Estimating The Mortality Rate For COVID-19 | COVID-19 Dashboards about covid19-dashboard HOT 4 CLOSED

@ebressert - thanks for the feedback.

I completely agree about testing biases. I spent the better part of yesterday afternoon attempting to (a) build a new model entirely around (normalized) tests per million (from this data source - https://ourworldindata.org/coronavirus#testing-for-covid-19) and (b) adding that as a country-level covariate to this existing model.

The up-shoot for (b) is that to date there are so few countries that report testing counts, that the covariate had no effect (beta estimate near zero). Further, because many of the worst-struck countries (Italy & China) now have some of the largest rates of testing, the covariate is a bit of an admixture between countries that have been good at testing from day 1 and countries that only deployed lots of testing after they were hammered by COVID-19.

The up-shoot for (a) is that I was trying to go directly after the true case counts for each country, using tests per million as the model input (and assuming the same hierarchical model for mortality rate). At the end of the day, I couldn't get the model sampling to converge, likely (again) because only ~30 countries currently report testing rates :(

Regarding days since lockdown -- that's a good idea. I'd have to think about how to model that (the model here wouldn't work very well, since it assumes the country-level covariates are static). Something that tried to measure the impact of various types of lockdown would be interesting.

from covid19-dashboard.

For testing biases you could have a normalized feature for number of tests per million people. Won't remove all biases, but it would help.

from covid19-dashboard.

Also, days since lockdown. You could have a sequential category feature where you have something like the following:

WFH
WFH + closure of restaurants/bars
WFH + closure of restaurants/bars + curfew
etc.

These should be government mandated directions. Countries where folks have to do this themselves at the organizational and sub levels (like the US) will be quite noisy.

from covid19-dashboard.

@jwrichar ☝️

from covid19-dashboard.