Coronavirus Update VIII: Blown Forecasts & Peak Performance

All the good stuff, caveats, code, data sources and explanations are linked, some in Update III, and the most important in Update II, Update IV, Update V, Update VI, Update VII, Bayes Theorem & Coronavirus, and the Sanity Check Perspective, so go to them first before asking what-about-this-and-that. Skip to the bottom for the latest model. Thanks to everybody emailing me sources, including Ted Poppke, Jeff Jorgensen, Jim Fedako, Joe Bastardi, Philip Pilkington, John Buckner, Harry Goff, John Goetz, Warren McGee. Sorry I’m slow answering emails.

These are growing too long, so I’m moving all non-statistical material out of the updates.

SECTIONS:

1. Our Model Performance
2. Expert Model Performance
3. Flu
4. Our Model Newest Guesses
5. USA! USA! USA!

Our Model Performance

The naive model, which I have been consistently warning us under-predicts until we get to the peak, on 9 March guessed 6,200 worldwide total deaths. Completely blown. Off by an order of magnitude. By the 17th, the model said 14,000—with the same caution given about under-predictions.

By the 24th, 39,000. Last week: 80,000. Now (see below): 130,000.

If you followed the naive model on a daily basis, you would have seen the same steady upward projection like in the week by week totals. But over the last week the numbers actually started decreasing, but only by a fraction. Because it appears we are at the peak, the point at which I said the model would start doing a better job.

Recall that this is only a naive model whose only assumption is that the reported numbers—and not necessarily actual numbers—follow a standard shape. That’s it. We started with it because so many people were going on and on about exponential increases and doubling times—everybody forgetting it is impossible for these increases to be sustained. The exponential shape is wrong.

Our shape is right (though for about a month the slope was off), and if I were to do it all over I’d certainly not estimate the model raw, and instead use prior estimates for the parameters—as a good Bayesian would—from an outbreak that looked like it might be similar. Like from Swine flu (see below). Ah well, I leave this as a class exercise.

Expert Model Performance

Maybe you don’t recall, but back in February, many were predicting tens of millions of deaths across the globe.

A model used at Johns Hopkins in a simulation of a generic outbreak (not the coronavirus) from late in 2019 predicted some 65 million deaths. This was used by some to predict coronavirus numbers, though it was not the model’s original intention. Other early models using “AI” (fancy statistical models with impressive sounding names), said 53 million. Bill Gates came in with the smallest early prediction of 33 million. Another IT pro said 43 million. The figure of 50 million was popular.

There were others. Such as the Ferguson model, which we have dissected before. Do you remember how it was said based on that model we’d have 1 to 2 million dead—if we didn’t follow instructions? In the USA alone.

If we follow instructions, the official model—whatever it was—-on 1 April said “White House projects 100K to 240K US deaths from virus“.

President Donald Trump on Tuesday warned Americans to brace for a “hell of a bad two weeks” ahead as the White House projected there could be 100,000 to 240,000 deaths in the U.S. from the coronavirus pandemic even if current social distancing guidelines are maintained.

Even if. Again, this Birx said “said pandemic forecasts initially predicted 1.5 million to 2.2 million deaths in the U.S.” If we didn’t obey.

People are beginning to question the models. One report complained:

White House officials have refused to explain how they generated the figure – a death toll bigger than the United States suffered in the Vietnam War or the Sept. 11, 2001, terrorist attacks. They have not provided the underlying data so others can assess its reliability, and have not provided long-term strategies to lower that death count…

At a task force meeting this week, according to two officials with direct knowledge of it, Anthony Fauci, director of the National Institute of Allergy and Infectious Diseases, told others that there are too many variables at play in the pandemic to make the models reliable: “I’ve looked at all the models. I’ve spent a lot of time on the models. They don’t tell you anything. You can’t really rely upon models.”

You can’t really rely upon models? Golly!

