{n} x_{i} x_{j} }}, \quad \forall \;j \ne i$$

where $x_{i}$ and $x_{j}$ are attribute values for NUTS regions $i$ and $j$, and $w_{ij}$ is the spatial weight between both regions. $n$ is the number of regions in the dataset. Then for inferential purposes, a z-score of $G$ is calculated.

The global Moran’s I statistic for spatial autocorrelation is given as

$$I = \frac{n}{{S_{0} }}\frac{{\mathop \sum \nolimits_{i = 1}^{n} \mathop \sum \nolimits_{j = 1}^{n} w_{ij} z_{i} z_{j} }}{{\mathop \sum \nolimits_{i = 1}^{n} z_{i}^{2} }}$$

where $z_{i}$ is the deviation of an attribute for region $i$ from its mean, is the spatial weight between regions regions $i$ and $j$. $n$ is the number of regions in the dataset. $S_{0}$ is the aggregate of all the spatial weights. Then a z-score of $I$ is calculated.

The Getis-Ord Gi* local statistic is calculated as follows

$$G_{i}^{*} = \frac{{\mathop \sum \nolimits_{j = 1}^{n} w_{ij} x_{j} - \bar{X}\mathop \sum \nolimits_{j = 1}^{n} w_{ij} }}{{S\sqrt {\frac{{n\mathop \sum \nolimits_{j = 1}^{n} w_{ij}^{2} - \left( {\mathop \sum \nolimits_{j = 1}^{n} w_{ij} } \right)^{2} }}{n - 1}} }}$$

where $x_{j}$ is the attribute value for region j, $w_{ij}$ is the spatial weight between region i and j, n is equal to the total number of regions and

$$\bar{X} = \frac{{\mathop \sum \nolimits_{j = 1}^{n} x_{j} }}{n}$$

$$S = \sqrt {\frac{{\mathop \sum \nolimits_{j = 1}^{n} x_{j}^{2} }}{n} - \bar{X}^{2} }$$

For statistically significant positive z-scores of $G_{i}^{*}$, the larger the z-score is, the more intense the cluster of high values (hot spot). For statistically significant z-scores, the smaller the z-score is, the more intense the cluster of low values (cold spot). A z-score near zero indicates no apparent spatial clustering.