COVID-19

[7Wannabe5]

In my realm there is a definite lag in reporting from hospital to county to state. When the numbers were going up fast I could roughly predict the upcoming state numbers based on prior day(s) hospital reporting numbers. I also noticed that the IHME model makes some sort of correction for this.

[tonyedgecombe]

We are seeing the same thing in the UK, it has been explained away as delays in reporting the deaths. It sounds like one of those thing that should be easy but turns out to be much more complex on closer examination.

[vexed87]

Admin work in hospitals are usually completed by 9-5 Monday to Friday office (non-clinical) staff. I'm not familiar with the reporting system for C-19, but databases I work with used for reporting to the govt. here in UK don't refresh until midnight. So we're probably looking at early Tuesday before the figures catch up after 5pm on Friday. Probably similar things going on here.

[steveo73]

That was when the policy was to let the virus run through the community until we attained herd immunity. That was deemed politically unacceptable hence we started to lock things down.

It was an alarmist prediction that was so far off base it wasn't funny. Alarmist predictions are poor science and show a lack of understanding of science.

I think you've also touched on a great point here - maybe those alarmist predictions caused more trouble because people weren't ready to call them out as being poor science. I'm not sold on this idea and I haven't really mentioned it because we were behind the eight ball from the beginning but we need to be real careful about trusting poor scientists and poor science.

[jacob]

I think this counts as the first natural disaster [in the US] during the pandemic. An emergency on top of an emergency.

Two aging dams broke in Michigan after this weekend's torrential rains in the Midwest. (The flood control we put in 5 years ago saved us, but practically all our neighbors that we know about had backflow water in their basement. One had two feet!) 10,000 have been asked to evacuate from 3,500 homes. Also hurricane season is about to kick off. Wanna go to a covid shelter?

[tonyedgecombe]

It was an alarmist prediction that was so far off base it wasn't funny. Alarmist predictions are poor science and show a lack of understanding of science.

You don't know that because we didn't let it run its course.

[7Wannabe5]

@jacob: The flooding situation is kind of surreal. I saw footage this morning of 2000 acre lake that was emptied; houses and docks surrounding a giant mud crater. Luckily, it’s a fairly affluent area, so not too many refugees stuck in shelters. Reinforces my permaculture informed conviction that you might think you live in a nation, state, city etc., but really you live in a watershed.

Also, cyclonic activity forced evacuation of 3 million yesterday, so relatively small potatoes.

[jacob]

There's now some concern in Sweden (foreign ministry) that Swedish passports might turn radioactive upon other European countries opening up their borders to each other after breaking the curve. Diplomatic work is being undertaken to convince the other nations that Sweden was and is not an open free-for-all virus fest, lest borders remain closed. (Sweden still hasn't broken the curve though.)

Something similar might happen to US residents depending on how things go.

[horsewoman]

The question of how much damage the virus and lockdowns are causing to the economy is not only hotly discussed here on the forums! The German Helmholz Institute published a study in which they tried to find out how the contact restrictions are affecting the German economy. There is no English translation for the study so far, but the "Translate to English" Feature of Google Chrome gave out a very good result - at least the page with the summary.

Since copying large amounts of text is not allowed here on the forum I took only the last sentence:

"In order to take the most economically advantageous route, which can be reconciled with a further containment of the epidemic, a slight easing should be economically preferred compared to the measures on April 20, 2020. Significant easing is not recommended from both a health and an economic point of view."

Here is the link to the summary for those interested:
https://idw-online.de/en/news747332

and the link to the full paper - in German but with graphs.
https://www.ifo.de/node/55371

[steveo73]

You don't know that because we didn't let it run its course.

Correct but it appears highly unlikely. The science was definitely incorrect. It was incorrect whenever a simplistic exponential model was utilized. The reality is that this virus progresses exponentially in certain situations but not across the total population. That point was missed by some laypeople and scientists.

