Yesterday, a group in Oxford released a paper that implied that a signifcant fraction of those in the UK may already have been infected. This was quickly picked up by numerous media outlets who highlighted that coronavirus could already have infected half the British population. James Annan has already discussed it in a couple of post, but I thought I would comment briefly myself.
To be clear, I certainly have no expertise in epidemiology, but I do have expertise in computational modelling. So, I coded up their model, which is described in Equations 1-4 in their paper. They were also doing a parameter estimation, while I’m simply going to run the model with their parameters.
The key parameter is , which is the proportion of the population that is at risk of severe disease, a fraction of whom will die (14%). They explicitly assume that only a very small proportion of the population is at risk of hospitalisable illness. Consequently, they focus on scenarios where the proportion requiring hospitalisation is 1% () and 0.1% (). The Figure on the right, which considers , , and , is from my model and seems to largely match what’s been presented in the paper.
The curves that start at 1 and then drop are the proportion of the population that is still susceptible (left-hand y-axis) while the diagonal straight lines are the logs of the cumulative deaths (right-hand y-axis). I’ve also shifted the models so that the latter overlap. This Figure illustrates why this study was picked up by the media. Cumulative deaths to date is just over 400. If the proportion of the population at risk of hospitalisation is small () then just over 30% of the total population would still be susceptible. In other words, more than half of the UK population would already have been infected. On the other hand, if the proportion at risk of hospitalisation is large () then the proportion susceptible is still large () and the fraction that has already been infected is small.
One way to estimate is from the date at which the first case is reported. If is small then the lag between the first case and the first death is larger than if is large. The paper implies that the current data is more consistent with a small than a large . The problem, as this critique highlights, is that this implies that this first case is the progenitor of most of the subsequent cases. Given the small numbers involved, this may well not be the case, since a localised outbreak may not have taken hold. Hence, there doesn’t really seem to be strong evidence in support of being small and, consequently, there is little evidence to suggest that a significant fraction of the UK population has already been infected.
Okay, despite the lengthy pre-amble, this is really what I wanted to focus on in this post. I think it’s perfectly fine to play around with models and to try and estimate various parameters. However, especially when the results have societal significance, it’s very important to be clear about what’s been done when presenting the work publicly. This research has not demonstrated that more than half the UK population has already been infected, it’s simply illustrated that it’s possible. Clearly if most of the UK population has already been infected, then this virtual lockdown could probably be relaxed. However, if is not small, then the lockdown would seem justified. As James points out in this post, even though the paper implies that the current data is consistent with being small, there do seem to be regions where this seems not to be the case.
So, I think it’s highly irresponsible to present a result like this without being extremely careful to minimise the chances of it being misconstrued. It’s clearly not possible to completely avoid research being misrepresented, but researchers do – in my view – have a responsibility to ensure that this not an easy thing to do. It would be great if the impact of this virus is far less severe than we currently think. However, until we have more evidence to support such a conclusion, we really should be very careful of presenting results that imply that this is the case.
This post ended up being much longer than I intended. I was mostly wanting to highlight how I think the presentation of this result was highly irresponsible. The first bit was just meant to illustrate what they’d done in their model. Since I’m not an expert in this field, and have no interest in spreading misinformation about an important topic, if any experts think I’ve made some kind of mistake, feel free to point it out.
I also wanted to post another figure, which is essentially the same as James highlighted in this post. The curves that rise and fall are the number of people who are infectious (left-hand y-axis) while the curves that rise and then level off are the cumulative deaths (right-hand y-axis).
This again illustrates (given that cumulative deaths to date is just over 400) that if the proportion requiring hospitalisation is small () then the number of people who have already been infected is already quite high, while if the proportion needing hospitalisation is large () then the number of people who have already been infected is much smaller. It also illustrates that the overall cumulative deaths depends quite strongly on this parameter; if we relax current conditions based on this work and it turns out that isn’t small, the impact could be substantial.
In the interests of transparency, if you would like to codes that produced the two figures, you can download them from here.
I’d just like to emphasise a small but important detail:
“This research has not demonstrated that more than half the UK population has already been infected, it’s simply illustrated that it’s possible.”
could be more pertinently written as
“This research has not demonstrated that more than half the UK population has already been infected, it’s simply illustrated that it’s possible *if we only look at the numbers of deaths alone and do not make any attempt to take account of numerous other sources of evidence*.”
It reminds me of some of the sillier climate sensitivity games people play: “based on this small set of useless data, we cannot rule out sensitivity being eleventy-one!” to which the appropriate response is not “oh look, sensitivity might be eleventy-one” but rather “use a better data set you numpty, we all know that sensitivity is 3.”
Ben McMillan just posted a comment on an earlier post, which highlighted this article about testing in Iceland.
As of March 15, they had results from 5571 tests, yielding 48 positive results. Additionally, they had identified 473 cases, with one death and 12 in hospital. This does not seem consistent with the proportion needing hospitalisation being as small as suggested in the paper.
You make a good point and the PM press conference today discussed the problems of inaccurate testing.
From the data I have seen, for example https://ourworldindata.org/covid-testing, currently about 10% of COVID-19 tests done are coming back positive. Since testing is being focused on the most symptomatic individuals, it seems unlikely that random testing would produce a higher percentage of positives, otherwise we would do better by random testing. Therefore 10% seems like a upper bound for COVID-19 frequency, and most likely a very high upper bound.
As time is of the essence in acting to identify and isolate coronavirus hot spots, fast data communication can help contain the pandemic, witness what has so far happened in South Korea:
Indeed. Also, as I understand it, the specificity of the test is about 90%. I always get slightly confused about false positive and false negatives, but I think this means that about 10% of those who are negative will test positive. If the true number of negatives is much larger than the true number of positives, then this would imply a large fraction of the positives are false positives (I hope I’ve got this right).
If a large percentage is already infected, then active cases would not only be flattening, they would be going down. Only one country has accomplished this, and it’s China, which practiced aggressive isolation methods. You can’t get the virus if you are effectively isolated.
Herd immunity and effective isolation are exactly alike in this respect: they both deny the virus a host where it can replicate.
So if you have a herd where a large percentage of the animals have acquired meaningful immunity, the UK, dying will rapidly slow and new and active cases will drop precipitously.
It’s China. They isolated the holy crap out of this thing and they stopped it. You can see where it ran out of new hosts. It could not get to them. Girl from Wuhan in a youtube video – did not leave her apartment for 43 straight days.
If it somehow matters, conduct a good random test of the population.
The biggest issue is that we don’t know what the denominators are. Meaning we don’t actually know how many are infected, and we simply won’t know that for a very long time now.
As far as I know there’s only one region that’s aggressively tested its population and aggressively traced contacts, and that’s South Korea. It also has a measured death rate of .6% buuut that’s skewed by is age demographics.
Smart countries are copying South Korea’s efforts. (I know we are in Alberta Canada.)
The US is currently trending to overtake any other nation for most cases. Good work guys, keep it up?
Japan is the place I find interesting. They ain’t doing much… yet the spread is low. Their culture already has social distancing component, they ‘don’t give offense’. Its super crowded yet they don’t get so close that they touch, even on transit.
I am not an expert either, but … the IC study also sucked majorly. Tens of millions hospitalized and several million deaths in the USA alone.
Do you think that these studies are rushed at all? Well, I do.
We’ve never seen something like this before, Given our current levels of technologies and communications. Peer review before we’ve even reached a global peak? Like Roulette, someone will make a correct guess, fame and fortune to follow.
Given the original global baseline of zero infections, the current doubling time is ~6.5 days.
