US Coronavirus deaths
16 responses | 0 likes
Started by pj - April 9, 2020, 4:47 p.m.


FWIW I ran the figures for reported deaths in the US. The starting point was where I could find the data. I calculated a weighted average of the rate of increase in new deaths through 4/8 to be 1.14. Using this rate to estimate the number total deaths out to 4/30 gives a total of about 300,000 deaths. This is most likely quite a bit too high, due to the number of new cases probably currently peaking and the lag in deaths vs cases.

I then tried calculating a least squares line though the rate of increase data. This came out with r= -.00813*day+ 1.21055. So decreasing very slightly at this point. Given that there major risks when you extrapolate, using the rates from this formula to calculate the day by day deaths, the number of total deaths by the end of April comes out to be about 88,000. Again given the lag and the number of new cases probably peaking, this is likely still a bit high, but probably closer to what will actually happen. 


daydate
new deathsrate of increase
total deaths

new deathsLe Sq rate of increasetotal deaths



actual

actual




13/27
394

1264




23/28
4561.16
1720




33/29
5751.26
2295




43/30
8301.44
3125




53/31
9281.12
4053




64/1
10491.13
5102




74/2
9680.92
6070




84/3
10511.09
7121




94/4
13311.27
8452




104/5
11650.88
9616




114/6
12551.08
10871




124/7
19701.57
12848




134/8
19400.981.1414788







projected

^ weighted avg
projected


projected

projected
144/9
2,220.841.14
17,008.84

2,143.351.10481861616,931.35
154/10
2,542.341.14
19,551.17

2,350.581.09668517519,281.93
164/11
2,910.371.14
22,461.55

2,558.731.08855173421,840.65
174/12
3,331.681.14
25,793.23

2,764.491.08041829324,605.15
184/13
3,813.991.14
29,607.22

2,964.331.07228485227,569.47
194/14
4,366.111.14
33,973.33

3,154.491.06415141130,723.96
204/15
4,998.161.14
38,971.49

3,331.201.0560179734,055.16
214/16
5,721.711.14
44,693.20

3,490.711.04788452937,545.87
224/17
6,550.001.14
51,243.21

3,629.471.03975108841,175.35
234/18
7,498.201.14
58,741.41

3,744.231.03161764744,919.57
244/19
8,583.661.14
67,325.07

3,832.161.02348420648,751.73
254/20
9,826.261.14
77,151.32

3,890.981.01535076552,642.71
264/21
11,248.731.14
88,400.06

3,919.071.00721732456,561.78
274/22
12,877.131.14
101,277.19

3,915.480.999083883460,477.26
284/23
14,741.261.14
116,018.45

3,880.040.990950442464,357.30
294/24
16,875.251.14
132,893.69

3,813.370.982817001468,170.67
304/25
19,318.161.14
152,211.85

3,716.830.974683560471,887.50
314/26
22,114.711.14
174,326.56

3,592.500.966550119475,480.00
324/27
25,316.101.14
199,642.66

3,443.120.958416678478,923.12
334/28
28,980.931.14
228,623.58

3,271.930.950283237482,195.05
344/29
33,176.291.14
261,799.87

3,082.650.942149796485,277.71
354/30
37,978.981.14
299,778.85

2,879.250.934016355488,156.95
Comments
By metmike - April 9, 2020, 5:45 p.m.
Like Reply

Thanks much pj!

Good stuff.

I'm betting there will be only half the 88,000 deaths that your model predicts.

The effects of human behavior with mitigation are impossible to model as indicated by all the recent models drastically cutting all earlier estimates by a great deal.

This will continue.

My last wild guess was 1,200,000 total cases and 30,000 deaths at the end of the month may be low on the deaths by  as much as 10,000 and high on the cases by 100,000 but I am using what is most likely to happen using critical thinking.

For instance, it's extremely likely that the rate of new cases will be dropping for the rest of the month. Not every day but most days. It's plausible that the daily rate could be well under 20,000 at the end of the month. 

That will take at least a week to show up in big drops in the deaths, which is why I probably need to increase that. 

I have also been using the Kinsa health data as a LEADING indicator.

https://www.marketforum.com/forum/topic/49867/


In some ways, it's like weather predicting.  We have dozens of weather models that use thousands of mathematical equations based on the physical laws of the atmosphere that do all the work and we could never try to forecast without. 

It would take a human their entire life to solve all the equations for just 1 model run!!

However, a good operational meteorologist can look at the model forecasts and recognize when the models are missing things. I note that the local NWS in most locations these days just uses the computer generated model output in their forecast vs trying to beat the models.

