abstract
- An important strategy for data classification consists in organising data points in clusters. The k-means is a traditional optimisation method applied to cluster data points. Using a labour market database, aiming the segmentation of this market taking into account the heterogeneity resulting from different unemployment characteristics observed along the Portuguese geographical space, we suggest the application of an alternative method based on the computation of the dominant eigenvalue of a matrix related with the distance among data points. This approach presents results consistent with the results obtained by the k-means.
- Using a cross-section database that observes the Portuguese labour market in two different phases of the business cycle, the present paper aims to address the issue of the segmentation of the Portuguese labour market taking into account the heterogeneity resulting from different unemployment characteristics observed along the Portuguese geographical space and applying two optimization clustering methods: the k-means and the spectral methods. The k-means is a traditional optimisation clustering method applied to cluster data observations. Spectral clustering is an alternative method based on the computation of the dominant eigenvectors of a matrix related with the distance among data points. The results obtained by the two methods are not identical but are very close and show that, apart the economic phase of the cycle, Portugal presents two very different profiles of registered unemployment. One of them can be considered problematic because it presents a higher percentage of unemployed women, long duration unemployed and unemployed with low levels of formal education—these are the groups that present more difficulties in the labour market and for which is more difficult to find a job after losing one. The segmentation of the labour market is a reality and the labour market is not adjusting to the business cycle.