As I get older, I get a glimmer into how odd I really am. Not that I think I’m all that different than anybody else in this regard, we’re each weird in our own way. For today’s post, I’m trying to remember that I simply process information in what is evidently an unusual way.
Example I’ve mentioned before: when we studied Euclid in freshman year lo these many decades ago, 9 times out of 10 I would look at the drawing, read the proposition – done. Working under stated premises and the rules of logic, you assert, for example, that the angles opposite equal sides of a triangle are equal? Sure. Prove it? Sure – hand me the chalk. It took me a good while to understand that the other students, generally very intelligent people, didn’t work this way, that the truth of the propositions was something they got to by working through the proofs step by step, and that they had little idea how to proceed with the proofs without first working through them.
Now, I’m an idiot when it comes to language and many other things – can’t seem to get even the basics, and forget them faster than I can learn them. But logic and Euclid? Evidently, I’m some sort of idiot savant.
I say this because of my growing frustration, where arguments over the virus seem to circle around and around the same irrelevancies and claims, while ignoring what, to me, are the glaringly obvious points and the inevitable conclusions to be drawn from them. Then I remember: I’m odd, perhaps these points are not obvious to others? Maybe I need to lay them out in some sort of rational order, and not skip any steps? So, today, for what seems to me to be the umpteenth time, let’s go over the basic issues.
1. Messy data, and what it tells you
Way, way back in January and February, we started getting information out of China about an epidemic. Over time, but before mid-March when the shutdown was imposed here, we got data showing case fatality rates from various areas in China. Most of the action was in Wuhan, but it had spread to other areas as well. The CFR for various regions were quite different, ranging from, if I recall, 4.5% in Wuhan proper to well under 1% in outlying areas.
The very first thing these numbers tell you: the data is very messy. These various outcomes CANNOT be caused by the virus alone. There must be – MUST BE – other factors at play. The next thing they tell you: if these numbers mean anything, they mean anyone’s chances of dying from the virus is heavily dependant on where they live.
I evidently need to harp on this: these CFR numbers, in themselves, don’t actually tell us how deadly COVID 19 is. To get to that point, you would need a whole bunch of additional information – information, it turns out, nobody has. I’ve harped on that in previous posts. But they do tell you that there is no one number that represents how deadly this virus is, that even so simple-minded a number as CFR varies enormously from place to place.
From the beginning to this day, the claim has been made that ‘we were acting on the best data available.’ I am here to say:
NO, ‘WE’ WERE NOT.
The best we knew, the clearest information we can get and could ever get from the early data or any of the subsequent data: the seriousness of the virus depends on where you live. There is no one CFR that expresses the seriousness of this infection.
The ‘information’ we acted upon, the evident cause of our panicked overreaction, was: dead Italians! 7.7% CFR! Ferguson’s model! Millions dead! If, instead, we had said: the data is very messy so it’s impossible to conclude much of anything from it, but, if it does mean anything, it means the CFR depends enormously on where you are when you get infected, we would then have asked different questions and proceeded differently.
But ‘we’ didn’t. We ignored the actual evidence in favor of wild, worse-case assumptions, that we then plugged into models, which, dutifully, produced worse-case numbers.
Garbage in, garbage out.
Model output in not data or evidence. Acting on model output is not acting on the evidence.
(This is why, by the way, I focus almost exclusively on deaths, and belabor how deaths are counted, and dismiss case counts and CFR as misleading – deaths are the ONLY way to meaningfully gauge the seriousness of the outbreak. And not that many people had died when our overreaction was inflicted. Some states were shutting down before a single death had happened within their borders.)
I’m putting a recap/expanded summary of what I mean by messy data and what that means to the numbers being tossed around in a footnote. The messiness of the data generally means one would need to be careful using any of them and caveat the hell out of any claims. In this case, with few exceptions, the messiness of the data tend strongly toward overstating the seriousness of the pandemic. I’ve been harping on this in previous posts, which I why I merely footnoting it here.