Reposting this link: this is a study from Iceland where a random sample of the population was tested. I estimated that it was consistent with a bit more than 10% of cases coming to the attention of the health system (usually through GPs or hospitals).
Thanks. I mentioned that in my response to James.
ATTP: yes, thanks, apologies for (deliberately) being repetitive!
If really 50% of the British population would be infected, how many of the Italian population must be infected, assuming the same ratio of death?
And in the Lombardy?
On the other hand, if we assume an upper limit of 50% of the population of the Lombardy is already infected, we can derive an upper limit of 0.866% of the British population is already infected. And the real value could be much lower.
Pop: UK: 60M, It: 60M, Lom: 10M
Death: UK 465, It 7503, Lom: 4474
I think the Italian situation just about fits their models with but it would imply that almost the entire population has now been infected. However, you’re right that Lombardy seems completely inconsistent.
It seems a lot like climate sensitivity in that it doesn’t greatly affect the rational decision if you can show that ecs *might* be low, a more compelling argument would be to rule out ecs being high (because the loss function isn’t linear – for Coronavirus, as soon as you go beyond the capacity of the health system, the losses rise very quickly)
Everett: I’m no expert either, but my wife is, and she’s currently working 12 hour days on this. All I really hear is that we don’t have viable numbers yet, but we do know that if this gets out of control it will overwhelm the health care system quickly.
Mostly folks are just freaking out when they should simply follow standard procedure…
Keep your distance.
Wash your hands.
Don’t touch your face.
Avoid plague ships.
I have no knowledge of epidemiology or those models. That said, I am a little surprised by how the model is formulated in the original paper, wrt. equation 1, which drives the change in the “proportion of infectious people”.
The way one enters in this condition is easy to understand – it is random and proportional to the number of susceptible people * infectious people. But the way the diseases gets “resolved” is less obvious to me: it is formulated as an exponential decay, such as for radioactive decay.
Equation 1 means, afaik, that at any time, sigma is the fraction of “sick people” that would “resolve” (die or recover, but no longer contaminate others). In a “population” of radioactive nucleus, that makes sense easily because nucleus have no ‘memory’, they just have a constant probability to disintegrate. But is a population of sick people following the same rule? I would have expected that it takes a certain time to recover, and does not occur randomly : in other words, with 1/sigma having a mean of 4.5 days as they indicate (with a rather small standard deviation = 1 day), if there are 10 people sick on day 0, then between day 4 and 5 there would be roughly 10 people “resolving”, ie, who dies or recovers by day 5 depends on who became sick on day 0. By contrast, the differential formulation that is in equation 1 implies that the number of people recovering between day 4 and 5 depends on the number of people sick by day 4. When those numbers are changing quickly, that could perhaps have an influence on the results? Perhaps that is a standard definition of the “infectious period”, but I am puzzled at how this can represent the actual process correctly, ie I wonder if disease resolution and radiactive decay really follow the same path as I suspect the equations imply. I guess that people modelling that kind of thing know whether this is stupid or not, but if not, then your python code appears able to explore the change by replacing sigma*y in eq. 1 with dz/dt at time (t-1/sigma).
And then there’s Japan:
With no explosion. 😉
Keep your distance = Something I’ve been practicing since age 8.
Keep your distance = Something I’ve really been practicing since age 60.
The only way that I can go deeper into my hole is to completely stop eating. Now where is that stock of canned goods …
The only realistic option is to shelter in place, I have my own doubts that most have truly sheltered in place, simply because they are not use to doing so, or are purposefully ignoring the blatant warning signs.
Philippe Marbaix is asking whether there is more to it than that. The Oxford group says they are solving a stochastic SIR compartmental model, which means that all the DiffEq parameters are allowed to vary in addition to possibly adding a noise term (see their Table 1), and then use a MCMC (Markov Chain Monte Carlo) simulation for fitting. This accounts for possible variation that occurs in the real world. We’re discussing this stuff over at an Azimuth Project thread that began in 2011. Some interesting ideas on how to estimate asymptotic values have emerged.
“From the data I have seen, for example https://ourworldindata.org/covid-testing, currently about 10% of COVID-19 tests done are coming back positive. Since testing is being focused on the most symptomatic individuals, it seems unlikely that random testing would produce a higher percentage of positives, otherwise we would do better by random testing. Therefore 10% seems like a upper bound for COVID-19 frequency, and most likely a very high upper bound.”
Infection is local, and the attack rate will be local. There MIGHT be a meaningful “global” average
But the data is very high spatial frequency so global metrics can mislead you.
Think rainfall data. The average over the country might be low, but that downpour causes a flood
that overflows your dam.
“Therefore 10% seems like a upper bound for COVID-19 frequency, and most likely a very high upper bound.”
Kinda meaningless to policy, which should be local.
NY is seeing a positive rate of 25%. Part of that is testing protocol. They only test with Symptoms.
So your positive rate is going to be a function of your test criteria. Who gets tested where
In Korea the test approach is different. I will illustrate by using the call center case.
Employee on 11th floor tests positive on march 8th.
All 207 employees in his company are tested.
All Contacts are tested
residents (553) from other floors in the building are tested
“From the call center building in Guro-gu, Seoul, no additional cases were confirmed. The current total is 158 confirmed cases since 8 March. Of the 158 confirmed cases, 97 are persons who worked in the building (11th floor = 94; 10th floor = 2; 9th floor = 1), and 61 are their contacts. The KCDC shared the interim result of their epidemiological investigation in collaboration with Seoul City, Incheon City, and Gyeonggi Province during the monitoring period of 9-22 March. The call center on the 11th floor had the highest infection rate (43.5%), compared to 7.5% and 0.5% for 10th and 9th floors, respectively. There was no confirmed case from other floors. Of the 226 persons identified as family members of the 97 confirmed cases who worked in the building, 34 (15.0%) were infected. Of the 97 confirmed cases, 8 (8.2%) were asymptomatic cases. Of the 16 persons identified as family members of the 8 asymptomatic confirmed cases, no confirmed case was found.”
The Korean testing data is not exactly random, since you chase down contacts. But they are running about 2.5% confirmed positive rate. probably better than just testing symptomatics
To reinforce the notion that infection is “local”, about 80% of all Chinese infection was family to family. if your dad tested positive he was sent off to live in a gym for 2 weeks
Not sure any of this makes it any more clear, but in general the national average stats don’t
really give you much information. Best example. calculate china 2 ways. With and without Wuhan.
denominators matter and the choices made ( who to test where) will change what you perceive.
duh. Or calculate Italy by region, or Korea by region. Hugely different infection rates.
Hmm. what would your reaction be if I calculated the national windspeed for America while
a hurricane was hitting the coast and remarked that the average was unexceptional?
This professor gets it. They stopped a contagious and lethal virus without relying on herd immunity or having a vaccine, The west is just in denial on this, and it is going to cost them dearly. Summer might stop it, but it also might not. It’s barley begun its onslaught through the 3rd world.
click on document 213 and 214 ( the green “n”)
you will find all the test data
mortality by age and sex
And cases by region
denominators matter. Glad to see alberta doing what works.
economic shut down is not a long term solution,
neither is head in the ground or screaming hoax as my conservative friends seem to think
For testing Korea ( and Singapore) have done massive tracking and Proactive HUNTING for the disease. Not reactive testing of symptomatics.
“In Daegu, testing has been completed for every person at high-risk facilities. Of the 32,990 test results, 224 (0.7%) were positive results.”
Ah they mentioned the APP we have in Korea.
