Read on Twitter
So a lot of you are sharing the projections from IHME ( https://covid19.healthdata.org/united-states-of-america). I want to draw your attention to a very optimistic assumption in their modeling.
1/
1/
Let's look at this graph of their Apr 10 projection of Italy's daily death rate. It drops by 45% on the day their current projection begins?! WTF. In reality, there were MORE deaths today than on Apr 9. 2/
In fact, all their charts show a very steep drop off in deaths after peak in a way that doesn't seem to match what we know from Italy, China, or South Korea even by eyeball. What gives? 3/
Well, if you look at their methods from their preprint ( https://www.medrxiv.org/content/10.1101/2020.03.27.20043752v1.full.pdf), you can read (pg4) that they're fitting the cumulative death curve to _a simple gaussian error function_. 4/
You might be more familiar with the derivative of that, which is best known as "the normal distribution". In other words, they've assumed daily death rates are symmetrical around the peak! 5/
For that to be true, the post-quarantine R-factor would need to be the exact reciprocal of the pre-quarantine R-factor (generally thought to
be about 3). That means the epidemic shrinking exactly as fast as it grew! 6/
be about 3). That means the epidemic shrinking exactly as fast as it grew! 6/
There's no biological/epidemiological reason for that. And just looking at Italy, it's almost certainly much higher than 1/3. Which means these
IHME projections are probably optimistic about when the peak will arrive here and how many people will die. 7/
IHME projections are probably optimistic about when the peak will arrive here and how many people will die. 7/
The cumulative fit will start to correct itself as we get weeks past the peak, but from where we are sitting in the US pre-peak, the built-in symmetry assumption of their model is painting too rosy a picture of the future. 8/
