The Volokh Conspiracy
Mostly law professors | Sometimes contrarian | Often libertarian | Always independent
"CDC Estimated a One-Year Decline in Life Expectancy in 2020. Not So — Try Five Days"
From Dr. Peter Bach at StatNews:
The Centers for Disease Control and Prevention made headlines last week when it announced that Covid-19 had reduced the average life expectancy of Americans in 2020 by a full year…. [But t]he pandemic's appalling toll could not have reduced life span by nearly that much. My own estimate is that when Covid-19's ravages in 2020 are averaged across the country's entire population, we each lost about five days of life.
The CDC's mistake? It calculated life expectancy using an assumption that is assuredly wrong, which yielded a statistic that was certain to be misunderstood….
People understood [the CDC's report] to mean that Covid-19 had shaved off a year from how long each of us will live on average. That is, after all, how people tend to think of life expectancy. The New York Times characterized the report as "the first full picture of the pandemic's effect on American expected life spans."
But wait. Analysts estimate that, on average, a death from Covid-19 robs its victim of around 12 years of life. Approximately 400,000 Americans died Covid-19 in 2020, meaning about 4.8 million years of life collectively vanished. Spread that ghastly number across the U.S. population of 330 million and it comes out to 0.014 years of life lost per person. That's 5.3 days. There were other excess deaths in 2020, so maybe the answer is seven days lost per person.
No matter how you look at it, the result is a far cry from what the CDC announced.
It's not that the agency made a math mistake. I checked the calculations myself, and even went over them with one of the CDC analysts. The error was more problematic in my view: The CDC relied on an assumption it had to know was wrong….
Read the whole thing, which strikes me as persuasive (and which explains the almost certainly incorrect assumption on which the CDC relied); and while I'm far from an expert on the subject, I ran it by some professors who are much more knowledgeable than I am, and they generally agreed with Dr. Bach's analysis.
Editor's Note: We invite comments and request that they be civil and on-topic. We do not moderate or assume any responsibility for comments, which are owned by the readers who post them. Comments do not represent the views of Reason.com or Reason Foundation. We reserve the right to delete any comment for any reason at any time. Comments may only be edited within 5 minutes of posting. Report abuses.
Please
to post comments
I recommend More or Less: Behind the States
It's a bit British-related, but has done wonders for my statistical literacy in the past few months.
Glad to hear! :-).
OMG!!
The CDC makes up numbers for political purposes and call them Science!!
Damn; whodathunkit?
L,
The numbers are not made up. They are the result a procedure that is well know to overestimate deaths in periods of transient mortality.
So you could say that they are more the result of laziness with an admixture of incompetence than outright lies or exaggerations.
DM
It doesn't overestimate deaths, it underestimates life expectancy due to the transience. But it is well understood why it happens during period with greater than normal death rates.
It over estimates death rates and thus underestimates life expectancy. Those are 100% correlated quantities. But the simple way up put it it probably clearer to most people.
"It over estimates death rates"
But again, why do you say that? Can you 100% tell me that the death rate in 2022 will be lower than that in 2020? The problem is, you can't do that.
What death rate do you put in for 2022? For 2023? For 2024? Based on what data? What estimates? What "models"? Did you model in the hypothetical war deaths in 2025? What about the neo-pandemic in 2030?
These are the problems you run into once you move away from the hard data, and a well precedented methodology, and into "models." Does the current system have potential flaws? Sure. But they are well understood, and everyone uses them, so the data is consistent? Once you start "guesswork estimates" of future death rates, the data becomes basically useless. Every country will overestimate its own future life expectancy.
"why do you say that? "
Because the statements are 100% equivalent.
I could ask you the same question about Underestimate life expectancy. Why do you say that.
Estimates are estimates. They must be used in most human endeavors to move from the past into the future.
I don't know if you use many extensive estimates in your work work. They are an integral part of mine. The use of estimates embodies risk. Therefore, the sound use of those estimates includes mitigation strategies both with regard to time and cost in tangible resources.