Mine is set to a 5 minute zone. 5 minutes by taxi. If a case gets reported within 5 minutes of me
I will get an “amber alert” a text message telling me what place to avoid. Went off twice this morning. That’s about average for a day. So the other day, some traveler from Poland presented
at my local hospital ( dumbshit) and tested positive. So I get a text and I imagine the hospital
receiving gets a disinfecting. he will be dispatched to a life treatment center, 7000 were opened,
for isolation and possible hospitalization. The government took control of 7000 locations nation wide
to house the infected people who don’t need a hospital
“So, I think it’s highly irresponsible to present a result like this without being extremely careful to minimise the chances of it being misconstrued. It’s clearly not possible to completely avoid research being misrepresented, but researchers do – in my view – have a responsibility to ensure that this not an easy thing to do.”
Perhaps they should have named them Representative Corona Pathways.
That’s an interesting point. If I get a chance, I’ll have a look at that (I think I need to try and do some of my own work today 🙂 ).
How singapore works
church is not a good thing
Contrast between South Korea and US couldn’t be starker. Infections are still outrunning our ability to test. Flying blind.
The problem is that we have two different imperatives and one common pathway.
The best way to finish the problem is to get herd immunity up to 70% as quickly as possible while protecting the old and frail.
IIRC correctly, you’re a doctor. Is that right? How do you get herd immunity up to 70% without over-stressing our existing healthcare? As I understand it, the current strategy is based on flattening the curve so that our healthcare systems can deal with the enhanced number of patients.
You let people die (provided it isn’t you, that is O.K.)?
Protecting the old and the frail is exactly the point of the current strategy – I don;t think the idea is to protect the healthy it is to stop healthy people from spreading the virus to the vulnerable.
That’s a intriguing comment. It leads me to wonder why the energy modelers aren’t using compartmental models to make projections of usage as epidemiologists are doing for contagion growth with their SIR compartment models. See if anybody is doing that:
Protecting the vulnerable – largest government project in the history of government. Basically talking about putting a large percentage of the population, the elderly and the caregivers, into nursing home prisons until either a vaccine appears or herd immunity, which, while likely, we don’t even know it is possible.
90% of the healthcare professionals and politicians in the west, plus 100,000% of the least helpful climate skeptics on the face of the earth, think South Korea and Japan have achieved herd immunity. Dangerous.
i seems quite a few people simply don’t have the cognitive ability to work thru the twin concepts of high communicability and a low (but relatively high) mortality rate
its the “trace” gas argument – people get fixated with %’s,
“its a small % isn’t it!!!!! – I mean lets get some common sense perspective on this” [/sarc]
There are a couple of papers linked from this Nature news piece which bear on it: What the cruise-ship outbreaks reveal about COVID-19.
Pretty much everyone was tested, some multiple times. Only 18% of positives were asymptomatic; and getting on for half of symptomatic positives need hospitalisation, from studies around the world. Yes cruise ships carry an older cohort, but 18% is a long way from 99% or 99.9%, and there were young crew members. The second one is a preprint which finds about 1% deaths from confirmed infections and an infection fatality rate – the proportion of all infections, including asymptomatic ones, that result in death – of about 0.5%. Which seems eerily familiar. Maybe there’s something to that consensus?
The third one is about infection transmission: higher on a ship as you’d expect, but less than 1 once people were quarantined in their cabins. Which presumably means it can’t spread through the A/C (they must have left it on or people in interior cabins would have died of heatstroke).
If you’re using molecular tests then you are testing who has it. If you are using serum tests you test who has had it. They measure entirely different things. The sort of calculations you are working would be best suited to serum tests. The isolation of infected individuals is best done with molecular tests. Just something to keep in mind. Make sure you check your test type.
COVID-19 Coronavirus Pandemic
This is my estimate, how may are already infected, including the untested and how completed the tests are.
Estimates for total cases, by scaling my model for Germany to other countries, by assuming the same age structure and death rate for all countries, as well as that the reported numbers of deaths are correct.
Cases are for 2020-03-26 00:00 GMT.
These numbers are higher than the official tested due to:
-Timing of testing, the testing is made days after infection. And the numbers increase rapidly.
-There are cases, especially mild ones, that are not tested at all.
I assume constant probability of testing, so that the growing rate of positive tested and total cases is the same. If testing is increased over last weak(s), than growing rate (and total cases) is overestimated, if testing is decreased, than growing rate (and total cases) is underestimated.
Death rate is estimated from 1.3 to 2 per cent, if including ALL infected people, including those not tested or with no or mild symptoms. Ratios of below 1.3 per cent of cumulative deaths and my total cases estimate are simply due to the time delay between infection and death and the rapid growth of infections.
Without adequate medical treatment the death rate would be even higher. And many people who survive with damaged lungs will need medical treatment for long time or there remaining life.
CSV Table of my results:
k means 1000
M means Millions
Country, case est, growth, tested cases, found, popul, infected, death1, death2, death3, death4
USA, 966k, 27%/d, 68k, 7%, 329M, 0.29%, 1027, 14482, 3455k, 10364k
+New York, 463k, 31%/d, 33k, 7%, 20M, 2.4%, 366, 6938, 205k, 616k
+other USA, 503k, 25%/d, 35k, 7%, 309M, 0.16%, 661, 7529, 3249k, 9748k
Spain, 887k, 20%/d, 50k, 6%, 47M, 1.9%, 3647, 13304, 495k, 1484k
Italy, 635k, 7%/d, 74k, 12%, 60M, 1.1%, 7503, 9524, 633k, 1899k
+(0)Lombardy, 231k, 3%/d, 32k, 14%, 10M, 2.3%, 4474, 3422, 106k, 317k
+other Italy, 404k, 11%/d, 42k, 11%, 50M, 0.79%, 3029, 5969, 527k, 1582k
France, 306k, 16%/d, 25k, 8%, 67M, 0.46%, 1331, 4595, 704k, 2111k
(1)Iran, 246k, 9%/d, 27k, 11%, 83M, 0.30%, 2077, 3683, 872k, 2615k
(2)China, ?215k?, 0%/d, 81k, 38%, 1401M, 0.02%, 3281, 3234, 14711k, 44132k
(3)Germany, 191k, 15%/d, 37k, 20%, 83M, 0.23%, 206, 2863, 873k, 2619k
UK, 137k, 17%/d, 9.5k, 7%, 66M, 0.21%, 465, 2061, 693k, 2079k
Netherlands, 94k, 19%/d, 6.4k, 7%, 17M, 0.54%, 356, 1406, 183k, 550k
Belgium, 73k, 25%/d, 4.9k, 7%, 12M, 0.63%, 178, 1090, 121k, 362k
Switzerland, 70k, 16%/d, 11k, 16%, 9M, 0.81%, 153, 1044, 90k, 270k
Turkey, 63k, 41%/d, 2.4k, 4%, 83M, 0.08%, 59, 942, 872k, 2615k
Portugal, 51k, 33%/d, 3.0k, 6%, 10M, 0.49%, 43, 759, 108k, 324k
Canada, 36k, 23%/d, 3.4k, 9%, 38M, 0.10%, 36, 542, 399k, 1196k
(4)Brazil, 28k, 23%/d, 2.6k, 9%, 211M, 0.01%, 59, 424, 2216k, 6647k
Sweden, 21k, 18%/d, 2.5k, 12%, 10M, 0.20%, 62, 310, 108k, 324k
Austria, 15k, 10%/d, 5.6k, 37%, 9M, 0.17%, 31, 228, 93k, 280k
Israel, 14k, 24%/d, 2.4k, 16%, 9M, 0.16%, 5, 217, 96k, 289k
Australia, 13k, 14%/d, 2.7k, 20%, 26M, 0.05%, 11, 197, 269k, 808k
Denmark, 12k, 17%/d, 1.7k, 15%, 6M, 0.20%, 34, 173, 61k, 183k
(4)Malaysia, 11k, 17%/d, 1.8k, 16%, 33M, 0.03%, 20, 167, 344k, 1031k
(5)S.Korea, 10k, 2%/d, 9.1k, 92%, 52M, 0.02%, 126, 149, 543k, 1629k
Ireland, 10k, 18%/d, 1.6k, 15%, 5M, 0.21%, 9, 154, 52k, 155k
Japan, 10k, 10%/d, 1.3k, 14%, 126M, 0.01%, 45, 140, 1323k, 3969k
San Marino, 2700, 11%/d, 208, 8%, 33k, 7.6%, 21, 38, 350, 1050
(4)WORLD, 4559k, 13%/d, 471k, 10%, 7774M, 0.06%, 21k, 68k, 82M, 245M
(0)Lombardy: number of death are already higher than estimated (death1 vs. death2), higher death rate may due to either more older people infected, or health treatment not more adequate for all who need this
(1)Iran: lower bound, may be much higher due to unreliable numbers
(2)China: Model not applicable, not in exponential growing, total may be approx 200k)
(3)Germany: model value, values of other countries are scaled with this
(4)different age structure, less older people, so model may be some what off, on the other hand countries with less older people have often a poorer health and health system, don’t know which effect wins
(5)S.Korea: They may have found almost all cases by testing and were able to stop spreading the epidemic.
