The New York Times obtained a CDC document indicating the likelihood of 3000 new deaths of COVID-19 a day by summer, and a total of 200,000 deaths or more.
The White House says
This is not a White House document nor has it been presented to the Coronavirus Task Force or gone through interagency vetting. This data is not reflective of any of the modeling done by the task force or data that the task force has analyzed
Additionally, White House officials say they are using a “cubic model” prepared by the Council of Economic Advisers, led by Trump adviser Kevin Hassett. Hassett is the fellow who predicted a Dow Industrial average of 36,000 just before the tech crash. The “cubic model” is said to predict that deaths will drop to zero by May 15.
Let’s take a look at that alleged CDC document.
The header designates the presentation as being for or by the Interagency task force, for a videoteleconference (VTC) presentation. The logos for the Departments of Health and Human Services, of which CDC is a part, and the Department of Homeland Security, of which FEMA is a part, appear on every slide. So it looks like an official presentation, although it might be a draft. I’ll leave out the footer on subsequent slides.
The presentation has some other interesting information, in forms useful for tracking the epidemic. Looks legit, from what I know of New Mexico’s situation.
I didn’t realize that positive test results were going down in Georgia and Florida. Good for them. We’ll see how long that holds. (I’ve left some slides out.)
Now, from the Data and Analytics Task Force (Reporters, who are these folks? And what model are they using?)
These two are marked “Official Use Only.” They are where the scary numbers come from.
We don’t know enough about the assumptions in this model to evaluate it, but the numbers don’t look that unreasonable to me, particularly with states easing restrictions. If we keep going the way we’re going now, with no big increase, we’ll hit 100,000 deaths by mid-May. That’s four months from the start. So two months more with increases from the rallies now being perpetrated by rightwing funders?
There’s more to the presentation. Check it out.
DAVID ANDERSON
Just an FYI on the “cubic model”; to get to near zero deaths by May 15th is to have very few if any newly infected folks after late last week. Covid kills slowly.
WaterGirl
@DAVID ANDERSON: Can you say more about that?
MisterForkbeard
This model is the same one that was described as a “catastrophic miss” by white house insiders and was specifically formulated to justify re-opening the economy.
wseattle
I wrote this over at Drum’s place but it fits here too. The big scary numbers should not be too surprising.
I think there’s been a lot of confusion around models trying to predict the time course of the pandemic. Ignoring explicit timing details, the following things are clear: the virus is quite contagious and will spread quickly if unchecked; U.S.-style lockdown is enough to reduce R to near 1, but not below 1 to shrink the number of infections (outside of NYC where the infection rate may be on the order of 30% based on preliminary serology tests); the infection fatality rate (not the case fatality rate which depends on testing prevalence) is likely around 0.5%, give or take.
Based on these reasonable estimates, the virus can be expected to spread throughout much of the population over the coming months to years barring continued lockdowns, more robust lockdowns with testing/tracing, or a vaccine. As a significant portion of the population gets infected like in NYC, it will become easier to reduce transmission even if numbers are well shy of herd immunity. Rough numbers suggest that unless something big changes in the coming months, we should be prepared for on the order of 500,000 U.S. deaths (320 million people x 30% infection rate x 0.5% mortality rate). Such rough estimates don’t say when the deaths will occur, but this is a reasonable baseline for people to come to terms with the scale of the problem. Obviously, numbers could be even worse until the other 70% of people are reliably protected.
None of my estimated numbers are firm… e.g. I see the Cuomo actually reports that 21% of NYC tests showed antibodies rather than 30%. But this just implies that the 0.5% fatality rate may be too low and the scale of the problem is more or less unchanged.
dmsilev
Let me guess. They put in the deaths vs. time series into Excel and used the polynomial regression tool, and then did what every basic book on data analysis tells you not to do and blindly extrapolated the polynomial beyond the range of the data. And, since the particular shape of the curve happened to yield a negative cubic term, that pretty quickly forces the extrapolated curve to zero in the near future.
David Anderson
@WaterGirl: Tomorrow’s post :)\
But basically we know that COVID takes 5 days from being infected to being symptomatic, and then another 7 to 8 days to go from symptomatic to hospitalized (12-13 days) and then several more days in the hospital to death. All of those are medians, so the people who are dying today were infected in the middle of April. The people who will be dying on May 15 are probably infected between April 24 and April 30.
It takes a while for a cohort of people to move through the entire COVID-19 possibility chain. Deaths lag infections by a while.