In the old days, operational meteorologists worth their oats could often whip the computer models.

The NWS does a wonderful job providing us with their products and service.

The huge difference between weather models and COVID-19 models is that one is based on the well known physical laws of the atmosphere...........the other one is greatly affected by uncertain human behavior, predicting the not so well known science(infections/deaths) of a brand new virus.  

The latest empirical data/observations will always count much more in the unknown situation because it immediately closes the gap between known and unknown for that data point...........which might have been pretty wide.

By pj - April 17, 2020, 1:21 a.m.
Like Reply

FWIW I updated the least squares fit I did last week. When calculating the daily rates of increase I left out the "probable cases" that were added a few days ago, which would have caused a spike for that day. My feeling, maybe they're learning/getting better at treatments, maybe some of the pressure is coming off the medical facilities will result in rates dropping a little faster than the least squares straight line projects.


datenew deathsrate of increase
total deaths





3/27394

1264
3/284561.157360
1720
3/295751.260965
2295
3/308301.443478
3125
3/319281.118072
4053
4/110491.130388
5102
4/29680.922784
6070
4/310511.085744
7121
4/413311.266413
8452
4/511650.875282
9616
4/612551.077253
10871
4/719701.569721
12848
4/819400.984772
14791
4/919000.979381
16712
4/1020351.071053
18747
4/1118300.899263
20577
4/1215280.834973
22105
4/1315351.004581
23640
4/1424071.568078
29825
4/1526181.087661
32443
4/1621710.829259
34614

Projected


new deathsLS rate of increasetotal deaths


1.025398
4/172,229.221.01666936,846.22
4/182,266.371.00794039,112.59
4/192,284.370.99921141,396.96
4/202,282.570.99048243,679.53
4/212,260.840.98175345,940.37
4/222,219.590.97302448,159.96
4/232,159.710.96429550,319.67
4/242,082.600.95556652,402.27
4/251,990.060.94683754,392.33
4/261,884.260.93810856,276.59
4/271,767.640.92937958,044.23
4/281,642.810.92065059,687.04
4/291,512.450.91192161,199.49
4/301,379.240.90319262,578.73



By metmike - April 17, 2020, 12:11 p.m.
Like Reply

Thanks very much pj. Great stuff!

The last 2 days has featured an increase in new cases again, possibly related to the record cold/dry weather pattern that hit at the end of last week/early this week.


If the weather is actually having an influence and this is seasonal(we don't know that) then our models will be way off.

https://www.marketforum.com/forum/topic/49391/

By pj - April 25, 2020, 1:44 a.m.
Like Reply

2nd update

Surprisingly (and no doubt at least partly by coincidence) the projected total deaths by 4/30 remains almost exactly the same as last week's analysis at 62,000+. Below a plot of the data with the least squares line. Note `it's slope is slightly declining and the projected rate is below 1. If this stays on track it means we may well have seen the peak in the (average) number of daily deaths.


datenew deathsrate of increase
total deaths





3/27394

1264
3/284561.157360
1720
3/295751.260965
2295
3/308301.443478
3125
3/319281.118072
4053
4/110491.130388
5102
4/29680.922784
6070
4/310511.085744
7121
4/413311.266413
8452
4/511650.875282
9616
4/612551.077253
10871
4/719701.569721
12848
4/819400.984772
14791
4/919000.979381
16712
4/1020351.071053
18747
4/1118300.899263
20577
4/1215280.834973
22105
4/1315351.004581
23640
4/1424071.568078
29825
4/1526181.087661
32443
4/1621710.829259
34619
4/1725351.167665
37154
4/1818670.736489
39014
4/1915610.836101
40575
4/2019391.242152
42514
4/2128041.446106
45318
4/2223410.834879
47892
4/2323421.000427
50234
4/2419510.833049
52185

Projected


new deathsLS rate of increasetotal deaths






0.973020
4/251,898.360.96503454,083.36
4/261,831.980.95704855,915.35
4/271,753.300.94906157,668.64
4/281,663.990.94107559,332.63
4/291,565.940.93308960,898.56
4/301,461.160.92510362,359.72



By metmike - April 25, 2020, 3:11 p.m.
Like Reply

Thanks very much pj. More on this later.

By pj - May 1, 2020, 5:49 p.m.
Like Reply

Update as of the end of April.

The past two projections for the end of April total deaths of 62,000+ turned out to be quite close to actual total of 63,856.

Below is the actual data through 4/30 with projections for May. The slope of the least squares line has flattened (become less negative) a little since the prior weeks' calculations, the projected rate (only) now dropping below 1 in the next few days. Obviously the projections are highly sensitive to whether/when the rate goes below 1.

