Visualizing observed deaths using polar coordinates
mcculley.github.io
mcculley.github.io
> Using this kind of graph one can more easily see how changes year-to-year in total observed deaths have been affected by COVID-19.
It’s important to look at and account for the number of deaths directly attributed to COVID, isn’t it? Is there some discrepancy between the excess YoY deaths implied here and the number directly attributed to COVID? Also would be good to normalize per-capita, just to fully factor out population growth?
> Using total observed deaths, one can disregard the (mis)attribution of deaths to COVID-19 and more clearly see the failure of federal and state governments.
Out of curiosity, what are the misattributions and failures being referred to? Just confused a little since the source data is federal (CDC) and has this exact same graph showing the excess death rate this year, just on a horizontal plot.
If that's important to you than just use Cartesian graph. Polar graph shows better seasonal patterns as it does not cut abruptly at the end of the each year.
A) Uncertainty about how deaths are effectively false positives, and how many "deaths due to COVID" weren't counted. This is a combination of plain uncertainty, not having testing available for everything, all the way up to people fudging numbers (perhaps by choosing not to test categories of deaths).
B) It's unclear (as in people will argue, on both sides) if "deaths with COVID" vs "deaths due to COVID (with some definition of 'due')" is the right measure.
The failures is just shit talking (shit talk can be both deserved and undeserved) the government responses. The basis of the shit talking is that the red and green lines should be far closer to the previous baselines.
As for shit talking, I have been frustrated since March of 2020 at the impotence and incompetence of our culture and government to respond and I'm tired of arguing with Trump and Biden apologists. I could probably word all of that better.
Yes, it's always a trade-off of course, but I generally prefer to sqrt() the radial axis of this kind of plot so it is equal-area.
Unfortunately it doesn’t make the numbers less tragic.
Whole circle (angle = 2 x pi) is pi x r^2
There was a lot of discussion around how many more deaths were caused by COVID versus normal expected causes. I wanted to make that clearer.
You make a good point about the area being misleadingly large. I had not thought of that.
As for population growth, I thought about that but I'm not sure where to get reliable numbers for each region. As is, one can see that in 2017-2019, the curves were showing steady growth as expected to correlate with population.
Some people don't trust that data. I've heard people tell me that when COVID started the hospitals started marking everyone who died as having COVID regardless of what the real cause was.
>Shared hundreds of times on Facebook, posts claim: “Every time a hospital admits, discharges, or loses a patient to Covid-19, they are compensated 15% more according to the CARES ACT, SEC 4409”. The posts also say that New York City hospitals are “inflating all of their #coronavirus numbers” to take advantage of this section of the act.
https://www.reuters.com/article/uk-factcheck-more-money-for-...
There is obvious borderline cases when COVID merely accelerated another disease (heart failure), but I bet in 99% of cases, the ultimate cause of death is one or multiple organ failure due to COVID.
In other words, even if you question their categorization of deaths, it’s impossible to deny that something out of the ordinary caused 600k excess deaths in the US, and that “something” happened to coincide perfectly with the COVID-19 surges.
To draw a parellel, if someone started getting more tips at the restaurn they worked at, or more Tinder matches than usual, they should notice "Hey, something's changed..." and wonder what the reason behind it is..
This creates the illusion, if one takes those countries media for real, that in other countries the disease is not as ravaging or non-existant. One should trust democratic countries data more and signal distrust on data/ positions / declarations from authoritarian regimes.
This however is not happening much on online plattforms, were one dictators "cheerful" lies are equally loud represented next to a loyal servant of a democratic state giving "grim" news.
The platforms thus help to creates the impression, that authoritarian regimes are dealing "better" with the situation and those wishing for a return to the "before" times, seeing those before times seemingly alive, thus steer towards totalitarian propaganda.
Lies for money subvert the system allowing to sell lies for money as a business model.
This problem solves itself.
If a government is caught lying, its previous statements to a fact should be auto-corrected and future statements be overwritten with a extrapolated truth, given by someone proofen to be trust worthy.
From what I've heard, it went both ways, depending on the point in time. E.g. Poland had an election coming up mid-pandemic, and around that period, many people believed the government is underreporting COVID deaths, by making doctors classify them as caused by any of the comorbidities present in the deceased (and past certain age, pretty much everyone had some comorbidity, e.g. a weak cardiovascular system, or cancer).
COVID statistics were always heavily politicized. I think the most reliable way of getting to the truth was, and still is, looking at the excess death counts (i.e. how many more people died in a given month over the average of deaths in that month in last few pre-pandemic years).
Hospitals got economic relief from Covid because they couldn't take in other patients for fear of the contagiousness of Covid. So they got "credit" for "Covid patients" for financial reimbursement claim purposes.
