THE SECOND OF TWO POSTS LOOKING AT THE STATISTICS WHICH INDICATE HOW VIRULENT AND HOW SERIOUS THE COVID-19 REALLY IS. [NOTE THAT THERE ARE OTHER POSTS IN THIS SERIES BY Dr. BRIGGS].
Coronavirus Update IV — Take A Deep Breath — If You Still Can
Apologies for the duplicate email! I hit the wrong button yesterday and published the incomplete update meant for today.
All the good stuff, caveats, and explanations are linked, some in Update III, and the most important in Update II, so go there first before complaining. Or skip to the bottom for the latest model.
Personal update: I haven’t seen any panic in NYC. Toilet paper still on shelves. On a flight to Florida on the weekend was one passenger wearing surgical gloves, and a mother insisting on wiping down everything her kids would come into contact with.
Condoms are selling out, because why? Because people are putting them on their fingers to avoid coronavirus. And this.
Panicked yet?
“Don’t joke, Briggs. You don’t understand. It’s much worse than they’re saying.”
That so? How do you know?
“Because the numbers they’re reporting are wrong, faked, too low.”
No kidding? How do you know?
“Because it’s much worse than they’re saying.’
That so? How do you know?…
Iterate ad nauseum.
Yes, because anything other than instantly reporting numbers on iffy tests in rambunctious medical circumstances points to a conspiracy. I’ve worked with medical data for twenty years, and the best you can say about it is it’s a mess. Look at our own data. It’s not like the entire world has got together to decide to release 100% accurate counts at 8 PM Eastern Savings Time. Numbers come in from all over, staggered and rough. We must account for this in our minds when wondering whether how much to trust the model.
Last Friday I did some numbers, which unfortunately I cannot update, because my number source stopped carrying totals for individual Chinese cities. A real pain in the kiester. Nevertheless, here’s what I did then. The idea is still sound:
Hubei province, Google tells me, has 58.5 million people. China (Monday morning, EST) had 80,735 reported cases, which are spread all over the country of one billion souls. Even if all cases were in Hubei, the population case rate is 0.14%. China deaths (also Monday morning) 3,119. This is a 3.8% in-case death rate, or 0.005% death rate for the whole population. Tuesday morning update: 80,754 cases, 3,136 deaths, almost no change.
And the numbers are slowing fast (get it?) in China. Cities nearby Hubei aren’t “going exponential”—the most favored phrase I read—except in the trivial sense that going from 1 to 2 to 4 cases is exponential, but not especially concerning.
Panicked yet?
Here’s another take:
In case you can’t see it, it reads:
Here’s another “As the coronavirus spreads, one study predicts that even the best-case scenario is 15 million dead and a $2.4 trillion hit to global GDP“. If you read the story, this is the “best-case scenario”, too.
On the other hand, some caution: Why Novel Coronavirus Fatality is Likely Overestimated.
CDC on the flu, week ending 29 February (these are always 1-2 week delayed):
A lot of people on our side of the divide are on Trump for on Monday pointing out last year saw 37,000 flu deaths in the USA. As if him saying it making its wrong or unimportant. Everything is political.
Italy and Iran had many more deaths. Iran might be slowly, Italy nearing its peak, probably.
Now you will remember what happened with SARS (see the links above where I keep repeating this). It was bad at the hot zone, and in a couple of other remote places, like Canada, which had very high, double-digit death rates. But it was almost benign in many other remote places. Similar kind of thing is happening with coronavirus. Whether is remains like this is only a guess. These guesses may be way off and this thing escalates like the AHA says. It hasn’t done so by now in China, but hey, we could have bad luck.
What has apparently (or might have) happened with coronavirus is that there are two strains, the worst in China and perhaps Italy and Iran, and the milder most everywhere else. I haven’t seen any confirmation this is so (about who has what), but it would explain the differences.
Incidentally, the tests people are using for diagnosing coronachan, and there are many, are not perfect. With the heightened publicity, i.e. panic, many more are going to be checked and surely some false positives are finding their way into the case numbers. That’s less likely with deaths, but not impossible. That means the true overall death rate might be higher than we think. But it also much less infectious than we think. We won’t know any of this for sure until long after, when it becomes the interest of obscure scholars.
Here’s another hot take. Barbie said, “Math is hard!”
No, wait. This one. This PhD said (click to see the whole thread):
So many more headlines! Like Ross Douthat’s “My Sunday column: The Coronavirus is Coming For Trump’s Presidency” in which he among other cringey things wrote the eye-rolling “Obviously the White House isn’t to blame for everything that’s gone wrong with the coronavirus response.” Rod Dreher is taking the Fr James Martin role. Never Trumpers in general are trying to score political points.