The actual rate on 4/27, being so high, may have skewed the line a bit and perhaps could be eliminated, as being an outlier. Though it's part of the pattern of larger drops over the weekend being made up for by bigger increases on the following week days. OTOH once you start "playing" with data, you run the risk of the whole exercise becoming arbitrary.

Using the data as is, and extrapolating the least squares line into May, gives expected total deaths of about 92,000 by 5/15 and 106,000 by 5/31. My gut feeling is that these projected totals will turn out to be on the high side.

datenew deathsrate of increase
total deaths





3/27394

1264
3/284561.157360
1720
3/295751.260965
2295
3/308301.443478
3125
3/319281.118072
4053
4/110491.130388
5102
4/29680.922784
6070
4/310511.085744
7121
4/413311.266413
8452
4/511650.875282
9616
4/612551.077253
10871
4/719701.569721
12848
4/819400.984772
14791
4/919000.979381
16712
4/1020351.071053
18747
4/1118300.899263
20577
4/1215280.834973
22105
4/1315351.004581
23640
4/1424071.568078
29825
4/1526181.087661
32443
4/1621710.829259
34619
4/1725351.167665
37154
4/1818670.736489
39014
4/1915610.836101
40575
4/2019391.242152
42514
4/2128041.446106
45318
4/2223410.834879
47892
4/2323421.000427
50234
4/2419590.836465
52191
4/2520651.054109
54256
4/2611570.560291
55413
4/2713841.196197
56796
4/2824701.784682
59265
4/2923900.967611
61655
4/3022010.920921
63856

7-day moving avg  1947

Projected


new deathsLS rate of increasetotal deaths


1.014592
5/11,974.981.01031465,830.98
5/21,995.351.00603667,826.32
5/32,007.391.00175869,833.71
5/42,010.920.99748071,844.63
5/52,005.850.99320373,850.49
5/61,992.220.98892575,842.71
5/71,970.150.98464777,812.86
5/81,939.910.98036979,752.77
5/91,901.820.97609181,654.59
5/101,856.350.97181383,510.94
5/111,804.030.96753585,314.97
5/121,745.460.96325787,060.43
5/131,681.330.95898088,741.76
5/141,612.360.95470290,354.12
5/151,539.320.95042491,893.44
5/161,463.010.94614693,356.45
5/171,384.220.94186894,740.67
5/181,303.750.93759096,044.42
5/191,222.380.93331297,266.81
5/201,140.870.92903498,407.68
5/211,059.900.92475699,467.58
5/22980.150.920479100,447.73
5/23902.210.916201101,349.94
5/24826.610.911923102,176.55
5/25753.800.907645102,930.35
5/26684.180.903367103,614.53
5/27618.070.899089104,232.60
5/28555.700.894811104,788.30
5/29497.250.890533105,285.54
5/30442.810.886255105,728.36
5/31392.450.881978106,120.80


Least Squares Rate = -0.004278 *(day) + 1.155762

By metmike - May 2, 2020, 1:09 p.m.
Like Reply

Thanks again pj.

Yes, your model did an amazing job at projecting deaths at the end of April. Congrats!

I was way too low for guessing  deaths but nailed the positive tests/infections....because this has been measuring only a fraction of the REAL infections. 

This tells me that there have been many more deaths/positives than I expected(and to take your model seriously) but the reality, now that we are finding out there could have been between 10 and 50 times more cases that never got tested is that the positives are extremely too low for measuring how many cases in the REAL WORLD exist out there.....but catching up, which is making it look like the infection rate is not dropping. 

Instead of 1 million measured, there may have been more than 10 million REAL infections.

The more data that we get, the better handle we get on it. 

By pj - May 2, 2020, 5:55 p.m.
Like Reply

"Yes, your model did an amazing job at projecting deaths at the end of April. Contrats!"

Thanks. Probably a significant amount of luck involved with it coming out that close. More particularly, the rate of the rate of change staying so stable over that period.

By pj - May 8, 2020, 7:08 p.m.
Like Reply

Another week another update. The projections continue to track actual deaths pretty well. As of 5/7, 77,813 projected vs 76,928 actual. 

Below the projections though the end of May. The least sq line has flattened a bit again (diminishing at a slower rate), no doubt skewed some by the high values on 4/28 and 5/7. Now still a few days before dropping below 1.

For reasons unknown, I was unable to copy in my table this time around.

