On Wednesday, May 20, 2020, the Centers for Disease Control and Prevention told The Atlantic magazine that it had been counting two different coronavirus tests as if they measured the same thing. A viral test, taken by nasal swab, shows whether someone is infected right now. An antibody test, drawn from blood, shows whether someone was infected weeks or months earlier. Two days before, the agency's own COVID Data Tracker had stated plainly that its numbers represented only viral tests. By Wednesday, with no announcement, that sentence had been replaced by vaguer wording that no longer ruled antibody results out.
The distinction mattered because states had built their reopening plans around one figure: the share of coronavirus tests coming back positive. A falling positivity rate was the signal that an outbreak had cooled enough to loosen restrictions, and the arithmetic assumes every test in the count is answering the same question. Ashish Jha, then director of Harvard's Global Health Institute, explained why blending the two types would distort it: antibody tests go mostly to people who are not currently sick, so they return far fewer positives than viral tests do, and folding the two together would drive the reported rate down without any change in how many people were actually infected.
The CDC's own reporting showed the pattern Jha was describing. On Monday, May 18, its tracker had recorded 10.2 million viral tests nationwide, 15 percent of them positive. By Wednesday, after the tracker's wording changed, it recorded 10.8 million tests of any type, and the positivity rate had dropped a full percentage point. Over those two days the agency's total grew by 630,205 tests, of which only 52,429 were positive, a swing in the direction Jha's mechanism predicted. Pennsylvania, Texas, Georgia, and Vermont were blending the two test types on their own dashboards; Vermont's health department later said it had not realized it was doing so. Virginia had done the same until reporters caught it a week earlier and the governor reversed course.
None of this required anyone to invent a result. What moved was the boundary drawn around which tests belonged in the same total. A CDC spokeswoman told reporters the agency would report the two test types separately within weeks. Texas's Republican governor denied blending the numbers when asked about it directly, even as local reporters found the state still doing it; Virginia's Democratic governor had already admitted the same practice and apologized for it.
A positivity rate is not a fact about a virus. It is a fact about a count, and a count means only what its definition says it means. The volunteers who spent that spring building an independent tracker to check the government's own numbers later described exactly this: states lumping antibody results in with diagnostic tests, inflating totals, and confusing the CDC along with everyone else.