But you need to base your estimates on something. You need to base them on hard data.
These life expectancy tables are hard data. They take clear data. Death rates, at a given age, in a given year.
They don't estimate. They don't project. They don't use assumptions. They are hard data. And that's why they are especially valuable.
Right, but you're modeling. If you think the "clear data" for death rates "at a given age, in a given year" is useful, you must think the "given year" is a proxy for something useful. Otherwise why not use 1919 as opposed to 2020? Ostensibly you think 2020 is a better proxy for 2021 than 1919 is. But that's a modeling decision.
"Right, but you’re modeling."
No, you're actually not modeling. That is the fundamental mistake being made here. "Life Expectancy" as it is defined here is a statistical measure. Not a model. I need to repeat this a few times. It is not a model. It is a statistical measure.
It (Period life expectancy at birth) is defined as the mean length of life of a hypothetical cohort assumed to be exposed, from birth through death, to the mortality rates observed at a given year
Again. Not a model. Statistical measure.
Period life expectancy isn't a model, but the decision to use any given period (for any reason) is modeling. You would not expect someone to say that COVID decreased the life expectancy of people using a 1918 period benchmark.
Period life expectancy is a great statistic except when it is not. Periods with unusually high mortality that ends abruptly (like pandemics and wars) are not reliable. The period life expectancy in the UK in 1917 was 51.4 for men. The next year it was 44.6. The next year it was 52.5. All that is, as you said, hard data. So how much did the events of 1918 decrease the life expectancy of the average male in the UK? You would never attempt to answer that using only period life expectancies; you'd want a cohort expectancy, too. The period expectancy is being dragged by the very event you're trying to evaluate the effects of.
Again. I don't know how many times I need to say this.
Period. Life. Expectancy. Is. Not. A. Model.
It is a STATISTIC.
If you're using it as a model, you're doing it wrong.
If you want a model, go to Cohort Life Expectancy and their future estimates. The Social Security tables are here.
https://www.ssa.gov/oact/TR/2012/lr5a4.html
But again. Repeatedly. Period life expectancy is not a model.
"These are the problems you run into once you move away from the hard data, and a well precedented methodology, and into 'models.'" Your dichotomy is a false one.
All projections into the future are based on models. The only "hard facts are the number of people whose deaths were attributed to COVID-19 each month. And even those carry a substandial uncertainly when one reports that the deaths were due to COVID-19.
Moreover, the use of a methodology with well known flaws is not excused just because it is frequency used in circumstances in which those flaws have a minimal effect. The deliberate use of an unsound methodology in circumstance in which the flaws make a large difference in the analysis is borderline academic misconduct UNLESS the limitation is clearly pointed out and discussed as part of the analysis.
"All projections into the future are based on models."
That's the flaw here. It's a misreading of what it actually being said. The life expectancy tables are not really a "projection into the future". It is defined as follows. "Period Life expectancy at Birth is the mean length of life of a hypothetical cohort assumed to be exposed, from birth through death, to the mortality rates observed at a given year."
No, that definition IS a modeling assumption. Does it have physical meaning? Not necessarily.
This entire issue is central in scientific data analysis. And if you quote, probability p and I quote (1-p), we are not talking about 2 different things. The physical interpretation of either is founded on a mathematical model that we are calculating the characteristics of.
That definition makes it a statistical measure. Not a model.
The error is in assuming it is a model.
You don't understand statistics of the physical world. You think statistics is arithmetic. If you read the actual report. VSRR-10, you will see that the model of the calculation has many underlying assumptions that may lead both to over and underestimates of the period life expectancy. As a lawyer, you may not understand that these tables are projections into the future, but that is their principal practical use. Insurance companies make money because they rely on on the tables being an accurate projection into the future for large populations with the same underlying characteristics as the population for which the statistic was applied. It is all a model of the future in the large, absent that meaning the tables are just arithmetic written in a particular neat form on a page.