2. There is nothing novel about novel viruses
Whenever it is pointed out that, you know, people and viruses have evolved together for millenia, and that we need new flu shots every year because every year we have to deal with new versions of flu viruses, we are told “COVID 19 is a novel virus! It’s not the flu! Nothing that has happened with flu viruses has any bearing on COVID 19 (and: you are an idiot to propose it does – I’ve been told this).
But wait – every new flu bug is a novel virus by definition. And, while, according to the current classification system, coronaviruses are not flu viruses, there are and have been plenty of coronaviruses floating around as long as people have been alive. Some common cold viruses are coronaviruses, for example.
Just because COVID 19 is caused by a novel virus doesn’t mean it is any more scary than next year’s flu bug, which will also be a novel virus, or the next cold you catch, which could very well also be a novel virus, and even a novel coronavirus.
Novel doesn’t equal ‘super-scary’
If we want to look for a recent, comparable virus, how about SARS in 2002? SARS is a closely related coronavirus, much harder to catch but more dangerous if caught. It died out once the weather got nice. It made a brief reappearance the next year, and then died out for good.
The claim that COVID 19 requires extraordinary, job-destroying precautions simply because it’s novel is absurd. Humanity has endured novel viruses for millenia, year after year after year. There is no evidence to support the idea this virus is going to be particularly worse than any of the others.
Buuut – if you mistake model output for evidence, and ignore what the actual evidence is, then ‘novel’ means this virus is way different than the usual. As more evidence rolls in, it becomes clearer and clearer that COVID 19 has been wildly and irresponsibly overhyped. ‘It’s just the flu’ looks more and more accurate as each day passes.
3. (Recklessly) Assumed Homogeneity
Assume a spherical cow of uniform density in a friction free environment. A blog post from Sarah Hoyt reminded me of this old joke, how complex situations must be simplified to be modeled – and thus, why models so frequently give gibberish answers.
The models necessarily assume a particular spherical cow of uniform density: that there is *one* rate of infection and *one* fatality rate, and one value for any of the other variables. (You apply the calculus after you’ve entered the values.) True, you could build multiple models representing multiple populations – the Imperial College model did two famous ones- one for Britain, one for the US – but that should engender an uncomfortable discussion of which I have heard nothing: how small do we slice it?
The logical answer, based on the very earliest evidence out of China, and reinforced with every new piece of information, is: very, very small. Not a country, not a state, not a city. How about a cruise ship, or a nursing home, or a particular Wuhan tenement? Or a California suburb, a grocery store, or a school? Does anybody pretend to believe you are getting remotely the same CFR in an Italian nursing home and a Southern California grammar school? Or from one nursing home or school and the next (unless it’s zero)?
I can hear the ‘buts’ – but people interact! But people don’t stay in their school or suburb! So? What does the model do? Assumes *another* spherical cow of uniform density: that everybody interacts the same way and amount, that some ‘average’ rate of transmission represents what’s really going on, that human interactions can magically be reduced to some (assumed ) number.
That is again, stupid, and *assumed to be false* by the very mitigation efforts currently being imposed on us: our faith in the belief that quarantines and various isolations and restrictions can stop the virus means we accept the notion that how people interact is different and effects how the virus spreads.
We could, therefore, have focused on exactly which interactions between exactly which people are the likely vectors, and striven to control them. We tragicomically did not: in our fine state, churches and garden centers are shut down; homeless encampments and public transit are not. People shopping outside in fine weather or sitting for an hour in a church – too dangerous. People riding around in subway cars and buses, or camped out in their own feces – acceptable risk. The homeless, in particular, then go mingle with social services personnel and the people they panhandle from. Yet homeless encampments were explicitly exempted from the rules. Sound prudent to you? Either California wants the homeless to die and doesn’t care how many other people they take with them – not politically viable – or they think this lockdown is a joke. Or? Similar insane steps seem to have been taken elsewhere. New York, last I heard, had not even yet shut down subways and elevators!
