This post is by Phil Price, not Andrew.
Frequent readers of this blog will already know about the wager between me and a commenter called Anoneuoid. Would the number of new COVID cases in the U.S. in the seven days ending 10/5 be lower than 500,000? If yes, I pay him $34. If no, he pays me $100. The number of new cases in those seven days was around 590K, so he owes me $100.
Anoneuoid, please send the money to Andrew (in small, unmarked bills); Andrew, please post your preferred address in the comments. Please use the money to buy donuts for the Stan team or for some other Stan-supporting activity that isn’t too much hassle for you.
Anon:
You can send it to me at Department of Statistics, Columbia University, New York 10027.
Is the full address on this page good?
https://www.stat.columbia.edu/~gelman/
yup
Anon:
Thanks. I received the $100 and, following Phil’s instructions, will use it to provide calories for valuable members of the Stan team.
Wow, what a wager.
Anoneuoid thanks for creating the dialogue.
How do we know that Anoneuoid is a “he”? Not sure about the gender breakdown of blog readers…
Wagering opportunity! I’ll offer 4:1 odds. If Anoneuoid is a male you pay me $400, otherwise you pay me $100. You win if Anon identifies as trans or non-binary or whatever the language.
Anon please don’t spoil it until we see if we have a bet.
Phil –
> If Anoneuoid is a male you pay me $400, otherwise you pay me $100
Heads I win and tails you lose?
Anyway, i’ll offer way better odds than that. I’ll offer 20-1 odds. Although I don’t have as much money to throw around. If anoneuoid is non-male I’ll pay $100. If they is male I’ll contribute the $5 to the donut fund.
Based on politics I suspect Anoneuoid isn’t non-binary (I hope that’s the right term).
“Whatever the language”
Cool. Glad to know inclusivity is valued.
Curiously, how close was he? That is, were the number of average weekly cases dropping (fast)? Would he have won, if the bet was made for the middle of October?
Also, don’t get donuts. There are so many good pastry shops in NYC. Go to a French bakery and pick up some fresh pastries. (If you haven’t lived outside of NYC in some time, you may not appreciate how wonderful it is to do this.)
Maybe I could pay someone $100 in cold cash to finish the damn Autograph project that’s burned through two or three students by now.
What is the autograph project
Not super far off. He was closer than the CDC projections were at the time the bet was made by quite a bit. If you go back to that previous post it looks like the point estimate for the CDC was around 1.1M cases and Anoneuoid said below 500k, actual about 590k
So listening to a random “crank” on the internet gets you within about 20% of the answer whereas the CDC is off by a factor of about 2
Daniel:
Anon’s no random crank! Anon’s the person who pointed out the problem with those un-age-adjusted death rates from a few years ago (see here). By some principle of transitivity, Anon deserves a Nobel prize.
But, to be clear, Anon did not guess 500K. Anon’s statement was, “it is a near certainty cases will be below 500k per week on Oct 1st.” So 500K was at the upper bound of Anon’s “near certainty” interval. I don’t know what Anon’s point prediction was. But if it was 250K, say, then Anon was off by a larger factor than the CDC.
Phil’s point prediction from that earlier post was 650K, which is closer to what actually happened than either the CDC or Anon. So maybe Phil’s the one who deserves the Nobel prize . . . but I guess $100 will have to do!
Andrew –
> So 500K was at the upper bound of Anon’s “near certainty” interval.
I said something quite similar to Anoneuoid here:
https://statmodeling.stat.columbia.edu/2021/09/16/wanna-bet-a-covid-19-example/#comment-2025148
And I also noted the slant in how he compared his number – as if it were a median number which it clearly wasn’t – to the CDC’s median number. He should have been comparing it to the lowest end of the CDC’s range.
Further –
I don’t know how Phil got 590, but at Worldometers the count for those dates is around 720,000 – and that number will go up as lags in counting in many states, but Florida in particular, are closed up over the coming weeks.
Hey what do you know, Anon was wrong. I am truly shocked.
I used the New York Times daily “new case” counts to get 590.
Honestly I don’t see how Anon could have come up with a point estimate far below 500k, as is implied by 500k being near the top of the uncertainty interval. Just getting below 500 would have required a faster decline from the peak than we had seen for any previous peak, and getting to something like 300k would have been way beyond previous experience. I can see how that could happen — maybe the companies that make tests would run into supply shortages or something, to give one example — but to be so sure it would happen seemed kinda nuts. I guess I’m just repeating myself.
