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, and Bayes Theorem & Coronavirus, 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, Jim Fedako, Joe Bastardi, Philip Pilkington, Willie Soon, Harry Goff, John Goetz, Warren McGee. Sorry I’m slow answering emails.
Do not miss the Perspective On The Coronavirus: Sanity Check which shows how the coronavirus fares with respect to other panics—-I mean pandemics.
Again I ask you to consider how watery our spines would become if the media picked something like flu to comment on, shrieking the day-by-day hourly by hourly case and death increases of everybody who happened to be infected with the flu virus—-not just killed by the virus, infected with it.
Did you not know flu stresses hospitals at its yearly peak? Did you not know it comes in clusters like this new virus? We could shout these predicaments from every platform!
If we really worked at it, we could have the world’s population at fever pitch at least half of every year.
This is happened after yesterday’s lock down post. People have lost their damned minds. Florida man (Governor) tried 15 May, but chickened out and later said he really mean 15 April. Virginia honorary black man (Governor) saw the 15 May and raised it to 10 June. Yes, the 10th of June, the biggest screw-you to citizens yet.
This has all the earmark’s of a typical purity spiral. Maybe we should have a pool. What public leader will be the first to breach 1 July? Answer in comments below.
Another point. Couple of weeks ago I told you about a doctor at one hospital in Washington state who moaned about being surrounded by death and devastation, back when there were fewer than 40 deaths in the entire state. I called him on it, and he deleted the tweet announcing his misery.
I’m seeing many more of these oh-woe-is-look-at-me anecdotes. One guy reported a weepy anecdotes from a nurse in a southwest state (the names aren’t important). Reported cases and deaths at the time were very low. I pointed out the nurse couldn’t have been right. The person still believed the nurse.
People want to be part of this crisis so much that they don’t see the harm in juicing their stories with extra color. This is too big not to tell good skin-crawling tales. “Lying” is not the right word for this embroidery. It’s much more akin to how the media paints every Olympic athlete a victim of some kind, so that otherwise boring events (like where they twirl those pieces of clothe while dancing) have some interest.
Try not to believe everything you hear.
Update. 7:50 AM
Sanghi deleted the tweets below, saying potential trouble with the data: see new tweet. The claim from other sources is the data is likely okay, but reporting difficulties (ahem: see below) make it suspect. However this shakes out, the question to us is still important: should we lock down if it saves lives? Thanks to Blessed Bradys for alerting us.
Want proof this lock down is working? Here from economist Siddhartha Sanghi is a graph (data from CDC) showing USA raw weekly death counts.
Down now about 10,000 per week, across all ages. The biggest relative drop is in 0-17 year olds.
Same thing for other age groups: see his whole thread for other breakdowns.
This is happening the same time coronavirus deaths are increasing. Meaning, overall many fewer people are being whacked.
Here from Nick Szabo is a graph showing the rate of other infectious diseases found in samples.
Flu, common cold, everything down. This is trickier, because this is the time of year when these things always tail down; flu especially drops to near 0 around now, or a couple of weeks from now. Depends on the weather—and spring is running late, especially in New York! Still, the rapid drop-off can almost surely be ascribed to the lock down.
We don’t need more evidence. This is sufficient to prove that locking people down saves lives. At least those lives lost to some diseases and to such things as car accidents.
Deaths down 10,000 per week! Think if we kept the lock down going for a whole year. That’s 520,000 lives saved, maybe more. Half a million! More than the coronavirus will ever get.
If your metric is lives lost, then lives lost are lives lost. We might—should—allow that losing an 85 year old with multiple comorbidities to an infectious disease is not as bad as losing a healthy 17 year old to a car crash. If I need to explain why, you have no heart—or commonsense.
It is indisputable lock downs save a huge number of lives, and even lives of the most value to a society. Save them for what? Never mind! The lock downs don’t necessarily stop infectious diseases, for this is a practical impossibility, but they slow them sure enough. Flu, for instance, kills 20-30,000+ a year in the States. Every year. Car crashes (we saw earlier) kill about 37,000. Every year.
So why don’t we lock down all the time? Well, we can’t lock everybody down quarantine style all the time. Starvation would set in. But we can greatly proscribe peoples’ activities. In the same way a modern 40 year old first-time mother helicopters over her only child, we can limit the places people go, we can dictate the modes of transportation, we can enforce curfews for all citizens. We can mandate—not just now, but every winter—masks must be worn at all times. No public—or private—displays of affection.