"As a lawyer, you may not understand that these tables are projections into the future, but that is their principal practical use."
No. This is incorrect.
If you want to forecast future use, you use the future estimates in the COHORT LIFE EXPECTANCY, as below.
https://www.ssa.gov/oact/TR/2012/lr5a4.html
Say it with me again. Period Life Expectancy is not a model.
AL,
You are clearly not knowledgeable about this business. But you are thick headed. You do not understand what a model is and what arithmetic is.
I cannot say more except that this is my professional business and you are incorrect. Try reading the entire report from CDC including the technical appendix. Come back when you have done so. and list all the Assumptions in their statistical model.
Verbatim from the CDC report: "Life expectancy at birth represents the average number of years that a group of infants would live if they were to experience throughout life the age specific death rates prevailing during a specified period. In the first half of 2020, life expectancy at birth for the total U.S. population was 77.8 years, declining by 1.0 year from 78.8 in 2019"
The problem here isn't with the CDC report - it is that reporters were skimming so fast they read the second sentence while missing the sentence immediately prior to it.
This is a technical report. There is a limit to how much the CDC can be expected to dumb it down for the innumerate and still be valuable to its intended audience.
Don,
Have you actually read the report. There are a couple assumptions that lead to potential error. But they all relate to the collecting of the death data to date. Sampling error. Just like you get for population statistics
After that, they're just plugged into some simple equations. They don't project at all. Read the report and the technical note.
"The life expectancy tables are not really a “projection into the future”. "
Kind of a shitty name, then.
Fertility rates aren't really how fertile women are. It's really a representation of birth rate in a specific population.
Gross domestic product doesn't really represent "product"
And you shouldn't really expect that your experiment will result in the 'expected value', either!
AL: FWIW, my degree is in statistics, and what you are saying in this thread seems spot on to me.
It is well understood to people who understand statistics and the methods.
However, it's not well understood by the general masses, even among the educated class.
It's a textbook example of a technical truth that is clearly deceptive when used in this way.
Incompetence *is* fraud.
Sorta makes you question the Global Warming numbers, doesn't it?
CDC is just a nest of Democrat spies and Deep State agents. I suggested Trump fire all of them, start over. Fauci is a notorious quack, who is an agent of the Clinton campaign. Trump was not getting good advice. But no one is listening over there.
Fauci does not work for the CDC.
True, Fauci works for the Democrat Party and the tech billionaires. They made $1.3 trillion more in 2020, from the quack lockdown, than in 2019.
I may regret having asked this, but how would anyone have made $1.3 trillion from the lockdowns? Who would have benefitted from the lockdowns?
Don't bother asking Behar that question. You'll just get more blather back from him.
That's lawyer denier talk.
You have proved me correct. You answer with ad hominem blather (intelligence denier talk, if you prefer).
That doesn't answer my question. Don, you're probably right, but I'm still curious as to where he's getting that number; I suspect from his ass.
Ignore the part about not answering my question; I posted that before I saw that you did.
Obviously Amazon was busy, as were Door Dash, Uber, Fed Ex, UPS. Most surprising was the biggest profit was made by the Walton family, owners of Walmart, from the mail business at Walmart and at Sam's Club, with its lower cost. They made a 7% in higher profit. Best of all, they got rid of Trump from the lockdown destroying the market and the economy. Trump caused massive wage pressure from shutting down immigration. The lowest paid workers had the biggest raises of any group, cutting into their profits.
Krychek. This publication is a far left propaganda outlet for the Chinese Communist Party. $1.3 Trillion in the US. $3.9 if you include the Chinese billionaire scum.
https://www.msn.com/en-us/money/markets/billionaires-have-added-2413-trillion-to-their-net-worths-during-the-pandemic-a-44-25-increase-from-march-2020/ar-BB1e1eK4
Very good. Be careful not to link to this post from Twitter or Facebook, they'll ban you for speaking too much truth about Covid.
