This Stanford study of coronavirus prevalence in Santa Clara County is really interesting but the population adjustment seems to be doing an awful lot of work medrxiv.org/content/10.110…
Raw result was that 1.5% of 3,330 people tested had antibodies to SARS-CoV-2. Population-weighted prevalence was 2.81%. Adjustments for "test performance characteristics" (i.e. how many false negatives there might be) take it to a range of 2.49% to 4.16%
That takes things from (1) in line with other estimates that actual infections range from about 4X confirmed cases to 25X to (2) the authors' claim that actual infections in Santa Clara County are 50X to 85X confirmed cases
The main thing that makes me suspicious of this result is that you'd think the undercount would be much more severe in NYC, where something close to half of Covid-19 results come in positive, than in a county where the positive rate is only 10.74% ...
… but if you multiply NYC's confirmed cases by 85X then 10 million of 8.4 New Yorkers have already had the coronavirus. I'm no expert, but that seems a little high.
The sample in the Santa Clara County study seems to have been very heavy on white people from the affluent suburbs of Los Altos and Los Gatos, so clearly they needed to do an adjustment, but ... 
It's clear that many more people have had Covid-19 than the confirmed case numbers show, and this study indicates that the undercount is more severe than previously thought. But for some reason I'm not quite ready to embrace the implication that 120% of NYers have had the disease
I *am* willing to believe that 15% or more have, though bloomberg.com/opinion/articl…
In conclusion, it's a paper that makes sweeping and far-out-of-consensus conclusions about the disease's prevalence and fatality rate based on 50 positive tests among 3,330 people who responded to Facebook ads, plus a bunch of model-based adjustments. Use with care
Here’s another critique that touches on some problems I missed https://t.co/ZpwUkguIug
Aaand, here's one of the Facebook ads used to recruit participants. You can adjust and weight all you want but there is no way the people who responded to this ad are representative of the general population technologyreview.com/2020/04/17/100…
