Local mortality estimates during the COVID-19 pandemic in Italy

Estimates of the real death toll of the COVID-19 pandemic have proven to be problematic in many countries, and Italy is no exception. Mortality estimates at the local level are even more uncertain as they require strict conditions, such as granularity and accuracy of the data at hand, which are rarely met. The ‘official’ approach adopted by public institutions to estimate the ‘excess of mortality’ during the pandemic is based on a comparison between observed all-cause mortality data for 2020 with an average of mortality figures in the past years for the same period. In this paper, we show that more sophisticated approaches such as counterfactual and machine learning techniques outperform the official method by improving prediction accuracy by up to 18%, thus providing a more realistic picture of local excess mortality. The predictive gain is particularly sizable for small- and medium-sized municipalities. After showing the superiority of data-driven statistical methods, we apply the best-performing algorithms to generate a municipality-level dataset of local excess mortality estimates during the COVID-19 pandemic. This dataset is publicly shared and will be periodically updated as new data become available.