Samples are drawn from Western Washington, Southern Florida, New York City, Connecticut, Utah, and Missouri, with additional locations—San Francisco, Louisiana, Minnesota, and Philadelphia—to come on line soon.pic.twitter.com/JyXlECxXUC
You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
Samples are drawn from Western Washington, Southern Florida, New York City, Connecticut, Utah, and Missouri, with additional locations—San Francisco, Louisiana, Minnesota, and Philadelphia—to come on line soon.pic.twitter.com/JyXlECxXUC
Measured seroprevalences range from a low of 1.1% in Seattle in late March to a high of 6.9% in NYC at the same time.pic.twitter.com/6DEltrBtFu
A full research paper posted to the medRxiv yesterday provides more details on methodology and results.https://www.medrxiv.org/content/10.1101/2020.06.25.20140384v1 …
I have no particular reason for skepticism regarding the results. Perhaps someone with appropriate expertise can comment on the appropriateness of the antibody assays.pic.twitter.com/x1NGjDwpVv
One thing to be aware of with the study is that it does not use a random sample of the population. That's not surprising. Obtaining blood draws from a truly random sample is notorious difficult. This is a convenience sample, i.e., they use what was available.pic.twitter.com/TwfLmzSiVF
The hope is that the people who had blood draws for non-COVID-related reasons have about the same likelihood of getting COVID as those who did not. One can think of many reasons why this might be incorrect—but still it seems a reasonable sample to consider.
So we have an estimate that roughly 10 times as many people have been infected as have tested positive. This is unsurprising. It's consistent with most other estimates obtained by a range of methods over the past three months.
Keep in mind that these samples were taken some time ago; that ratio could drop as testing increases. We have 2.5M reported cases now. If that ratio is remains correct, that would mean there have been about 25 million cases in the US, or about 7.5% of the population infected.
With about 125K deaths to date, this gives a fatality rate of >0.5 percent, because total deaths will climb somewhat as some of the currently infected people die. Again, no surprise: this is in the 0.5-1.0% range that most epidemiologists have been suggesting for months now.
One thing to keep in mind as you look at the paper (other than possible bias due to convenience sampling) is that the confidence intervals in Figure 1 are not posterior confidence intervals but rather bootstrap confidence intervals.pic.twitter.com/I6oZiS2Yev
They do not account for sampling bias and other sources of uncertainty; they tell you about the variability in the samples themselves and the degree to which corrections for demographics and false positives/negatives could be influencing the estimates.
I have not looked into the methodology underlying the corrections for false positives and false negatives. Especially for the lower-incidence ares, this could be important and we've seen it done improperly in some previous analyses.
Overall, the paper pretty confirms a lot of what we already believed without challenging much of anything. That may not be the most exciting result, but it's useful to have confirmation from another study, across multiple areas, conducted on this scale.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.