Meaning of the columns:
Country: Name of country
case est: my estimate of total cases until today 2020-03-26 0:00 GMT
growth: growth rate in per cent per day
tested cases: number of official positive tested cases
found: per cent of (my estimated) cases found in the tests, if this is low, a large number of cases is undetected
popul: population M=Million
infected: per cent of population already infected
death1: number of deaths up to now
death2: number of deaths, if 1.5% death rate and if no more infections immediately
death3: number of deaths, if 70% of population is infected so slowly, that health system can give adequate treatment to all
death4: number of deaths, if 70% of population is infected rapidly, assumes a 3 times higher death rate (4.5%) when the health system breaks down.
Note: k means 1000s, M means Millions
A lot of people are trying to paint the handling of COVID-19 in absolute terms of deaths, or eliminating the virus. You just can’t do that.
What we’re really trying to do is make sure the health care system can handle what is happening, and cope with the cases as they come up. Once that happens we need to get our economies back on their knees.
I was kinda stunned you were the only one to use that approach.
“The problem is that we have two different imperatives and one common pathway.
The best way to finish the problem is to get herd immunity up to 70% as quickly as possible while protecting the old and frail.”
easier said than done.
same with cutting c02
“Contrast between South Korea and US couldn’t be starker. Infections are still outrunning our ability to test. Flying blind.”
Point of care testing has just been submitted to the FDA, 15 minute results.
we might see a world where you have to be tested to travel, work etc.
just replace hog/pig with human being
Can’t say for sure that it exists and is well-known but under a different name. That’s the problem with seeding mathematical ideas across disciplines. This issue came up on a twitter thread that popped up on my timeline this morning. I suggested that applied category theory may help here. Computational structures can be “categorized” according to computational flow, and thus structures & algorithms used in different disciplines but with conflicting names can be pattern matched via the data-flow structure and then re-applied. This would be with potentially less effort since likely so much effort has gone into application areas where the categorical structure has been proven useful.
Read this entire thread:
If someone is actually interested in doing something like this, category theory central (John Carlos Baez & company) is the place to pursue it.
I was pointed out to this website on twitter and I think it’s interesting enough to share it here: https://www.sciencemediacentre.org/expert-reaction-to-unpublished-paper-modelling-what-percentage-of-the-uk-population-may-have-been-exposed-to-covid-19/
> We’re discussing this stuff over at an Azimuth Project thread that began in 2011.
You rather necromanced a 2011 thread to plug your own research, Web. I doubt PaulM (_ppmv) will need category theory for what he’s doing. Have you ever used it?
Drive-by done, btw.
Recovery does not mean virus free:
Give it 2 more weeks.
Yes, the churches on Easter Sunday won’t be safe unless the prior 14 days have zero positive tests. So, around 100,000 cases have to clear in the next day or so.
Checking roughly the data for the USA, by Easter or a few days later there won’t be enough hospital beds. A rough estimate places the apex at April 18th or so. That’s just the apex. Then there is the long, long tail…
JCH. its one of the clearest ways to see the data
Total vs New is similar to the Hubbert Linearization I showed
I just do not believe a sieve can stop it. Drove by a McDonalds today that is next to large construction site. Lobby was full of construction workers. So was the outdoor seating area. Side by side. Misled. Just horrifically misled.
Neighbors working at home with maids coming once a week. The other days of the week, they maid is going to somebody else’s house. Mess up like that on a farm and 20% ’em are dead on 5 farms.
Anyway, Kevin Pluck linked to it on twitter.
I take it the wacky South Korea line is the virus chewing into the general population?
A very nice video modelling the spread of the disease is found here (3blue1brown):
Confusingly, there’s another Imperial College group (from the engineering department) who’ve done a quick-and-dirty analysis matching second derivatives of total deaths to China. Unfortunately, the UK media is reporting it as if it was as as authoritative as the Ferguson team, or indeed from the Ferguson team, who do this for a living, who have access to government data, and who are using a sophisticated multi-layered model developed at their leisure back in the mid-2000s where all they have to do now is plug in new matching values for well-understood parameters (well-understood, but not as well constrained as for influenza because we don’t have as much data). Having said that, a real novel influenza strain would have had different parameters from seasonal flu anyway, so that will have been anticipated in the model design.
I’m wary of the way they allow themselves the freedom to slide the curves through time to arbitrarily datum each two time series. Although perhaps they would argue the early history is water under the bridge and they’re only comparing post-lockdown trends. More important I think, “start of lockdown” means different things in different countries. As do further increments of lockdown. The UK lacks the infrastructure to do a China or South Korea, even if it wanted to. And crucially, the only country that is far enough down the time-evolution path to show that it can be driven to near-zero is South Korea, which adopted the same aggressive contact-tracing, testing and isolation as China. Nowhere in the West is doing that, or AFAICS planning to do that. Or could: Hubei, for example, has 5G everywhere so control centres get reports from villages in real time. Nor do they account for the available-ICU-beds vs. saturated-ICU-beds-and-triage stages. The China reference data appears to be from after the period when Wuhan ran out of ICU beds, from when they had the spread under control and had more beds available. The idea that Italy and Spain will get away with about 25,000 deaths seems far-fetched, given that the south of Italy, for example, is following Lombardy. And that the USA will get away with 20,000, given what we’re seeing in New York and elsewhere and the mixed messages from Washington over lockdown, fantasy.
Yes holger I was just watching that.
‘I take it the wacky South Korea line is the virus chewing into the general population?”
one cluster , follow ups from contact tracing and Imports.
“From Manmin Central Church in Guro-gu, Seoul, 7 cases (members = 4; contacts = 3) have been confirmed since 25 March and contacts are under investigation. From Hyosarang Nursing Home in Gyeonggi province, 3 additional cases (resident at the facilities= 2; Staff= 1) from cohort quarantine were confirmed. The current total is 20 confirmed cases (Residents= 15; staff= 5) since 19 March. From Second Mi-Ju hospital in Daegu, 13 additional cases were confirmed. In total 75 cases (inpatient = 74, staff =1) have been confirmed as of 28th March.”