Like I said earlier and you can see it over the course of how this pandemic has developed. People with a good grasp of science (lay people and scientists) didn't come up with crazy predictions and even when they inflated the figures articulated that they were probably wrong and were erring on the side of caution. Compare that approach to people who don't understand science and state comments like it's the science or it's the math. It is a significant difference in philosophical outlook which represents an understanding of science compare to a complete lack of understanding of science.

Interestingly I heard Niall Ferguson referred to as disgraced recently. I think that is a good thing. When you get something so wrong it's good to take a step back and reflect on your lack of understanding on a certain topic. Hopefully people learn from this experience and we have much more informed debates about other scientific discussions that may impact society.

[tonyedgecombe]

Here is the link to the summary for those interested:
https://idw-online.de/en/news747332

and the link to the full paper - in German but with graphs.
https://www.ifo.de/node/55371

Klement has some a piece on it at https://klementoninvesting.substack.com ... f-lockdown

[tonyedgecombe]

The science was definitely incorrect.

No, you don't know that.

[steveo73]

No, you don't know that.

Of course it was. It was also extremely incorrect. Just break it down into simplistic terms. Were 500k deaths predicted in the UK and 2 million deaths predicted in the USA ? Yes. Did that occur. No.

Ergo:- those predictions were completely incorrect and a great example of poor science.

I'm really struggling with why you fail to recognize this. It'd be good to understand where you are coming from and in particular why you can't admit those predictions were definitely alarmist.

There have been plenty of good scientists and lay people who understood the science a lot more and didn't make crazy predictions.

Maybe a good question to ask you is why did some scientists make much better predictions of the virus progression and why do some scientists not suggest strict lock downs as the right way forward ? Is the science different in these situations ?

Why were some laypeople and scientists so wrong and yet some have been able to remain consistent in their approach. Just look at this thread. Why did some people get it really wrong whereas some people didn't ? I believe it's because of a lack of understanding of science and in particular a lack of understanding of the difference between a mathematical model and the science. The mathematical model is just a way to represent our understanding of the current process (science) and these models are going to deviate from reality. They aren't set in stone predictions. They can be wildly off base. I think anyone who understands science should realize this because predictions of the future are typically way off the mark.

So poor scientists believe in their models implicitly and act as it their models represent reality. Good scientists see the flaws within their models and are open to criticism. You get certainty from poor scientists. You hear complexity and risks when you talk to good scientists.

[IlliniDave]

Daily cases in Alabama still on the increase, but per my calibrated eyeball looks more linear than exponential. Still, not good, especially since tomorrow restaurants can start serving dine-in customers provided they are at the lower of half seating capacity or whatever it takes to maintain the spacing guidelines. I won't be dining out any time soon though.

Some smart left leaning people riffing on science pertinent to this topic.
https://www.youtube.com/watch?v=p6-xwaHU4f4

Be forewarned, they meander through several topics and are critical of the right, POTUS, etc. No easy, simple memes or sound bites.

[jacob]

The difference is that there are two interpretations of what "prediction" means here and consequentially people are talking across each other.

The first one is "there will be 500k deaths in the UK if the country does not lock down immediately". This is a contingent prediction telling what will happen if such and such actions are not taken. It is in the same category as a smoke detector going off to tell you that if you don't put the fire out now, the house risks burning down.

The second one is "there will be 500k deaths in the UK". This is a forecast specifying what will happen in the future. Such a forecast would tell what will happen including what actions would be taken. A weather forecast is a good example. The weather happens regardless of what we try to do about it because in reality it is out of our hands. Another example of a [looser] forecast is that "the stock market goes up in the long run".

Insofar one does not understand the difference, confusion obtains.

The epidemiological predictions being made were, as far as I know, all contingent predictions. This is normal and expected from scientists since they are not in the business of second-guessing policy reactions. Their job is to tell decision makers what the outcomes are of specific actions so that's what they did and still do. As decisions are made, now contingency models are being made reflecting what is known.

