I haven't used chatgpt yet, so I'm interested to see what I can do! My email is [email protected]

I'd also be happy to look at your code at some point this weekend if that would be helpful

I'm also a little surprised by how much the positive oversample pulls down the upper end of the CI for p, although again that might be a bootstrap artifact

It generally looks good to me! The one thing that stands out are the confidence intervals, especially sensitivity and specificity - they seem a little tighter than I'd expect and the CI for sensitivity doesn't even include the true value. Is that just an artefact of bootstrapping?

It definitely seems like someone should have! But I don't know where. It's pretty far outside what I typically read, though, so that's not particularly informative. If you come across something or write it up yourself, please send it along, I'd love to read a more formal treatment!

Yeah, depending on your priors and how many re-tests you do, your new confidence interval for the prevalence is probably something like 0.039 - 0.047, and that also means that your specificity has a CI of something like ~0.953/0.96 - ~1

But you need significantly more gold-standard tests than 1/prevalence in order to estimate the prevalence in the first place.

I need to write it out more formally to check, but I think that if you have a good measure of the prevalence, then yes. In the extreme scenario where you know the prevalence exactly and get a positive retest on all of the cheap test's positives, then you know that the false negative rate must be 0.

I don't think that it is possible to reliably say anything about the specificity in this scenario. If 1/1000 people have the disease even with a false negative rate of 100% and all 200 tests used in the negative group, there is still over an 80% chance of observing 0 false negatives in your sample.

The conscious embrace and enabling of anti-vax by the GOP leaders, the ones who know better, is probably one of the most unambiguously evil things I’ve seen in my lifetime.

water supplies are a) not a problem in the east b) a big problem in the west that could be more or less solved overnight by cutting livestock feed agriculture by like a fifth. data centers are a rounding error