I've been thinking hard about test positivity rates and how they impact on difficulty of estimating incidence and prevalence when there aren't random samples from the population. Blog post with #rstats at http://freerangestats.info/blog/2020/05/09/covid-population-incidence.

We know that test positivity rates can be used as a pragmatic indicator for malaria. But the relationship between these rates and actual incidence is non-linear. This accords with common sense that higher positive rates probably mean we'd find more cases if we did more tests.

In the case of COVID-19 in USA, test positivity rates have been very high, and people like @NateSilver538 are rightly monitoring them as a pragmatic indicator of incidence and data quality. Is it possible to adjust confirmed case counts for positivity to get better pop estimates?

I think we *can* adjust how we extrapolate from confirmed cases to pop estimates and even Reff by taking positivity into account, but it's only a minor improvement and can't fix big problems of data quality and missingness. More in the blog at http://freerangestats.info/blog/2020/05/09/covid-population-incidence