Forecasts for 2021 Covid-19 deaths in the USA

This saved version only includes forecasts and rationales from these people, all of whom have agreed to share these among ourselves:

Dawna Coutant
Lou DeMarco
Seamus Fagan
Dima Klenchin
Carolyn Meinel
Dwight Smith
Richard Stenberg
Linchuan Zhang

CONTEXT:
Questions in this set ask you to forecast real or hypothetical events in 2021. All questions in this set will be scored using the reciprocal scoring method. Click here for more information on this method. For more information about how to forecast values instead of probabilities (like many of you are used to from ACE and FOCUS), see this document.

QUESTION:
How many total deaths attributed to COVID-19 will have been reported by Johns Hopkins University (JHU) for the United States from January 1st, 2021 through December 31, 2021 (inclusive). 

Resolution Criteria: The outcome will be determined based on data provided by JHU on 31 December 2021 describing Covid-19 deaths during 2021 (excluding deaths in 2020). See Coronavirus JHU


LZ
Linchuan Zhang

LAST UPDATED: 8/9/2020 6:41:53 PM

12,500	60,000	125,000	270,000	750,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Adjusted my 5th percentile upwards and adjusted my 95th percentile downwards, partially as epistemic humility updates towards what others believe, and partially with the understanding that it's unlikely 2021 will be worse than 2020, given better treatments etc.
DC
Dawna Coutant

LAST UPDATED: 8/8/2020 5:28:24 PM

40,000	80,000	200,000	350,000	410,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Best case scenario, people start taking the virus seriously in the next couple of months and their survival instincts kick in and the vaccine is on the market early and is effective. My worst case scenario equals my worst case scenario for 2020. A significant number of people do NOT take it seriously so we continue to have hot spot flare ups and reinfections. The vaccine is not-effective or not available until late in the year. We have learned more about how to treat the virus, so that should lower deaths in 2021 from 2020 levels. But... the virus really didn't get got in the US until March, so 2021 will have the virus producing deaths for a couple of extra months more than in 2021. My 50 percentile bin assumes there is still lots of non-compliance, but slowly more people accept this as the new normal and a vaccine is effective and slowly becomes available to a significant section of the population by the middle of the year.
1 RESPONSE  
BL
Barry Leybovich

LAST UPDATED: 8/11/2020 4:07:01 PM

As a thought experiment: in what percentile would you categorize the US response to covid in 2020 (where 0 is perfect, and 100 is the worst). The edges might look something like New Zealand and Japan or South Korea on one side, and something like UK, Brazil, Mexico on the other hand.

Now, thinking about things like the US election, what percentile would you categorize as most likely in 2021?
1 RESPONSE  
DC
Dawna Coutant

LAST UPDATED: 8/12/2020 8:57:13 AM
US response in 2020 I would put at around 85-90 (If Brazil, and Mexico are 100). Not because of the national response, but because we have better medical infrastructure and because of individual state responses. It is still terrible because of open borders between states and misinformation from the US administration, but there are bright spots in the country, i.e. my own state of New Mexico. [source 1] 
In 2021, If we continue with Trump I expect it to continue that way. He does not appear to learn from data. If Biden comes in then I think we will probably drop/improve to 40-45%. Biden will follow the science, but a significant number of Americans will continue to see it as a political issue and resist their 'right' to a rugged individualistic flaunting of Public health.
0 RESPONSES LISTENRESPOND
SOURCES
[source 1]

CM
Carolyn Meinel

LAST UPDATED: 8/8/2020 11:05:24 AM

80,000	103,770	183,113	240,000	400,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Influenced heavily by Seamus Fagan and by early signs that most schools and academic institutions will take a hard line on fighting COVID-19.
SF
Seamus Fagan

LAST UPDATED: 8/8/2020 8:18:00 AM

27,375	91,250	182,500	292,000	511,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Some level of herd immunity should kick in next year, also learned behaviours and better treatment and possible vaccination.
365*75=27375
365*250=91250
365*500=182500
365*800=292000
365*1400=511000