Certainly not Fauci’s models: “Dr. Fauci claimed that AIDS might be transmissible by ‘routine close contact.’ (May 5, 1983, issue of the Journal of the American Medical Association). (Michael Fumento, The Myth of Heterosexual AIDS, p. 237).”

These did not age well. pic.twitter.com/T8zBZzeuiO

— BrendanEich (@BrendanEich) April 6, 2020

You’ll have to click on that to maximize the pics, but it appears the median peak deaths per day was projected at 2,271, and another from about 1,200 to 3,500 (I’m eyeballing these).

Okay, enough of all that. Save it all for another day: we still have to cover blown hospitalization models, blown ventilators models, etc.

Here’s the gist of all this.

If in the States we don’t get about 100-240 thousand deaths, because we are now obeying, the Official Model can be considered blown.

Therefore, the model believers cannot claimed that they saved us from 1 to 2.2 million deaths. Because that claim is based on the model being accurate. And it isn’t.

Yet you and I both know they will still claim credit. I heard the news conference Monday night at which they were already starting to take credit.

Who takes credit for the flu going away? How does it go away without massive intervention? How much more would totals be for coronavirus without intervention?

Flu

Here, straight from the CDC, are the weekly flu deaths from 2013-2020, up through Week 12 this year.

See 2017-2018? Over 1,500 dead a week at the peak. Hospital- and ICU-stressing deaths. Painful awful deaths. Most of which—all of which?—were unreported by the media. Did they therefore happen? Are these deaths like the tree in the forest which falls outside earshot?

Another thing to note is how typical 2020 is.

Now the weekly deaths due to pneumonia.

These are raw numbers, and are declining, but since population is increasing, the picture is better than suggested. Again, 2020 looks normal-ish. By which I mean, there’s a lot of people keeling over from pneumonia.

Now the weird one: all deaths.

Yes, that drop is in the data: we went from 52,000 in Week 10, to 48,000 in Week 11, to 40,000 in Week 10. Some say this is the CDC printing incomplete data, which given this is the government is easy to believe. It matches the date of the most current flu update, though. And all the other numbers look kosher. Come back in a week and we’ll see.

If it’s right, then the lock downs are working. Turns out firing people from their jobs and locking them inside keeps them from dying of things like car crashes. All we have to do, if we want to save lives, is lock people away forever. Right?

Look how many people would be saved from flu deaths if we locked everybody down, and not just the elderly or sickest. Why don’t we? Why not every year do a lock down? Answer below.

Locking down annually will certainly save medical resources, which we now know are all critically important, as this CDC plots of illnesses due to flu prove (central estimate +/- 95% confidence bounds):

Twenty to fifty million a year. Millions, yes. Every year.

Now hospitalizations:

Hundreds of thousands, to even millions a year. A lot of people in the ER and ICU!

Now deaths:

Median of 38,00 per year, each and every year, and anywhere from 12,00 to 61,000. Maybe even up to 90+ thousand.

Would you say that’s a lot? Would you say it’s a lot if COVID-19 deaths turn about to be ~20,000 in the States?

Final 2020 numbers aren’t in. As of 28 March, CDC said “so far this season there have been at least 39 million flu illnesses, 400,000 hospitalizations and 24,000 deaths from flu.”

Incidentally, it isn’t just the USA. Here’s Europe (the distracting lines are model averages).

Our Model Newest Guesses

Remember: this is a model of reports, not of actual numbers. We are modeling the reporting process, not actual cases and deaths.

One more caveat, which I’ve made before, but not highlighted. If there is a third peak, this naive model can’t see it. Such a peak might come (here’s the highlight) in the Southern Hemisphere winter, when they’re amidst their own flu season.

The Southern Hemisphere has about 800 million souls, and the Northern Hemisphere has the rest of the 7.7 billion. So if our model below is right, and the NH infections burn out around 150 thousand deaths, then a simple extrapolation for a third peak would suggest another 18,000 or so deaths from the SH. That’s a crude, back-of-the-envelope guess based on a substantial if. Don’t make book on it.

