I had some very nice emails after yesterday’s enormously long diary entry. I thought this comment in particular encapsulated the entire “mission statement” of what I’m doing – thank you, Peter.
Thank you for all your hard work and thoughtful analysis. They say that information can help diminish fear, and your diary is useful food for thought.
I certainly don’t expect people to consistently agree with my interpretations (and Peter very politely avoided saying he does, above!), but I do hope the raw data I link to, and the reasoning behind my interpretations, helps you to set your own understandings.
I also realized an interesting thing when looking at the several negative responses. These were from people objecting to my explanation why the statistical projections of what shape our coronavirus pandemic will take are variously wrong and extremely wrong.
Some disagreements were along the lines of “these are very well respected people making these projections”. This is a variation of the classic “Trust me, I’m a doctor” line – “trust them, they are doctors/statisticians”. It is also akin to the opposite of an ad hominem attack. With an ad hominem attack, you say “this is a bad person, therefore what he is saying is wrong”. That’s on any list ever compiled of logical fallacies – bad people aren’t necessarily stupid or wrong.
The contrapositive statement would be “this is a good (respected) person, therefore what he is saying is right” and is no more valid than the classic ad hominem statement. Good and intelligent people make mistakes, too.
There’s also the difficulty of assessing who is good and who is bad – classic case in point being our President, with about half the country placing him far to one side of the good/bad spectrum and the other about half far on the other side, and both using their assessment of his character, and imputed sense or lack thereof, as a filter through which to view/accept/reject just about anything and everything he says.
Another example, if needed – the social prominence of the Kardashians would seem to validate their status as knowledgeable experts if we use “well liked/respected/popular” as a measure of the validity of “trust me” statements. (Yes, I know there’s a difference between “trailer trash” loving the Kardashians and fellow scientists admiring their peers, although it could be said in both cases, it is peers expressing opinions on their peers!) See also Taleb’s tweet on this point, mentioned below.
One person had been writing to me a week ago and referred, then, to the projection that was revised from 500k deaths to 20k deaths as being a much critiqued projection, but in expressing disagreement with my comments yesterday, was now claiming the authors were all extremely respected. He seems to have forgotten that just a week ago he too was joining in the criticism. I can’t understand that selective memory at all.
I’ve also uncovered two further “proofs” of my statement that many/most/all of the projections we are being given are of low or no value whatsoever.
The first is hard to follow and understand and is more theoretical in nature (while citing fascinating past examples to support the logic). It is an essay by Nassim Nicholas Taleb, the author of fascinating books about unusual events that he calls “Black Swan” events. His article talks in complicated detail about where statistics and modeling is valid and where it is not valid. The quick summary is that he has pandemics in the category of events that do not lend themselves well to such techniques.
Taleb was one of the very first to call out the danger of the Covid-19 virus, in a paper he published on 26 January, pointing out that initial estimates were likely to be unrealistically optimistic, initial responses would quickly be overwhelmed, and calling for urgent restrictions on social interaction and mobility because we couldn’t afford to get it wrong. It is already a tragedy that his paper was ignored for over a month before being drastically shown to be prophetic. I’ve saved a copy of the paper, and here’s a link to it.
Oh, he also cruelly describes the computer model used by the 500k -> 20k death guy (who bizarrely continues to refuse to allow anyone to inspect his model while demanding we all accept it) in this tweet as
this is not a model usable for real world risk & decisions–rather something with the FRAGILITY of a house built with matches to impress some tenure committee.
More of his critique is in this formal paper.
Taleb also points out the disconnect between a person who is tremendously respected in the academic community and one who is actually competent and proficient in this tweet.
The second “proof” is easy to understand and directly linked to the Covid-19 problem. It is an article from the fivethirtyeight.com website – a site best known for aggregating and averaging political opinion polls. It has done the same for various projections about what the future of the virus infection will be in the US.