Anoneuoid –
> Honestly I don’t see how Anon could have come up with a point estimate far below 500k
I think they seems to have spelled that out.
They thinks that the CDC (and by extension *all of the* the modelers contributing to their ensemble) completely “ignored” any variation due to “seasonality.”
So perhaps they thinks that “seasonality” had a particularly large influence during this period compared to other drops (I can think of some ways to make such a theoretical postulation – such as unusually dramatic seasonal changes during this period in the areas with the highest case rates, such as Florida – although I don’t know if any of them would actually apply).
I don’t think that’s nuts, just based on sone questionable mind-reading combined with a likely over-estimation of the influence of “seasonality.” Although according to Anoneuoid, apparently they was very far off even though they actually *underestimated* the impact of “seasonality,” as described here:
> As you can see, there is an exceptional correlation between the average temperature and change in covid cases. I did not expect it would be this good,
https://statmodeling.stat.columbia.edu/2021/09/16/wanna-bet-a-covid-19-example/#comment-2025131
hmmmm.
Sorry – that was for Phil…
I really just thought cases would drop by half, which would be the same rate as they rose through august. So you can imagine a very narrow uncertainty interval around 500k.
I started looking into what happened a bit here. For cases, I used the covidData R package which in turn uses JHU data. It seems to differ from the CDC data by undercounting during the rises and overcounting during the drops by up to ~10%, probably due to different backfilling strategies (eg, weekend counts are very low in the JHU data):
https://statmodeling.stat.columbia.edu/2021/09/16/wanna-bet-a-covid-19-example/#comment-2025131
The change in cases was clearly correlated with temperature. Eg, by avg Sept-2020 temperature (F) this is the fraction of states that gained cases during Sept-2021:
40-50: 1/1 (100%)
50-60: 10/12 (83%)
60-70: 6/25 (24%)
70-80: 0/10 (0%)
80-90: 0/2 (0%)*
* Assuming Hawaii was over 80 F, which is what I saw in other data. For some reason Hawaii is left out of the NOAA temperature data I used.
You can see the charts in my linked post for the magnitude of these changes. Magnitude-wise my prediction worked for the south, but cases in colder states started rising while the intermediate temperature states tended to only drop at about half my expected rate.
It is possible that in a couple weeks we will still see total cases below 500k, but after seeing the colder states start rising already I suspect this wave lasted too late in the year for that.
Anoneuoid –
> So you can imagine a very narrow uncertainty interval around 500k.
As they say, hindsight is 20/20. You can call a very narrow uncertainty interval after the fact but it’d have been much more convincing had you stated one ahead of time.
But in a way it doesn’t really help you out much since as you narrow the uncertainty interval you’re effectively increasing your certainty about an estimate that was pretty far off (I’m thinking the @720k+ number will hold up).
> Magnitude-wise my prediction worked for the south, but cases in colder states started rising while the intermediate temperature states tended to only drop at about half my expected rate.
So in other words you overestimated the impact of “seasonality.” It’s like you want to have your cake and eat it too. Getting the impact of seasonality in only one set of states suggests to me you’re missing confounding variable or a mediator/moderator.
“A very narrow uncertainty window around 500K” wouldn’t do it, since that would imply about a 50% chance of being over 500K, which you were almost certain wouldn’t happen. Maybe you meant “a very narrow uncertainty window around 400K”? But why would the window be so narrow?
Outdoor temperature surely has an effect but…well, just look at the plots of cases or hospitalizations or deaths: last year the summertime nationwide peak was in mid-July, this year it was in early September. And the peaks in individual states were at different times from each other, and differed from year to year. The picture really isn’t consistent with seasonality being the only thing that matters — not at all; and even if it were, the confident belief that cases would fall off faster than they increased — the opposite behavior of the previous four local maxima in case counts — seems hard to justify.
The outcome of one bet doesn’t prove anything but in this case it’s not just the outcome that I find fault with, I think the reasoning was faulty. That’s not a given just because you lost the bet: If you offered 100:1 odds that the next set of winning Powerball numbers wouldn’t include 5, 6, and 7, but you end up losing that bet, it’s not because your reasoning was wrong but because you got unlucky. 1-in-1000 events happen sometimes; if they didn’t then they’d be 0-in-1000 events. But I don’t think this is one of those cases. I think you could know before the wager that a number below 500,000 was not ‘almost certain.’