“The KCDC urged all inbound travelers to follow the precautionary measures with emphasizing on continuous cases of inflows from Europe and others. There were 41 imported cases(from Europe= 25, from America = 11, from Asia =4) which are 28.1% out of 146 newly confirmed cases yesterday. 168 cases were confirmed in the airport screening within the last 2 week out of 363 imported cases in total.
– Persons entering from Europe or the US should return home straight after arriving at the airport, refrain from using public transportation, travel in their own vehicle, and wear a face-mask during movement. Beginning 28 March, airport limousine bus or KTX train will be provided for inbound travelers who cannot access to their own vehicle”
A lot of hospitals are 1 staff and multiple patients— super spreader staff.
One thing about K hospitals– anecdotal observation. They are NOT big on the “blue glove” changing that you haven in a US hospital. I had an operation before and when I went into
ER NONE of the girls taking blood wore those disposable gloves you see in a USA facility
Seriously the girl drawing blood got my blood all over her hands and was totally non chalant
about it. So, they may have this testing thing down but no system is perfect and this critter
is apparently hard to wipe out.
Korea has two regions that are still on an exponential path. So I expect more crackdowns.
The government has started to crack down on quarantine violators. The last was a 19 year
old girl returning from the US. She was symptomatic, and went to Jeju Island with her mom
Gave it to her mom. word is she will be fined,
“According to the Jeju government, the woman and three other people, including her mother, arrived on Jeju March 20 and ― despite her coronavirus symptoms ― traveled around the island before being confirmed to have the disease at a Seoul health center on March 24.
As a result, 38 people, who had been in close contact with her, are now under quarantine in their homes. Most of them are residents of Jeju, an island with a population of 670,000.
Their movement information, which is available on the Jeju government website, shows they stayed at Hanwha Resort Jeju for the first two nights and at Haevichi Hotel and Resort Jeju for the other two. All places they visited have been disinfected.”
“can be driven to near-zero is South Korea, which adopted the same aggressive contact-tracing, testing and isolation as China.”
Isolation in china was the iron fist. very different.
When Wuhan was closed down, they literally closed it down. No flights out, no trains out.
Road blocks. and all though out the province small towns put up literal road blocks.
Some got out
China quarantine was different than Korea. in China you got separated from your family and put
in a big room with others. Lots of scenes from those places on social media. Old ladies dancing
waiting out their 14 days. Korean “Lock down” was a voluntary thing.
Now its getting more stringent.
I will say this. watching the Koreans play wack a mole with this thing indicates to me that
voluntary measures even with a super compliant community is eventually going to break down
if just a matter of time until we have another super spreader who will set off a 1000 person outbreak. HK, Singapore, Taiwan.. all having to play wack a mole now.
Thanks for the local insights Steven. Yes I was over-generalising, with contact-tracing being the main difference from Europe. Although the impression from here is that there was also a more rapid lockdown and more compliance (apart from the church group, which was traced). Regardless, “Western lockdowns will be as effective as China’s”, or “lockdowns without full tracing and testing will be as effective as lockdowns with”, would seem somewhat optimistic assumptions. We’ll see whether China also has to play whack-a-mole. The advantage they have is a very large, very heavy hammer and a surveillance society already in place. Kinda like having your garden already loaded with motion cameras and sound detectors so you can be on the mole when it first peeks above the surface.
I posted some of the WHO China links on another thread and will re-post here. The Geneva press conference in particular has some eye-opening insights into the on-the-ground social situation in China as well as the medical stuff. As in “why is everyone staying indoors with no soldiers on the streets?”.
Geneva news conference (an hour, and an hour of Q&A).
Beijing press conference the previous day (transcript)
Report of the WHO-China Joint Mission on Coronavirus Disease 2019 (COVID-19)
The best way to get herd immunity is with vaccination. Until a vaccine is developed that can be distributed world wide we will be fighting the disease. The object of quarantine is to minimize death and disruption before that time.
Re the preprint from the optimistic IC team.
I did my own quick-and-dirty (five minutes). Scary, and sad. Assume the US case number is meaningless because too few are being identified, and assume a constant mortality rate. US cumulative deaths form as near-perfect a log-lin plot as I’ve ever seen in real data. Perhaps that’s just a feature of the large sample number and the fact that the limited interventions have had no effect and it’s doubling at a constant rate. The doubling time is about three days. From the publications I’ve seen, mean time from infection to death is about two weeks (I think the most recent paper I read said 17 days but it depends on age profile which will be different between countries), so let’s say five doublings. Nothing substantial was done two weeks ago so those infections are already locked in. By my reckoning the US is already committed to about 50,000 deaths. And that’s assuming the ICUs don’t get any more overwhelmed than they are now. The 20,000 prediction should be exceeded next weekend.
Another five minutes. The paper mentioned in the Nature News article I linked to above proposed a 0.5% death rate, allowing for both asymptomatic and undetected cases. Intermediate between the WHO and the Oxford models. Then based on yesterday’s 1700 cumulative deaths, the US had 300,000+ cases a couple of weeks ago. They’d detected just over 2,000, less than 1%. They’ll have getting on for ten million in a couple of weeks unless something changes pronto. Since most are in places like New York which have imposed varying degrees of lockdown, it will hopefully be better than that and their curve will be flattening by then. If not, we’d be looking at more than a million deaths by the end of April. If the 0.5% mortality rate is right, the US should be approaching herd immunity by then and it should be levelling off, and the hotspots should already have herd immunity (assuming that infection does actually confer immunity). OTOH that death toll assumes the death rate doesn’t go up dramatically as ICUs can’t cope.
Actually that took longer than ten minutes because I checked and rechecked my calculations. Somebody please tell me I’ve made a mistake. The Ferguson models make me fear I haven’t, and that it will be a matter of how many millions.
I’m getting more like 2.5-to-2.6 days per doubling for the USA (I do 3, 5, 7 and 9 day centered moving log-linear regressions, sort of like LOWESS). Europe is also getting hammered. There appears to be some upward trending of the doubling times, but nowhere’s near what is needed to stay out of say six figure territories. 😦
James Annan’s latest post also suggests that it’s less than 3 days per doubling.
Right now there is no safe and effective vaccine for any coronavirus that is known to infect human beings. Hopefully there will be for this one, but it is not guaranteed.
There is no hog cholera, a viral disease that started in the United States and spread throughout the world, in many western countries because they elected to eradicate it. Some countries can ship hogs into the United States, like Canada, because they also have eradicated the virus from within their borders. The ones who have not, cannot.
As SM indicated, we may have vastly different rules for traveling between countries in the future because this may not be over for a very long time.
China is indicating they believe their country is approaching a virus-free status. They are at ~3200 today, far down from March 14. Nobody knows more about SARS than they do; herd immunity was not in their playbook:
Only way they’re wrong is if there was a wide spread in their country, but they obviously do not believe that, and the their actions and results on the ground are totally incongruent with wide spread.
Rush Limbaugh just told his audience not to expect more than 20,000 US fatalities .
Because climate models:
I’m not an epidemiologist, but my best guesses are:
The ICUs definitely wouldn’t cope with a largely uncontrolled epidemic: Lombardy was partly controlled, and their hospitals were completely overwhelmed. I think they had ~0.1% excess deaths just due to CV.
