COVID-19

[fiby41]

The discussion on the last few pages especially the long form answers have been educational to me.

India crossed the 1,00,000 total infected mark so percentage figures become easier to understand hence this update.
4.3% of all tested are positive
57.9% active
39.2% recovered
3.156% deceased.
First day of more than 5k newly infected saw 8 states and 4 union territories reporting no new cases for atleast 24 hours on the following day.
3rd in terms of newly infected per day.

[jacob]

Here are the data points for US covid deaths for March prior to US lockdown measures becoming effective and changing the death rates. Also included is a curve fit to the exponential solution predicted by the simple SIR-model for the asymptotic case where S~1 (the beginning of an epidemic). I used https://mycurvefit.com/ to make it.



It's science and it's the math, and it looks pretty good to me.

Post lockdown, people stopped using constant doubling time exponential functions to predict further time evolution for obvious (at least to those with an understanding of the science) reasons.

[7Wannabe5]

@steveo73:

Forget about exponential. If a bathtub that holds 40 gallons is filling at the rate of 4 gallons/minute, how long until it overflows? Obviously, some wisecracker 6th grader can reply “Never, if I pull the plug.”, but that doesn’t change the method for solving the problem. You can construct a problem or model that does include the likelihood of mitigation. For instance, “What is expectation of final water level given 80% uniformly distributed probability that somebody will enter bathroom and pull plug over the course of the next 10 minutes?”

If the problem is clearly stated or the model clearly describes assumptions then it is only meant to be understood as more or less correct within the confines of its description inclusive of data and assumptions. It’s really rare to come across models where assumptions or data are hidden. The problem is that most people just pounce on the conclusion regardless of context.

In fact, if you go to the trouble of really looking under the hood of a model then you can readily make use of it in conjunction with further assumptions of your own choosing. You can accept that I am reflecting reality when I tell you how big the tub is and how fast it is filling and accept that I know how to express those realities in conjunction mathematically, and then mix in your own special sauce in the form of estimation of arrival of plug-puller, etc etc etc etc etc etc etc

ETA: What Jacob said.

[George the original one]

How many people on here have promoted the line that COVID-19 progresses as per an exponential model

Dude, it spreads exponentially until something reduces the number of available hosts, like lockdowns or behavior changes. You can easily, easily see this on a log-log chart like Covid Trends. Doubling time of 2-3 days.

https://aatishb.com/covidtrends/?locati ... on=Vietnam

Edit: Let's look at Russia & Brazil. Russia has reached something of a peak in new cases because the urban centers have become saturated and people are taking the virus more seriously; most likely new infections will trend downward now, but slowly. Brazil has public awareness, but virtually no orders, so their infections are going to double in 12-13 days. 246k cases in Brazil right now, so by May 31 they'll have 492k cases unless something changes... if you doubt it is exponential, then my prediction should be way off, not just by a day or three.

[steveo73]

If the problem is clearly stated or the model clearly describes assumptions then it is only meant to be understood as more or less correct within the confines of its description inclusive of data and assumptions. It’s really rare to come across models where assumptions or data are hidden. The problem is that most people just pounce on the conclusion regardless of context.

I agree - herein lies one part of the problem. It comes back to my point earlier on the validity of some study stating that something is statistically significant. Some people appear to believe that if something is statistically significant it holds true for the complete population when that is not necessarily true.

It's not a layperson problem or better put not just a layperson problem. I think it's a scientific philosophy or framework issue as much as it as layperson issue. I have a solution for it and there doesn't appear to be any disagreement to the solution. In fact I think you believe that you disagree with me but you are agreeing with me.

[steveo73]

Dude, it spreads exponentially until something reduces the number of available hosts, like lockdowns or behavior changes. You can easily, easily see this on a log-log chart like Covid Trends. Doubling time of 2-3 days.

Exactly, It's why you can't just utliize a simplistic exponential model because it's not. It's the complexity that people don't understand when assessing the science.

[steveo73]

It's science and it's the math, and it looks pretty good to me.

Post lockdown, people stopped using constant doubling time exponential functions to predict further time evolution for obvious (at least to those with an understanding of the science) reasons.

Bingo. All the alarmists who didn't understand the science predicted very different results compared to what has occurred. The model did not represent reality. It works within a given context but it doesn't hold true in all contexts. Patterns change.

I actually think it's a group of people (scientist and lay people) who haven't had a robust framework with which to analyze the science that have really
struggled with how this pandemic has evolved. It doesn't fit a simple model. The alarmist hysteria was way over the top.

Plenty of good scientists (and laypeople) have recognized the issue is complex and have been able to think this event through without the hysteria. I watch ABC TV and there is a doctor on who has provided calm, rational, objective discussions about the issue. I've heard plenty of great scientists talk and there have been some great links provided within this thread.

[tonyedgecombe]

Bingo. All the alarmists who didn't understand the science predicted very different results compared to what has occurred. The model did not represent reality. It works within a given context but it doesn't hold true in all contexts. Patterns change.

In the UK it has been worse than the scientific predictions, back in March we were told total deaths would be about 20,000 but it's way beyond that already.

[7Wannabe5]

@steveo73:

I have no problem with the notion ofcreating more complex models inclusive of risk management techniques. What I was attempting to communicate is that doing it on the fly in your head is unlikely to produce accurate results. For instance, resorting to simple black/white heuristic of “This expert was wrong and fooled me, therefore the opposite must be true!”

What I am also attempting to communicate is that it is easy/safe to shoot down somebody else’s work, much more difficult to construct your own predictions, clearly express them, and update/revise them in accordance with an overall strategy which is also clearly described.