Keith P.
They’re going to get some lackey to say most of the deaths have a different cause than COVID
khead
If there were any open casinos I’d bet $100 on OVER 300K deaths by Labor Day.
dmsilev
@WaterGirl: Take the time needed to go from infection to hospitalization, through ICU and ventilator support, to death. It’s typically a few weeks. For there to be no new deaths beyond ten days from now, the number of new infections would have to have stopped a week or so ago.
Kelly
The parents of a river trip buddy were diagnosed with the virus today. First cases among my friends. She is asymptomatic and he says he feel OK but has a runny nose. Elderly, live in an assisted living community. Early diagnosis seems like a plus.
My buddy and I have been doing river trips since 1980. Fingers crossed. The virus has been light in Oregon so plenty of medical capacity.
WaterGirl
@David Anderson: Thank you! Looking forward to reading what you put up tomorrow.
randy khan
Did I hear that right? A cubic zirconia model?
For the reasons that Mr. Anderson mentions, any model that has deaths going to zero by May 15 almost certainly is way off.
debbie
@dmsilev:
I don’t think you need to know anything about anything to know that zero deaths in 11 days is just plain ridiculous.
Omnes Omnibus
@randy khan: I am not a data guy by any means, but getting to zero deaths in the next two weeks is patently absurd.
Gravenstone
@DAVID ANDERSON: Yup, basically no one currently I’ll can die. And basically no one else can get sick. Let us all know how that works out for you guys.
Either that, or they neglected to mention they meant 5/15/2021…
rikyrah
dmsilev
@debbie:
People with that on their resume are rejected by this White House as being overqualified for the position.
jl
Thanks for highlighting the report, Cheryl. I am not sure the map looks reasonable for California, which has areas that CDC claims are steady or rising when maps put out by local public health departments show, for example, 7 out of the 9 SF Bay Area counties clearly dropping. Problem is that there are big spikes jutting out the epidemic curve for SF Bay Area that seem to indicate big bursts of community surveillance testing. How well did the CDC people adjust for that?
Until we understand what model they used, and what scenarios, really hard to understand what the forecasts mean. Are they assuming that the states that are reopening too early will forge ahead with their reopens no matter what?
For the forecasts, I don’t know what ‘range estimates’ means, that is some sort of uncertainty interval, but doesn’t point to a specific thing, like a confidence interval or a Bayesian credible interval. Whatever they are, the ‘range estimates’ are very wide, and consistent with much better and much worse outcomes than the point forecasts. (that mismatch in accuracy in the point estimates is so odd, I wonder if I am just making a blunder in interpreting that graph.)
How come the model seems to be much less well calibrated for deaths than for cases? The model seems very biased toward lower end of expected deaths, but not for cases. I don’t understand that at all. Looks more like a presentation designed to troubleshoot a preliminary attempt at model calibration than a finished product. Anyway, I can’t make it all add up. Maybe someone with more inside info on what they are doing can help.
And, ‘cubic’ model? What the blank is that? I guess time for the epidemiologists to try out stock market forecasting.
schrodingers_cat
By my quick mental calculation if we have 3000 deaths a day we will be up to 150,000 dead by the first week of June. 200,000 seems like a low estimate.
schrodingers_cat
OT: Does anyone have Samsung Galaxy A50 and how do they like it? I currently have MotoG 5.
moops
@Omnes Omnibus: It has no reality in it, but that is not important. It tells Trump and his cabal what they want to hear.
Jeffro
When we are closing in on the 500,000 dead estimate by Labor Day (per wseattle’s comment), and going back into strict lockdown just before school/college/football seasons, all I ask is that the national snooze media at LEAST VISIT SOME DIFFERENT MIDWESTERN DINERS for comments from ‘real’ Americans.
I think a half million dead Americans is too steep a price to pay for better reporting, but maybe that’s just me.
And I hope they ask those different diners: exactly what about the trumpov administration’s “approach” to handling this crisis inspires them to give him another four years? Let us know in detail, please.
MattF
@dmsilev: I thought— ‘Some sort of mis-applied polynomial fit’. Well, since Excel does it, what more do you need to know? Bah.
For those who don’t know… polynomial fitted curves are notoriously wiggly, so one has to choose between a not-so-good fit and a fit with too many undetermined parameters. Of course if you’re a Trumpian scientist you just try different stuff until you get the answer you-know-who wants.