Another spherical cow, a somewhat more subtle one: that grouping people by age or any other characteristic makes all those in the class effectively the same. The class of, say, people 75 to 84 years old is somehow homogenous, thus any person in that class stands, according the CFR for that age cohort, a 10.32% (or whatever) chance of dying if they catch COVID 19.
Nonsense. Within that group are some older, some younger, and, much more important, some healthier and some sicker individuals. Some already have breathing and heart troubles, some don’t, and all this is a matter of degree. Lumping them together by age fatally muddies the answer to the underlying question: how likely is COVID 19 to kill them?
One subset is very likely to die, much more than any other. People are abandoned to die in nursing homes by the millions every year, with the people doing the abandoning feeling more or less bad about it. Some even visit. I guess all that daycay when we were kids gave us a high tolerance for other people’s misery. Be that as it may, it is BLINDLY OBVIOUS that the populations in nursing homes are way more likely to be seriously ill and ARE WAY MORE LIKELY TO DIE than their age cohorts in the outside world.
I asked a relative who worked for years in a nursing home. The staff knows that, likely as not, some little thing – a cold, a stomach flu, an infected scratch – will kill those people sooner or later. Probably sooner, if they are very weakened; maybe later, if they are stronger.
Dying after catching a viral infection is an extremely common way people in nursing homes to eventually check out. The virus would not kill them if they were not already old, sick, and weak.
Reports are that about 20-25% of all the COVID 19 deaths are people in nursing homes. This may – may – be understated, as it’s possible the decedent got sick and was shipped off to a hospital to die. Also – and I don’t know, but it seems very likely – some of those who die at home very well might be under hospice care – another (better) way we treat those who we expect to die soon.
So ‘where you are’ must also contain caveats about where you are healthwise. This information, which was available at the time of the US lockdowns, was effectively ignored. What happened in Italy, a country with a history of ‘excess deaths’ for even just seasonal flu, is that COVID 19 ripped through some nursing homes, then through some hospitals, and then started to peter out as soon as it had claimed, not random 81 year olds, but very sick 81 year olds and other people sick enough to be packed into a nursing home or hospital.
The point of all this: Now, it’s common among the educated to claim that our lockdown – and our sheep-like surrender of our constitutionally guaranteed right to free assembly – was merely prudent, based on what was known at the time, and that caution should be exercised in lifting restrictions. That’s a far as you can go and stay in the Kool Kids Klub. Further, other considerations, ones that were not available at the time the initial decisions were made, don’t retroactively make the initial decisions made without them somehow more reasonable. Excess deaths in (tightly-packed, #WohanStrong, subway & elevator ridden) New York and its suburbs NOW does not contradict or stand against what was know back in March: that packed conditions where people jostle about and breath the same recycled air in closed buildings, elevators and subways is *pretty probably a high-risk area* and that prudent steps should be taken there.
But NYC ain’t Laramie, WY or suburban California. A subway isn’t a suburban park. An elevator isn’t a garden supply store. Yet we model and set policy as if they are, not in accord with, but in defiance of, what was known at the time.
“Where is Every Body”: Fermi’s Ghost in China
Take a look at this picture, and note the date:
China started its by now famously effective and credulity-straining lock down on January 23, but didn’t get around to a general travel ban until January 30. The virus had spread all over China, presumably before January 30, such that cases were found everywhere. (Note that the size of the hexagons reflects cases, but more obviously, reflect local population densities: The huge one is of course Wuhan; the tiny ones are where few people live all spread out; the big ones where many people live in large cities. The 1.3 billion Chinese in China are not evenly distributed, and do not live in one uniform fashion or under one uniform climate or geography, etc.)
By the January 30 lockdown, COVID 19 had had about 2 months to spread out from Wuhan – and spread it did, as you can see by the map above.
