First Section updated For Sorry Ari Fans
Many have asked me to comment on The New Atlantis article “Not Like the Flu, Not Like Car Crashes, Not Like…” by Ari Schulman, Brendan Foht and Samuel Matlack.
My comment is thwwwbbbppppppt! Read on if you want more detail.
Say! Welcome Schulman fans. I see that in response to this column our Ari has tweeted out some clever jokes—more than rivals of my own. Like saying he shouldn’t bother answering criticisms from a guy (me) “who writes like an 8th grader who just read his first Dave Barry column.”
Probably realizing this wasn’t enough, he repeated his mistake by saying his mistake wasn’t a mistake: “The one substantive methodological criticism the author offers is factually wrong, plain and simple. The other lines are not estimates, not smoothed, not modeled. They chart actual reports of actual deaths.”
No they’re not.
Take the blue line below, labeled “2017-2018 flu”. Those are not weekly actual counts of flu deaths, but numbers from the “excess death” model. Flu just isn’t counted that well in death certificates, which is why the have to resort to excess death models.
The funniest line is the “1957-58 Asian flu (with pnemonia).” Those aren’t actual counts, either. Sorry Ari.
Come to think of it, Sorry Ari is a good nickname.
The line is funny because the 58 Asian flu was estimated to kill about 2 million worldwide—at a time when the population was much smaller. Sorry Ari’s line makes it look like a bad case of hemorrhoids compared to coronavirus.
Unlike Sorry Air, we’ve done this right, in two ways. The first is in the perspective post, which compares pandemics through the ages. Post is a month old, mind, but it still holds up. I’ve updated the main plot here. COVID doesn’t beat the median estimates of Swine Flu, though it’s likely to come in around that total.
Here, Sorry Ari, is the way to do it: (This is only one plot of many there.)
Now tell the truth. Does Asian flu look better or worse than coronadoom?
The other way is to compare actual all-cause weekly death counts. We did that for England & Wales and the US. All the details are explained here.
The whiz bang finale is this plot, the per capita death rates through time of the US:
The drops are due to incomplete data. Since it’s still incomplete, coronavirus is going to have to work hard to beat the 2017-2018 flu seasons. But hope springs eternal.
Sorry, Ari. You blew it, and served only to gin up needless fear.
Under the wheels
Do high schools still show those nightmare-inducing driver training videos? You know the ones.
Sally straps herself into a two-ton murder missile on wheels, sets off for the dress boutique. She reaches down to adjust her heels…and…WHAM! Chalk up one Sally—and one innocent family of four, who moments before were on their way to Church to attend a rally for your favorite victim group, but who are now lazing about all over the highway. Or, as Ray Stevens once sang, aaaaaaallllll over the highway.
Five deaths. A mere five of about 37,000 people a year who end their fleshly careers fertilizing the flowers along our nation’s highways and byways.
One would not ordinarily dismiss as statistical artifacts tens of thousands of deaths that turn the stomachs of even grave-robbing vivisectionists. But these are not ordinary times.
For now the dreaded coronavirus is upon us!
People who die from that bug are special. True, a flu victim and a coronavirus victim will often end their earth-bound existences strapped to gurneys sucking oxygen from the only plastic straws still legal. And there will be as many flu victims—each and every year—as there will be coronavirus victims this year.
But dying from the flu just isn’t comparable!
Why? Well, because we have not scared ourselves witless—this is a family blog, else I would have used the more popular, and the much more accurate, word—by flu or car crash deaths. Or heart attack, or cancer, or ebola, or zika, or giant Asians wasps, or indeed deaths of any other kind.
We often tell ourselves campfire stories about the ways to be pushed over the final cliff of life. Theses are excellent ways to relieve tedium and bring eyeballs to advertisements. Only this year we took these morbid tales seriously. We have behaved like sugar-addled children. We don’t dare crawl into bed before peaking under it to see if the escaped maniac ax killer with the hook for hand lurks there.
Coronavirus victims die a martyr’s death. People who die of other causes deserve what’s coming to them, or don’t matter.
Not only that, but everybody is now an expert on the statistics of causes of death.
One flu over the regression
The reader may be surprised to learn that nobody knows how many people died of the flu last year. Or the year before that. Or any year.
Nobody know the exact number, that is. The best we can ever do is an estimate of the real number, a guess from a statistical model.
One paper explains it thusly:
In non-pandemic years, influenza-associated death is mainly restricted to the elderly and people with underlying chronic illnesses. However, analyses of death certificates show that clinicians often do not attribute influenza-related deaths to influenza, but rather to a pre-existing underlying condition. Influenza-associated deaths may therefore be hidden not only among cases of pneumonia but among other causes of death such as cardiovascular events or metabolic disorders. Hence, all-cause mortality has been found to be more complete and accurate for assessing the total impact of influenza on mortality.
In this pandemic year, as we have seen, the underlying condition is ignored and the DBI (death batted in) of coronavirus is instead credited.
Since the true tally of flu deaths isn’t known, it has to be estimated. The idea is to plot up all deaths by time, which looks like this, the CDC’s latest weekly numbers.
Ignore the plummeting at the end. That’s probably because these are government-compiled numbers; as such, they suffer from delays. Almost certainly these will pop back up in time, after government employees diligently return to (I hesitate to write it) work.
The yearly cycle with deaths peaking in winter is plain. We had a peak in January of this year, after which the curve started its downward course, as usual. Lockdowns began in mid-March, after the decrease began, on the downslope. Suggesting that lockdowns were not as efficacious as those who used them to destroy the lives of millions are hoping. Imagine if they were wrong about that!