We can crack down. We can make mothering the law of the land. We can save lives. Why don’t we? If we, for instance, banned abortion we could save over 600,000 lives per year. Alas, in a matriarchy, women are confused about whether their own children should live (hence waiting until almost too late to let one kid escape the womb). Abortion won’t be banned.
In all other cases, we could wrap ourselves in a blanket of fear, shudder in the dark, frightened of every threat or perceived threat. We could magnify imperilments in our mind and move to protect ourselves from them, no matter how unlikely. After all, if our actions saves even one life, it will be worth it.
Life in quantity is superior to life in quality.
Which I believe is the motto of slaves everywhere.
I quote the peroration of Rusty Reno’s rousing reproach:
Alexander Solzhenitsyn resolutely rejected the materialist principle of “survival at any price.” It strips us of our humanity. This holds true for a judgment about the fate of others as much as it does for ourselves. We must reject the specious moralism that places fear of death at the center of life.
Fear of death and causing death is pervasive—stoked by a materialistic view of survival at any price and unchecked by Christian leaders who in all likelihood secretly accept the materialist assumptions of our age. As long as we allow fear to reign, it will cause nearly all believers to fail to do as Christ commands in Matthew 25. It already is.
I quote myself:
Since this is the internet, I am forced to write: Saying do not hysterically overreact is not equivalent to saying do nothing. Wear a mask, wash your hands, don’t be reckless with old folks. Do I have to be your mother? Do not panic.
A model is a list of premises, which usually includes observational statements, from which are deduced conclusions. The conclusions are often stated in probabilisitic form. If the premises of the model are necessary truths (like 1+1=2), the model itself is true.
The further the model’s premises are from necessary truths, i.e. the greater the number of unproved or hopeful assumptions, the greater the suspicion the model may be false, or of no use. Models which are not true in the strict sense may still be useful. The only way to prove a non-true model’s mettle is to test it against Reality.
Regular readers will recall we have done this with hundreds of expert models through the years, concluding Murphy’s Law applies to even the smartest person’s theories. We have been doing that test with our own naive model, and found it wanting in certain aspects, which we’ll discuss below.
How about the expert coronavirus models?
The ubiquitous Fauci has a model which says there will be 100-200,000 deaths in the USA alone. That’s 50 times more than what we have seen so far, and far more than in China (about whether they are lying, see the special section below).
A different model than Fauci’s (I presume) says “COVID-19 could lead to more than 80,000 deaths”. It’s worth noting some details of this second model:
Forecasters at the Institute for Health Metrics and Evaluation (IHME) at the University of Washington’s School of Medicine analysed the latest COVID-19 data at a local, national and international level.
These include hospitalisation and mortality rates, as well as patient date in terms of age, gender and pre-existing health problems.
Specifically, they looked at the time lag between the first fatal cases and public interventions such as shuttering schools and businesses.
They then looked at each American state’s ICU bed and ventilator capacity.
It isn’t wrong to try approaches like these. But they quickly grow in complexity, and should grow in uncertainty, because none of the numbers and assumptions fed in are themselves certain. This is so in physical models, and even truer in models of human behavior. The authors of this model appear to understand that, and say “Estimates ranged between 38,000 and more than 160,000” dead bodies, which is a large window.
Now this window incorporates the outer neighborhood of 2009’s swine flu, which killed anywhere from 10,000-18,000 Americans (see Update VI for details). See what I mean, though, about medical numbers? Even now this death count is an estimate and not a hard count. We’re acting like we have hard counts on coronavirus, which isn’t likely, as we’ve seen in Italy: deaths of people with coronavirus are not equivalent to deaths from coronavirus. Numbers for this pandemic will be estimated and re-estimated in the years to come, long after the public has lost interest.
There was no lock down with swine flu—nor for the flus from the 50s or 60s, which killed ten times as many people, and which most have forgotten even happened. Yet they all still went away. That there was no lock down meant that taking credit for their disappearance was difficult—but not impossible. Somewhere out there, at this moment, is a politician taking credit for the sun rising in the east.
This Fauci, in a White House press conference I watched on Sunday evening, is already taking credit, saying 2.2 million would have died if it were not for our actions. Here’s a breathless article restating that. He got this alarming figure from his model. The number saved is a counterfactual, meaning there is no way to prove it unless we accept Facui’s model is perfect. Stop to grasp this.
If at least 100 thousand Americans don’t die from coronavirus, we’ll know Fauci’s credit taking was wrong, because his model has proved itself a failure.