The first US case was reported on January 20. The first furtive steps at control took place with travel restrictions in January, but full on, government-led efforts were not taken until March 16.
So: China and the US both had about 2 month of spread before strong steps were taken. China, with a more densely packed population in many places, poorer sanitation and personal hygene practices, and an outbreak in the dead of winter, might – might – be supposed to be a more favorable environment in which the virus could spread. Be that as it may, on March 31:
WASHINGTON (AP) — 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.https://apnews.com/6ed70e9db88b80439a087fdad8238009
When you build models, one thing you routinely do is a reality check: is my output reasonable? Are there any existing cases against which I can try my model to see if it makes sense, given what actually happened? This is simple prudence regardless of the type of model. When I used to use financial pricing models, I knew or could easily find out what was happening in the market – what people were charging for equipment financing, and what kind of yields the finance companies were getting (and cost and tax assumptions, expense allocations, cash flow timings and all sorts of related trivia). So, if I modeled a case where the output was wildly different to what was happening in the real world, I’d look into it hard. No way am I just going to use output that contradict reality. I’d get fired.
So, when his team told Trump a bit over 2 months into this, that between 100K and 240K Americans were going to die, even with all the restrictions in place, I have to ask: where is every body? where are all the Chinese dead?
Because China was 5 months into it by this point, had similarly had 2 months for the virus to spread unchecked, had if anything more favorable conditions for its spread – and yet, as of today, is reporting under 5,000 deaths. Would not logic dictate that, since there are 4 times the number of Chinese as Americans, and, as you can see from the map above, COVID 19 was just about everywhere in China, that something like 400,000 to a million dead bodies should be piling up by now? Under the assumptions used to predict 100K to 240K dead Americans?
So, this forcast fails the sniff test rather badly. Of course, I think the Chinese communists lie, and think they could probably lie their way around an order of magnitude more deaths – 40K, say. But 400K? A million? Possible – these are the people who offed something like 65M in the Great Leap Forward, after all – but that took years, before spies, satellite surveillance and a semi-open country. Seems kind of hard…
Or, perhaps they are not lying, and instead are only counting people where COVID 19 clearly killed them? As Neil Ferguson, the guy behind the model, recently said about deaths being attributed to COVID 19 actually being from other causes:
It might be as much as half or two thirds of the deaths we see, because these are people at the end of their lives or have underlying conditions so these are considerations.https://www.telegraph.co.uk/news/2020/03/25/two-thirds-patients-die-coronavirus-would-have-died-year-anyway/
I don’t know, but either their numbers or ours are bogus, or both are and something in between is happening: the Chinese are hiding some deaths, and we are wildly overstating ours.
More to the point: at the time when this “100K – 240K” claim was made, it could not have been based on ‘what was known at the time’. It was not based on science or even a good-faith effort. All the supposed data available makes it nonsensical. Instead, like all the doom and gloom projections so far, it seems based on wildly pessimistic assumptions that nobody sniff-tested. SOMEBODY needed to ask: what about China? And get a straight answer. Trump’s got the surveillance satellites and spies, after all.
Springtime for COVID!
Finally, one last thing that keeps popping up: there are some viruses that seem to survive the spring and summer months better than others. Not very many that anybody has heard of. In fact, the only ones I’ve heard of are the Spanish Flu, which survived for about 2 years in conditions – WWI, primitive medical care – that could hardly be farther from what we have now, and the 1957 Asian flu, which lasted about a year and a half. Each of these flus somehow survived the disinfectant effect of sunshine, and so came back for an encore once the weather got cold again.
But, we are constantly reminded, COVID 19 isn’t the flu! It’s a *novel* coronavirus. Way different! So, do we have any coronaviruses to compare it to? Why yes, yes, we do: the SARS outbreak of 2002-2003. That SARS virus is closely related to our current bug. Infections broke out in November 2002; the WHO declared it over in July of 2003. Then, for unclear reasons 251 cases were identified in Toronto in 2003-2004. That micro-outbreak ended in June, 2004, with an appalling 43 deaths, or 6% of all the deaths.