In the UK, there are ~4000 ICU beds (although more are hastily being set up). About twice as many need ICU as eventually die, and if 50% are infected, are assuming 0.5% mortality, that’s 300 thousand people needing (but not getting) ICU in the UK.
This kind of maths probably explains the government’s pivot from a herd-immunity strategy to trying to suppress the epidemic.
Pingback: Kontrasting Corona – Symptoms Of The Universe
Hot of the press (includes thru t2020-03-28 JHU dailies) …
These are 9-day moving averages for doubling times. World-CNKR = World-(China+South Korea)
“We’ll see whether China also has to play whack-a-mole. The advantage they have is a very large, very heavy hammer and a surveillance society already in place. Kinda like having your garden already loaded with motion cameras and sound detectors so you can be on the mole when it first peeks above the surface.”
yes every building in China will get a designation of Red green yellow. To get access you need to
swipe the QR code which will log you at the location.
Aids in Contact tracing.
I think Korea is losing the battle. Clusters continue to pop.
Seoul is at 400 cases and still growing.. sub exponential, so its a rather steady number
per day. It is just a matter of time. ticking time bomb. People will let their guard down
Nightclubs in Gangdam, PC rooms, Nora bang ( karaoke) The youth cases are huge here
and its just a matter of time before we get another explosive case.
Singapores approach is much more workable for open society. It’s a very cool Bluetooth based
System. They open sourced this app. urge you governments to consider it.
Korea has something similar, but not as good as the Singapore system
1. you download the app and OPT in.
2. you turn your blue tooth on.
3. As you walk around your phone will Ping other users, normal BT ping.
4. based on Signal strength you
will collect a record of who you were close to ( More on this later cause its cool)
5. If you come down with covid, then you consent to release the data
6. The people you were within 6 feet of you get a text. Show up for testing.
The use of blue tooth is rather cool. Long ago I worked with a covert communication system.
basically it took advantage of the lack of any good atmospheric window for that wavelength.
That meant I could talk to you if you were close, but no one else could overhear.
In addition there were some cool things doable in passive ranging on received signal strength.
basically you work the radar range equation backwards, and infer range from signal strength
and transmission losses etc
So in the blue tooth application since difference blue tooth transceivers have different power characteristics they had to take a bunch of different model phones to the anechoic chamber to
test signal strength. If you didn’t do this anyone with a powerful Bluetooth would light up people outside the 6 foot window. So you get pinged, you look up the model of the phone, you calculate range and record if within 6 feet.
pretty cool. It doesn’t use GPS so it doesn’t TRACK your movement, it just tracks your
“contacts” so as you bounce around town, you will leave bread crumbs with people you passed within 6 feet of.
Thank Steven. The bluetooth thing is indeed cool, especially in societies where everyone has a smartphone. Only loophole is it doesn’t identify surfaces you’ve contaminated for someone else to touch. There’s a video of someone who went back to join his wife in Shanghai which showed some of the measures: CCTV everywhere, face recognition, phone tracking, Big Data AI putting it all together so they can back-trace contacts once a case is identified. We’re all watching to see how the early countries do with suppression of pop-up moles and with getting back to some semblance of normality. Especially the more open societies where Big Brother isn’t looking over your shoulder all the time.
I made the US doubling rate for total deaths closer to 3 days than 2.5. However I was consistently getting 2.5 for the UK when the government was saying 3 or 4. I wondered if they were using the doubling rate for new cases or deaths, not totals. I have a feeling that should be more sensitive to a decline in spread once the numbers get up, because you’re subtracting a baseline and just looking at anomalies, kinda like with temperature stations. Their advisers will want the earliest possible warning that measures are working or not working
Actually, Russell, Limbaugh is playing the ozone hole or Y2K card. We were told it would be bad and went to a lot of trouble to prevent it, and now it’s not been bad, it was a waste of effort preventing it because that shows it was a nothingburger after all.
Is it off-limits to say there are some people for whom the Darwin Awards were invented? Wonder how many are holding Coronavirus Parties?
Oh well, I had to do it for the UK. Doubling time 2.5 days. Around 20,000 deaths committed to as of 23rd March when the lockdown came in. The government sources saying that’s an optimistic outcome were probably told that’s the floor.
I live in a wealthy neighborhood that is mostly staying at home, though, continuous shopping trips.
But the thing they are doing that just drives me nuts is their maids. Some twice a week. On the other days, the maids are presumably cleaning somebody else’s house: Typhoid Maidy.
(Obviously, they should be paying their maids to stay home and they should be cleaning their own houses.)
US-style stay at home is a sieve because the message from our incompetent leader resembles the ramblings of schizophrenic street preacher. I doubt it will work, and it was possible for it to work. He’s horrible at everything.
Steven Mosher –
Thank you for posting this over at WUWT:
How did you come across it?
My level of respect for MIC Lewis dropped significantly with his absolute crap posts at WUWT and Judith’s about death rates from. Coronavirus.
How ’bout you?
I suggest everyone view that video – it presents a nice contrast between how science is performed and what usually takes place at sites like WUWT.
A few more cents on this… If you look at the trajectory curves for COVID-19, there’s a less obvious issue. All the measurements are different, and not exactly comparable. It is what it is, but testing and policy are what is driving the measurements, are anything but standard.
Mosher says South Korea may be losing control. I’m not sure about that, but still their surveillance efforts are based largely on contact tracing, so its going to have issues with pop up clusters. But its clear that its not a problem for them.
More than anything else, you have to realize that all global efforts are to make sure the health care system can handle case load. A vaccine or something may come later, but that’s not guaranteed, and that’s not now.
If you look at countries with flattened infection loads, then you can see that their health care systems would be far from overloaded. The vast majority of infections recover in days. If you look at the US, then you can see that its health care system is about to be whacked really hard. If you look at Japan, its infection rate is stubbornly low so its health care system may in fact just be coping just fine.
Urns. 5,000 urns. I can’t believe it. They ordered 5,000 urns. What on earth would a province of 58.5 million commie liars need 5,000 urns?
I think, thanks to Fauci, they’re getting ready to ship emergency funeral supplies to the United States.
Anyway, if you look at the Chinese provinces that are directly across the water from South Korea and Japan, they typically have far lower numbers than South Korea, and almost all of them have lower numbers than Japan as well.
Seems to me that Nic is doing pretty much the same here as he does with his climate sensitivity work. Very confident about his results. Focuses on one dataset. Makes assumptions that tend to lean in one direction. I notice, for example, that he didn’t seem to correct for false positives. As I understand it the sensitivity and specificity of the test are about the same (~90%). So, the fraction of those who are positive, but test negative, will be about the same as the fraction of those who are negative, but test positive. However, if there are many more who are negative than positive, then the number of false positives will be bigger than the number of false negatives (I hope I’ve got that the right way around). Nic, corrects for the false negatives (by increasing the number who were probably positive) but seems to completely ignore the false positives.
On Nic’s analysis:
1) It would be reasonable to assume that the deaths on Diamond Princess reported to be due to COVID, with consistent timing, were in fact due to COVID. Claiming that many of them were really due to other causes, while not impossible, is a bit of a stretch. i.e., this is ‘making assumptions as favorable as possible for your point of view’.
2) 10 deaths (an extra 2 since Nic’s analysis) is not enough to make strong conclusions on mortality, but the last thing you should do is split this data into even smaller subgroups. The age profile of mortality is anyway well constrained by other data.
3) Diamond Princess passengers are unrepresentative of general community, even accounting for age profile (e.g. because they are likely to be both healthier and wealthier).
4) As ATTP stated, you should account for both false positives+negatives, but this is hard, so mostly just increases error bars, rather than pushing the result in one direction.
