The Washington Times (not to be confused with the Washington Post) has published two recent articles with that “hoax” story. This is one of them: Washington Times
This article makes two points: first, that the models forecasting a large number of Covid-19 deaths in the US were wildly wrong, and second, that the real number of Covid-19 deaths will not be many more than in a bad flu year. The stories imply that the fear caused by the virus and the hugely costly political response of lockdown were unnecessary reactions to a mass media hoax. Are these stories right?.
The Models are Neither Right Nor Wrong
On the issue of the models, all very early forecasts of deaths from such models are likely to be inaccurate because of limited data and because they have to make assumptions about the future based on the present. But that inaccuracy doesn’t make them “wrong”, any more than the model would be “right” if the modeller had made a lucky guess and the forecast was quite accurate.
You cannot simply judge a model, after the fact, as having made right or wrong “predictions”. Models are not simple conclusions. They are hypothetical ‘if……then…..” statements.
In the coronavirus context this would mean something like: ‘If the death rates in China and Italy, with certain adjustments made, are applied to the US, with certain other adjustments made, and if the US takes no action to reduce the impact of the virus, then the estimated range of deaths in the US will likely be ……’ Of course, if death reports from China and Italy are inaccurate, and the adjustments made are inappropriate, and if the US does take drastic action to control the virus, the “then” part of the statement will also be inappropriate.
Once the US took drastic action to reduce the impacts of the virus, the entire “if…..then….” statement became obsolete, because such government action was intentionally excluded from the “if” assumptions.
I don’t suggest that models cannot be criticized. Some climate change models, for example, have been criticized for running too hot, showing more warming in model output numbers than in actual temperature observations. Similarly, the COVID-19 widely publicized and influential models from Imperial College (London, UK) can be criticized. But criticism with the advantage of hindsight of changed circumstances is unfair. Fair criticism would be showing that the assumptions, and the adjustments to raw data in creating the model, were unrealistic at the time they were made. But that is not what the Washington Times has done.
COVID-19 is Not Like the Seasonal Flu.
The most glaring error in the Washington Times story is its invalid conclusion that the deaths from the coronavirus and the deaths from the seasonal flu prove that the coronavirus is no more serious than a bad seasonal flu year.
The US is not put into a total lockdown with massive testing every year when the seasonal flu comes around. If that had been done annually, not only would there have been far fewer deaths from the seasonal flu but the US economy have been destroyed long ago. In short, it is wrong to compare two incomparable situations.
We will never know what would have happened if the US had done nothing about the coronavirus. There is no way to know, today, whether the original model output would have been accurate without the strong action taken. A model rendered obsolete by subsequent events cannot now be judged as having been right or wrong.
If the Washington Times had wanted to provide a reasonable criticism of the original Imperial College model it would have had to look into the model itself: to judge its selection of the raw data, the adjustments made to that data, the further assumptions, and then the computer code used to run the model. But the Washington Times didn’t do that. It simply – and wrongly – compared the COVID virus deaths after massive government intervention with the flu deaths after no government intervention.
If the stories claiming that the coronavirus is the biggest political hoax in history are true, I have yet to see any evidence of that.
>>>>>>>>>>>>>>>>>>>>>>>>
If you would like to be notified of my future blog posts please enter your email address above the FOLLOW button on the left side of the page and then click on the button.
If you like my blog please click LIKE and share it via Twitter, Facebook or email.
Categories: coronavirus, COVID-19, Deaths, Deaths per Capita, Emergency, Government, Uncategorized
Andrew's Views 
Scott Adams of Dilbert fame was a financial modeller in his earlier life and has an interesting viewpoint on models (Podcast #903): essentially, that they’re not for generating information but for persuading people. I understand and concur with your point that we can’t judge models when our actions change the assumptions on which they were based, but Canadian officials gave a range of 4,400 deaths (if we moved heaven and earth) to 44,000 deaths (if we did very little) and an “expected-case” of 11,000 to 22,000 deaths that did assume physical distancing and an extensive lockdown. Now they’re projecting less than half the lower end of that range. So I think it’s reasonable to ask, “What the heck?”
LikeLike
The fact that the range of model output deaths was variable by multiple of 10 tells you something. There was very little data and a lot of guesswork in that original estimate. Similarly, the “expected case” is somewhat of a misnomer because it is only expected as another model output, based on limited data, particularly because there is no way to tell how good a job the public will do at physical distancing. This is also guesswork.
There are two important numbers used for these models, neither of which can be calculated with accuracy.
The first is the infection fatality rate: the percentage of infected people who will die. With very limited testing, mostly of people who are already very sick, the number of infected persons will be greatly underestimated – perhaps as much as by a factor of 50 according to a recent study by Stanford University epidemiologist Doctor John Ioannidis. A short video is here: https://www.youtube.com/watch?v=T-saAuXaPok
His study concludes that the infection fatality rate has been greatly overestimated because the denominator, the number of people infected, has been greatly underestimated. For the US epicenter of infection, New York City, he estimates that the risk of dying from COVID-19 is about the same as the risk of dying in a car accident when driving to or from work.
The other important number is the infection rate R. If R at the beginning was estimated to be 2.0 this means every infected person will infect 2.0 others. Again, this number is largely guesswork. However, when this number goes down to less than 1.0 the virus will gradually disappear. Obviously, a change of even a decimal point will make a huge difference.
When relatively minor changes in key numbers make major changes in the final output the result will be much volatility and widely varying estimates. That is just the nature of the beast and we can’t do anything about it.
LikeLike
And while I’m here . . . the USA death projections (from the Institute for Health Metrics and Evaluation which, regrettably, at that time did not run a model for Canada) were first quoted by Dr. Birx (31 Mar) as being between 100,000 and 240,000. Then they were revised to 93,000 (2 days later), 81,000 (3 days after that), and 60,000 (4 days after that). It’s hard for me to keep up or to understand what it means, although fewer is better, for sure. What’s pertinent here, I think, is that the revisions do not seem to be the effect of the lockdown, because these estimates have all assumed physical distancing. That’s even what it says at the top of the IHME site. The experts have been clear that these numbers could/would be much worse if we let up.
LikeLike
Modelling is the same process regardless of the country. It starts with a mixture of data and estimates or guesses and applies adjustments and then equations to produce output. It is essentially the same as a verbal essay but written with numbers. Therefore, it is the product of the person doing the modelling. A model output is not a fact, it is an opinion.
The reason for revisions merely days later is because both new numbers were coming in and new modellers were producing other models with different outputs. There is a tendency for multiple models to be interpreted as being the most reasonable around the midpoint, although sometimes with wide variations around that point.
LikeLike
Correction? Final paragraph, you likely intended to use Washington Times but wrote Post. Washington Post might not agree. Thanks for the article.
LikeLike
Thank you. Correction made.
LikeLike
<> Well said. I’m still astounded by how these opinions have driven public policy. And how they’ve been communicated.
LikeLike
Thanks for the link to Dr. Ioannidis’ video. A reasonable and thoughtful approach.
LikeLike