4 ms·
I totally agree that the confirmed case counts are way too low, because even most people who were symptomatic weren't able to get tested (e.g. me), let alone ra
by CydeWeys 7y ago
I totally agree that the confirmed case counts are way too low, because even most people who were symptomatic weren't able to get tested (e.g. me), let alone random people who were asymptomatic.
But the 21% study is seriously flawed because it didn't do a random sampling of the population. We need that at a minimum to know with any certainty what the actual exposure rate is. The figures that are coming back from studies using random samples in other places have been much lower.
- SpicyLemonZest 7y agoIt did a random sampling. We should do followups to screen off possible biases people have proposed, but stopping random people in the grocery store is by any reasonable standard randomization.
- matwood 7y agoNo. It’s a random sampling of people who are out during lockdown. It can’t being extrapolated to the whole population when large parts are not leaving their homes.
- SpicyLemonZest 7y agoAnd if you go around to people's homes, you'll oversample people who aren't out during the lockdown. There's no silver bullet here.
- matwood 7y agoYou do both. This is why study design matters. And it's one of the reasons all of the early antibody studies have issues (the other being test accuracy).
- CydeWeys 7y agoIt's almost like you need to do your random sample based on a list of all residents, and not just go out and try to find people at various locations.
- SpicyLemonZest 7y agoYou say "almost like", but scientific studies rarely sample the population this way because researchers generally don't have access to a list of all residents.
- joshuamorton 7y agoWell, the type of information you're trying to gather is rather unique. Usually we just wait for a virus to run course then test lots of people to see the resulting case counts. But we can't do that here. Normally you just use a control and test group, but that doesn't work for figuring out underlying infection rates. There are some tests trying to sample everyone in a geographic area (SF Mission census block) but the data isn't out yet because they're conducting tests as we speak.
- SpicyLemonZest 7y agoI guarantee you that when the data comes out: * It will also show an undercount of at least an order of magnitude. * Commenters will still pop up to explain why the results can't be trusted and which further studies are absolutely required before we believe them.
- joshuamorton 7y agoI guess we'll see. I don't share your certainty. Although I'd love to be able to go outside sooner. Also elsewhere in this thread it's mentioned that the Florida and Santa Clara results could be entirely explained by high type II error in the test. The Florida test appeared to have a false positive rate of ~15% when independently validated, which is basically the infection rate they found. In other words, this is a specific form of base rate fallacy where the test accuracy is really low.
- suzzer99 7y agoDoes anyone not think case count is an order of magnitude less than infected count?