On Monday and Tuesday of this week, the site asked some of the most prominent experts to predict how many total cases of the virus would be reported on Sunday (29 March) – in other words, to calculate ahead a mere five or six days. Interestingly, as you can see in the chart below, the data on US cases has been surprisingly consistent and steady in its upward path both prior to and subsequent to that time, and there was a useful time series of data to base the future projection on, so a prediction should be simple.
The best point of this is that these experts were being asked to make a prediction that would very quickly be able to be measured against reality. Unlike most of their predictions which are essentially unaccountable for if they will ever be right or wrong, here was a really simple thing for them to predict and then be judged by, and at the same time being an essential element of all their other, harder to test, predictions.
Look at the extraordinary divergence of predictions the experts handed in, and, in some cases, the stunning range between their high and low estimates. In most of the public predictions, we never see high and low ranges, just “exact” numbers.
Two of the experts each gave a range where their low estimate and high estimate were separated by a factor of ten. So on Monday or Tuesday, with US case counts being then 44,000 or 55,000 respectively, these “experts” were saying “By Sunday, we predict a total case count will now be between 50,000 and 500,000”.
Already, on Saturday, the total case count is probably going to end the day at about 125,000 and the “right answer” for Sunday is likely to be 145,000 – 150,000, compared to the averaged prediction of 117,000.
The wide range of different predictions, the huge “plus or minus” factors that most of the estimates include, and the likely variation between predicted numbers and reality tomorrow show the unreliability of these predictions, even when the people being asked to make a very short term prediction and were given a lot of previous data.
If you keep reading the article, it gets better/worse. One week prior to that, on 16/17 March, the same experts were also asked to predict the total for this Sunday. That was a harder prediction to make – a shorter prior data series, and being asked to predict nearly two weeks into the future. At the time, the total case count was 4,400 and 6,100. The average prediction for this Sunday, back then, was 19,000. That number was exceeded four days later, when on 21 March cases totalled 23,720.
So, a stunning fail back then, and a significant miss for the shorter prediction this week.
What part of that gives us any confidence in anything else these experts are predicting, especially the more complicated issues, and more than five days into the future? Oh yes, these experts are very respected. But also, clearly, very wrong.
Two new cases sees the Vatican City returning to top place, and Luxembourg moves up above Iceland. The seven states of any size with the highest infection rates are :
- Vatican City/6 cases/7,491 cases per million
- San Marino/224 cases/the equivalent of 6,602 cases per million people
- Faero Islands/155/3,172
Here are the top six major countries, showing death rates per million of population in the country :
- Italy/10,023 deaths/166 deaths per million
To put those numbers into context, the death rates per million in US/UK/Canada are 7/15/2.
For major countries and/or outbreaks, and in general :
|Total Deaths/Percent of all Resolved Cases
|Active Cases (ie not yet died or cured)
|US Cases/Deaths/Case rate per million
|UK Cases/Deaths/Case rate per million
|Canada Cases/Deaths/Case rate per million
|Worst affected major country/case rate
|Second worst country affected
As we expected, Italy dropped another place and now is the third worst affected country. Spain not only overtook Italy, but is also gaining on Switzerland. Belgium shot up the rankings today and now appears as number 5; displacing Norway to #6. Noting also the rapid rise of cases in Germany, we’d not be surprised to see Germany reach #6 in the next day or two.
We think we have an explanation for the puzzling anomalous steady rise in the percentage of deaths, now at almost 18%. We think this is the result of some countries not detecting or recording cases until a person has suddenly died, that being the very first that they have appeared on that country’s statistics. And as health systems get more overwhelmed, it seems sadly possible that this type of situation is becoming more prevalent.
Today we look at the latest version of my cumulative daily death chart, with superimposed data for eight different countries (I added the Netherlands since last showing this chart because it has the next highest number of deaths, making it valid and relevant to follow).