Mary G
I’ve been meaning to send Anne Laurie a link to this Children’s Hospital of Philadelphia PolicyLab site where you can enter your county and they show you their predictions of how things will go in the next 120 days with different levels of social isolating. I don’t know anything about statistics or what model they’re using, but it’s interesting to see how even partial mitigation helps.
It says, for example, that if we in Orange County, CA continue sheltering in place as we are, which they say is 65% or more of the population, cases will disappear entirely around July 15. If we go to 50% on May 15, cases go down to low single digits by August. If we go to 33% on May 15, we go up on a shallow curve until there are three times as many when the model ends in August. If we wait until June 1, cases will slowly go down to not quite zero mid-August at 50%, and still go down but not near zero at 33%. So two weeks makes a big difference. Of course, I don’t think people will hold out that well. The beach really is almost a religion.
Also, too, we currently have only 52 people in ICU in a county of 3 million+ so it’ll probably be politically impossible to get people to cooperate if it’s only going up to 150 or so. We have 700 ICU beds, so it looks like medical system overwhelm isn’t going to happen by factors of a lot.
But all of this is just guesswork because we still don’t know hardly anything about this disease yet, or what people will do. Some of the protesters are anti-vaxxers who will die to prove their point.
Redshift
It is darkly hilarious that the White House thinks this statement is a blow against the credibility of this document.
WaterGirl
@Jeffro:
I just wanted to see that again.
WaterGirl
@MattF: It’s probably easy if you work backwards from the answer you want, no?
Geoboy
Just checked on Wikipedia. Total US armed forces killed/missing in World War II was 407,316 over 3 1/2 years. Trump looks to do more than that by Election Day (eight months). Hell of a job, Donny!
Grd0Hro
Per today’s Washington Post . . .”White House officials have been relying on other models to make decisions on reopening, including the IHME model and a “cubic model” prepared by Trump adviser and economist Kevin Hassett and the Council of Economic Advisers.
On Monday, however, the IHME model — widely used by states and heavily relied upon in the past by the White House — also revised its deaths significantly upward to reflect the reopenings in several states.
The IHME model — created by the Institute for Health Metrics and Evaluation at the University of Washington — is now estimating that the United States will reach nearly 135,000 deaths by August 1. That number is significantly higher than its mid-April estimate of 60,308 deaths.
The IHME’s new higher projections “reflect the effect of premature relaxation of restrictions,” said its creator Christopher Murray. “In this era where those mandates are being relaxed, people should be aware the risk of infection is still there.”
The number of deaths estimated by the IHME model by June 1 is still much lower than those in the draft government report. But the IHME model is considered among the more optimistic projections.
Even more optimistic than that, however, is the “cubic model” prepared by Trump adviser and economist Kevin Hassett. People with knowledge of that model say it shows deaths dropping precipitously in May — and essentially going to zero by May 15.”
https://www.washingtonpost.com/health/government-report-predicts-covid-19-cases-will-reach-200000-a-day-by-june-1/2020/05/04/02fe743e-8e27-11ea-a9c0-73b93422d691_story.html
dmsilev
@MattF: It’s the sort of thing where, if I see it on a freshman lab report, it comes back to the student circled in red ink with a big “NO” written over the graph. Or the modern electronic equivalent.
I hope I’m wrong with that guess, but I kind of doubt it.
Roger Moore
@Keith P.:
If we have hundreds of thousands of deaths, it’s not going to matter what they say. People will know someone who died, and most of them won’t accept, “Oh, grandma died of something other than COVID-19, even though she had all the symptoms” as an answer.
rikyrah
Mary G
Insert obligatory Twain quote about lies, damn lies, and statistics.
MattF
@WaterGirl: Right. There’s a well-known source of bias that comes from ‘fixing’ your data analysis until you get the ‘right’ answer. Actual scientists go to great lengths to avoid this.
schrodingers_cat
@dmsilev: Garbage in Garbage Out.
Redshift
@Jeffro: I saw some wingnut on a Twitter thread a while back who was sure it must be media people causing these outbreaks across the country, because obviously they are the only ones traveling all over, to do these stories. You can’t make up anything crazier than what they actually believe.
Jess
I wonder why central Mass (where I live) is dark orange–the rate of infection has been relatively modest here compared to eastern Mass, and while we have our share of morons here, they don’t seem to be as abundant or as extreme as they are in other places. People do seem to be pretty good about precautions here.
rikyrah
rikyrah
Mary G
@Roger Moore: I wish I thought that was true. The majority of deaths are low income, many black and brown or in jail. I heard my neighbors talking over the fence blaming poor people who don’t take care of themselves or are lazy nursing home employees who are killing all the good upstanding old white people. And they’re Republicans who were definitely not all in on Trump. The plethora of yard signs they had up for GWB, McCain, and RMoney were absent in 2016.
