Fuzzy approach to discrete data reduction: an application in economics for assessing the skill premium

Abstract

Measures of stock of skills alternative to human capital have raised fresh difficulties, especially in data managing. We propose to empirically compare the efficiency of a hierarchical cluster analysis and a fuzzy clustering in reducing discrete skill data. The outcomes of both methods are subsequently used to measure the impact of skills on earnings in addition to human capital. The proposed methodological comparison was made using an original dataset of retail bankers’ skills assessed by supervisors. Empirical evidence shows that the fuzzy approach is more efficient than the hierarchical clustering: the resulting clusters are fewer and easier to interpret. Furthermore, the earnings equation enriched with skill variables allowed us to correct the education premium, and provides information on monetary incentives related to individual skills. Our paper attempts to raise researchers’ and practitioners’ awareness of data reducing methods, and their implications for wage determinants.