Should be re-written as:
"Preliminary test results suggest 21% of of people approached in a crowded grocery store and who would agree to give blood have Covid antibodies"
All of these tests suffer the same sample bias, and that sample bias is massive.
People who are already in a grocery store (risky behavior), who are willing to give a blood sample (believes they may have been previously exposed) are not the same as people who are leaving their house only for very limited purposes and ordering food online.
If we assume that the % of people shopping in grocery stores and willing to take the test are in risk group A, and the people not shopping are in risk group B. We can run the following quick analysis.
First, assume risk group A is 5 times as likely to have had COVID than Risk group B (but plug any assumption in there necessary)
Then assume that risk group A is only 10% of the population, then rerun the numbers as follows:
For every 100 people in risk group A who tested, 21 were positive.
Risk group A is 100 people, and risk group B is 900.
Risk group A's 100 were 21 prior positive. Risk group B's 900 could be estimated that 4.2 people per 100 were positive, so in total of the 900 people 37.8 were positive.
21+37.8 = 59 people of 1000, or 5.8%.
Plug any numbers you want in the analysis, but the assumptions drive huge variability in the % of New Yorkers infected. Without a less biased sample, we really don't know much other than that way less than 21% of New Yorkers have anti-bodies.