The WSJ had a headline I’m too lazy to find that ran something like “How deadly plagues will become more common”, which was redolent of global warming concern.
Finally, a word on cancellations. I said it last time, but it bears emphasis. Many companies and politicians are banning travel, forbidding gatherings and the like. Some of this is surely concern and appropriate. But I’d bet much is fear—-fear of the crowd. Just what would happen to MegaCorp Inc. Ltd. if an employee gets coronavirus on an official business trip when all the other companies have stopped flying?
What would happen to the politician who wasn’t seen spending billions on brother-in-law contracts and sees an outbreak in his constituency? To ask is to answer. All this adds to the panic.
One thing that we cannot do is this, which I see everywhere. Given “I don’t know what’s really going on” we cannot conclude “It’s really bad out there!” This is Talebism. The only thing you can say if you don’t know what’s really going on is that you don’t know what will happen. To say anything else is to take a black swan dive.
Here is the code and the data (see earlier posts for details). All numbers from 8 PM EST, Monday night. You MUST read updates II and III for the code notes if you’re playing along; all the caveats, and there are many, are there.
Here is the naive model applied to the total cases and deaths:
The guess for total cases is ~160,000; total deaths ~6,200. This is higher than last week, but I changed the date of the second peak start to make the totals higher, since it seemed model was catching the top of the secondary peak. That it might be is seen in the next picture.
Here’s the daily cases:
Have we reached the secondary peak? If so, then the model is probably not terrible, though I’d guess it’s an underestimate, as it was during the primary peak. If the same pattern holds, multiply everything by (today) 1.1 to 1.3. That multiplicative adjustments goes to 1 as we approach the secondary peak, naturally.
What if there’s a third peak? Obviously, this naive model can’t see it. But neither can we see it in the data. It’s only a guess one way or another. We have to wait and find out.
Here’s the daily deaths:
Just as we cautioned last week, the daily deaths necessarily had to lag daily cases, for you can’t die until after you get sick. Again, look for the media to tout the deaths and not the cases. But because the deaths haven’t yet reached the secondary peak, the daily forecasts are surely too low. Also, deaths are more than in the primary peak, which is of interest, and suggests there might be a third peak, but smaller.
Looks like we have a couple more weeks of madness ahead of us.
Addendum
The daily data per country is here; see China, for instance. The model can be used on the country data, too, and it would make a great exercise to do so.
I also said this:
People are taking actions to prevent spread of the disease (especially hyper vigilant actions) as absolute evidence of the prevalence of the disease. This is like saying the number of people who buckle up makes a good estimate of number hurt or killed in car crashes.
Many normally sober people with expertise who are making predictions are very scared, it seems, of being on the wrong side. They’re fearful of hearing “You said it wouldn’t be bad, but look how many died!”
But they’re willing to hear “You way over-forecast.”
This is wishcasting and wrong.
The decision people make based on the forecast can be weighted lopsidedly; very willing to suffer false positives, say.
But the best forecast is always the most accurate, regardless of the cost-loss of decisions.
We’re aiming for accuracy here with the most unsophisticated model we can make, taking only into account the “shape” of viral outbreaks (this is also a stats class post). It hasn’t been terrible. It might turn terrible, but then it also might stay un-terrible.
Of course, I have to guard against the opposite, and be careful not to under-forecast because of all the over-forecasts.
All the good stuff, caveats, and explanations are linked, some in Update III, and the most important in Update II, so go there first before complaining. Or skip to the bottom for the latest model.
Personal update: I haven’t seen any panic in NYC. Toilet paper still on shelves. On a flight to Florida on the weekend was one passenger wearing surgical gloves, and a mother insisting on wiping down everything her kids would come into contact with.
Condoms are selling out, because why? Because people are putting them on their fingers to avoid coronavirus. And this.
Panicked yet?
“Don’t joke, Briggs. You don’t understand. It’s much worse than they’re saying.”
That so? How do you know?
“Because the numbers they’re reporting are wrong, faked, too low.”
No kidding? How do you know?
“Because it’s much worse than they’re saying.’
That so? How do you know?…
Iterate ad nauseum.
Yes, because anything other than instantly reporting numbers on iffy tests in rambunctious medical circumstances points to a conspiracy. I’ve worked with medical data for twenty years, and the best you can say about it is it’s a mess. Look at our own data. It’s not like the entire world has got together to decide to release 100% accurate counts at 8 PM Eastern Savings Time. Numbers come in from all over, staggered and rough. We must account for this in our minds when wondering whether how much to trust the model.
