1\\ A bit of a speculative thread on why the @IHME_UW model has failed so catastrophically

To recap: hospitalizations for COVID-19 fell outside the IHME's 95% confidence interval *twice* in one week, despite the IHME revising their model

The chances of this happening are like 0%

2\\ The common rejoinder is that "The model didn't fail, social distancing just worked really well"

That's cute bullshit. The IHME models assumed "full social distancing" from the outset

And if your model isn't falsifiable, it isn't a model

So, how DID the model fail so badly?

3\\ IHME built their model around two core assumptions, one reasonable and one disastrous:

A) That only countries past their peak in deaths should be considered. This is fair.

B) Lockdowns work, so since the US is locking down, only consider countries with lockdowns.

4\\ That "lockdowns work" seems too obvious to dwell on. But consider that lockdowns have costs as well as benefits:

Benefit: enforce social distancing outside the home

Cost: keep everyone at home, which is where a respiratory disease spreads most readily

5\\ Now, the crucial observation:

Most social distancing benefits of lockdown had ALREADY BEEN ADOPTED VOLUNTARILY

People are not totally dumb. When you scare them about a disease, they wash their hands more and avoid restaurants https://twitter.com/DKThomp/status/1248432702757302280?s=20

6\\ So, lockdowns force people to remain at home, where they infect each other

This may explain why lockdowns and death counts are *positively* correlated

Lockdowns are not a cause of improvement, they are a political reaction to large death counts https://twitter.com/boriquagato/status/1248264375267282944?s=20

7\\ Consider what that means for the IHME model:

Demography, climate, and lifestyle meant that Italy/Wuhan/Spain had bad outbreaks, so they were the first to lock down

So, *only countries with bad outbreaks* were included in the model! https://twitter.com/ElonBachman/status/1248360411017236480?s=20

8\\ As I've quipped elsewhere, IHME's approach to modeling has been like basing an actuarial table off the Kennedy family

They are assuming the US is exactly like Wuhan, Lombardy, and Madrid, but nothing like Japan, Korea, Germany

No wonder their model fails within 48 hours!

9\\ The IHME model has another problem: it uses deaths on a "when reported" basis rather than a "when occurred" basis

Because many municipalities experienced data backlogs in late March, early April has seen a global "catch up" in reporting: https://twitter.com/ElonBachman/status/1248383934036996107?s=20

10\\ Summary: the @IHME_UW model is sophomoric hackwork. They will revise their US death forecast again this week, to below 60,000