Cluster-based measures of regional concentration. Critical overview

This paper provides an overview of the available cluster-based measures of geographical and sectoral concentration (often referred to as specialisation) and tests their statistical behaviour using the Monte Carlo simulation. The study proves that the degree of the aggregation of the dataset matters in the result and that this sensitivity to the Modifiable Areal Unit Problem is inherited in most of the measures. Gini, Krugmann, Theil’s H, and Ogive together with other analysed indices are proved to be non-absolute measures that are dependent on the values in the surroundings. Two regions with the same internal industrial (sectoral) structure but with a different share in the overall volume will have a different sectoral concentration index, which limits the inter-regional comparability of these measures. The results also indicate that the information capacity of the measures could be the same, mainly due to the construction of the measures. Thus, in regional comparisons a justified selection of the measures from the different information clusters is a necessity. The empirical ranges of the measures are narrower than the expected theoretical ranges, which causes the interpretation to be more restrictive. The Mantel test and the correlation analysis show, that the innovations in the input data, such as rescaling or permutation, do not alter the results significantly.