Last Friday I did some numbers, which unfortunately I cannot update, because my number source stopped carrying totals for individual Chinese cities. A real pain in the kiester. Nevertheless, here’s what I did then. The idea is still sound:
Diamond Princess carried 2,670 and 1,100 crew, and had 696 coronavirus cases [still 696], or 18%. 6 dead [it’s now 7, but I’ll leave this as it is], which is 0.8% among cases, or 0.16% in toto.China’s numbers have long since slowed, and almost stopped (not of flu, year by year!)—the rate of increase has slowed, I mean. The totals necessarily can only increase. The Wu Flu began in December, and it’s not growing worse at the hot zone, or really anywhere in China.
This ship makes nice upper estimates: consider the tight quarters and mandatory mingling and isolation.
Then look at Shagnhai (close to Wuhan), which had Monday about 337 cases and 3 deaths—months after the outbreak. Shaghai has 24.24 million souls.
Thus: 0.14% case rate, 1% in-case death rate, 0.00001% death rate in toto.
A nice lower estimate.
A little too low, though. Applying Shanghai to world [7.7 billion] gives ~110,000 total cases, ~1,000 deaths.
We now have ~100,000 cases, ~3,400 deaths. But near the secondary peak.
Princess to world: ~142 million cases, ~12.3 million deaths. With no indication data trending that way.
Hubei province, Google tells me, has 58.5 million people. China (Monday morning, EST) had 80,735 reported cases, which are spread all over the country of one billion souls. Even if all cases were in Hubei, the population case rate is 0.14%. China deaths (also Monday morning) 3,119. This is a 3.8% in-case death rate, or 0.005% death rate for the whole population. Tuesday morning update: 80,754 cases, 3,136 deaths, almost no change.
And the numbers are slowing fast (get it?) in China. Cities nearby Hubei aren’t “going exponential”—the most favored phrase I read—except in the trivial sense that going from 1 to 2 to 4 cases is exponential, but not especially concerning.
Panicked yet?
Here’s another take:
In case you can’t see it, it reads:
American Hospital Association “Best Guess Epidemiology” for #codiv19 over next 2 months:Those stats are for the USA alone, not the world. Though I did see an ackshually guy say it’s not the whole AHA but just one professor. These kind of numbers are not uncommon from all kinds of sources, however.
96,000,000 infections
4,800,000 hospitalizations
1,900,000 ICU admissions
480,000 deaths
vs flu in 2019:
35,500,000 infections
490,600 hospitalizations
49,000 ICU admissions
34,200 deaths
Here’s another “As the coronavirus spreads, one study predicts that even the best-case scenario is 15 million dead and a $2.4 trillion hit to global GDP“. If you read the story, this is the “best-case scenario”, too.
On the other hand, some caution: Why Novel Coronavirus Fatality is Likely Overestimated.
CDC on the flu, week ending 29 February (these are always 1-2 week delayed):
CDC estimates that so far this season there have been at least 34 million flu illnesses, 350,000 hospitalizations and 20,000 deaths from flu.USA deaths from coronachan: 22. Germany, incidentally, which had more than twice the number of USA cases had 0 deaths (so far).
A lot of people on our side of the divide are on Trump for on Monday pointing out last year saw 37,000 flu deaths in the USA. As if him saying it making its wrong or unimportant. Everything is political.
Italy and Iran had many more deaths. Iran might be slowly, Italy nearing its peak, probably.
Now you will remember what happened with SARS (see the links above where I keep repeating this). It was bad at the hot zone, and in a couple of other remote places, like Canada, which had very high, double-digit death rates. But it was almost benign in many other remote places. Similar kind of thing is happening with coronavirus. Whether is remains like this is only a guess. These guesses may be way off and this thing escalates like the AHA says. It hasn’t done so by now in China, but hey, we could have bad luck.
What has apparently (or might have) happened with coronavirus is that there are two strains, the worst in China and perhaps Italy and Iran, and the milder most everywhere else. I haven’t seen any confirmation this is so (about who has what), but it would explain the differences.
Incidentally, the tests people are using for diagnosing coronachan, and there are many, are not perfect. With the heightened publicity, i.e. panic, many more are going to be checked and surely some false positives are finding their way into the case numbers. That’s less likely with deaths, but not impossible. That means the true overall death rate might be higher than we think. But it also much less infectious than we think. We won’t know any of this for sure until long after, when it becomes the interest of obscure scholars.
Here’s another hot take. Barbie said, “Math is hard!”