Sure, even on its own terms, the number was a joke. They took the worst month of the pandemic, basically, and extrapolate what would happen if it went on forever.
More to the point, if you didn't get Covid, or did, but it was a mild case, it didn't take any time off YOUR life, at all.
It's not like everybody is going to die a week sooner than they were. Most people were not effected at all, while the people who did die are already dead.
Mostly when you think of things that reduce your life expectancy, you're thinking personally; Nobody who wasn't involved thinks of the war with Iraq as having taken x number of months off their life, though people certainly died in it.
I went through chemo back in 2010, THAT took some years off my life, in the literal sense that I can now expect to die several years sooner than otherwise due to the health consequences. I had covid a month ago, it probably took nothing off my life expectancy, at all.
Actually, it very well might have. But this calculation is too early to measure that.
It was a pretty mild case, comparable to a bad head cold, so probably not.
But, sure, it could be that we discover to our horror in a few years that everybody who'd had Covid is aging twice as fast as normal, or is subject to a dramatically higher probability of lung cancer, or some such. No particular reason to expect that, though.
If you still have no sense of smell, it may be lodged in your brain.
David,
In fact it may even have increased your life expectancy by giving you a modicum of immunity to a more virulent coronavirus. There is no way of knowing.
"In fact it may even have increased your life expectancy by giving you a modicum of immunity to a more virulent coronavirus."
Good point. The one thing we can say with some confidence about someone who recovered from a mild case of covid that that they're probably less likely to die of covid.
The one-year number is clearly too high, for the reasons given, but the five-day number is indefensibly low. COVID-19 during 2020 reduced the average American's life expectancy by perhaps five days -- but we have already lost several more days in 2021 using the same logic. There were roughly 360,000 US deaths attributed to COVID-19 through January 1st, and now there are 160,000 more.
The report is specifically about 2020, 2021 isn't particularly relevant.
It's actually pretty close.
5 days loss of life PER YEAR, over 80 years (average life expectancy) is 400 days. Or pretty close to 1 year.
Yes, it's pretty close if you assume the pandemic will continue at a peak death rate forever. That's the big flaw here.
That's how life expectancy is calculated though. It MUST assume the current annual death rate going forward. It's how it is always calculated. It cannot "estimate" future death rates.
You always see dips in the historic life expectancy during major pandemics or wars. See below.
The error here is in people interpreting the data incorrectly.
"The error here is in people interpreting the data incorrectly."
The "error" is the CDC releasing, and the media publicizing, data that they KNOW most people will not interpret correctly.
I've already had people tell me we are all going to die a year earlier even we didn't get sick, which is an even more incorrect interpretation
The CDC always releases life expectancy numbers. Every year. Here it is for 2018. Always using the same metrics and calculations.
https://www.cdc.gov/nchs/data/databriefs/db355-h.pdf
If people don't properly understand what that means, that's a different literacy issue. If newspapers deliberately mis-state what it is, that's a bigger issue.
By the way, AL, Do you think that insurance companies blindly apply the CDC calculation to get their actuarial tables? Of course not.
They use the same basic data and then know how to correct the raw statistics in a way that does not increase their risk profile.
"Do you think that insurance companies blindly apply the CDC calculation to get their actuarial tables?"
No. But they USE the same basic data. If you change how a statistical measure (And that's what period life expectancy IS) is defined or measured, then all types of problems result.
They make money and compile their actuarial tables from mortality (or accident, or disease) data just as does the CDC. They are not experiencing "all sorts of problems" that you fear.
The source of problems are the sometimes large uncertainties and frequent inconsistencies among portions of the data set. The assumption the all the data in the set are consistent and collected in the same manner and have the same systematic biases IS a model of the population.
The arithmetic is just a set of calculations by a computer. The results are as meaningful or as meaningless as the validity of the underlying model of the population. That applies to people as much as it applies to processes and inputs in a chemical plant.
re: "That’s how life expectancy is calculated though."