Jess
Josh Marshall is skeptical about the CDC report, BTW.
https://talkingpointsmemo.com/edblog/trying-to-make-sense-of-that-leaked-cdc-model
scav
A cubic model from this White House? Doesn’t this mean whatever words dropped out of the nearest blockheaded beach blond?
moops
@rikyrah: this blocking of testifying has got to be stopped.
Jeffro
@Redshift: Hofstadter would be like, “it was meant to be an ANALYSIS, not an EFFIN’ PLAYBOOK!!!1″
Hey GOP – when you hammer on people’s fear and paranoia buttons for 2, 3, 4 solid decades, what did you expect? Hell, what did WE expect?
“The media is ‘seeding’ the virus amongst us real ‘Muricans and now we cain’t have our rallies…and worse yet…Applebees is closed!!”
We really do need to get out of the way and let them pound hydroxywhatever, drink bleach, rally and hug their brains out, and just sit tight.
JPL
@rikyrah: Well of course they are. Red State govs will also block updates on the spread.
cain
@Keith P.:
Yes pages and pages of obituaries suddenly, but no, it is not corona – we’ve just been hit by a large systemic plague of different diseases. /s
Jeffro
@rikyrah: Like he/they always do.
Dems really ought to step up and publicly call it out: “Will NO ONE STAND UP FOR AMERICA, and simply come down here to testify about all the AMAZING work that the trumpov maladministration has been doing these past 4-5 months, to save us from this pandemic? The president*, after all, is now saying he knew it all along…let’s HEAR MORE ABOUT THAT.”
“Also…WHERE DID ALL THE HIJACKED PPE GO, JARED?”
Tenar Arha
@Jess: Prisons, jails, nursing homes, rehabs, assisted living locations maybe
Roger Moore
@rikyrah:
This is one of the days when I’m not so proud of my hometown.
Benw
@WaterGirl: I think this. Someone set their fit boundary condition at 0 deaths in mid May and tried to fit the current curve. It took some weird non physical cubic polynomial to reconcile everything
lamh36
Wanna hear some real-time COVID tesing news?
So, at my hospital, we have 2 POCT Abbot COVID IDNOW machines. We were doing like up 20 or more of those a day, but when the false negative results became a thing with Abbott, we implemented the Cepheid COVID PCR method. Cephied is 50 min, Abbott is 15min. So now we have to do Cephied testing for ER, Same Day Surgery and Outpatient surgery along with the IDNOW testing if ordered by the physicians.
So yesterday we has at least 80 Cepheid test between inpatient, surgery (inpatient and outpatient) and ER specimens. We have 8 Cepheid modules (although our machine is able to allow 16. But we have to get the additional modules installed) ONLY, and we also are doing CDIFF toxin PCR as well. So imagine 80 test a day average, only 8 hourse each shift. ER and SDS are priority, so if we get an ER specimen then we have to do those before the surgical specimens.
Now though, the Pathologist Director wants up to leave 2 of the 8 modules always open for ER specimens! Which now means we only have 6 modules to use for the other Cepheid COVID testing AND the Cdiff which by the way is a 45 minut test!
We are “in the line” to get more modules, but the other problem now, aside from increased Cepheid workload and not enough modules, is that the DOD has sequestered supplies for the Cepheid device and they are the ones siphoning out those supplies to providers. So, we are likely to run out of supplies by the end of this weekend if DOD doesn’t allow us to receive more!!
Roger Moore
@MattF:
Good scientists avoid this. Unfortunately, science, like any field of endeavor, has people who are always looking for a faster, easier way of doing things. The whole process of manipulating data analysis to get favorable results is well enough established that is has a name: p-hacking.
cain
Fixed.
jl
At this point, I am not even sure what the point of unconditional forecasts are. It is not like the initial big upward wave of an epidemic towards the peak when not really much that can be done but shut everyone up in their houses as much as possible, and hope the peak can be squished down somewhat. Seems like too much depends on policy decisions made over next few months.