DK
Dima Klenchin

LAST UPDATED: 8/6/2020 9:53:01 PM

3,000	8,000	15,000	35,000	80,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Revising: 

1. Huge uncertainty on several fronts still exist - has a potential to get very low or very high. 

2. After much thinking about it, I continue to be optimistic overall. Many things weight toward the bright side. None of them is super-likely but each has potential to considerably reduce deaths. And they all can be ~ additive: 

- Strong enough innate immunity in significant % of the population (majority of especially vulnerable infected and dead in 2020 already)
- Immunity acquired in 2020 lasts > 6 month. Not necessarily 'lasts' to prevent infection but remains good enough to reduce severity. 
- Vaccine. 
- Finally, effective treatment(s). Even an HCQ in the absence of Trump might do the trick. If there are no nasty ADEs, monoclonal neutralizing antibodies are almost a slam dunk for the second part of the year in treatment of the death-prone populations. 
- 50% of population actively adopting the rules of 'new normal' (e.g., lots of offices likely stay as home offices). Maybe even active measures to protect elderly become a norm. 

Bottom line is, unlike the many of common colds in the past (which all occasionally cause pneumonia and deaths. e.g., RSV at ~ 15K annually), the attention will be laser-focused on SARS-2, and that alone should cause a reduction of deaths caused by it. 

So: setting median at ~ an average flu year, setting 2sigma tails ~ an order of magnitude from that, and a geometric mean between the tails and median as quartile values:

4, 11, 35, 100, 300K


*******************************************************************************************
Previous rationale: 

Oh, I don't know. We should be mostly burned through by the end of 2020. Not really buying super-short immunity either. Maybe a treatment, finally. Maybe even a vaccine. Shouldn't be worse than a bad flu but hard could easily be much better (SARS appears to be an easier target than flu). But probably not but more than 10-fold better? ~35K is a geometric mean of 1968 and 2009 flu deaths in the USA. The median value is tricky but for now going with something between 3.5 and 35, ~ 15K 

May be too optimistic. Will revisit. 
SF
Seamus Fagan

LAST UPDATED: 8/3/2020 4:43:12 AM

27,375	73,000	146,000	292,000	511,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Getting worried about my guestimates being too low for 2021. What should we use as baserates - 2020 covid deaths? annual flu deaths? Looking at figures for cases in other countries that I thought had it under control, this does not seem to be the case. What will change in US in 2021 - improved treatment? will help but no panacea. I am not overly optimistic that vaccine will make a big difference in 2021. How far along will US be towards herd immunity in 2021?

Changing to 75,200,400,800 and 1400 per day.

SF
Seamus Fagan

LAST UPDATED: 7/31/2020 4:07:38 PM

14,600	36,500	54,750	146,000	365,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Probably a bit conservative on my higher end, rising to 1000 per day. 
75 percentile to 400/day
LZ
Linchuan Zhang

LAST UPDATED: 7/31/2020 4:49:07 AM

7,500	60,000	125,000	270,000	1,800,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

My uncertainty ranges over 2 orders of magnitude. The 5th percentile follows my 5th percentile trajectory from 2020 (where after this wave, the US somehow miraculously keeps Rt<0.95). 50th percentile is relative early vaccines starting to kick in late 2020/early 2021, and having a very prominent effect by fall 2021 (herd immunity from vaccine). 75th percentile is where we do a few things right and a few things wrong. 95th percentile is where the country just kind of gives up and throws in the metaphorical towel. (It'd be worse except that maybe in those worlds we also would be 'saved' by bad testing).
1 RESPONSE  
LZ
Linchuan Zhang

LAST UPDATED: 8/5/2020 7:27:55 PM
As far as I can tell, the fatality rate is currently decreasing over time primarily due to changing demographics in who gets infected (younger and more hispanic population, less horror stories in retirement homes), and secondarily due to better treatment (lot of medicines work a little, we no longer put people on ventilators). I think hospital overwhelm will cancel out the better treatment, and large case spikes will start overwhelming some of the demographic factors, though of course I'm not confident in this.
LZ
Linchuan Zhang

LAST UPDATED: 8/5/2020 7:18:57 PM

I think you (and others) are probably right that my numbers are too high on that end. 