There are two fascinating things to see. The first is that after matching appropriate start dates, six of the countries follow astonishingly similar curves. The second observation is that the other two countries (Iran and China) are totally different, but very similar between themselves.
What is it that China and Iran are doing differently from the other countries? Call me a cynic, but the most obvious difference is that in the case of these two countries, the data is almost certainly extremely under-counted – in China’s case as an official policy, and in Iran’s case, as likely as not due to internal chaos as official disinformation.
This is an interesting article that seems to offer solid evidence that points to China’s enormous under-counting of fatalities. Contrast the suggestions of hundreds/thousands of continuing deaths with the official claims – for the last ten days China has never reported even as many as ten deaths per day (the official count for today is 3).
This leads to a suggested rule of thumb when looking at any sort of predictions. If they use any of the Chinese data to base any of their assumptions and projections, then they are almost certainly wrong.
Who Should Pay?
Did anyone else note that the $2 trillion bailout is now being described as a $2.2 trillion bailout?
Timings And Numbers
Mayor De Blasio is now warning that “more than half” of New York city will become infected with the virus at any one time. That implies that in total, many more than half will have the virus, which puts his prediction fairly far into the pessimistic range of projections. He also anticipates the city may remain essentially closed through the end of May.
Maybe he should tighten up some more on the current lockdown in his city so as to avoid those outcomes.
We tried to make sense out of the wide difference between Norway and Sweden yesterday (a difference that has persisted another day today). We may or may not have partially succeeded in that. Here’s an interesting article that looks at the difference in experiences between Los Angeles and New York – stark differences in case numbers.
We do accept that there is more “social distancing” in Los Angeles as a normal state of affairs (people in their personal cars rather than on the subway lines and buses, and single family residences rather than apartment buildings for example). But is that the only variable? Neither the article nor I am sure on that.
There are lots of things we don’t yet understand about the virus, and I noticed – in a different context – an interesting speculation by one analyst that perhaps there are genetic elements to a person’s susceptibility. If that is so, and if Italians are more at risk, that would help explain New York’s greater numbers.
Here’s a dangerously nonsensical statement, from Bill Gates of all people (need I add, a highly successful and widely respected guy with a top notch intellect). In warning of a ten week lockdown being needed, he says – “If we do it right, we’ll only have to do it once for 6 to 10 weeks”.
Well, yes, we might have possibly eradicated the virus within the US (and his “if we do it right” sort of assumes we did do it right, which I’d rate as colossally improbable). But what happens the minute we start allowing people to cross in and out of the US again. Are we going to impose rigid two week quarantines on every person entering the US? Would that be possible? Let’s look at the numbers.
There are about 77 million visitors to the US each year, and we’ll guess this is more or less matched by, let’s say, the same number of Americans traveling out of the country and returning. So 154 million people a year, each requiring two week quarantines. Not self-quarantines, because those are imperfect. A “bullet-proof “100% secure path from the airport jetway to the quarantine facility. That would see, on any average day, 6 million people in quarantine facilities. We don’t have anything like that, anywhere in the country, at present. Indeed, the entire count of hotel rooms in the US (which couldn’t be used for quarantining anyway) comes to a bit less than 6 million.
And how many people would be required to service and support all these quarantine rooms? To provide meals and other requirements?
In any event, it all gets back to the inconvenient truth that many wish to ignore. We can possibly institute some types of controls on the legal entrants to our country. But we also have an unknown number of illegals entering the country every day. It only takes one of them to bring the infection with them, and all of a sudden, we’re at the start of another entire crisis cycle, and soon enough, another multi-month lockdown.
What is the point in enduring a 6 – 10 week lockdown if it is not guaranteed to succeed, right from the get-go, and will fail as soon as the first infected person next enters the US undetected?
Well, there is a possible justification – the flattening the curve concept, meaning we don’t overload our healthcare system and have time to bulk it up. Plus “buying us time” to develop invaluable palliatives – treatments for infected patients, and the ultimate cure – a vaccine.