Basically, in contingency modelling, there's a clean separation between the calculators and the deciders. The decision makers rely on the scientists providing accurate information and not trying to second-guess them. There was some noise about Ferguson maybe exaggerating the numbers(?) and if so that would have been very bad ... but the reason for his downfall was that he broke isolation to go hang out with his GF despite telling everybody else to lock down. That didn't go over well. Meanwhile, other contingency predictions did pretty well. I plotted the graph of a simple one above. This is not that hard.

An alarmist contingency model would be one that turned out in retrospect to predict far worse outcomes than it should have. (We can see if a model was alarmist by rerunning the model with more/better data.) However, a contingency model is not alarmist just because problems didn't happen after actions were taken to prevent it ... just like a smoke detector is not alarmist just because the house didn't burn down after warning the residents to put the fire out. An alarmist smoke detector would be one that goes off everytime someone farts.

Actual forecasting is much much harder because it also requires predicting human response behavior. Typically, what people do are scenarios which turn the forecast into the familiar contingency predictions instead. For example, behavior A leads to outcome X; behavior B leads to outcome Y; C leads to Z, and so on. Such a hybrid doesn't try to predict the behavior but it does give predictions. People can then do their own weighings according to how likely the think the behaviors are. Many investment forecasts are like this.

It can be tricky to evaluate the expectation if there's a with an extremely low probability but high cost outcome. (Think asteroid strike.) Here the conventional (linear) way of thinking about costs no longer hold.

It's also complicated because one has to specify the "behaviors". If one gets the behaviors (which are input parameters) wrong (incomplete), then the contingency predictions are kinda useless. Many sciences are developed to a point where there is more uncertainty in chosing the behavior than the outcome of that behavior. For example, it's harder to predict whether a patient will take his medicine than what the outcome of taking the medicine respectively not taking it will be. Same here ... it was harder to predict what Trump would do than it was to predict what the virus would do on population aggregated---it's novel, but it's not that novel.

An alarmist forecast can be checked after the fact. Was the outcome far better than predicted? IHME is an example of an actual forecast (because human decision making is part of its innards). Despite all the beatings the IHME is receiving, it's actually been within the 100-240k US uncertainty band since late March now. So has the IFR from the CDC which has been on the money since January.

[steveo73]

Insofar one does not understand the difference, confusion obtains.

Jacob - it's not that simple. Those predictions were predictions that were stated with confidence and I remember people stating it's just the maths. That shows a complete lack of understanding of the science.

If you knew that at the time for instance when you posted your simple model you should have said this is a simple model that at best represents a tiny piece of the issue that in reality isn't worth even talking about when it comes to discussing policy decisions and how this virus develops. It's a really poor model of how this pandemic will progress and it shouldn't be used in any way to form policy decisions because the science is bunk science.

An alarmist contingency model would be one that turned out in retrospect to predict far worse outcomes than it should have.

So 500k and 2 million numbers weren't exaggerated ?

Actual forecasting is much much harder because it also requires predicting human response behavior.

I know that but I also act with integrity in that I don't make stupid nonsensical predictions are try to state it's the science or the maths.

It can be tricky to evaluate the expectation if there's a with an extremely low probability but high cost outcome. (Think asteroid strike.) Here the conventional (linear) way of thinking about costs no longer hold.

We need to use risk management techniques to analyze these scenarios but we need to get rid of the poor science to do that. It's when assessing situation like this pandemic or climate change and we have people (scientists and laypeople) who don't understand the science and how make alarmist predictions (plenty of people on here and scientists). So the people who came up with simplistic models to evaluate this situation and didn't state the problems with their predictions do not provide an accurate representation of the risks.

It's also complicated because one has to specify the "behaviors".
...
An alarmist forecast can be checked after the fact.

I find this pretty hilarious. Yes modelling is fraught with dangers. Going forward let's all remember that and not be so sure of the science or math going forward,.

