MODERATOR NOTE: coronavirus resources & updates: https://wolfr.am/coronavirus
2020-05-08—preliminar-statistik-over-doda-inkl-eng.xlsx (469 KB)
10 Likes
– you have earned Featured Contributor Badge 
Your exceptional post has been selected for our editorial column Staff Picks http://wolfr.am/StaffPicks and Your Profile is now distinguished by a Featured Contributor Badge and is displayed on the Featured Contributor Board. Thank you!
1 Like
Jan!
Good work. My own Swedish estimates are rather “black”. It’s going to be very bad with the Swedish “open” strategy. Based on a previous SEIR model published in Mathematica, I predict around 14000 deaths in Sweden by September… Check the file… Any comments are welcomed!
Best regards
Christos
Corona SWE.nb (372 KB)
Thanks Christos. It all depends on how many that have actually been infected this far (with this one could to a good estimate of the Swedish fatality rate) and how many are needed for herd immunity (with this one can estimate how many that will actually get infected eventually).
Just by looking at your graph comparing estimated with observed deaths it is quite obvious that your model is not correctly tuned. Does this mean that your total in September will be wrong? Not necessarily, but I would say it will quite likely be wrong.
You are using a very high beta resulting in 3.6 infected per infected person, while data shows that Sweden has been just above 1 for most of the time, and now has had an R value below 1 for at least a week (I don’t remember exactly for how long).
Thanks Jan. I am very aware of that and have tried with much lower beta-values and a large variation of the other parameters and time as well. The above parameters were the best I could get to approximate the death numbers, so far. The death statistics are the only ones that are true, all others are not observed. To argue that the Swedish Rho is about 1 right now might be wrong too, since there are very few tests in Sweden, people are outside and nobody knows how many each (unkown) infected person infects. What we do know is that out of about 200000 tests so far, there have been slightly more than 30000 positive (i.e. a ratio of 15%). This ratio is almost equal to the Italian ratio 6 weeks ago. I therefore doubt if our Rho is so low. By the way, according to FHM, they believe that the ratio of infected in Sweden right now is about 15-17%. My dynamic graph does estimate it to about 1.8 million people (18%) when t = 82.
Great work (+1). Economically it seems that so far the Swedish experiment has not brought tremendous advantages: Pandemic, Shutdown and Consumer Spending: Lessons
from Scandinavian Policy Responses to COVID-19?
2005.04630.pdf (1.27 MB)
1 Like
Interesting, tough not very surprising. I would assume that both countries have a big part of their consumer consumption online, which is probably not that dependent on the lock down. Of course, things like restaurants, are mainly an onsite transaction (even though a lot has moved to take away), but most other businesses are no longer really in need of any physical interaction.
The question I think, is not the momentary difference, but the difference over time. Assume that Sweden goes through the Pandemic in half the time then Denmark, then the total impact on consumer consumption will probably be substantial. However, we do not know that yet. What happens if it is the other way around (yes, not that likely, but still with limited data there is still a small possibility for this)?
A probably more important factor is unemployment rates. Unfortunately, I don’t think it is possible to find rates for April yet, and there will also be a lag due to employment regulations (in both countries).
As a side note, they point out that they have only looked at card payments, and mention that as a limitation for the study. For Sweden, and probably Denmark, I would guess that this a minor problem as cash is hardly used (at least in Sweden). When we don’t use card we pay with the phone. I would guess that I have spent less than 50€ in cash during the last couple of years in Sweden.
2 Likes
Well, death numbers in general are not as reliable as people think in most countries (see https://www.nytimes.com/interactive/2020/04/21/world/coronavirus-missing-deaths.html). This combined with the fact that you point out yourself, that the number of tests varies from country to country makes it completely pointless to compare case fatality rates at this point.
Sweden has taken way more tests then is needed to get reliable data for calculating the current R (not R0, nobody knows R0). As long as you have control over your test strategy and do enough tests, then you don’t need anywhere near complete testing to get a reliable number (similar to that you do not need to poll the while population in order to get a reliable answer on e.g. political opinions). But even more important than this is that you can now clearly see in the ICU numbers (steadily going down) and death rates (same there) that those claiming that R was under 1 since a few weeks back were most likely right. Reference: COVID-19 i svensk intensivvård
1 Like
Another thing that I missed mentioning, and that is really important too is which type of government support businesses and consumers have been provided, and how that affects e.g. the national dept. There’s a lot of things to consider…
2 Likes
Thanks Jan!
I hope I am very wrong and my death estimates are too high! But I am very worried with the Swedish strategy. By the way, my initial R is about 3.84 = (0.275/0.0715) solvS[t] /. solv, if t = 0, i.e. if S[0] = 1. Today, when t = 84, the R-ratio is about 1.23 since the susceptible is about 32%, and in 4 days, it will fall to 0.99. By the end of September it will fall to less than 0.08, since the model predicts that there will be about 2% of the population as potential susceptible during the fall/winter.
There something wrong in your model. If we assume an R0 of 3.84 (which I guess you mean with initial R), then herd immunity is easy to calculate and you would reach it when 1/R0 are susceptible, i.e. 26% of the population is susceptible. To get down to only 2% susceptible you need either a much higher R0 or that you run towards the herd immunity, so the herd immunity does not have enough tome to slow down the growth. Considering that you talk about a time frame for this fall/winter then we are apparently not running towards the herd immunity.
See herd immunity here: Herd immunity - Wikipedia
1 Like
Ok Jan!