Even model failure will not stop those in power everywhere from asserting it was they who saved your skin. Remember to ask them, as they reward themselves, how is it the other, worse diseases went away without their supervision? Their response will like that WHO bureaucrat who was asked about Taiwan (he pretended not to have heard the question, then hung up on the journalist), or they will insist this Wu-flu was indescribably worse than the other pandemics, and it was only thanks to their efforts that it didn’t rage out of control. I base this prediction on my own model of human behavior. Wait and see if I’m right.
Most of you will have heard of the ups and downs of Neil Ferguson, the Brit who first predicted doom for England using a theoretical model which was, and “remains the model being followed by world governments“.
Ferguson and his model originally predicted some 250,000-510,00 deaths, in a population of 66 million. That more-or-less matched Fauci’s one or two million in the USA, adjusted for population size differences. Same model, says this source.
The Telegraph did some digging on Ferguson (no link; article was sent to me):
He [Ferguson] was behind disputed research that sparked the mass culling of farm animals during the 2001 epidemic of foot and mouth disease, a crisis which cost the country billions of pounds.
And separately he also predicted that up to 150,000 people could die from bovine spongiform encephalopathy (BSE, or ‘mad cow disease’) and its equivalent in sheep if it made the leap to humans. To date there have been fewer than 200 deaths from the human form of BSE and none resulting from sheep to human transmission.
Mr Ferguson’s foot and mouth disease (FMD) research has been the focus of two highly critical academic papers which identified allegedly problematic assumptions in his mathematical modelling.
The scientist has robustly defended his work, saying that he had worked with limited data and limited time so the models weren’t 100 per cent right — but that the conclusions it reached were valid.
Ah. It was wrong but it was right.
What Ferguson might mean is that the conclusions were deduced from the model, and so, conditional on the model, they were true conclusions. This is what I call in Uncertainty a local truth, as opposed to a universal or necessary truth, which is a proposition that is always true, even if you don’t want it to be.
These local truths are why scientists—smart people!—are so in love with their models. They put as much sweat and care into them as Pygmalion did his sculpture. How can they abandon them, even after the fail to conform to Reality? They cannot: they instead distrust Reality. I do not jest. Ever notice how global warming modelers don’t blame their models for forecasting failures, but instead look to the historical data and “adjust” it, sure history can’t be right, since it if were the models, which are beloved, would work. Don’t you believe the same thing won’t happen here, since death counts will be re-estimated.
This goes for statisticians, too! I keep reminding myself to call my model “the naive” model so that I don’t develop undue fondness for it.
This Birx, in apparent contradiction to Fauci, said “Models are models. When people start talking about 20% of a population getting infected, it’s very scary, but we don’t have data that matches that based on our experience.”
She said the media should not “make the implication that when they need a hospital bed it’s not going to be there, or a ventilator, it’s not going to be there, we don’t have evidence of that.”
“It’s our job collectively to assure the American people,” she also said. “There is no model right now — no reality on the ground where we can see that 60% to 70% of Americans are going to get infected in the next eight to 12 weeks. I want to be clear about that.”
She later walked that back, saying 100-200,000 is a “best case” scenario. Politics is still politics.
The modelers are getting pretty specific. Here’s one headline: America’s darkest day: Modelling predicts 2,271 Americans will die from coronavirus on April 15 and the pandemic will stretch past June. Not 2,272, but 2,271. Sheesh.
When this is all over, we’ll be able to look back on these models and discover precisely why they failed—if they fail. Model failure, if it exists, will likely be for the usual reasons: too heavy a reliance on uncertain premises, and the reification of certain mathematical equations. criticism has already begun, and a meatier criticism here. (I don’t want to do this until it’s all over, because most people won’t care why the models failed.)
Ferguson, Fauci and others are all modeling what they say are cases and deaths caused by coronavirus. Just what are we modeling? Something different! I quote myself again:
The naive model we have been using, which started as a class project, is confusing some. Rather, the numbers are. What are they?
Reports. Our model is a model of the reporting of numbers. That’s it, and nothing more. To the extent these numbers represent real cases, accurately ascribed deaths, and diligent records, then our model will attempt to describe the real extent of the outbreak. It’s “attempt” because, of course, the model is far from perfect.
Now, even in the case of measurement error in counting cases, improperly ascribed deaths, and chaos and inconsistencies in reporting, the model will still attempt to describe what numbers are being reported. Reporting is a very human process, and that’s what we’re modeling, the process.