Actual

new deaths     rate of increase     total deaths

5/1     1,897             0.861881             65,753
5/2     1,691             0.891407             67,444
5/3     1,153             0.681845             68,597
5/4     1,324             1.148309             69,921
5/5     2,350             1.774924             72,271
5/6     2,528             1.075745             74,799
5/7     2,129             0.842168             76,928

Projected
new deaths      rate of increase    total deaths
5/8     1890              1.009142              78818
5/9     1907              1.006159              80725
5/10   1919              1.003176              82645
5/11   1925              1.000194              84570
5/12   1926              0.997211              86495
5/13   1920              0.994228              88415
5/14   1909              0.991245              90325
5/15   1892              0.988262              92217
5/16   1870              0.985279              94087
5/17   1843              0.982296              95930
5/18   1810              0.979313              97740
5/19   1773              0.976330              99512
5/20   1731              0.973347            101243
5/21   1685              0.970364            102928
5/22   1635              0.967381            104562
5/23   1581              0.964398            106143
5/24   1525              0.961415            107668
5/25   1466              0.958432            109135
5/26   1405              0.955449            110540
5/27   1343              0.952466            111882
5/28   1279              0.949484            113161
5/29   1214              0.946501            114375
5/30   1149              0.943518            115524
5/31   1084              0.940535            116609

Least squares rate = -0.002983  x (day) + 1.131443

 

By TimNew - May 9, 2020, 7:03 a.m.
Like Reply

This is amazing stuff PJ.  Thanks!!

By bear - May 13, 2020, 11:08 p.m.
Like Reply

please allow me to explain why the "reported" deaths figure probably is fraudulent.  

you can find other info on this topic at armstroneconomics.com

since the u.s. govt has supposedly made the proposal to pay hospitals for covid deaths,  now all deaths are covid deaths.  doctors at hospitals are being told (by administrators) to put down covid, even when people are dying of other causes.  and in most cases, definitive testing has NOT been done to determine cause of death.  but it is still listed as a covid death anyway. 

if a person dies of heart disease,  then the hospital has to fight to get payment from the family of the deceased.  but if they put down covid,  then they expect to automatically get paid by uncle sam.  

one funeral home in new york said that recently ALL death certificates say covid as cause of death.  NO one is dying of heart disease, or cancer, or flu, or any other cause.  

ask yourself... what are the odds ?  


so you will probably get a more accurate covid death rate in a place like iceland, where there is NO motivation for hospitals to commit fraud to boost their revenue.  

in iceland where they did a large scale test,... 50% of the population had covid, but most had No symptoms.  4% of the population showed symptoms,  far less than one tenth of one percent died of covid.  (no worse than the flu). 


the point here... in the u.s., we dont know if the figures are overstated by 50%, or 100%, or 200%, etc.  NO ONE is doing an honest assessment of how many of those death are actually covid (as compared to a flu death, or a heart attack death, etc).  

By bear - May 13, 2020, 11:11 p.m.
Like Reply

my econ professor always told us... figures don't lie,  but liers figure.  

remember, here in the u.s., politicians get their death figures from the health dept.  health depts get their figures from hospitals.  and hospitals have a motivation to Lie about cause of death.  $$$$ follow the money $$$$.

By pj - June 1, 2020, 3:57 p.m.
Like Reply

6/1 update. 

The projection made on 5/1 (above) for the total deaths in the US by the end of May, again fortuitously, was very close to the actual number. Projected 106,121, actual 106,195.

Proving that being that close was in part luck, was the update done a week later which projected 116,109. I believe that one was thrown off by weekend drop off and subsequent bounce way up, with the methodology, using squares of the data, accentuating the phenomenon.

Again FWIW at this point, using all the data and the same method, the projections for June are 118,900 by 6/15, 126,100 by 6/30.

By metmike - June 1, 2020, 7:27 p.m.
Like Reply

pj,

Congrats again. Your estimates have been better than any model from any source that I've seen this year.........hands down. You say that luck is party responsible. Maybe so but luck runs out, skill continues. 

You will note that I stopped trying to make predictions a long time ago............I'll stick to making predictions in a  field of expertise that I actually have some skill at...........weather and climate!

Too bad that you aren't an atmospheric/climate scientist (-: .............actually, anytime you want to take a crack at predicting global temperatures for the next 100 years,  I'll take  it seriously!


Before my kids play in chess tournaments, I will sometimes tell them, "don't have good luck, use skill to win"

By pj - June 2, 2020, 1:39 p.m.
Like Reply

Thanks again for your kind words. Wish I could come close to predicting futures prices.

By metmike - June 2, 2020, 1:58 p.m.
Like Reply

But if you had that great talent in predicting futures prices and were making millions of dollars, you wouldn't be associating with us here (-:



You might even vote for Donald Trump