It is not my own model. It is posted by Arnoud Buzing, Wolfram Research, 2020-03-22 , "The SIR Model for the Corona Pandemy and modified it slightly. If you check the differential equations you see the initial R is very high, when all 10 million people in Sweden are susceptible, and it falls overtime when the S[t] falls so that with the same beta & gamma parameters there are fewer non-infected people to be infected. The infection and death rate fall (accumulated deaths reach about 13600 by the end of september). I have tried with a various length of t but the best I one I found was about 220 days, i.e. by the end of September. I repeat, I hope I am wrong! By the way, the deaths today were only 19!
A model is never better than the data it is based on, the initialization, and parameter values. In this case
As far as I can see your parameter setting do not take account at all for the impact that policies had on R0. You are using an R0 of 3.65. This might be correct (no one knows). However, we know that there has been several policies introduced and that these have affected the effective “R0”. We know that this effective R0 i closer to 1 than 3.65, see COVID-19 - R0 and Herd Immunity - are we getting closer? and COVID-19 policy simulator: can you find the perfect policy? .
1 Like
Btw, your number of 16,000 in itself is quite reasonable. Reasonable numbers is easy to calculate by just multiplying fatality rate (0.2 - 2% perhaps) with number of infected (50 - 90% perhaps) giving between 10,000 to 180,000 deaths depending in which figures apply. Of course, the fatality rate is the most important factor here (I think they believe it’s around 0.2% in Stockholm to date). That’s why it is so important to protect risk groups.
Anyways, the number in itself is not that strange, but the parameter settings you have in your model simply don’t add up with the current development, so the trajectory will most likely not be as in your simulation while the end result might be the same.
1 Like
Thanks for the work. I see a lot of “excess of deaths” this year. And a lot of them in countries with lockdown protocols, Something that almost no one is talking about is the fact that, if we keep away from sun, we are goona have a lack of vitamin D, that is fundamental for the inmmune system. Also we are isoleting our body, this is another way to depress our inmmune system, it is call “hygienist hypothesis”.
Another issue is that, people arround the globe, where quarantine have been apllied, is under a cronic stress; they are loosing their jobs, therir companies etc. This condition, stress, is a strong apigenerator, it means that modifies the DNA expression pattern, it have been assoieted with the development diseasess like cancer, coronary diseasess, mental diseasess etc.
The idea of use a medieval strategy, in order to avoid the health collapse, is probably going to be the cause of more diseasess and death in the near future. Unfortunately we can not make a model with those conssiderations.
1 Like
Nice post! I like the way you very “conversationally” retrieve data and visualize them!
The time series plot very clearly shows the increased number of deaths in April 2020, but that observation does not come across in the subsequent bar charts. Making a 2D bar chart with an additional breakdown by month shows that April 2020 really stands out:
chartData =
TakeList[#, {31, 29, 31, 30}] & /@ dailyDeaths[[All, 1 ;; 121]] //
Map[DeleteMissing /* Total, #, {2}] &;
BarChart[chartData,
ChartLabels -> {{2015, 2016, 2017, 2018, 2019, 2020, "Average"},
None}, ImageSize -> Large, BarSpacing -> {0.25, 2},
PlotLabel -> "Sweden monthly deaths",
ChartLegends -> {"Jan", "Feb", "Mar", "Apr"}]
The effect is not lost looking at the population-adjusted deaths:
chartData =
TakeList[#, {31, 29, 31, 30}] & /@ dailyDeaths[[All, 1 ;; 121]] //
Map[DeleteMissing /* Total, #, {2}] &;
chartData = MapThread[Divide, {chartData, pop}];
BarChart[1000 chartData,
ChartLabels -> {{2015, 2016, 2017, 2018, 2019, 2020, "Average"},
None}, ImageSize -> Large, BarSpacing -> {0.25, 2},
PlotLabel -> "Sweden monthly deaths (per 1,000)",
ChartLegends -> {"Jan", "Feb", "Mar", "Apr"}]
It’s probably going to take years to sort out all the effects of this pandemic. The problem will be hampered by siloed data, which will compounded by too much political influence on the collection and reporting of those data.
3 Likes
So far, the Swedish open society experiment has not been very successful compared to the majority of European countries that have applied much restrictive lockdown policies. For instance, if the Swedish Song Contest for Eurovision that took place on March 19 in front of 30000 people at Friends Arena in Stockholm was without public, precisely as in Denmark, it could have reduced the number of infections and deaths.
lockdown.nb (279 KB)
1 Like
Thanks Rob, I completely agree with you.I’m skeptical to anyone that draws too big and definite conclusions at this very early stage. Data will be added bit by bit, and as we learn to understand it and understand how reliable it is, we can start making more precise conclusions.
1 Like
The first relevant antibody tests for Sweden came just a couple of days ago. As expected the largest proportion of positive antibody tests are found in Stockholm. A total of 7.3 percent of the blood samples from people in Stockholm were positive. For the southern region of Skåne (that borders to Denmark), the proportion of positive samples was 4.2 percent and for Västra Götaland, where the second largest city, Gothenburg is located, the figure was 3.7 percent.
The analyzes also show that there is a big difference between age groups. Antibodies were most common in people aged 20-64, a total of 6.7 percent of the samples in the group were positive. In the 65-70 age group, the proportion of infected persons was less, 2.7 percent. This indicates that protection of elderly has worked to some extent at least.
The tests were taken end of April, so more people will have antibodies by now. It is my understanding that it can take a few weeks after an infection before you have built antibodies. If that is the case, it means that means that even more people have been sick at this point. Does someone know? @Robert Nachbar perhaps?
1 Like
As a comparison Denmark’s first tests show that only 1% of the Danes have antibodies yet. Just as in the Swedish case, it is based on quite few samples (but more than Sweden) that are not completely randomized either, so the data might not be that reliable yet. In both cases the studies continues of course.
2 Likes