FindingApr 14, 2020

Estimating the Fraction of Unreported COVID-19 Infections

Ali Hortaçsu, Jiarui Liu, Timothy Schwieg
For every COVID-19 case reported in the US in early March, there are likely 6 to 24 unreported cases (accounting for reporting lags).
Estimated Reported Infections by County

The novel coronavirus outbreak was declared a national emergency in the US beginning March 1, 2020, with states imposing various levels of lockdown measures. By April 13, there were nearly 550,000 confirmed cases in the US, with deaths approaching 22,000.[1] While this is clearly a major health crisis, the country is also facing a deep and possibly long-lasting economic recession. One crucial question looming over both the health and economic effects is how many people have actually contracted COVID-19 and the actual mortality rate; that is, while the number of confirmed cases is known, there are likely a large number of cases that have not been confirmed and, likewise, some deaths that have not been attributed to COVID-19.

To address this crucial knowledge gap, the authors have developed a unique strategy to estimate the likely real impact of the COVID-19 pandemic on the US. This strategy is based on the variation in travel from the epicenter of an outbreak to other locations that were not previously infected. Through a series of estimates based on known infection rates and expected rates of transmission, and incorporating the likely effect of travel from an epicenter of an outbreak to other areas, the authors estimate the percentage of unreported cases. The results are striking: for example, on March 13, across major metro areas, the authors estimate that on average only 4.16% of total infections were reported across the US with an eight-day reporting lag, meaning that for every case there were 23 unreported cases. The range of results across model assumptions and time periods utilized vary between 6 to 24 unreported cases.

Finally, while the authors stress that their results are dependent on strong assumptions and reliable data, they believe their methodological strategy is a solid start that can fuel additional research.