1/ I'm going to try to explain this once more for all the blue-checks insisting that they KNOW lockdowns work.

The models failed. In real time. Even in NY state, the April 1 iteration of @IHME_UW model estimated four times the hospitalizations last week as were taking place.

2/ But it's crucial to understand WHY they failed.

There is a TIME LAG between infection, symptoms, hospitalization, ICU care, and death. 5 days on average from infection to symptoms, several more days to hospitalization, 2-3 more to ICU care. These are averages, but correct.

3/ The models assumed (not unreasonably) that in places like New York, lockdowns had come TOO LATE. Between March 17 and March 25, hospitalizations in NY rose 7X. The models assumed that because of the time lag, the rise would go on UNTIL THE EFFECTS OF THE LOCKDOWN KICKED IN...

4/ By that point, >5,000 new patients every day would need hospitalization - and ultimately New York would need 140,000 hospital beds and 30,000-40,000 ventilators. (Remember those numbers?)

5/ BUT THE MODELS WERE WRONG. After March 25, new hospitalizations remained roughly flat in NY for the next 10 days. Now they have plunged. That’s why the overflow in NYC is basically empty. (Outside NY, the trends are similar, and it doesn't matter when the lockdowns started.)

6/ Now, there are several possible reasonable arguments around this, including that voluntary social distancing had some effect before the lockdown. And there’s a much broader question as to what is happening right now in Spain and Italy – or Japan, or Sweden.

7/ But before we can have that debate, everyone needs to underestimate that the models didn’t fail last week because they UNDERESTIMATED the effect of the lockdowns. They failed because they OVERESTIMATED the effect. Anyone who says otherwise just doesn't understand the lag.

Understand, not underestimate. Sorry.