I guess what it comes down to is: would you want to bet on the number of reported cases during the winter wave?
I’ll bet on anything, given the right odds!
My uncertainty in the number of winter COVID cases (or hospitalizations, or deaths) is enormous. How many more of the unvaccinated will get vaccinated? Will another variant come up that is even worse than Delta? To what extent does vaccine effectiveness wane with time? How many people will get booster shots? How many areas that have been hit hard are approaching herd immunity the old-fashioned way? (Indeed, what fraction of Americans have thus far been infected with COVID)? What will happen with testing rates, to what extent (if any) will they decrease as employers require employees to be vaccinated rather than requiring testing?
I wonder if there’s any literature (or an obvious answer) about how to come up with a fair bet based on competing distributions? Ideally I would draw my statistical distribution (for, say, number of reported cases from Nov 1 through Feb 28) and you would do the same, and we would apply a rule to come up with a bet or a family of bets based on the differences between our distributions.
A bet on hospitalizations or deaths would also be possible and would take some of the free parameters out of it. Although then one of us would be in the macabre situation of “hoping” for more deaths. That doesn’t bother me but I note it anyway.
Phil –
Don’t forget to add consideration of distribution of new therapeutics – as thee seems to be one that’s very effective in the pipeline. Obviously, thar would affect hospitalizations and deaths.
Also, it looks like rapid testing may finally be more and less expensively accessible which could significantly affect spread.
I’d be more concerned with the reports from the UK and India of many people with covid symptoms testing negative on PCR but positive for antigen. This was also reported in France last year.
Anyway, I would bet on the reported cases going back over 1 million/week by the end of next Feb. We should also be more clear about the data source since JHU, CDC, NYT, and worldometers all seem to differ by substantial amounts.
Phil –
I don’t know if you saw this paper, a link to which I posted on the other thread – but I think it lays out a pretty interesting analysis of how “air drying capaxity” and UV interact with temperature and humidity to describe help wexomain the influence of seasonality.
Of course I can’t evaluate the math or the physics or the related virology – but what I like is that (unlike, it seems to me, Anoneuoid) they’re careful to place potential effects of seasonality as an environmental factor within a network of multiple causal influences (in fact referencing socio-economic influences as being “dominant”).
https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2021GH000413
Assuming any correlation to state-to-state variation in temperature, maps on to a predictive relationship of *temporal* variation in temperature, is a pretty huge error. It’s like saying that because white people average higher income in the US, you’d expect people to lose money if they get a tan.
Zhou –
> Assuming any correlation to state-to-state variation in temperature, maps on to a predictive relationship of *temporal* variation in temperature, is a pretty huge error.
For me – considering the climatic and environmental heterogeneity in a state like California, and the heterogeneity across states with similar climates on factors like population density – I tend to agree.
But since you seem like a smart fella, I’m curious to know more about your reasoning. There surely does seem to be some predictable elements related to “seasonality” at the regional and climate zone level…
For me having done a bunch of ecology, it basically comes to “phenotypic plasticity”. Even if the environmental conditions are the driving factor (and it’s not a spurious correlating driven by e.g. conservatives being bad at anti-covid measures), people living at different lattitudes are adapted to different environments, and thus have different and varied responses to changes in climatological conditions. A texan exposed to weather that is more like new york does not turn into a new yorker. They turn into “a texan in cold weather”, and in terms of how a texan in cold weather acts with respect to covid19 spread, we have only 1 replicate in our dataset.
Please excuse a possibly dumb question (I’m new around here) but why do case counts matter to people so much? Is there a strong correlation with severe illness and/or death and hence it’s a useful predictor? If so, have people been able to make robust predictions about what’s like to arrive (I’ve not seen that made easily available to the public)?
Case counts don’t matter much to “people around here” but if you follow the link for the history of the bet, including the link before that one, you’ll find that this all started because of the discussion of a particular CDC projection that happened to involve “cases.”
Given the amount of testing that is occurring these days the gap between infections and cases is not as wide as it once was, although these are still far from the same thing. At any rate many people’s activities, as well as government policies, seem to be partly governed by what is happening with the case counts, so it makes sense to cate about them. Plus there’s the principle of “if you can’t be with the one you love, love the one you’re with”: what we’d really like to know is infections, or at any rate that’s one of the things we’d like to know, but since we don’t have that we look at cases instead.
eddie –
> Is there a strong correlation with severe illness and/or death…
I for sure ain’t no epidemiologist but from my observations, in the whole hospitalizations, ICU admissions, and deaths have tracked pretty well with numbers of positive tests all over the globe throughout the pandemic.