Last week the model said this week would be the peak—though our naive model has been consistently under-predicting the peak and totals. Here’s the latest:

Just under 2 million reported cases, and about 131 thousand reported deaths.

Same thing without the log scale.

It’s difficult to tell from these pics, unless you’re used to looking at similar one, but the inflection point has been passed. Which we can see in the dailies.

First, how accurate are reported cases and deaths to actual? Probably not very, as this plot shows: deaths/cases.

If this ratio represented the actual mortality rate, the line would be steady, averaged across the globe, assuming constant virus “strength”. That this plot is still rocketing northwards means (1) cases are under-reported (lack of measurement), (2) deaths over-attributed (dying with is not same as dying from), or (3) both. My guess is (3), with (1) being the most important part.

If this is right, and it is noticed, we may see a ramp up of measurements—which the model cannot see—leading to a big boost in reported cases. But deaths won’t rise as much. We’ll see.

Daily reported new cases.

Somebody asked me elsewhere about the variability, saying, for example, France reported on one day a multiple-day cumulative total right where it looked like we were at the peak, and could we smooth this. I said no, because that’s cheating. We’re not reporting actual cases, but reports. And the choppiness is part of the reporting process.

Again, if no third peak comes now, and which may never come, WE’RE ABOUT TO GO EXPONENTIAL! Everybody remember words like those? Works not just going up, but the other way, too.

Now the logic of modeling the reporting process might be clearer. Those who govern us can see these same numbers—usually on a country-by-country basis. At what point will they issue the PANIC LESS bulletin? Surely not right the peak, because the fear will be it isn’t a true peak. Politicians fear false negatives far more than false positives. My guess is another week, depending on locality.

That would be the point at which the peak becomes indisputable. And the point where the weather starts cutting us a break. Forecast for NYC for Friday night, for instance, is for upper 30s. Thirties! Rain throughout the week. Longer-range says next week in the 60s, and some virus-killing sunshine. LA just had a record amount of daily rain,.

This will probably be the last time we need to show the acceleration plots.

The zero line has been breached. The peak is upon us, or even past. Says the data and the model.

Now daily death reports.

That odd up-down up-down up-down is still with us, suggesting tomorrow is a down day. Which is good, in the sense we didn’t catch down day today (Monday night), else predictions would look too good.

Same story with the acceleration, and easier to see up-down.

Zero line clearly breached.

Finally, here’s the same global totals, now per capita.

Model says reported cases will be 0.025% of population, and deaths about 0.002%. Flu, WHO tells us, kills about 650,000 in a bad year, which is 0.008%. The Wu flu scores a quarter of that, a number which puts us solidly in the Swine Flu category.

Recall Swine Flu estimates were 152-575 thousand dead. And no panic. No global freak out or lock down. Many horrible deaths, but none televised, so they didn’t count.

USA! USA! USA!

Totals for USA, using the same model structure (though of course estimated only using USA numbers) and data from this source (numbers vary across sources: remember how messy medical data is?):

Total reported cases: 510,000. Total reported deaths: 18,500. Which is almost exactly what Swine “What Me, Panic?” Flu was. Other sources, using complex models assuming “full social distancing” say 82,000 USA deaths. (Like ours, they update this routinely, so click over now if you want to see.)

Daily reported cases:

Looks like the peak. Smells like the peak. Feels like the peak. But is it? Hey, maybe, maybe not. I’d bet yes.

I won’t show the accelerations, since I’m betting on the peak. Here’s daily reported deaths.

We’re hearing this will be worst week, which if they include that top number will be. Maybe close or tied to last week. That’s just by looking at the data, not the model. The more complex model linked above says the 16th will be the peak with 3,130 deaths per day. USA alone.

Could it continue to rise as they have it? Hey, maybe, maybe not. I’m betting not. If the weather forecasts are right.

To support this site and its wholly independent host using credit card or PayPal (in any amount) click here