Measuring organised crime in Italy: dual composite indicators integrating text mining and official statistics

This study introduces a dual, data-driven framework for mapping organised crime in Italy by constructing and cross-validating two complementary, per-capita indices aggregated through a non-compensatory composite methodology. The municipal index mines more than a decade of ANSA (Agenzia Nazionale Stampa Associata) news (2012–2023) using Natural Language Processing to extract and geolocate mafia-related events at the municipality level, and complements these signals with computer-vision/optical character recognition (OCR) extraction of clan presence from DIA (Direzione Investigativa Antimafia) investigative maps. The provincial index relies on official records of mafia-related offenses, the dissolution of local administrations due to mafia infiltration and the number of clans, all normalised by population. Across provinces, the composite indices align strongly (a correlation ≈ 0.75). Moreover, for the two offenses unambiguously linked to mafia activity (Type I offenses), news-based counts correlate closely with official records (0.92 for mafia association and 0.88 for mafia-type murders/attempted murders), supporting the use of media narratives as a proxy for underreported activity.