So we’re not arguing against lockdowns, but we are saying, be honest about it. Don’t promise a totally impossible outcome because all that does is make the population much less cooperative next time.
We mentioned the administration’s desire to start monitoring county by county rather than state by state in the US. We also understand it is monitoring by zip code (there are almost 42,000 zip codes in the US, 13 times more than the number of counties). Here’s a great county by county map, much of the way down this page. It is updated twice a day.
The source data can be seen on this map, but the visualization is not so clear.
We’d also make the point that while it is helpful to know total case counts by county, it is even more helpful to know the case counts expressed in proportion to the county populations. A county with 100 cases is clearly in an extreme situation if the county only has 1,000 people within it, but if it has 1,000,000 people, it is “getting off lightly” with fewer than the average number of cases. Only this type of visualization helps us see where the virus is most intense.
Here’s an interesting survey that shows 40% of companies expect business travel to resume within three months. Another 17% think it will be more like six months, and 40% don’t know.
We’re in the “don’t know” category.
States and counties and townships are getting more aggressive at keeping “outsiders” away. In Rhode Island, the police are going door to door looking for New Yorkers, and pulling NY plated cars over on the freeways.
We’re worried about our food supply. Not terrified, but anxious and keeping a watch on predictors of problems. Here’s an article that doesn’t reassure us – clearly nothing like acceptable social distancing is happening in cases like that.
Reminding us that the food supply chain has many links is this article, pointing out another point of concern.
Logic? What Logic?
Remember what we said, not far above, that any predictions based on China’s numbers should be treated with enormous doubt and suspicion? Here’s an article, basing its findings on some terribly insubstantial assumptions, including an analysis of the published Wuhan data, that concludes with the finding that, if this model is correct, then generally accepted predictions for total fatalities may be way too high.
The article authors – respected professors of medicine at Stanford (and remember our opinion about validating opinions by rating how respected the opinion-holders are…..) are projecting perhaps total deaths in the US in the range of 20,000 – 40,000, rather than, well, who knows – maybe many millions according to some of the other projections.
We hope they’re right, but withhold an opinion on the matter for now. But with US deaths already on the high side of 2,000, and with death counts being somewhat predictable into the next week or two based on current case counts, it won’t take long to see if we’re likely to pass 20,000 or not.
We also find ourselves shaking our heads. If the experts can’t agree among themselves, isn’t this further validation of our suggestion that they’re probably all wrong!
Virus? What Virus?
Here is an article that laments the lost opportunities in the past to do more to anticipate and prepare for the Covid-19 pandemic.
A second article observes that – by their own metrics – researchers concede that 87.5% of medical research is wasted or inefficient. To be fair, that’s a hallmark of much research – just like how a virus, when it mutates, is more likely to come up with a “bad-for-the-virus” mutation than a good one, much research necessarily leads to dead ends.
But the article is a good read, and the innate inaccountability of some research shouldn’t then give a blanket indemnity for the stupid research, the trivial research, and the projects never completed. We also loved its list of ridiciulous topics that money has been spent on – for example, $3 million spent to study why gay women were more overweight than gay men, or a study that ended up with the hardly stunning revelation – people who walk more each day are likely to be healthier than people with sedentary lifestyles.
Very important : An interesting article appeared today pointing out what has always been implied but never really studied or debated. The 6′ separation recommendation is a huge compromise, not an ideal distance. In reality, infected droplets from a cough can travel 26 ft, at a speed of as much as 60 mph.
That’s not all. As we’ve also regularly pointed out, the droplets can be aerosolized, and so can “float” in the air for several hours.
The bottom line is that your visit to a half empty supermarket is actually way more risky than you might think. You not only should keep greater distance from your fellow shoppers and the store servers, but you also have to consider everyone else who has been in the same space as you during the last couple of hours or more.