We should all going forward learn from this. It's a real life example of poor science and how some people don't understand science. No more models being described as reality and we are doomed. It's a risk and all predictions are subject to our lack of understanding of the science and the problems inherent within statistical modelling.

I'd add that good scientists or rational thinking people who have some self-awareness will learn from this if they got it completely wrong but it starts with a recognition of the flaws within the science and modelling. Stop portraying certainty that doesn't exist.

[7Wannabe5]

Personal behavior aside, some of the critiques I’ve read of Neal Ferguson’s science are ridiculous. For instance, one article referenced by conservative think tank dissed him for making use of stochastic modeling, because results in science should be replicable and the results generated by a stochastic model aren’t

[IlliniDave]

Personal behavior aside, some of the critiques I’ve read of Neal Ferguson’s science are ridiculous. For instance, one article referenced by conservative think tank dissed him for making use of stochastic modeling, because results in science should be replicable and the results generated by a stochastic model aren’t

Results of stochastic models, barring malfunction of the underlying computer, are replicable. Not much of a think tank there, ha!

I think the bigger question is: "Is modeling science?" The answer is that it depends on the motivation for it's use, and how it's used. Generally, as a professional modeler, I say no, modeling is not science. It's just a glorified chalk board when you are using it to chase something you don't have a good grasp of, and a means to illustrate science when it's something you have a good grasp of.

[jacob]

@iDave - From the perspective of an ex-physical scientist and ex-financial modeller, I think it's fair to say that "all science is based on models but not all models are based on science". For example, suppose I wanted to build a tracker-predictor, for example, a targeting radar for tracking falling objects.

An example of a model that was not based on science would be a Kalman filter (I presume you're familiar---if not, it's a purely mathematical construct for updating a variable with some prescription under uncertainty ... like an algorithmic expression of an ARIMA fit) that presumed that the velocity was unchanging (which we know is wrong since Galileo, but bear with me). This filter would still track and update the velocity and it would even work reasonably well because the changing velocity was updated during the error correction. The position would of course be equal to the last position plus the estimated velocity times the interval time. However, this model knows nothing about gravity.

An example of a model of the same tracker-predictor that was based on science would simply be s=s0+v0*t+1/2g*t^2. It's not a perfect model since it doesn't include friction, but it is science based and pretty decent as long as the air is thin and velocities are not too close to terminal.

(Experts would of course combine the two and use the Newtonian kinetics as the physical equation for the Kalman filter.)

I'm not aware of any science that's not some kind of model. Unlike an amateur who would know one or less models for a given phenomena , an expert [scientist] would know at least two or more models for the same scientific phenomena and have a way of evaluating their relative usefulness. They'd also know under which circumstances and situations the various models are valid. For example, Newtonian models for light objects at low speeds; special relativity for fast objects close to the speed of light; general relative for heavy objects and those close to the speed of light; and quantum mechanics et al for tiny objects. They even know where the white spaces on the map of models are. Fundamentally, models are thus ways of describing reality and scientific models are models that have been subject to the scientific method.

[Gilberto de Piento]

I suppose I'm just repeating what Jacob said in a different way but I'll add my two cents anyway. There are a ton of models that are meant to make predictions that will not come true. For example, let's say I make a model that predicts who will become a criminal. Say 10% of the group we are working with is predicted to have a criminal record after some amount of time.

If we then use that model to deploy an intervention that causes some of these would be criminals to stay out of trouble our prediction would now be wrong because only 5% become a criminal. We wouldn't complain that the model doesn't work and many scientists are bad. Instead we would be happy that the model is a tool that can be used as part of a larger program for reducing crime.

The virus models work the same way. They aren't a failure because the predictions didn't come true. They did their job by predicting what would happen if no action was taken.

I'm going to leave it at that, this conversation where one person argues against all comers with a few repeated, nonspecific points reminds me too much of the climate change threads.