How accurate are the numbers in reflecting Reality? It wouldn’t surprise anybody who has worked with medical data over a long period of time to say “not very”.
This approach avoids a lot of difficulties, such as trying to estimate the mortality rate. This is because there isn’t one. There is a R0, there is a chance of getting it, there isn’t a chance of anything. All these are conditional, and conditions vary dramatically. I want to avoid this to maximum extent possible, hence focusing only on global reports.
Pants On Fire
The assumption is that China is lying about death counts, because, well, they’re China, and they’re evil. The virus must have been worse there, because look how scared I am.
I have no idea if they are lying. Since we’re modeling reports it doesn’t really matter. I do know dubious attempts to prove lying when I see them. This is worth doing because it focuses on how everybody’s a sudden expert.
Take this curious story. “Estimates Show Wuhan Death Toll Far Higher Than Official Figure.”
Since the start of the week, seven large funeral homes in Wuhan have been handing out the cremated remains of around 500 people to their families every day, suggesting that far more people died than ever made the official statistics.
“It can’t be right … because the incinerators have been working round the clock, so how can so few people have died?” an Wuhan resident surnamed Zhang told RFA on Friday….
The source said Wuhan saw 28,000 cremations in the space of a single month, suggesting that the online estimates over a two-and-a-half month period weren’t excessive.
Gist is these urn models (inside probabilist joke) result “in an estimated 46,800 deaths.”
To counter this, I’m going to use Uncle Mike’s numbers from Update II, which as far as I can tell are accurate.
Hubei Province includes Wuhan and “a huge conurbation of three cities: Hankou, Wuchang and Hanyang.” Has some 58.5 million souls. Quoting Uncle Mike, the non-abortion “death rate in China was 7.29 deaths per thousand population in 2019”. That gives about 1,000 to 1,200 deaths a day from all non-abortion causes.
Now large Chinese crematoriums have a capacity of about 20 (these are big buildings), and it takes two hours to burn up a body—there are some bones left; the burning is not as complete as in the West. That makes about 240 crispy corpses a day per crematorium.
Thus, at least 5 crematoriums, and thousands upon thousands of urns a week carried in spooky urn-carrying trucks, must be going 24/7 with no bathroom breaks or inefficiencies just to keep up with normal demand.
Here the math ends, because I have no idea how many crematoriums the Wuhan area has. Has to be more than 5. I know the Chinese keep crematoriums on the outskirts of cities, because it’s bad luck to go by one.
There is so much speculation, and so few hard numbers, that to conclude definitely, or even with high probability, that China hid some 40,000 extra bodies over the course of a month or two because of urn tracking, when nobody who is not an expert in the Chinese mortuary business knows what he doesn’t know about urns, is not warranted.
Speaking of instant expertise. Couple of weeks back, nobody had ever heard of a ventilator or respirator, or anything like that, except what they might have seen on a prime time soap opera. Suddenly, overnight, everybody knew the precise number of ventilators that should be placed in every ICU in the country, they were certain about all the ins and outs of ER patient flow and bed capacity, they knew everything there was to know about the manufacture of ventilators—from the supply chain, time to build, loans to fund the process, to delivery mechanisms—-and they knew there were damn unhappy that there weren’t enough ventilators near them. We learn so fast!
The reason I never speak about ventilators is that I know squat diddly about them. Only time I’ve seen them is in passing through the ER, or in the NICU, where I was once involved in an ultrasound study to measure some skull and spine thingees in premature infants.
This instant expertise is the Curse of Democracy. Not only is everybody an expert on everything, everybody has to be, because you might be called on to vote about it.
As I have been saying from the beginning, our naive model has been under-predicting when it predicts before the peak has been reached. It under-predicts the timing of the peak, and so it necessarily under-predicts the totals. Once the peak shows, as it did in China, the model performed well, and we expect it will perform well when the secondary peak shows. Last week we guessed, using the model, that the secondary peak would hit in about a week. Didn’t quite happen, it now seems.
The great benefit of this model is that it is simple. It’s only assumption is that the reported numbers will resemble numbers typically reported in outbreaks. Reported cases and deaths cannot remain “exponential”. Time it takes to “double” cases and deaths must rise to infinity—meaning new cases and deaths must necessarily drop to zero. Instead, outbreaks resemble logistic curves: and that’s all we’re doing.
Here is the latest model run, using data current as of Monday 8PM EST (same source).
You can see the data just starting to decelerate, on which more below. Can it instead re-accelerate? Yes. Yes of course. This model cannot see peaks beyond the present one. It is a dirt-simple model whose only assumption is that reported numbers resemble a logistic curve, as in all prior outbreaks.