But that wasn't the "cause of death" on the death certificate. The COD was listed as whatever it was.
There looked to be some deliberate attempt to conflate these assignments in the minds of people by some bad actors last year.
But again, that's not saying that it didn't happen. But when it was being talked about last year it seemed like it was being used for political purposes.
However, connecting the end of one year to the start of the next is not what's important. Having a undistorted line graph that plots each year would be much more informative.
For this data, I'm not at all clear on why a standard line chart isn't better. I'll confess, I don't find these charts bad; I just don't see them as clearer than the standard charts. What is the specific advantage?
The commentary on lag of reporting being a surprise is, itself, surprising at this point. That was a big point of contention earlier, but I'm not sure I have seen it pan out in a way that matters.
I'm torn on the commentary of the failure of governments. Policy choices have been very hard with more emotion driving choices than analyzed results. I'm personally in the bend of "be overly cautious," but I also get an uneasy feeling in just how little control there is in comparisons. About the only thing I know not to do now, is get cocky for any region and assume that anyone is already through this.
EDIT: Months to be precise.
Splitting not by month, but by season, would be neat to see. But I'm sceptical there are any learnings to be had here.
> ... well enough.
My, "shows trends lines well enough" should be assumed as followed with "such that I do not see anything new in the post charts."
For myself, I am not convinced we know what the world would have looked like given any alternative response. I do not think we should give up. I just don't know what has actually worked. Hard to argue against more testing, though.
Edit: also, to be clear, I am not trying to argue you shouldn't do the charts this way. Rather, saying I probably need some training to align to what you think this makes more clear. And I believe it could just be different licks, and all that.
This is the kind of feedback I was looking for. Thank you.
Is there a visualization that tells the story that deaths due to covid weren't a big issue?
The only way(other than just making up numbers) I have seen this done is to look at the last few months and use the reporting lag to mislead people into thinking there were fewer deaths than past years.
That said, I haven't seen that one attempted. Unless it is to get and show good non uniform the risk involved is.
The main question/story seems to be around just how preventable this was. And on that one, stories are all over the place.
A plot of Case Fatality Rate would go down over time as the number of cases goes up. A somewhat related, slightly confusing example is cumulative deaths on a log plot. A handful of people would look at the line without much focus on the axes or title.
A plot of Covid deaths by age group could make it more likely that young people engage in riskier activities, since it downplays the risk for "most" people---well, except someone else's grandparents who were gonna die soon anyway.
And, in the WTF category: https://images.theconversation.com/files/330304/original/fil...
Yes, sort of. There was a great paper in CHI this year discussing how heterodox communities use visualizations (https://dl.acm.org/doi/fullHtml/10.1145/3411764.3445211). The rough conclusion is that visualizations do not just empower people to express the truth but are also used in skilled ways by heterodox communities.
"The radius of each sector would be proportional to the square root of the death count for the month, so the area of a sector represents the number of deaths in a month."
It's hard to go back and forth between the graph and the legend to tell what year it is. One way to improve the readability is to have the color shift uniformly between one PAST color and one PRESENT color so I can follow the line from the past to the present.
While animation rarely improves graphs, it really can if time is the 3rd dimension your 2d visualization (image) needs. Something like this http://www.climate-lab-book.ac.uk/spirals/ (first gif)
I looked for a CSV of the data that created the chart, but I failed.
And perhaps the polar chart gives a false impression of higher values being much worse than they are, because the area is skewed so large the further out it is.
The CSV[0] comes from the CDC website[1] and you can find it in the code[2].
[0]: https://data.cdc.gov/api/views/xkkf-xrst/rows.csv?accessType...
[1]: https://www.cdc.gov/nchs/nvss/vsrr/covid19/excess_deaths.htm
[2]: https://github.com/mcculley/ObservedDeathVisualizer/blob/mai...
What caught my eye was that while many states almost returned to pre-pandemic death rate, Florida and Texas have been running above average while their governor's are fighting masking mandates.
USA population growth is about 0.6%/year. Adjust total population on a per month basis and show the deaths per population instead of the total deaths.
Would love to see a bunch of age-stratified versions.
https://euromomo.eu/graphs-and-maps/ has similar data for most of Europe and Israel, and they adjust for population growth and a seasonal component (looks like a sine wave of period 1 year).
Wonder what that means. Maybe that everyone who was gonna die from COVID-19 exposure already did. Maybe something else.
We can be sure that "December January February 2022" will show some excess.
What happened was that people were overly careful - in all parts of life, e.g. meeting people, wearing masks (less of a flu season = less deaths) and going to a doctor when sick - hospital s greatly increased their capacity and afaik Germany never had the dramatic situations where people couldn't get oxygen or staff was at the edge of quitting.