So, again, based on the evidence we actually have and not on worse-case assumptions about what *might* happen, the logical thing to assume is that what seems most definitely to be happening – nice weather is killing off the virus – is, in fact, what IS happening, and that, based on what happened with the closely-related SARS virus, little or no recurrence is likely to happen in the fall.
Again: evidence. If you want to claim that we need to cower in fear that this virus will, like the dream of postbellum South, rise again, then point to a similar virus that did so. No fair harping on how COVID 19 isn’t the flu, then turning around and using *the flu* as your example of Undead Viruses.
Conclusion: I’m appalled, as any reader of this blog knows, by misuses of the term science in connection with every hair-brained bit of panic-mongering that crosses the illustrious pages of our esteemed media. In a remotely just world, this pandemic will be remembered as a cautionary tale, and go down as yet another abuse of science from what future historians (one hopes, not in charcoal scratching on a cave wall) will call the Age of the Great Scientific Frauds.
The data is messy. Let me count the ways:
Some of the more important ways, at any rate.
- The definition of a ‘case’ is not clear or consistent from place to place, and changes over time. Cases are not reported in an orderly or consistent manner. Cases may or may not include those diagnosed from symptoms alone without a confirming test. Cases are unlikely to include very many asymptomatic people. Cases are also dependent somewhat on how much testing is being done. What this means: case counts across time and space give us only a vague idea of the virus’s spread.
- The definition of a ‘death’ is not clear or consistent from place to place, and changes over time, for many of the reasons given above. Filling out a death certificate is not simple. Often, the immediate cause of death is not clear. Further, nowhere, with the possible exception of China, is death by COVID 19 counted in the manne any reasonable person would consider fair, namely: did COVID 19 kill this person? Would he have lived if he hadn’t caught it? Estimate of how many deaths classified as caused by COVID 19 that could pass this common sense definition range from half – Ferguson’s high end estimate – to 12% or less by some Italian doctors. The US, Britain, Italy and France explicitly encourage or insist upon COVID 19 being listed as a cause of death if the decedent tested positive or could reasonably be supposed to have COVID 19, even if it is at most a minor cause. E.g., Ventura County reported 2 COVID 19 deaths a couple days ago: a 99 year old man (life expectancy in years = 0) and a 37 year old man, who died of a drug overdose but had tested positive for the virus. No reasonable person would count those as COVID 19 deaths. What this means: Death counts are not a fair representation of the number of people who were killed by COVID 19. It seems likely to be too high by 50% or more.
- Cases are not infections. Nobody knows, and nobody ever will know, how many infections there are or were. Cases will always understate infections, often severely, unless the disease is near 100% fatal or near 100% generates unique, serious symptoms, or we accurately test everybody in the world. Otherwise, mild or asymptomatic infections will constitute a probably large number of infections that do not become cases. Early own, using the then-available information, I estimated that infections outnumbered cases by at least 400%. Since then, wider testing in, for example, New York and Miami, suggest at least 15 times as many infections as cases. What this means: The virus is much more widely spread, and therefore much less dangerous, than would be suggested by the number of cases.
- Case Fatality Rate – CFR – is a) not the real fatality rate even in theory, b) can never be established with any confidence, given the uncertainty in the case and death counts, and c) needs to be measured over a more or less homogeneous population to mean much of anything. What this means: For the reasons above, the CFR may be 30 times or more higher than the true fatality rate: e.g., half as many deaths divided by 15 times the number of cases = 1/30 the CFR. if the number of infections is 15 times higher than the number of cases.
In conclusion: none of this is particularly hard to figure out, it should be obvious to any competent epidemiologist or model builder.