Of course there are important things to take into consideration, such as testing rates and record keeping, and the ratio of cases to severe illness and death has obviously altered along with the demographics of who’s being tested and with increased prevalence of vaccination. But with all of the noise there’s been about a “casedemic” and “false positives” (people who are pre- or post-infectious who have tested positive), the basic linkage between cases and morbidity and mortality has sustained.
There are time series plots of new cases, hospitalizations, and deaths at https://www.nytimes.com/interactive/2021/us/covid-cases.html They do indeed track quite well…maybe better than I would have expected, given how many at-risk people have been vaccinated now compared to a year ago. You can easily see that the first six months of the pandemic didn’t have as much testing — the number of deaths per case was much higher — but you can use any of these to track the pandemic at the timescale of a month or so. At shorter timescales, cases is a good leading indicator of deaths. So it’s useful in spite of its deficiencies.
I’m a little confused–I thought the final weekly tally was supposed to be based on CDC estimates. It looks like these estimates are closer to 700k cases for the week. This is because the daily average around this time is about 100,000. It would be good if the “final” value were formally provided in the original post, for posterity.
No, the bet is based on the actual number of cases reported during that week. The bet was motivated by Anoneuoid’s mockery of the CDC’s forecasts, but was not dependent on their forecasts.
Phil, I agree that actual number of cases reported during the week were supposed to be used. I thought the estimates were supposed to be obtained from the CDC website.
Anon: “I’d say use the data shown on the cdc forecasting site. The only problem with your bet proposal is theres gotta be people who will give better odds. I mean, what do you think the odds are?”
https://statmodeling.stat.columbia.edu/2021/09/02/there-are-no-equal-opportunity-infectors-epidemiological-modelers-must-rethink-our-approach-to-inequality-in-infection-risk/#comment-2022582
Then your response was: “Could you post a link to the cdc page that shows the number of new cases by day? I’m happy to use such a page.”
Then Anon’s response was a link to: https://www.cdc.gov/coronavirus/2019-ncov/science/forecasting/forecasts-cases.html
Clearly, daily estimates from CDC were supposed to be used. I thought this would have been taken from the average daily cases reported from the 7-day moving average that CDC reports on some of their other pages. As Joshua has written, it looks like the value is closer to 700k than 590k.
No, the original bet was supposed to be based off the CDC numbers. On September 3rd, Anon wrote:
“I’d say use the data shown on the cdc forecasting site.”
Then Phil replied, “Could you post a link to the cdc page that shows the number of new cases by day? I’m happy to use such a page.”
Then Anon supplied a link to the CDC forecasting page.
Then Phil replied that they went to the link but didn’t see daily totals.
Based on the left-most graph from the original link, looking at the solid black points (reported outcomes rather than forecasts) it appears there were about 700,000 cases for the week of Oct 1.
From there, I presume other links would have been followed to calculate daily cases for the week of interest.
For what it’s worth, the CDC daily cases are listed here.
https://covid.cdc.gov/covid-data-tracker/#trends_dailycases
Scroll down and expand the datatable.
Indeed, there were 686,000 cases for the week ending October 5th.
Unanon –
> Indeed, there were 686,000 cases for the week ending October 5th.
And that number will likely increase as the lag in the recording of deaths gets filled in over time.
Total as of now, per the CDC, is up to 725,398.
:
Interestingly, while Worldometers has a roughly similar total of 722,911, the distribution is very different:
For example:
At the CDC website: (
9/29) 108,827;
(9/20) j107,255
At Worldometers: (
9/29) 125,172; (
9/30) 123,189
NYTi has yet another number (doesn’t get any better for Anoneuoid, however): 753,958
@ Phil, or anyone else
Are you interested in betting on whether cases rise back over 1 million per week by the end of Feb 2022?
US cases were that high last winter, but back down by the end of February 2021. If they run a wide booster campaign in the coming winter, that may repeat. Deaths ought to be a lot fewer than last year, though.
(I’m not interested in betting.)
P.S.: Has Andrew ever publicly confirmed the receipt of the wager?
