Thread by @jnordvig

There is a lag between cases & deaths, and as we see acceleration in cases, there is always a debate about whether fatalities will follow. That debate was there in the US second wave in the summer, in the European 2nd wave in recent weeks, & now again in the US 3rd wave

=thread

US cases are exploding, we all know that...

https://twitter.com/ExanteData/status/1328231655195217922

But fatalities are growing much more slowly. Does it mean that the 3rd wave is totally different?

https://twitter.com/ExanteData/status/1328231651810439168

The simplest way to think about forecasting fatalities a few weeks out is to look at Case Fatality Rates (CFRs). And as we have argued in the US 2nd wave, it can make a big difference if you look at these in the aggregate, or in a disaggregated sense.

This grid (you can zoom in to look at each state more closely) has the CFR calculated dynamically, assuming standard 3-week lags between positive test and fatality (and using 14-day smoothing). It shows that CFRs came down a lot in the spring, but have stabilized in recent months

If we zoom in on the part of the grid with South Dakota and New York, we can see that many states now have CFRs that are converging to the national median, which is around 1.6%. New York is below, however, which is logical since there is a lot of of testing (=lower positivity)

The key point is that CFRs are stabilizing, around 1.6%, rather than continuing to the decline. And by the way, this experience has been similar in Europe (at some point the CFR stops going down).

What does this mean for fatality projections for the next few weeks? If cases are running at around 150K/day and the CFR is 1.6% on average, you should expect deaths to reach 2400/day. If you do a disaggregated calculation, you get somewhat different numbers, but no huge diff.

Our daily update, that does the projection based on disaggregated CFRs, is slightly higher (around 2600/day) in three weeks time. The explanation is that we have concentration of cases in states with higher CFRs (places w bigger unrecorded outbreaks)

https://twitter.com/ExanteData/status/1328292788253888512

You can look at a snapshot of case fatality rates by states here. When batches of fatalities are recorded, the CFR can be temporarily fluky (such as Georgia now). Ignore the outliers. The key point is that a lot of CFRs are in the 1-2% range & the time series above show stability

Based on the case data (and CFRs) we have available now, there will be a tick up in daily fatalities to 2000-2500 in coming weeks. And whether that will be the actual peak in this cycle will depend on which policy action is taken in the short-term to slow the case growth.

The experience in Europe over the last few weeks (and in the US over the summer) shows that 'lockdown lite' can have a major impact on case growth. Hence, the policy choice is not black and white, especially not if tough decisions are taken early on.

But if action is too soft, or too late, the outbreak will generate explosive case growth; and the stable CFRs suggests that fatalities will then eventually catch up. Likely not at the speed of the spring wave. But very significant nevertheless. I will leave it at that. END