Evaluation of occupational safety and health incentives Understanding the territorial bias

The EU-OSHA (2013) underlines that OSH interventions are not evaluated by means of rigorous evidence- based research. Moreover, the greatest majority of interventions concerns regulations. In this panorama, the Italian case is particularly interesting. In 2008, a system of economic incentives has been introduced. This represents a revolution because it leverages corporate social responsibility towards workers. Since 2010, Inail has financed OSH projects, which go under the name of "ISI calls", for about 2 billion euros. Despite this represents a big budget and quite long pilot experimentation, at the political level the balance is still leaning in favour of sticks rather than of carrots. Scientific evidence on the impact of incentives for investments on the reduction of accidents and diseases could help to solve this political dilemma. We overview the main theoretical and methodological aspects in counterfactual evaluation of OSH policies. Then we will present some empirical spatial analysis trying to find proxies of the attitudes of firms towards safety in different regions. The variables that explain the risk of occupational incidents are size and sector. If these variables were able to explain the whole variability of risk, it would be possible to estimate rather precisely the expected value of the incident frequency in a territory, by using the composition of its economic system following these variables. On the other hand, a difference between the expected and the observed frequency index, which we could label "territorial bias", would give a measure of unobservable determinants connected to the characteristics of the territory. We rely on very detailed data on occupational safety to estimate the territorial bias. Then we will do a preliminary impact assessment of ISI incentives on occupational safety. The ISI calls are appropriate for counterfactual impact evaluation, since they are based on the click-day, which configures as a natural experiment, with a control group identified among the participants that were not selected by the click. Nevertheless, the evaluation design of this policy is not exempt of difficulties, including identification of proper outcomes, identification of the causal relation, the problem of self-candidacy into the experiment and external validity. We will limit our analysis to a subset of actions in which the identification of the outcome and of the causal relation between the incentive and the outcome is clear. As for the problem of external validity, a fundamental contribution could come from the analysis of the territorial bias.