Yes, it was confirmed at the top of this thread.
Also, the first dose of Pfizer vaccine was reported to cause lymphocytopenia for about 1 week in healthy adults during the phase II trial. IIRC, it was also more common in the younger age group. This type data was never published again, but it seems likely it would also do that in children.
If you immunosuppress millions of children and send them to school I would expect that leads to even more cases this winter. Not just of covid, but RSV, flu, etc.
I understand lymphocytopenia happens temporarily as a result of an infection (such as catching a cold) as the body uses up lymphocytes to fight it; it’s not surprising that this would happen after a vaccination, as this is intended to simulate an infection. If you don’t keep your kids at home when they have a cold, you probably shouldn’t worry.
I wouldn’t label this an immunosuppression because the immune system remains fully functional. If someone had lymphocytopenia without an infection (or vaccination), I’d be worried.
It may not be surprising but if your immune cells are leaving the blood because they are busy with one thing, what do you expect to happen?
I would expect more infections along with whatever else they would otherwise be preventing.
Indeed, that is what we’ve seen for covid. More infections in the first week after the first dose than otherwise.
You are repeating your point without addressing mine.
My points were:
1) “more infections” equates to the same kind of risk that a kid would have if suffering from a common cold, which isn’t going to concern many parents. From someone countering this point, I’d like to see “more” quantified and compared to the number of infections kids are expected to undergo each winter anyway. “More” doesn’t mean much if it’s just a little more, and not very dangerous.
2) You can protect your kids by keeping them at home for the week. Super easy to do if the school vaccinates the whole class at once and puts it on a distance learning schedule for the week after. Obviously that would only happen if the “more” (see point 1) is actually scary.
For healthy adults it was about 25% had lymphocytopenia according to the standard threshold. Then there was ~1.5x increased chance of covid in that first week. The latter is “real world” data so who really knows, but that is what we have.
Yes, you could. If they were made aware of this risk, any parent that wants their child to wear a mask would probably do this.
Either way, that is a reason to expect extra covid cases this winter.
> ~1.5x increased chance of covid in that first week.
Which amounts to 4 extra days of exposure, easily balanced out by the reduced chance of Covid you have for much longer once you’re fully vaccinated.
Not scary.
But thank you for the numbers.
This turned out to be a myth mainly spread by journalists and politicians. It was backed by a
batch of crappy studies that all ignored different testing rates between vaccinated vs not, compared unvaccinated rates in the winter to vaccinated in the spring, etc.
Pfizer et al certainly never claimed their product was stopping infection. It was known beforehand that was highly unlikely.
> This turned out to be a myth mainly spread by journalists and politicians.
And the state of Washington DOH data every month recently.
https://www.doh.wa.gov/Portals/1/Documents/1600/coronavirus/data-tables/421-010-CasesInNotFullyVaccinated.pdf is the latest.
Even if you think they’re missing infections at a rate of 1:5, hospitalisations being down 1:14 is good enough for me.
You should also account for an ~3x “healthy vaccinee bias”.
https://www.cdc.gov/mmwr/volumes/70/wr/mm7043e2.htm
If we assume vaccinated are ~5x less likely to get tested and ~3x healthier, that could alone explain a 15x difference in hospitalizations.
I tend to think the vaccines should be somewhat protective against severe outcomes, so that probably overstates the total bias. But you can see it is no issue to get apparent 90+% protection from biased data alone.
FYI, this is the current CDC guideline on testing:
https://www.cdc.gov/coronavirus/2019-ncov/hcp/testing-overview.html
I have no idea how they come up with this stuff or how much it is actually followed, but it used to be not to test vaccinated at all. So a 5x difference in cases due to testing is quite plausible. I’d say it is at least 2x, even in hospital settings where screening is more likely.
Anoneuoid –
> I have no idea […] how much it is actually followed…
But don’t let that get in the way of you reverse engineering as if you do know, to explain statistics in a way to support favored conclusions.
I agree. That is why I don’t try to tell people what they should be doing based off this quality of evidence.
The best available evidence is probably all cause mortality from the RCTs.
Anoneuoid –
> I agree. That is why I don’t try to tell people what they should be doing based off this quality of evidence.
+1
> The best available evidence is probably all cause mortality from the RCTs.
I don’t like the whole “best available evidence” framing. Seems subjective to me and vulnerable to bias. You take all the evidence you have and put it together as an integrated whole. The individual lines of evidence aren’t in competition with each other.