5) What you actually get is wide bounds on possible mortality, so presenting only the central estimate is highly misleading. Given random and systematic errors, the range of plausible answers probably covers an order of magnitude.
But even if you believe Nic’s low estimate, this still corresponds to a Lombardy situation everywhere unless you flatten the curve: health services overwhelmed and massive excess death, mostly of the elderly.
Also, the more basic problem is, the contrarians have decided to live in a parallel universe on a whole range of topics. This is not ‘I have good reason to believe science is wrong on a specific topic’: it is ‘I am going to ignore any science I don’t like’.
Nic’s post takes an inherently non-random sample, from an outlier group (even controlling for age people on board a cruise are quite likely SES outliers, and thus not typical in baseline health status, access to healthcare, health behaviors, etc.,), who are exposed to a non-random and non-normal treatment (isolated in cabins for example), and tries to extrapolate as if the sample abs conditions are generalizable.
He then follows up in response to questions by expressing his “opinions,” such as that people on a cruise are likely in lower than average health .status – with absolutely no evidence presented to support his opinions.
It’s skockingkly bad. It isn’t science. I can’t evaluate his analyses of climate science – but his epidemiological analyses is stunning in its arrogance and lack of insight as to the basic science rules it violates.
Steve Mosher pointed out as much with Ioannidis’ article about the cruise data as well. WTF? His work wasn’t quite as bad, but really? Ioannidis?
It does seem as though quite a lot of people who have expertise with research, or data analysis, suddenly think it’s important for them to make a contribution to the discussion about COVID-19. I find this rather unfortunate. I think there is a difference between contributing to promoting the views of actual experts, and expressing one’s own views. The former can help to spread actual information, the latter could lead to the spread of misinformation.
Yes, I also thought that if one was to properly account for uncertainties, the results from an analysis using the Diamond Princess would probably be consistent with the UK estimates.
As here (preprint): Estimating the infection and case fatality ratio for COVID-19 using age-adjusted data from the outbreak on the Diamond Princess cruise ship.
Infection fatality rate 1.2% (0.39% – 2.7%), case (i.e. symptomatic) fatality rate 2.3% (0.75% – 5.3%). About 1% of those infected, and about 2% of detected cases, but with an order of magnitude uncertainty range.
The IC engineering prof was quoted on the radio this morning as now saying his initial effort was an underestimate. As Neil Ferguson said on BBC Radio 4 this morning: “suddenly, everyone’s an epidemiologist”. My turn. The official Chinese fatality rates are about twice as high (even after adjusting from the naive fatality rate), although their asymptomatic ratio is similar at about a half. I guess one reason for that is that they’re comparing age-stratified data from both places. While the average cruise ship passenger is older and less healthy than the population average, the average 75-year-old cruise ship passenger is sure to be wealthier than the average 75-year-old Wuhan resident, so perhaps healthier.
“4) As ATTP stated, you should account for both false positives+negatives, but this is hard, so mostly just increases error bars, rather than pushing the result in one direction.”
However that is still an excellent reason for doing it anyway. Decision making under uncertainty requires proper characterisation of the uncertainties.
WRT assumptions – an advantage of the Bayesian approach is that you state them and quantify them, via prior distributions, and means that if someone doesn’t accept the assumptions, there is a clear way of expressing your dissatisfcation with the conclusion. I wouldn’t attack Nic for that – stating your assumptions is good science, not bad, it facilitates questioning of those assumptions. In frequentist analyses those assumptions are often still present, but not clearly quantified, and often not explicitly stated.
“I think there is a difference between contributing to promoting the views of actual experts, and expressing one’s own views.”
IMHO that is a rather dicey proposition.
So, for example, 2.2 million deaths as a do nothing proposition for the USA in the very face of every other countries doing something. That one, do nothing. would be political suicide IMHO. It is like saying, everyone that goes into that building comes out with a broken arm, I wonder if should I go in there. Further, you have to actually do nothing to see if your model is even in the ballpark. So that, bottom line, we will never know how good the model was in a true do nothing situation.
Now we move on to shelter-in-place solutions, those would appear to work better than the early modeling results suggested, assuming that people actually shelter-in-place properly. Will we ever assume static beds again? I don’t think so. Will we take more drastic actions next time? I sure do hope so. Will we have better stockpiles or manufacturing capabilities next time? I really hope so.
Hindsight is 20/20, so let’s see how those models can be improved given rigorous postmortem efforts (assumes we have good data from all significant (confirmed+deaths) countries.
FWIW I (as a machine learning/statistics) chap have been invited to join in with efforts to work on the data science side of COVID-19. So far I have decided not to because adding lots of unqualified/inexpert people is known to be a often a risky (if not actually bad) management strategy when deadlines are close (c.f. Fred Brooks ““Adding people to a late software project makes it later.” – I suspect from his book “The Mythical Man Month”).
I’ve completed the request for help with modelling. However, I do know some of those on the steering committee and I would only plan to get involved if the modelling effort were lead by people with relevant expertise and they really did seem to need some additional help.
WRT assumptions – an advantage of the Bayesian approach is that you state them and quantify them, via prior distributions,
> I wouldn’t attack Nic for that – stating your assumptions is good science, not bad, it facilitates questioning of those assumptions.
Nic didn’t show how he accounted for his assumptions in the aspects I talked about, nor did he present evidence in support of those assumptions.
Perhaps if he had included a “limitations” section in his article…. he didn’t…
When you do the kind of generalizations that Nic did, you need a random sample that is also representative (in demographics and with respect to the treatment conditions your exploring) so as to support broader generalizations.
It’s hard to do that perfectly, of course, but you should establish and quantify those qualities in your sample and the treatment condition q or discuss the implications of your inability to do so.
Otherwise your conclusions are meaningless. Nic wasn’t merely reporting on associations. He was arguing for conclusions about causal mechanisms.
ATTP, good approach. I have some other reasons why I might not be a good emergency collaborator, so perhaps a special case. I’d still be happy to give my opinion/advice if asked, but I think it is vital that people don’t over-estimate their expertise and end up getting in the way.
“Nic didn’t show how he accounted for his assumptions in the aspects I talked about, nor did he present evidence in support of those assumptions.”
That is not uncommon – but at least he facilitated criticism by stating his assumptions. While it is possible to have an objective (in a technical sense) Bayesian analysis, there is nothing wrong with subjective Bayesianism (provided you are clear that is what it is), in which case your priors can simply encode your opinions.
“When you do the kind of generalizations that Nic did, you need a random sample that is also representative (in demographics and with respect to the treatment conditions your exploring) so as to support broader generalizations.”
I don’t think that is true. I machine learning there is an active research field called “covariate shift” which deals with situations where your calibration sample is not representative of operational conditions. I suspect a lot of it is a rediscovery of existing classical statistcal practices.
The dangers of nonexperts venturing outside their field of expertise:
ATTP – Glad to see Chris Bishop involved, just the sort of person to be in that sort of role.
Ben, LOL, I’ll need to share that around!
“At this point, my partner who works at a hospital was laughing at me,”
> I don’t think that is true. I machine learning there is an active research field called “covariate shift” which deals with situations where your calibration sample is not representative of operational conditions. I suspect a lot of it is a rediscovery of existing classical statistcal practices.
I don’t understand that, nor do I think I’m capable of doing so…. so I’ll consider it an uncertainty.
On Cesspool Etc. I said Johnny Uioannitus just threw 10 picks in the Super Bowl and has destroyed his reputation, and Professor Curry deleted it.
