“Our best estimate right now is for every case reported there were actually 10 other infections,” Dr. Redfield said.
The source for his claim was unclear at that time. The C.D.C. later posted the data on its website and on MedRxiv, a repository for scientific results that have not yet been vetted by peer review.
The C.D.C. researchers tested samples from 11,933 people across six regions in the United States during discrete periods from March 23 through May 3: The Puget Sound region of Washington where the first Covid-19 case in the country was diagnosed, as well as New York City, South Florida, Missouri, Utah and Connecticut.
The samples were collected at commercial laboratories from people who came in for routine screening, such as cholesterol tests, and were evaluated for the presence of antibodies to the virus — which would indicate previous infection even in the absence of symptoms.
The researchers then estimated the number of infections in each area. New York City, for example, reported 53,803 cases by April 1, but the actual number of infections was 12 times higher, nearly 642,000.
The city’s prevalence of 6.93 percent is well below the 21 percent estimated by the state’s survey in April. That number was based on people recruited at supermarkets, and so the results would have been biased toward people who would be out shopping during a pandemic — young people, or those who had already had the virus and felt safe, experts said.
Saskia Popescu, an epidemiologist at the University of Arizona, said the C.D.C. survey might also be slanted by people with chronic conditions who are more likely to visit commercial labs. Still, it is more representative of the general population than other surveys because it included everyone who came to the labs for a variety of purposes, instead of limiting it to specific groups such as health care workers who felt sick from the coronavirus.