How is death for any reason in an RCT subjective and vulnerable to bias?
Anoneuoid –
> How is death for any reason in an RCT subjective and vulnerable to bias
I would try to explain but that’s so obviously far from what I was saying I’m not going to even bother trying to clarify.
Try again. Put on your thinking cap and see if you can rephrase my point in a way I would agee captures it well. If you can’t then just say you don’t understand what I was saying. Your bad faith straw man doesn’t deserve a more elaborated response.
https://rationalwiki.org/wiki/Rapoport%27s_Rules
> If we assume vaccinated are ~5x less likely to get tested and ~3x healthier, that could alone explain a 15x difference in hospitalizations.
There’s a logical error here. Someone who has trouble breathing isn’t going to stay out of the hospital just because they didn’t get tested. And once they show up, if they haven’t been tested for Covid yet, they will be, pretty much immediately. So any difference in the testing rate won’t affect the hospitalization rate much, at least not in the younger age group.
This means that you can’t multiply 5×3, it’s just your claimed 3× bias against the WA DoH’s 14× difference in hospitalisation.
I have to admit I also don’t understand the MMWR paper you linked. First, it says that people selected for vaccination contributed to “unvaccinated” time pre-injection, so the groups aren’t really separate. Secondly, the study did not include any anti-vexxers: “eligible unvaccinated persons were selected from among those who received ≥1 dose of influenza vaccine in the last 2 years”. Thirdly, both the number of people in the vaccinated and unvaccinated groups were roughly similar, with similar demographics, and the numbers of deaths as well (6300:6700, 5:7 for kids), so I don’t really get where the 3× SMR difference comes from.
I found a 2015 study on bias in influenza studies, Frequency and impact of confounding by indication and healthy vaccinee bias in observational studies assessing influenza vaccine effectiveness: A systematic review, and they found people with co-morbidities were more likely to get vaccinated than healthy people in most studies. I’d expect the same to happen with Covid vaccinations, since co-morbidities have a well-communicated effect on the severity of the symptoms, and the vaccination helps here.
I’m still positive that the benefits of the vaccination outweigh the drawbacks of immune system being busy for a week.
It is because they compared the rate in person-years rather than per person.
From the footnote:
Anoneuoid wrote:
> It is because they compared the rate in person-years rather than per person.
Yes, I saw that. I understand the mechanism you quoted to increase the number of person-years in the control arm and decrease the number of person-years in the vaccination arm. Since that’s the denominator of the rate, it should’ve made the controls’ death rate lower, with the numerators (#deaths) being similar, but the SMRs given are opposite. It makes no sense to me.
Ah, I see what you mean. There must be about 3x more person-time in the vaccinated group then?
Then from table 2 we see there were 4.4x more deaths after 2 doses than 1 dose for pfizer and 3.7x more for moderna. Rate per person-yr was similar after 1 vs 2 doses. Also, pfizer was supposed to have a 3 week interval and moderna 4 weeks.
The avg person years post-dose #2 should then be ~4.4*3 = 13.2 weeks for pfizer, and 3.7*4= 14.8 weeks for moderna.
Round this to 14 weeks or 3.5 months. The middle of December (the Pfizer EUA was about Dec-15) to the end of May is 5.5 months.
Then we can guess the average date of second dose was about Feb-15th and first dose then a month earlier… which gets us to about Jan-15.
So on average there was ~1 month pre-vaccination and ~3.5 months post-vaccination. That is about the 3x mentioned above.
If this comparison is primarily between pre-vaccination and post-vaccination (very few actual unvaccinated) then it could make sense. Please check the math/logic though.
Also, I see that for 85+ year olds in 2019, there were ~80k deaths in Jan vs ~70k in May. For 2021, it was ~110k vs 67k.
For 75-84 year olds the 2019 mortality was 62k in Jan vs 57k in May.
https://data.cdc.gov/NCHS/AH-Monthly-Provisional-Counts-of-Deaths-by-Age-Gro/ezfr-g6hf/data
So comparing non-covid Dec-Jan mortality to Feb-May mortality (as they appear to do there) should only be different by somewhat less than 10%.
Here is another reason to expect more cases. That Q498R mutation has finally showed up:
https://github.com/cov-lineages/pango-designation/issues/343
This variant looks pretty bad in theory.