It’s funny that Judith takes all her praise as an un-biased moderator as a point of pride.
BTW – she’s been hoovering up my comments which are not at all aggressive – while leaving online many comments from Don et. al which are science free and insult laden.
Yes, she maintains a safe space for her little snowflake thugs.
I’m sure I’ve said this before, but post- match analysis of Judy’s is very dull, as well as being off topic.
A radical suggestion that if you don’t like a site’s moderation, don’t post there?
You owe me a new keyboard 8^D!
Her site, her moderarion. That’s the way the cookie crumbles.
But her pretense that it’s even-handed is just silly.
I was thinking of making a social distancing alarm. Something simple, like an EM field detector wired to an air horn in your pants.
“Mosher says South Korea may be losing control. I’m not sure about that, but still their surveillance efforts are based largely on contact tracing, so its going to have issues with pop up clusters. But its clear that its not a problem for them.”
our growth in cases in linear. wack a mole. find a cluster quash it. I have concerns that
this is not sustainable.. there’s that word again. People are hyper vigilant. that can’t last.
“Steve Mosher pointed out as much with Ioannidis’ article about the cruise data as well. WTF? His work wasn’t quite as bad, but really? Ioannidis?”
I have not been back to comment, but anyone who uses the diamond princess is wacked.
for patients over 80 they had like 1 death out of 51. CFR rate in Korea is ~18% for
that age range. If anything it’s likely the cruise ship had no people with co morbidities.
anyway,as in climate studies skeptics will always find a small dataset to exploit.
Then they will argue that this small dataset is somehow perfect
Reblogged this on In the Dark and commented:
One of the things I’ve written about on this blog quite frequently is how important the treatment of uncertainty is in science, both in the application of the scientific method itself and in the communication of results to a wider audience. This blog post makes a similar point about the presentation of results from modelling the spread of Covid-19.
In a similar vein, I don’t quite understand the comment “The key parameter is \rho, which is the proportion of the population that is at risk of severe disease, a fraction of whom will die (14%).” How certain is that last figure? What is the uncertainty in a given region? As I understand it, the fraction of infected people who dies from the virus varies significantly quite a bit from region to region and from country to country
Their prior on that was 14% +- 0.7%. So, a pretty narrow range. You’re quite right that it would probably vary wildly. In some sense, I guess it’s just some fraction of those who require hospitalisation and they’re considering various fractions that require hospitalisation. However, they are implying that you can use the date between the first infection and the first death to infer what fraction require hospitalisation and, hence, how many have already been infected. I haven’t had a chance to check, but this may then depend on their assumption about how many of those who require hospitalisation then die.
cormac, Anders: Actual epidemiologists are saying we don’t really know, other than ‘really bad’. There are just so many factors going into infection rates and survival outcomes.
I update my estimation of total number of Corona virus cases, including untested cases.
improving the model using different age structure and using new information about the number of tests.
The range given is the range of model estimates, the actual uncertainty may be higher.
The expected death rate is about 2% in European countries with a large number of old people, and about 1% in countries with less old people, assuming the age structure of the infected is the same as the total population. Both has an very large uncertainty. Under additional assumption, death rates above 3.6% or below 0.4% seams inconsistent with the observations so far.
If the percentage of infected older people is much lower than younger people the death rate would be lower.
Case estimates are for 2020-04-04 00:00 GMT.
status: fast growing (more than 20%/day), growing(10-20%/day), slowed down(0-10%/day), already peaked(new cases reduced)
Country, case est., status
USA, 2.1M-7.7M, growing
France, 870k-10.3M, fast growing
Spain, 740k-1.7M, slowed down
UK, 470k-1.4M, growing
Italy, 470k-1.2M, already peaked? (21.-26. Mar)
Iran, 310k-1.4M?, slowed down?
Germany, 310k-690k, slowed down
Turkey, 210k-560k, growing
Belgium, 130k-590k, slowed down?
Brazil, 110k-570k, growing
Netherlands, 100k-330k, slowed down
China, 170k-230k, already peaked (4.-13. Feb)
Philippines, 37k-410k, growing
India, 41k-270k, fast growing
Sweden, 57k-170k, growing
Canada, 79k-110k, growing
Switzerland, 65k-110k, slowed down
Portugal, 53k-120k, slowed down?
Algeria, 26k-210k, fast growing
Indonesia, 22k-210k, slowed down?
Ireland, 33k-93k, growing
Egypt, 13k-230k, growing
Romania, 30k-86k, growing
Israel, 34k-70k, growing
Poland, 26k-87k, growing
Ecuador, 20k-110k, slowed down
Saudi Arabia, 13k-150k, growing
Pakistan, 16k-110k, growing
Denmark, 26k-61k, growing
Peru, 19k-73k, fast growing
Serbia, 17k-74k, fast growing
Japan, 15k-77k, growing
Russia, 31k, fast growing
Mexico, 11k-81k, growing
Dominican Rep, 9k-87k, slowed down?
Austria, 23k-36k, already peaked? (26. Mar)
South Korea, 10k-12k, already peaked (3. Mar)
Australia, 7k-8k, already peaked (22.-29. Mar)
WORLD, 7.7M-35M, partly growing/partly slowed down?
Update for the update.
Case estimates are for 2020-04-05 00:00 GMT.
Country, case est, status
USA, 2.5M-7.6M, growing
+New York, 900k-2.5M, growing
+other USA, 1.6M-5.1M, growing
France, 1.0M-3.8M, growing
Spain, 790k-2.4M, slowed down
UK, 510k-1.7M, growing
Italy, 540k-1.5M, slowed down
Iran, 320k-1.2M?, slowed down?
Turkey, 250k-830k, growing
Germany, 310k-580k, slowed down
Brazil, 140k-800k, growing
Belgium, 150k-480k, slowed down?
China, 200k-250k, already peaked (4.-13. Feb)
Netherlands, 110k-330k, slowed down
Canada, 98k-190k, growing
India, 57k-290k, fast growing
Switzerland, 74k-160k, slowed down
Sweden, 48k-110k, slowed down
Portugal, 49k-95k, slowed down
Romania, 33k-100k, growing
Saudi Arabia, 15k-190k, growing
Ireland, 34k-81k, slowed down?
Algeria, 16k-150k, growing
Indonesia, 16k-150k, slowed down
Mexico, 17k-110k, growing
Denmark, 29k-62k, growing
Peru, 20k-85k, growing
Morocco, 15k-110k, growing
Russia, 36k-44k, fast growing
Israel, 30k-45k, slowed down
Chile, 22k-61k, growing
Ecuador, 16k-73k, slowed down
Moldova, 11k-100k, fast growing
Japan, 15k-58k, slowed down
Poland, 18k-46k, slowed down?
Iraq, 7k-120k, slowed down?
Phillipines, 8k-96k, slowed down?
Austria, 20k-36k, already peaked? (26. Mar)
Colombia, 14k-53k, growing
Ukraine, 10k-74k, growing
Pakistan, 10k-60k, slowed down?
Czechia, 17k-34k, slowed down?
Egypt, 12k-48k, growing
Panama,12k-47k, slowed down?
Dominican Rep, 9k-56k, slowed down
Malaysia, 13k-38k, slowed down
Serbia, 10k-48k, growing
Finland, 15k-32k, growing
UAE, 20k-23k, fast growing
South Korea, 10k-16k, already peaked (3. Mar)
Australia, 6k-7k, already peaked (22.-29. Mar)
WORLD, 8.4M-30M, slowed down?
Pingback: Science in the Time of COVID-19 | …and Then There's Physics