Supply and demand of digital skills in Europe: a statistical analysis of regional labour markets through latent variable and spatial regression models

This thesis explores the distribution of digital skills in Europe, adopting a labour market perspective. On the supply side, attention was focused on Eurostat’s data on ICT usage. A Multidimensional Item Response Theory (MIRT) analysis was performed on the items selected by Eurostat to compute its Digital Skills Indicators, conceptually based on the “Digital Competence framework for citizens” (DigComp). The estimation of exploratory MIRT models suggested a five-dimensional latent structure, different from DigComp. This data driven structure was used to compute five count indicators, representing individuals' proficiency in the corresponding dimensions, serving as manifest variables in a Multilevel Latent Class Analysis. With the help of selected covariates, five clusters of individuals, i.e. attitudes towards ICTs, and four groups of regions, i.e. levels of diffusion of digital skills, were obtained. Some issues related to the goodness-of-fit assessmentof MIRT models for binary variables were addressed in a model-based simulation study. The effectiveness of three decision strategies was evaluated across eighteen scenarios. Quantifying the degree of fit along a continuum via fit indices seemed to be more effective than testing the hypothesis of exact fit. For these indices, more restrictive cutoff values than those used in factor analysis were proposed. Lastly, data from online job advertisements collected by Cedefop in 2022 enabled to integrate information on the demand side. On the basis of the regional number of mentions of digital skills, a Percentage index of Digital skill Imbalance (PDI) was defined. The study of the determinants of PDI was undertaken in a spatial data analysis. An original route map which compared spatial econometric specifications, Geographically Weighted Regression (GWR) and Generalized Additive Models (GAMs) was adopted. The final model was a multi-scale GWR which distinguished two groups of covariates on the basis of the scale, i.e. local vs global, of their effect on PDI.