There’s an urban legend about a Texas man who takes a rifle to the side of his barn and sprays bullets across the wall, more or less at random. Then he finds the densest clusters of holes and paints a bull’s-eye around each one. Later, a passerby, impressed by this display, trots off in search of the marksman. In a reversal of cause and effect, the Texas Sharpshooter is born.
The Sharpshooter Fallacy is often used by scientists to illustrate our tendency to narrativize data after the fact. We may observe an unusual grouping of cancer cases and back into an explanation for it, cherry-picking statistics and ignoring the vagaries of chance. As we muddle through COVID-19’s winter surge, the story holds a deeper lesson about the perils of interpreting data without a full appreciation of the context. Omicron, because of its extraordinary contagiousness and its relative mildness, has transformed the risks and the consequences of infection, but not our reading of the statistics that have been guiding us through the pandemic. Do the numbers still mean what we think they mean?
A coronavirus infection isn’t what it once was. Studies suggest that, compared with Delta, Omicron is a third to half as likely to send someone to the hospital; by some estimates, the chance that an older, vaccinated person will die of COVID is now lower than the risk posed by the seasonal flu. And yet the variant is exacting a punishing toll—medical, social, economic. (Omicron still presents a major threat to people who are unvaccinated.) The United States is recording, on average, more than eight hundred thousand coronavirus cases a day, three times last winter’s peak. Given the growing use of at-home tests, this official count greatly underestimates the true number of infections. We don’t know how many rapid tests are used each day, or what proportion return positive, rendering unreliable traditional metrics, such as a community’s test-positivity rate,…