Good news – a new test is about to be released into the market which promises to detect an infection in 5 minutes. But the manufacturer’s claim (Abbott) to be planning to make 50,000 tests a day when production is at full speed needs to be measured against the US population of 330 million people.
Fortunately, this is not the only new test kit being developed and made, and indeed, Abbot says that in total, it will ramp up to 165,000 tests a day – a mix of this new test and an existing test.
We desperately need more tests. Indeed, if we had to choose between tests and ventilators, I’d choose tests, for two reasons.
First, when we start to get some meaningful test data, we can then start to shape our multi-trillion-dollar costing public health policies more exactly and shift from the “from an abundance of caution” approach and instead realistically respond. Part of that improved understanding of course is a knowledge of exactly how many ventilators we’ll need, and where they will be needed.
Second, here’s a generally overlooked truth. The country currently has about 150,000 ventilators in total, and there is a growing clamor for more and more. New York alone claims it needs another 30,000 (we’d love to see the exact details of that calculation – indeed, we think the state should be obliged to carefully explain why it is demanding so many more rather than just asking for them as if it is their right to expect the federal government to give it anything it asks for, without question).
The good news is that more are being made and will soon start appearing in massive quantities – tens of thousands.
But – and here’s the point which the people asking for more ventilators don’t seem to think of. We can quickly get more ventilators. But – how about the specialist medical teams that monitor each ventilated patient? As I’ve been regularly saying, in many respects, the critical constraint in our healthcare system is not machine, but people. One source I found suggests the country has enough specialist staff to man 100,000 ventilators, so we already have 50,000 more ventilators than staff.
We can get and will be getting more ventilators in just a few quick weeks. But how long does it take to train a new team of specialists to manage the equipment and monitor the patient? That’s the real crisis. So, a question to Mayor De Blasio – when you get your extra 30,000 ventilators, what will you do with them? Do you have enough trained staff on hand to deploy them? And if you’re scrambling to train staff, what other healthcare services will now suffer the loss of their staff who have shifted to ventilator care?
Each day brings more good news about hydroxychloroquine and its potential role in helping infected patients get clear of the disease with fewer problems and faster recovery. Yes, none of the data is yet rigorous to a “five nines” 99.999% certainty, but equally yes, while some people inexplicably insist we should withhold treatment using this drug – a drug generally recognized as safe and one of the most commonly used drugs in the entire world – until that ridiculous level of certainty is achieved, the real world is moving quickly to embrace it.
Here’s an article citing positive results in Belgium and Bahrain. France has now officially approved its use as a Covid-19 treatment. And the Indian Council for Medical Research has gone even further, and rather than any sort of reluctantly neutral statement, it is actively recommending taking the drug to help reduce the risk of getting the infection in the first place. Recommending prophylactic use is an enormous step further forward than permitting its use as a treatment.
We acknowledge the inadequate nature of some of the trials to date, and eager await news of the New York trials, which we think might start being released as early as Monday next week. But on the basis of “absolutely doesn’t hurt, and potentially might help”, well, as you might have previous read some days ago, we’d use the drug in an instant if we were affected, and have arranged to acquire a modest supply in case of that eventuality.
No Wall St activity over the weekend.
We understand from articles such as this that President Trump may be considering establishing a quarantine in NY, NJ, and parts of CT.
A correspondent tells us that NC will be closing its borders on Monday at 5pm.
This is a great series of visualizations of how an infection can be gained in one location and then spread to many other locations in just a few days. The visualizations convey an obvious message, but the reality is actually much much worse. You are just shown how people in one area spread around – it doesn’t add to the visualization the secondary infections from the people initially selected to track. Then the tertiary spread of people infected by people infected by the first group. And so on.
Thanks to reader Paul for giving us a positive note to end with, today. He reports a sign posted on a Richardson, Tx marquee: Single man with t.p. seeks single woman with hand sanitizer for a clean relationship.
Please stay happy and healthy; all going well, I’ll be back again tomorrow.