1) Immune escape
The N-terminal domain (NTD) is basically mutated beyond recognition, so the antibodies directed towards that will be ineffective. This includes an insert that apparently came from the human TMEM245 gene. Looking it up, I see this (phosphomimetic) EPE peptide has been used to interfere with ERK signalling (blocks transport into the nucleus). So there could possibly be an immunosuppressive effect or even lead to the spike protein getting into the nucleus.
2) Increased affinity for ACE2
The theoretical affinity ceiling (lower values indicate higher affinity) for antibodies is ~100 pM: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC42497/
The affinity of the original receptor binding domain (RBD) for ACE2 was 1,700-17,000 pM, but if mutated to E484K + Q498R + N501Y this dropped to 55-170 pM. With a few further mutations it can drop to 3.5-16 pM affinity: https://www.nature.com/articles/s41564-021-00954-4
This variant has E484A + Q498R + N501Y mutations. The uncharged alanine rather than positively charged lysine at position 484 will hopefully reduce the affinity for ACE2 since the binding site is negatively charged. The only way to tell is actually do the measurement, but if it drops below ~100 pM that could be a real problem for adaptive immunity.
3) Increased furin cleavage
While it does not have the delta variant’s P681R mutation in the furin cleavage site, it does have the alpha variant’s P681H mutation along with a nearby N679K mutation. This looks like it offers some advantage related to cleavage, but, like that E484A mutation, the only way to really tell is run the experiments.
It is always hard to be certain how a bunch of mutations interact with each other, but we have every reason to expect this one to outcompete delta. It could be theoretically worse (E484K/P681R rather than E484A/P6814H), but not by much and that is only two mutations away.
After reading the discussion on this variant so far I bet this is the risk that goes ignored. It could be a big deal. When the virus infects a cell, first the spike protein binds to ACE2 and gets stuck in the open conformation. That then reveals a couple cleavage sites. Once revealed, the S1 region (that includes this EBE motif) gets cleaved off. The now exposed S2 region (still attached to the virus) then inserts itself into the membrane to allow actual infection of the cell.
But what is happening to that cleaved off S1 peptide? If gets into the cytoplasm of nearby immune cells and messes with ERK signalling and/or gets translocated into their nucleus (doing who knows what) that could be a problem.
I hope they at least do some in vitro studies on this using relevant cell lines before releasing the variant vaccines.
And now we are back over 1 million cases/week.
Would anyone want to bet on when cases back below 500k/week?
Have you looked at the cases in Florida?
Must be that wicked polar vortex they’ve had the past couple of days.
I’m not in Florida, but after months of near perfect weather it did get cold enough to wear a jacket (mostly at night) in the last week or so. Unsurprisingly, cases started rising again.
Maybe if you check Florida weather (not avg, but minimum) you will discover the same?
Check here for historical Florida heating degree days. You can see this is the time they typically start closing the windows and turning on the heat.
https://www.ncdc.noaa.gov/cag/statewide/time-series/8/hdd/all/11/2010-2021
The rate of cases has skyrocketed. Has the tend in degree days skyrocketed or was the trend gradual?
Last year cases dropped or were flat from mid-August to mid-October, then rose sharply from mid-October to early January. Then dropped like a stone (continuing until early March 2021).
This year they were flat or dropped from mid-May to early-July, then shot up until late August, then dropped like a stone until mid-October, then were flat until early December and have shot up since.
Sure, there’s a broad seasonal effect. But your lockstep correlation with degree days is clearly inconsistent with the year over year trend in Florida, as well as in many other states.
Arizona last year, early November to mid-December a significant increase. This year something of a drop during the same period.
There are myriad examples.
This is why I hesitate to respond to you. Please look at the chart I shared? I won’t be responding again for a bit though.
I wrote:
So far, I’m on track with that prediction, even without any temperature charts. ;-)
Dec 1, 2020-Feb 28, 2021 had ~230k Covid deaths, based on the Wikipedia timeline. Given that we’ve seen ~35k Covid deaths in the US in December so far, I anticipate Covid deaths for this winter (Dec 1 – Feb 28) to run to 135-150k. Two months ago, I had expected less than that because of vaccinations; right now, I’m hoping that the omicron mortality rate is low enough to push fatalities below 50% of last year’s death toll.
130k Covid deaths so far, by Feb 28 that’s going to exceed my 150k prediction of Dec 29, but still be “a lot fewer than last year”.