That said, addressing specifically what you said: Infected Fatality Rate is not set in stone, or solely/mainly a function of biological factors vs chemical factors (virology/antibodies/idealized treatment). I was imagining that if things get really bad we'd have more hospital overwhelms than we saw in NYC, in a population that broadly is less healthy than the New York City population (NYC obesity rate ~22%, US obesity rate ~39%). I expect hospital overwhelms to be much more likely than virological mutations to lethality(which in the grand scheme of things are probably fairly rare?)
LZ
Linchuan Zhang

LAST UPDATED: 7/31/2020 7:41:49 PM
I disagree with the certainty of this range. 

At the 5th percentile, we can imagine Western European levels (though not quite East Asian levels) of bringing the virus almost entirely under control.

At the 95th percentile, re-infection becomes a serious concern. [source 1]
1 RESPONSE  
SF
Seamus Fagan

LAST UPDATED: 8/1/2020 2:16:42 AM
At 95% a much more virulent mutation - either in terms of transmission or mortality rates(hopefully not both!) must also be considered.
CM
Carolyn Meinel

LAST UPDATED: 7/31/2020 4:28:18 AM

80,000	200,000	300,000	650,000	1,800,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Adjusting for my misunderstanding of the 2021 questions, subtracting the 200K deaths for 2020 that I believe are baked in.
2 RESPONSES  
SF
Seamus Fagan

LAST UPDATED: 7/31/2020 3:58:46 PM
Do we think there is a one in twenty chance that 5000 per day will die every day next year from covid?
CM
Carolyn Meinel

LAST UPDATED: 7/31/2020 8:17:18 PM
The numbers in that post do not reflect my current numbers. For rightmost bin, 768,889

I'm forecasting a fat right tail because if Trump is reelected, he is an agent of chaos who has a genius for control of the Federal bureaucracy.
1 RESPONSE  
SF
Seamus Fagan

LAST UPDATED: 8/1/2020 2:03:36 AM
Agree re Trump - amazing the way he was able to deflect from GDP numbers on Thursday by tweeting about postponing election. Also how well he has managed to vilify WHO as agents of evil.
RS
Rich Stenberg

LAST UPDATED: 7/30/2020 11:53:27 PM

44,826	394,466	573,769	627,559	806,906
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Thanks to Dawna Coutant, I substantially reduced all my estimates when I saw the question is actually from 1/1/21 - 12/31/21

BL
Barry Leybovich

LAST UPDATED: 8/3/2020 12:25:14 PM

46,000	55,000	61,000	85,000	143,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Probably the hardest question so far! What does this amount to? For the upper bound, I chose 25% over the 95% interval of the worst flu season of the past decade. I took the 'worst' flu 5th percentile of the past decade, as well as the worst 50th percentile. Then I dropped the 25th and 75th percentiles somewhere in between.

https://www.cdc.gov/flu/about/burden/past-seasons.html

Edit: upping 95th percentile to be 50% worse than the 97.5th* percentile of previous decade. Need to consider some things like enhancement.

Edit: further upping 95th percentile by 35k for things like shedding immunity which may play a big role, and there's a lot of uncertainty on.

RS
Rich Stenberg

LAST UPDATED: 7/30/2020 11:50:15 PM

I am sorry (anonymized) but this is ridiculous, seriously. Maybe if Berlusconi were still in charge, the Italian numbers might maybe have something to do with what the U.S. could expect. You cannot compare a country that has demonstrated responsible leadership and reasonably functional political system (yes, I mean Italy) with a country that has neither. 
1 RESPONSE  LISTEN RESPOND

Dawna Coutant

LAST UPDATED: 7/30/2020 4:26:51 PM

80,000	125,000	170,000	265,000	360,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

I goofed on my initial forecast by misreading the question. I thought it was cumulative deaths through 2021. Just caught that it is JUST for 2021. So please ignore my first forecast.