No, wait. This one. This PhD said (click to see the whole thread):
We’re looking at about 1M US cases by the end of April, 2M by ~May 5, 4M by ~May 11, and so on. Exponentials are hard to grasp, but this is how they go. 4/nI pointed out:
Exponentials sure are hard to grasp. Logistic curves are even harder.I point this kind of thing out, and some get it. Some, though, are almost angry, as if any good news about this virus is unwelcome.
Infections resemble logistic curves and not exponentials, for the excellent reason that exponentials always predict *everyone* will be infected in time. And this never happens. 1/
Look, even CHINA (a billion souls) does NOT have MILLIONS infected, even after 2-3 months of the outbreak. How is it we’re going to beat their numbers so far from the hot zone? 2/
So many more headlines! Like Ross Douthat’s “My Sunday column: The Coronavirus is Coming For Trump’s Presidency” in which he among other cringey things wrote the eye-rolling “Obviously the White House isn’t to blame for everything that’s gone wrong with the coronavirus response.” Rod Dreher is taking the Fr James Martin role. Never Trumpers in general are trying to score political points.
The WSJ had a headline I’m too lazy to find that ran something like “How deadly plagues will become more common”, which was redolent of global warming concern.
Finally, a word on cancellations. I said it last time, but it bears emphasis. Many companies and politicians are banning travel, forbidding gatherings and the like. Some of this is surely concern and appropriate. But I’d bet much is fear—-fear of the crowd. Just what would happen to MegaCorp Inc. Ltd. if an employee gets coronavirus on an official business trip when all the other companies have stopped flying?
What would happen to the politician who wasn’t seen spending billions on brother-in-law contracts and sees an outbreak in his constituency? To ask is to answer. All this adds to the panic.
Onto the Numbers!
We had the initial peak, then the spread with the expected secondary peak. Will there be third and subsequent peaks? Will the coronavirus be like the flu or the ordinary coronavirus and be with us perennially? Will this really explode like it did in China everywhere? Hey, maybe, maybe not. Nobody knows. Not so likely though, because deadly contagious diseases tend to fade out or only pop up from time to time, like ebola or MERS.One thing that we cannot do is this, which I see everywhere. Given “I don’t know what’s really going on” we cannot conclude “It’s really bad out there!” This is Talebism. The only thing you can say if you don’t know what’s really going on is that you don’t know what will happen. To say anything else is to take a black swan dive.
Here is the code and the data (see earlier posts for details). All numbers from 8 PM EST, Monday night. You MUST read updates II and III for the code notes if you’re playing along; all the caveats, and there are many, are there.
Here is the naive model applied to the total cases and deaths:
The guess for total cases is ~160,000; total deaths ~6,200. This is higher than last week, but I changed the date of the second peak start to make the totals higher, since it seemed model was catching the top of the secondary peak. That it might be is seen in the next picture.
Here’s the daily cases:
Have we reached the secondary peak? If so, then the model is probably not terrible, though I’d guess it’s an underestimate, as it was during the primary peak. If the same pattern holds, multiply everything by (today) 1.1 to 1.3. That multiplicative adjustments goes to 1 as we approach the secondary peak, naturally.
What if there’s a third peak? Obviously, this naive model can’t see it. But neither can we see it in the data. It’s only a guess one way or another. We have to wait and find out.
Here’s the daily deaths:
Just as we cautioned last week, the daily deaths necessarily had to lag daily cases, for you can’t die until after you get sick. Again, look for the media to tout the deaths and not the cases. But because the deaths haven’t yet reached the secondary peak, the daily forecasts are surely too low. Also, deaths are more than in the primary peak, which is of interest, and suggests there might be a third peak, but smaller.
Looks like we have a couple more weeks of madness ahead of us.
Addendum
The daily data per country is here; see China, for instance. The model can be used on the country data, too, and it would make a great exercise to do so.
I also said this:
People are taking actions to prevent spread of the disease (especially hyper vigilant actions) as absolute evidence of the prevalence of the disease. This is like saying the number of people who buckle up makes a good estimate of number hurt or killed in car crashes.
Many normally sober people with expertise who are making predictions are very scared, it seems, of being on the wrong side. They’re fearful of hearing “You said it wouldn’t be bad, but look how many died!”
But they’re willing to hear “You way over-forecast.”
This is wishcasting and wrong.
The decision people make based on the forecast can be weighted lopsidedly; very willing to suffer false positives, say.
But the best forecast is always the most accurate, regardless of the cost-loss of decisions.
We’re aiming for accuracy here with the most unsophisticated model we can make, taking only into account the “shape” of viral outbreaks (this is also a stats class post). It hasn’t been terrible. It might turn terrible, but then it also might stay un-terrible.
Of course, I have to guard against the opposite, and be careful not to under-forecast because of all the over-forecasts.
No comments:
Post a Comment