US cases are still high, but have been falling off sharply for a good two weeks, so “back down by the end of February” will mean ~half a million new cases weekly.
It looks like my Nov 19 prediction will turn out to be spot on, despite being based only on how last year went, plus vaccines.
Joshua got me on the lag in death reporting before. There are also obviously many things changing. But I hope you are correct.
So much for “cases rise back over 1 million per week by the end of Feb 2022”, they were under half a million again by then, as I thought.
Deaths (as per the Wikipedia numbers) were around 160k from 12/1 to 3/1, 30% down from 230k the year before.
I won’t be surprised if that repeats next winter, although I don’t expect the death toll to shrink as much the next time.
Cases rose to over 1 million per day between Oct 2021 and Feb 2022, what in the world are you talking about?
And yes, that is the normal seasonal cycle that none of the cdc models incorporated.
If the news continues to barely talk about covid, I’d expect mortality to drop by about half next season. The hysterical response was just as bad as the virus. I doubt ADE, etc could be worse than that.
This is the new CDC forecast:
https://www.cdc.gov/coronavirus/2019-ncov/images/science/forecasting/cases/november2021/National-Forecast-Incident-Cases-2021-11-15.jpg
Does anyone really believe that is what will happen?
@Phil, I’ve been lazy but I am going to try to do the state (or maybe it ends up regional) level degree-days and power consumption by this weekend.
What do the models say that were best at predicting Oct 1?
I did not look into that, the cdc site has moved (or removed?) the info on the individual models. It would be interesting to know.
I’m not sure what you are looking for, but is here (the .csv files)?:
https://www.cdc.gov/coronavirus/2019-ncov/science/forecasting/forecasting-us-cases-previous.html
Scroll down to Oct 13th. The chart and xls file previously showed the individual model predictions. Now it is only ensemble.
The csv still contains county level predictions by model though. I would assume these are the same models and you could add it up to get the national forecasts, but I have not checked it.
And now:
https://www.cdc.gov/coronavirus/2019-ncov/science/forecasting/forecasts-cases.html
If they just included the obvious and expected seasonality this type of behavior would not be necessary.
Sure, I believe the new predictions could happen.
I’m curious why you seem to believe they will not happen. Well, I’m not **that** curious, since you have a track record of being (very) incorrect for this sort of thing.
Because as it gets colder people close their windows and turn on the heat. Just like when it gets very hot they close the windows and turn on the AC. Based on what we’ve seen, this must drastically raise the R0. Masks and vaccines appear to have negligible impact in comparison.
Anoneuoid –
> Masks and vaccines appear to have negligible impact in comparison
Obviously, masks and vaccines (along with ventilation in association with AC and indoor heating) aren’t the only relevant variables.
Ventilation appears to be 50-80% of it. We have heard some about it, but very little in comparison to the “subscription service” type solutions.
I’ve heard plenty about it. But you’re still ‘ignoring” critical factors.
https://www.cdc.gov/coronavirus/2019-ncov/images/science/forecasting/cases/november2021/National-Forecast-Incident-Cases-2021-11-29.jpg?_=09688
Now the forecast is two weeks instead of four. And obviously most of the models ignore reduced testing/reporting during holidays.
Anonydood: You made a post on 11/18/21 and provided a link for the CDC projections made on 11/15/21, which projected out to ~12/15. You thought these projection were wrong based on reverse-engineering what you thought would happen, using cherry-picked evidence that was synthesized to align with your preexisting bias. The projections made on 11/15/21 for 12/2/21 ended up being generally correct. More-so, the projections you now link to, made on 11/29/21, still project a net “flat” trend in cases through 12/13! This is on top of you *already* losing your bet with Phil, and cowardly not taking other people up on the same bet during that time. Talk about a short memory.
Yet you go on and on, spewing bad science into this blog’s void, ignore the dozens of people who doggedly tell you you’re overreaching or outright wrong, some of whom even waste their time to formalize sophisticated rebuttals and find links for you to read. You’re not worth the time! Speaking of time, instead of griping about not having time to perform your temperature vs COVID cases regression, why not stop obsessively posting on the blog for a while and actually do the work? Maybe it will be an interesting analysis! Phil’s even offered to write the post for you–what a nice guy!
Why are you so angry?
The next value on that chart is 850,061 (it is based on JHU weekly data that updates Saturday nights).