For example, does this country have the ability to get adequate clinical help to nursing homes so they don’t turn into charnel houses as soon as they get one case? Will national guard be called out to force workers at gun point into meat packing plants to keep hamberders flowing? Will states that reopen, responsibly, be able to get up an adequate outbreak control program? Will states that try to reopen while they seem to be still in middle of an epidemic wave have the sense to pull back?
Steeplejack
@schrodingers_cat:
Maybe not what you’re looking for, but I have a Galaxy S10e. Love, love, love it. Upgraded a year ago from a four-year-old first-gen Moto X.
TriassicSands
@DAVID ANDERSON:
Not at all, David, you see, it’s just going to “magically” disappear.
POOF! No more COVID-19.
The world bows before the omniscient, Donald Trump, slayer of SARS-CoV-2.
Meanwhile, back in the real world, cue Jack Nicholson repeatedly shouting “YOU CAN’T HANDLE THE TRUTH!”
That message goes out to the American people, millions of whom are proving themselves to be remarkably, but predictably, immature and irresponsible.* Whiners With Weapons™ across the country who are unwilling to do their part in facing this pandemic. The Constitution is not just a suicide pact, but millions of Americans are fine with it being a murder-suicide pact.
* This is no surprise at all. If Thomas Jefferson could see what has become of the American people, he’d probably wish he’d never written the Declaration of Independence. He might well recognize that a country with so many ignorant**, selfish, stupid people could not govern themselves and would turn to a authoritarian demigogue. Those same people would lack the discipline to maintain an appropriate response to the pandemic and would gladly sacrifice the lives of others in order to service their own needs. Choosing money over lives has become a part of the American ethos.
** “If a nation expects to be ignorant and free, in a state of civilization, it expects what never was and never will be.” Jefferson was no saint and there are many quotes of his that are far less wise than this one. But this one fits 2020 America.
Roger Moore
@cain:
I thought Lex Luthor was the junior senator from Florida.
Steeplejack
@Mary G:
Useful site. Thanks for the link.
John Revolta
This is not a White House document nor has it been presented to the Coronavirus Task Force or gone through interagency vetting. This data is not reflective of any of the modeling done by the task force or data that the task force has analyzed
HAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAH
I know I for one ain’t gonna take the CDC Report seriously until I hear what Mike Pence, Larry Fucking Kudlow and the Mnuch have to say about it.
cain
Let’s not insult Lex by comparing him to clearly low quality humans. :D Lex is more at McConnell’s level.
TriassicSands
@schrodingers_cat:
Take a look at a Google Pixel 3a. One of its nicest features is Android One — a clean Android installation without the bloatware that comes from Samsung and others, guaranteed fast Android and security updates, and if you care about the camera, it has the best camera available on a non-flagship phone. In fact, according to some reviewers the camera is better than those on phones costing more than twice as much.
Search and read reviews. Oh, and Pixel 3a phones are unlocked and will work with any major carrier (and probably with smaller outfits, as well. I’m using mine with a non-major provider).
moops
sadly, some good CDC scientists are going to have their careers ruined in the coming week.
Dan B
@Mary G: A good friend sent that model. It was from another source and seems to be in circulation to epidemiologists and other health professionals who advise governments. The differences between ending the lockdowns early and delaying was astounding. Opening up the country today could result in over a million deaths. Keeping the lockdiwn in place until mid-August would yield less than 90,000 deaths.
Feathers
@Jess: My understanding is that it is nursing homes, especially that VA hospital where they hid things for as long as they could.
I haven’t been watching the local news, but one of the last times I did there was a woman loudly complaining about her father being moved to another nursing home without her approval. Apparently, Mass was consolidating smaller facilities to admit coronavirus patients who left the hospital but were too sick to go home. The woman was going on about how her father had dementia and would be very confused and she couldn’t visit him to explain. I just wanted to shake her. Yes, this sucks! It especially sucks for you and your dad! But would you rather have coronavirus patients sent back to his nursing home because they weren’t sick enough for the hospital?
Also, we need an emergency appropriation so that caregivers get a living wage working only one job and don’t have to work at multiple nursing homes to string together enough money to live on.
MomSense
I feel such a profound sadness. Anguish? Usually I try to either be busy or find a diversion, but tonight I think I need to just feel it.
Another Scott
Thanks for the link to the report. Some quick thoughts:
Thanks.
Cheers,
Scott.
Bill Arnold
@dmsilev:
In normal usage that I’ve heard in the sciences, a cubic model is exactly that.
If so, then Hassett needs to be tried and publicly executed should be rewarded with wealth beyond his imagination for providing justification for the premature reopening of the economy.