Never mind. The CDC might adjust the January-March numbers higher. Who knows. They are beside the point—for the moment.
What happens with flu is that a model is fit to the yearly cycle, and to the upward trend which you can also see. That’s caused by population increase. These are not per-capita numbers.
The result is a model that emulates but greatly smooths the yearly cycle and upward trend. Epidemiologists argue over the niceties of these regressions; some use weather as a variable (which I like), and some don’t. Some use FFTs (fast Fourier transforms), some don’t. And so on.
The numbers that “stick out” above these smoothed series are called—and I swear I am not making this up—“excess deaths.” It’s not just flu, but “excess deaths” are estimated for a host of supposed causes.
Anyway, these excess deaths are summed for a season, and that total, with an accompanying plus or minus bound, which we saw before and we you can see again below, becomes the estimated “flu deaths” for the year.
Our current year isn’t in that plot yet. More CDC delays. Actually more than delays, which we’ll discuss tomorrow. Has to do with juicing the numbers.
This plot is yearly, so it’s peaky and spiky. But you could do the same for weekly data, and as anybody who has ever used a statistical model will tell you, it will be much smoother. We’ll see examples below.
Up until this year, these kinds of methods for estimating flu deaths didn’t receive much public scrutiny, because people usually don’t frighten themselves into manic tear-fests over flu. Not only that, but the numbers were not that crucial to know with any exactness. Ask yourself whether you’ve ever cared before in your life how many people died yearly of the flu.
We finally come to The New Atlantis article, the gist of which is captured in this chart (they do one similar for New York, which has the same shape and faces the same criticisms).
Weekly reported deaths for the dreaded coronavirus are in red. Heart disease and cancer weekly averages, i.e. estimated, deaths are on top. Car crash estimates are at the bottom, and other flu year estimates are in the middle.
The writers of the article say “It’s about the spike.”
It sure is. It can’t be compared with smoothed estimates! But it is compared. The comparison misleads.
Comparing actual counts, even assuming those counts are accurate, which they almost certainly are not, and which are likely juiced, with smoothed estimates from models is always bound to make the actual counts look worse in comparison.
Where by “worse” I mean more variable. More variable data always has the chance to make greater flights, up or down, than less variable data. This is the definition of variability.
The way the picture is constructed makes the dreaded coronavirus look worse, and not only worse but far worse, than the 1957-1958 Asian flu. This is asinine. Asian flu, the estimates tell us, slaughtered some 2 million souls. Two million.
It’s not over, but the dreaded coronvarius looks like it might tie with 2009’s Swine Flu, which isn’t even pictured. Swine flu estimates are between 152 and 575 thousand dead, with a median of 364 thousand. The 1968 Hong Kong chop suey fluy isn’t pictured, either, which rubbed out an estimated one million.
These omissions matter, as does the breathless hand-wringing daily reporting of dreaded coronavirus numbers. Here’s an illustration why. I had an exchange with a well know Twitterer and blogger who was passing around exaggerated stories of horror about the dreaded coronavirus, and in which I told him the Asian flu was much worse. The gentleman was incredulous and said he hadn’t even heard of Asian flu. Yet he was sure, and suggesting to his readers, the dreaded coronavirus was to be our doom. I said this was nonsense. He has since ceased communicating with me.
Imagine if the Asian flu’s numbers were reported daily and for months on end, accompanied by nonstop media coverage. The terror which strikes our hearts would be four times worse, at least.
Why we chose this bug to collectively lose our minds over, and not other, deadlier ones, is of course the real question. Let us know your theories below.
Echo echo echo
What makes these criticism of the NA article odd, is its authors made the same ones!
They acknowledge flu deaths are estimates and admit “Determining this number is not as simple as counting up death certificates listing influenza as the cause.” The own up to difficulties in the plot, too:
Instead of using estimates, on our U.S. chart, we have shown one line for deaths in which influenza was listed as a cause (which undercounts influenza-associated deaths); and another line showing deaths in which influenza or pneumonia was a cause (which overcounts flu deaths, as many pneumonia cases are not caused by flu).
The also fess up to the over-counts in dreaded coronavirus deaths “Skeptics of the reported death counts have argued that most people who die with the virus don’t die of the virus. This is surely true in some cases, as it is with the flu. In many deaths, webs of causation are tangled.”
But then they fret about possible under-counts, saying something about bodies being found in the streets and in homes in greater numbers than usual. They forget everybody is locked down and cowering inside their apartments. Folks aren’t checking on their neighbors or on lumps on street as much as normal.
They also admit that “Comparing [dreaded coronavirus] deaths to, say, an entire year of deaths from car crashes or influenza is not meaningful.” If it’s not meaningful, they don’t explain why the still do it.
And then they come back to their central mistake: the “spike”. “Amid the statistical noise is a powerful signal. The question is whether we choose to see it.”
Perhaps the most noticeable feature of both graphs is the Covid-19 spike — the rapid growth in deaths since the pandemic began. Car crashes, by contrast, show little variation week to week. And even compared to past flu seasons or pandemics, the rate of increase in Covid-19 deaths is markedly faster.
Forget their dismissing car crash and other deaths as unimportant because they happen more-or-less regularly. Why deaths that don’t make headlines don’t count as significant deaths can only be explained by reference to our panic. Anyway, you can’t help but see the daily counts spike amidst all those smoothed estimates!
It’s the wrong comparison. Making the right one is impossible, since daily deaths for the other pandemics are nowhere available. Their chart is no different than a ghost story.
To support this site and its wholly independent host using credit card or PayPal (in any amount) click here