As a starting place: I found 2 models that extended as late as Nov. 1 (IMHE and a machine model). [source 3] [source 2] [source 3] I took the top and bottom predictions from each model for Nov. 1 then looked at deaths/month for the last month (Oct) for each model and extended both model's predictions to Dec. 31, 2021. I ended with a high score of roughly 360,000 and, a low score of 80,000 dead.



Normally I would split the difference, but I thought I would try Lyle's suggestion of the Fermi estimation and rounded to 170,000. I set that as the middle bin. I guesstimated a rough split on the 25th and 75th percentile. I haven't worked with forecasting percentiles before, am I correct that the most likely number should be in the 50% bin?
DS
Dwight Smith

LAST UPDATED: 7/30/2020 2:54:59 PM

150,000	175,000	200,000	300,000	450,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

My baseline for all relevant questions is as follows. For the previous year, 2020, I am assuming a death rate of about 300,000, which has increased to about 1,000 a day. Given significant virus fighting fatigue, it is quite likely this number will be at least somewhat higher in 2021 without a vaccine. So I am putting my estimate at 1,200 per day or a total of 440,000. But given a virus is well circulated by early spring (say April 30th), we could begin to see a significant drop in case rates during spring and summer, leading to a very low 200-300 per day. So for the first four months, it would be 120,000, then a total of about 190,000 for the year. Resounding success in vaccine distribution and efficacy could drop the number much lower, perhaps down to about 150,000. 
SF
Seamus Fagan

LAST UPDATED: 7/30/2020 3:45:59 AM

290,600	312,500	330,750	367,250	495,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Using 276,000 as baseline forecast from Q1
Deaths in 2021: Trying to account for possible treatments and vaccination protocols as well as change in government etc etc.
Going with baseline of about 150 deaths per day. 
276000+ (365*150) = 330750
Low, if everything goes well, is 40 per day so 290,600
276000+365*100= 312500
276000+365*250=367250
276000+365*600=495000
DC
Dawna Coutant

LAST UPDATED: 7/29/2020 12:35:14 PM

274,000	356,000	438,000	569,000	700,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

As a starting place: I found 2 models that extended as late as Nov. 1 (IMHE and a machine model). [source 1] [source 2]I took the top and bottom predictions from each model for Nov. 1 then looked at deaths/month for the last month (Oct) for each model and extended both model's predictions to Dec. 31, 2021. I ended with a high score of roughly 685,000 and rounded to 700,000, a low score of 274,000 dead.



Normally I would split the difference, but I thought I would try Lyle's suggestion of the Fermi estimation and came up with 437,500. I set that as the middle bin. I guesstimated a rough split on the 25th and 75th percentile. I haven't worked with forecasting percentiles before, am I correct that the most likely number should be in the 50% bin?
My revised forecast considers a wider range of outcomes. I've also come to the conclusion that it will take much longer to get the virus under control.
DK
Dima Klenchin

LAST UPDATED: 7/29/2020 10:42:38 PM

3,000	8,000	15,000	25,000	50,000
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Oh, I don't know. We should be mostly burned through by the end of 2020. Not really buying super-short immunity either. Maybe a treatment, finally. Maybe even a vaccine. Shouldn't be worse than a bad flu but hard could easily be much better (SARS appears to be an easier target than flu). But probably not but more than 10-fold better? ~35K is a geometric mean of 1968 and 2009 flu deaths in the USA. The median value is tricky but for now going with something between 3.5 and 35, ~ 15K 

May be too optimistic. Will revisit. 
RS
Rich Stenberg

LAST UPDATED: 7/27/2020 10:02:38 PM

314,765	776,235	1,146,847	1,264,398	1,536,880
5th Percentile	25th Percentile	50th Percentile	75th Percentile	95th Percentile

Given current trends this is really just a question about what herd immunity will look like. My guesses are reflected in the three upper bins. The lower bins reflect the most chance we take effective actions sooner, and/or there is a vaccine in time to matter.

Edited to reflect greater interventions occur at 5 percent probability.