[saving-10-years]

@tonyedgecombe
Well that 20,000 figure for UK deaths was expressed as though it _might_ be attainable: "If we can keep deaths below 20,000, we will have done very well (Powys, mid March)." If only we had.

https://www.bbc.co.uk/news/stories-52419218

Some of the US models, (particularly the IHME estimates) have been much more pessimistic and have now adjusted and adjusted and seem to be finally somewhat realistic. (I now understand from this forum that this is because of the approach to modelling taken.

For comparison with the 20k prediction: The IMHE on 7th April (as far back as I can go using wayback machine) predicted UK deaths of 66,314 by August. Current IHME estimate is 43,479.

The peak death rate for UK (as predicted by the IHME) was 2,825 in mid-April (within a range of 934 and 9833). The actual peak was 960.
That is a huge level of uncertainty and I am rather glad that we were not then relying on that model to plan anything. I stopped paying attention but I'm starting to watch it again as its getting a bit more realistic.

I am a novice at understanding models, I just want something more international to compare the UK-generated UK stats to.

https://web.archive.org/web/20200407173 ... ed-kingdom - 8 April
https://covid19.healthdata.org/united-kingdom - current info

[7Wannabe5]

Most weather forecasters err somewhat on the side of pessimistic in their forecasts. Most stock market analysts err greatly on the side of optimistic in their forecasts.

[steveo73]

In the UK it has been worse than the scientific predictions, back in March we were told total deaths would be about 20,000 but it's way beyond that already.

Interesting. I think the alarmist models were extremely inaccurate. I'm pretty sure that Neal Ferguson contributed to this and he is a UK scientist. It was predicted that there would be 500k deaths in the UK. It's at about 35k now.

[7Wannabe5]

First interesting note from my reading of “Introduction to Infectious Disease Modeling” by Vynnycky and White.

Theoretical threshold for herd immunity (1-1/R0) is based on assumption of randomly mixing population, so notion that you can selectively seclude 30% vulnerable proportion of population and infect virtually everybody else and then mix vulnerable population back in without further spread is not theoretically sound. If the elderly in the U.S. were cordoned off until almost everybody else was infected, they would still be vulnerable in almost exactly the same way that every new batch of preschoolers would be very vulnerable to measles, mumps etc if they were not vaccinated.

[CS]

Interesting. I think the alarmist models were extremely inaccurate. I'm pretty sure that Neal Ferguson contributed to this and he is a UK scientist. It was predicted that there would be 500k deaths in the UK. It's at about 35k now.

And there is the irony. It is early days and your skepticism about the outcome, if widely held, will make it true.

[jacob]

First interesting note from my reading of “Introduction to Infectious Disease Modeling” by Vynnycky and White.

Theoretical threshold for herd immunity (1-1/R0) is based on assumption of randomly mixing population, ...

So you have the book already? This assumption goes away in agent-based models by construction. However, it should also be possible in multi-dimensional SIR type models (split by demographics or cities or vocation or whatever) by changing the cross-infection coefficients between the various groups.

Also see my two most recent answers to JL13 for how reignition works once herd immunity has been reached.

[George the original one]

#1 USA
#2 Russia
#3 Brazil NEW TODAY!!

[Quadalupe]

First interesting note from my reading of “Introduction to Infectious Disease Modeling” by Vynnycky and White.

Just finished the introduction and the book is quite readable so far. It seems like a nice introduction for laymen of both math and epidemiology. Li's book seems more useful for people who are more used to mathematics, specifically calculus.

[jacob]

Mass media routinely portray information about COVID-19 deaths on logarithmic graphs. But do their readers understand them? Alessandro Romano, Chiara Sotis, Goran Dominioni, and Sebastián Guidi carried out an experiment which suggests that they don’t. What is perhaps more relevant: respondents looking at a linear scale graph have different attitudes and policy preferences towards the pandemic than those shown the same data on a logarithmic graph. Consequently, merely changing the scale on which the data is presented can alter public policy preferences and the level of worry, even at a time when people are routinely exposed to a lot of COVID-19 related information. Based on these findings, they call for the use of linear scale graphs by media and government agencies.

https://blogs.lse.ac.uk/covid19/2020/05 ... -covid-19/

Keep in mind that 30% of Americans (similar numbers for other countries ranging from 8% in Japan to 32% in Italy) are not sufficiently numerate to grok linear graphs either. https://nces.ed.gov/pubs2014/2014008.pdf (Fig 2-B on pg 11 and Ex 3-B in App B-7) Similar communications problems exist for increasingly complex representations like trying to understand or explain what a model prediction is and tries to do.

[The Old Man]

Keep in mind that 30% of Americans (similar numbers for other countries ranging from 8% in Japan to 32% in Italy) are not sufficiently numerate to grok linear graphs either.

Many people in the USA have difficulties with fractions.

[7Wannabe5]

@jacob:

I downloaded the book. I grokked what you wrote in response to JL13. It would also be possible to drop assumption of random mixing in multidimensional SEIR model, but this sort of adjustment is not evident in consideration (hacked armchair model)of those who are suggesting strategy of “let it burn until herd immunity!”

I was trying to think of an analogy that might be more obvious. The best I could come up with is people sending around a Super Cute Kitty meme during a period when internet service has been blacked out for everybody over age 60. Even if everybody under age 60 is totally over Super Cute Kitty by the time the internet is restored to senior citizens, it will still spread among the seniors once their internet service is restored, because most seniors would have large number of other seniors as contacts.

So, in order to pull off this sketchy strategy without adjusting Rt piecewise upwards, senior citizens would first have to be nearly completely isolated from all young people until all the young people were infected and then they would have to be almost completely isolated from each other indefinitely.

@Quadalupe:

Yes, it is quite readable and is meant to be accessible to anybody with fairly thorough grounding in high school level math. However, the pace does pick up a bit in Chapter 2. Apparently there is some squishy usage of the notation within the field, so that lends some confusion.