Dan B
Picked up my partner’s brother at SeaTac this afternoon. The flight was the only one at arrivals. Surreal to have a total of 20 people waiting for rides.
BTW they are heading home to Bellingham, WA, otherwise known as headline grabbing, “Why does 2020 need to be worse?” Murder Hornet hot spot. Fun times!
Scamp Dog
@Another Scott: Hey, I’m also an Edward Tufte fan!
LeftCoastYankee
Cubic model means they got all the dice they could find and threw them on the floor and had some intern add up the rolls.
That intern is will be the next interim cabinet secretary.
dmsilev
I went ahead and blindly did a cubic fit, just like a good Republican apparatchik.
https://imgur.com/gRFVD71
Via careful choice of when to start the fit range (March 29 in this case), it’s pretty easy to get it to go to zero on the 15th. Of course, it swings hard negative after that, so we do have a zombie apocalypse to look forward to in a week or so.
TS (the original)
Rachel – speaking of COVID projections and the White House – “It’s pitiful”.
You could say that about every last piece of information coming from the trump sycophants.
Geoboy
@Another Scott: Ah Tufte! Often immitated, never equaled. Just think what Napolean could have done otherwise.
Keith P.
@cain: It doesn’t need to make sense… It just needs to give his supporters something to hang on to.
NYCMT
@Bill Arnold: He fitted cubic splines to the data. It’s total nonsense.
(NYCMT stands for New York City Math Teacher)
Steeplejack
@Roger Moore:
What does the p stand for in “p-hacking”?
Cheryl Rofer
@dmsilev: Nate Silver at 538 is doing the same things you are!
Steeplejack
@Cheryl Rofer:
LOL. From the comments:
“How long until COVID starts reviving the dead?”
“Give it a week or two.”
LongHairedWeirdo
@dmsilev: I honestly don’t know what they’re thinking, but “cubic” model reminds me of a modeling method called the “cubic spline” – a spline is a tool for drawing curves, so the method was to break the graph into pieces and define a different cubic equation for each piece.
I can’t think of anything *stupider* than trying to model an exponential function with a cubic. It’s… sigh.
The hallmark of exponential growth is, it’s growing based on its current size. And in the early stage of an epidemic, you can see how *perfectly* it works – if you have 5 wandering carriers, you’ll probably get 5x as many new infections as if you just had one.
This is why there tends to be a doubling time mentioned. Since it grows twice as fast, when it’s twice as large, its doubling time should be constant.
Lockdown and distancing slow the growth, and slow the doubling time, and that has happened, but there *is still a doubling time* suggested by the data. That means it’s still exponential, and, eventually, its growth will overwhelm the ability of a cubic equation to model it.
Using a cubic equation to model case counts is so astoundingly stupid that it reminds me of the professor quoted as saying “this isn’t right. This isn’t even *wrong*”.
You see, with epidemics, you’re sometimes interested in, “how fast are new cases growing? And, how fast is the *increase* of new cases growing? And, how fast is the increase, of the increase, of new cases growing?” A cubic can get that right no more than 3 times – how fast it’s growing is a quadratic, how fast the rate of increase is growing is linear, and the rate, of the rate, of increase, can be constant.
With exponential growth, *all* of those things are growing, at exponential speed. So when new cases are growing exponentially, the cases are increasing; and the rate of increase is increasing; and, the rate of the rate of increasing is increasing, etc.. The further away from your last known data you get, the far, far, worse a cubic is at estimating it.
For the record, I was not at the point at which I’d have imagined someone using a cubic, to model an epidemic, as a good *fictional* plot twist; it’s so abysmally ill informed it just wouldn’t have entered my consciousness as a possibility.
If that is, in fact, what happened, it doesn’t surprise me that it’s an economist who would work with Trump who did this – misused a modeling tool in a way that’s so stupid, it proves they had no idea what they were doing.
@MattF: In point of fact, that is why the cubic spline was developed – polynomials tend to oscillate too much, so a 3rd degree is an attempt at a “just right” fit.
rekoob
@Steeplejack: @Roger Moore: Without getting overly complicated, p-hacking refers to messing with the parameters of a statistical analysis to aver statistical significance when none (may) exists. I say may because it’s in the context of validating whether your initial premise (the null hypothesis) is likely to be correct.
As George Stigler once said, “we have tortured the data, but it has yet to confess”.
Steeplejack (phone)
@rekoob:
Thanks.