[burger]

Was just looking at the IHME site and it looks like it’s projections are holding up relatively well across most states and countries I checked out. CA is actually almost dead on the predictions, if not under. I like the IHME data because it is totally concerned about deaths and projected resources needed on the treatment side, not case growth impossible to know with such limited testing. Deaths are firm outcomes that can be tracked and projected much more accurately than number of infections. I’m sure the modeling incorporates some measure of case growth, but given the wildly different testing practices depending on where you live, displaying that data doesn’t help much.

The IHME predictions only look good because the model is constantly updated to fit the past data.  If you try to predict ahead with this model, it's basically useless.  Someone did a paper (which I'll never find now) looking at how good the models were just two days out, and most states were well outside the model predictions. 

Another issue is that the models show relatively quick declines in the disease, but if you look at numbers of cases and deaths in places like Italy and NYC, etc., the actual data is showing a relatively long tail,  meaning the disease is taking longer than expected to peter out.  Look at the curve for Italy: https://covid19.healthdata.org/italy  They predict a swift decline, but the numbers of deaths has been pretty steady since coming down off the peak.  You can see similar patterns elsewhere.

This isn't just an academic issue--it's going to get people killed.  If states start reopening things based on model predictions that are overly optimistic, we're going to have a big problem.  There will be another round of community spread and exponential growth again.

[burger]

Was just looking at the IHME site and it looks like it’s projections are holding up relatively well across most states and countries I checked out. CA is actually almost dead on the predictions, if not under. I like the IHME data because it is totally concerned about deaths and projected resources needed on the treatment side, not case growth impossible to know with such limited testing. Deaths are firm outcomes that can be tracked and projected much more accurately than number of infections. I’m sure the modeling incorporates some measure of case growth, but given the wildly different testing practices depending on where you live, displaying that data doesn’t help much.

The IHME predictions only look good because the model is constantly updated to fit the past data.  If you try to predict ahead with this model, it's basically useless.  Someone did a paper (which I'll never find now) looking at how good the models were just two days out, and most states were well outside the model predictions. 

Another issue is that the models show relatively quick declines in the disease, but if you look at numbers of cases and deaths in places like Italy and NYC, etc., the actual data is showing a relatively long tail,  meaning the disease is taking longer than expected to peter out.  Look at the curve for Italy: https://covid19.healthdata.org/italy  They predict a swift decline, but the numbers of deaths has been pretty steady since coming down off the peak.  You can see similar patterns elsewhere.

This isn't just an academic issue--it's going to get people killed.  If states start reopening things based on model predictions that are overly optimistic, we're going to have a big problem.  There will be another round of community spread and exponential growth again.