Estimating COVID-19 under-reporting through stochastic frontier analysis and official statistics: A case study of São Paulo State, Brazil

During outbreaks, natural disasters, and any unexpected event, it is usual that public authorities face issues tracking the outcomes, and the available reports of casualties tend to be lower than actual numbers. Using weekly data at the municipality level in São Paulo State (the most densely populated region in Brazil), we employ stochastic frontier methods to fit the dynamics of the COVID-19 outbreak spanning from March 2020 to December 2021. The empirical model incorporates the inverse hyperbolic sine transformation to address the issue of zero reporting of COVID-19 deaths. Furthermore, we utilize a flexible frontier method to capture the S-shaped epidemic curve in two distinct waves of infections/deaths. Our results reveal that the actual death toll resulting from COVID-19 is, on average, 1.24 times higher than the officially reported figures. Consequently, these findings hold significant implications for public authorities in identifying regions characterized by substantial under-reporting of COVID-19 fatalities. Moreover, this study presents an empirical framework that can be used for other municipalities, states, or countries confronting outbreak scenarios.