I am on a task force of local healthcare professionals and epidemiologists (don’t ask why someone thought I was competent enough to join something like that). I asked about false positives of tests and learned something interesting. Others can comment and that would be helpful 1n

I was trying to understand in my head what testing means, so I wrote down bayes rule. This is from my mixtape. I can never keep straight which conditional probability in the denominator is sensitivity or specificity (or neither). 2/n

But in my own preferred language, what is the Pr(+ test result | covid infected) and what is the Pr(+ | not infected). Here’s what the ceo of a local public hospital told me. The best data, he said, comes from Wuhan. They appear to use multiple symptoms as testing was endogenous

So they would test if prior belief due to symptoms was very high. I think that meant thinks like fevers and other symptoms. Then if prior belief of infection was high, tests came back positive 70% of the time. 4/n

That’s the true positive. I didn’t get an answer on the false positive of Pr(+ | not infected), which you need for the full decomposition you know Pr(Covid infections | +). Does anyone know? I also learned of a new test with “specificity and sensitivity in the mid-90s”. 5/n

That was a comment made by I think an epidemiologist or public health professional. So I take it there are several tests each of which has different Pr(+|not infected), but you can see in the Bayesian decomposition that prevalence is important for inferring prevalence. 6/n