Workplace accidents, economic determinants and underreporting: an empirical analysis in Italy

Abstract

Purpose

This paper proposes an analysis of occupational accidents in Italy at the regional level. For this purpose, our panel is composed of 20 regions over the 2010–2019 time span.

Design/methodology/approach

We apply different econometric estimation techniques (pooled OLS model, panel fixed and random effects models and semiparametric fixed model) using INAIL and ISTAT data. Our models investigate workplace accidents at the regional level by accounting for socioeconomic, labour market and productive system variables and controlling for possible underreporting bias.

Findings

Overall results reveal the existence of a relevant under-notification phenomenon of accidents at work with respect to moderate accidents, that is higher especially for the southern regions of Italy. However, when considering as outcome variable an alternative set of more severe workplace accidents our model specification remains highly jointly statistically significant. Among our main findings, the analysis shows that worker skills (blue collar) strongly affect the regional pattern of workplace accidents, i.e. an increase of 1% of low paid employees generates about an increase of 1.8 severe workplace accidents per thousand workers. Moreover, we provide evidence that the size of the firm is inversely related to the occupational accident rates. Finally, our results highlight a nonlinear relationship between GDP and occupational accidents for the Italian regional context, confirmed by the high statistical significance of the quadratic term in all the estimated linear models and by the semi-parametric analysis.

Originality/value

A first element of originality of our study consists of investigating the macro determinants of occupation accidents at a regional Italian level. Second, the empirical literature (Boone and Van Ours, 2006) highlights the possible bias of underreporting behaviours on nonfatal accidents in contrast to fatal accidents that are always reported. From this perspective, we have identified a few analyses (namely, Boone et al., 2011) considering different accident sets characterised by different severity degrees. Thus, this paper contributes to the literature considering five alternative subsets of accidents stratified by degree of severity (i.e. moderate, severe, moderate plus severe, severe plus fatal and total accident rates) to test for possible underreporting bias affecting our econometric model.