Last week, the model estimated about 40,000 total deaths. It’s now saying 80,000 (twice as many!), with reports lasting for an additional week to ten days, ending about 30 April. As I have been warning, the model has been under-predicting from the beginning until the peak has been reached, as in China’s reports, after which is does rather well. If we are not at the peak, then it’s likely the model will continue to under-predict, because it didn’t nail the upward slope. If we are at the peak, then the under-prediction, if any, is likely to be small.
I’ve been getting requests to show non-logged numbers. Here they are:
Easier to see reported cases top out at 1.5 million. I remind us: this is reported cases, and not necessarily actual cases. That it is likely not actual cases is suggested in this next plot, which is the reported deaths divided by reported cases, a naive measure of mortality rate.
Look at that sucker soar! Almost 5%!
The initial up and down we discussed many times before. The secondary increase, too. But how about that plateau and that rapid increase thereafter? There are several things that might be happening.
(1) Cases, especially mild cases, aren’t being reported, because not tested for. China, the claim is, stopped reporting cases after deaths ceased. The denominator is too small. (2) Deaths due to coronavirus are being over-reported. That is, people with coronavirus are reported as having died from coronavirus too often. This is happening, but the extent we don’t know. The numerator is too big.
And (3), everything is accurate. But since we don’t have that many bodies stacked up, that implies two things: (a) it’s much harder to get the virus than reports have it, (b) when you get it, you’re much more likely to die, which doesn’t match other reports. Making it much more likely (1) or (2) or both are true.
Now the daily cases:
A week ago the model said we’d be at or near the peak in about a week, though the model under-predicted the slope of increase. The data now appears—I stress the word—the peak is upon is. Yes, this could be a hitch in reporting, which happens, because reporting is a human process. The data could very well continue scaling upwards. Or it could be the real peak. We’ll know in a week!
About the secondary peak. If there’s a third, the model can’t see it, as I have long warned us.
For whatever reason, even though I gather data every day the same time, it has a curious strict up and downness to it. One day higher, the next lower, which we’ll see in the acceleration charts.
Recall that this gives an indication of peak prediction. When the dashed line crosses 0, we have hit the peak. The data is bouncy bouncy, increasing in variability. That’s because the numbers grow larger, and because new places start reporting increases, and old places are on the way down. These are global numbers.
Now the daily deaths:
It should be the case that the reported deaths peak lags the reported cases peak: you can’t die from it until you get it. There is a lag here, but it’s smaller than would expected, unless the virus is a one-day killer, of which there is no evidence. More likely, again, cases are under-reported.
There are, as with cases, suggestions the peak is close, or near. The model thinks so. But the model has been under-predicting. Let’s look for more clues in the acceleration.
Same interpretation as with cases. The variability increase is probably for the same reason. But that strict up and down, up and down, day after day is screwy, and makes me nervous. Are authorities getting together to release stats day by day? Or are some countries reporting more on a 48-hour basis, and not 24? Purposeful manipulation wouldn’t happen, would it? Nah.
I’ve tried to warn us, repeatedly, that medical data is a mess. I’ve tried to warn that tests are inaccurate (got called naughty names for this one; see the Bayes link above). I have tried to warn us about massive over-certainty in all these things. Listen to these warnings and be less certain!
Next full update next Tuesday.
Addendum Doing a USA model, too. Amazing (well, not to me) how different the numbers are depending on the source. Will report this soon. Friday, probably, since others posts in queue.
John D Stats reminded me that it would be fun to look at all this as a percent of global population. This ties it in with the Perspective On The Coronavirus: Sanity Check, which looked at body count percentages for all known (to Wiki, anyway) epidemics.
Roughly, the model now predicts 0.02% of all 7.7 billion of us will be reported to have a novel coronavirus, and 0.002% of us will be reported to have died of it. If the model is under-predicting, these will rise. Double them—hell, quadruple them.
That makes a 0.008% death rate. That puts coronavirus in line with 2009’s Swine flu, which nabbed almost the same. Still way behind the ’58 Asian flu and the ’69 Hong Kong Chop Suey fluy. Here’s only the last of three pictures from that post (enlarge):
That 0.008% is among the smallest. Only this is the first ever worldwide lock down. Of course, the coronavirus might fall in line with those “best case” models from Fauci et al. Who knows.
To support this site and its wholly independent host using credit card or PayPal (in any amount) click here