No. That's how life expectancy is calculated when the risk is in steady state. For example, when cars are invented you have to assume that they won't be uninvented and that car-related deaths will probably continue. Pandemic risk doesn't work that way. Pandemics sweep through a population, then drop to much lower levels of new infection following the Gompertz Curve. Calculation of changes to life expectancy in response to single spikes are not calculated as you allege but by the way the authors in the article above work.
Period life expectancy at birth is the mean length of life of a hypothetical cohort assumed to be exposed, from birth through death, to the mortality rates observed at a given year.
That is the literal definition.
And if you leave off the word "period" and / or don't explain that to the average newspaper reader, then you are misleading your readership.
You're telling them that your "spherical chicken" is the same kind of chicken that they buy in the supermarket.
Then blame the NYT. But this is how it's been done for decades.
"Life expectancy" refers to the period life expectancy. As opposed to the cohort life expectancy (which is different, and only really works for hindsight).
So you applaud being deliberately imprecise to the extent to mislead.
You would not get away with that in a scientific journal with good reviewers.
A scientific journal with good reviewers would have an implicit understanding of what life expectancy in this context (specifically period life expectancy) means, and wouldn't blink at it.
The issue is a lack of understanding in the general population.
No. 5 days lost per lifetime.
Assuming the pandemic doesn't continue for another 80 years. Or we don't see other pandemics....
With that kind of contorted twisting, there's a position in politics awaiting you.
No, Al.
You may not do that you have to integrated over the entire probability distribution given in the tables weighted by the demographic distribution function.
Yes, CDC was clearly wrong, not that it really needed this article to show that. Hopefully CDC will do the right thing and own up to its goof, and the media will learn not to be so credulous.
But as you point out, the article's calculations don't make all that much sense either. I don't see how taking the aggregate loss in lifespan from people who died from infection or a related cause (which is easy, just subtract their ages at death from their actuarial life expectancies and sum) and arbitrarily averaging that across the total population provides any sort of meaningful statistic about life expectancies.
What I'd really be interesting in knowing is, for the folks affected in nonfatal ways during 2020, e.g. nonlethal infections, stress, depression, lack of exercise, bad diet, etc., how did that impact their life expectancies? Nobody seems to be working on a statistic like that, which might actually be informative.
"I'm far from not an expert on the subject"
So... you are an expert?
Funny, sorry, fixed! I plead Russian being my native language; double negatives are the way we do things, e.g., "I not know nothing" / "я не знаю ни чего."
(Actually, the "far from an expert" locution wouldn't be double negative even in Russian; really, this was just an editing glitch.)
I hate to correct a native speaker, and Eugene is a fine grammarian (how could he not be???), and it's certainly true that when one learns Russian, the teachers say that the "... не [verb] ни [indefinite pronoun]” is a double-negative, BUT, it's not true. It's wrong when they say that. Actually the "ни" means something more like "any", not so much "not". As in "Что бы ни случилось, оставайся дома!" ("Whatever may happen, stay home!")
So the correct literal word-by-word translation of “я не знаю ни чего.” would be "I not know anything", not as Eugene says "I not know nothing".
(I should know. I was actually thrown out of the Kremlin for singing, once. I bet Eugene can't say the same!)
An even more telling example is the way Russians say "anybody" (or "anything"). They actually say it in two ways: if you know a bit about who the "anybody" might be, you say "who-that" (кто-то), but if you don't know anything about the "anybody", including whether there will actually be anybody, then you say "who-any-be", meaning, "whoever it might be", or, literally, "be whoever", or, more idiomatically in English: "be who it may". and the Russian phrase is "кто-нибудь". As in "If anyone hits me, I'll tell the boss." : "Если меня кто-нибудь ударит, я скажу шефу. " The phrase, кто-нибудь, word by word, is
кто = who
ни = ever/any
будь = be! (imperative).
The "ни" here is "ever" or "any", and there is no negative in the sentence at all.
There's a famous joke which may or may not be true, but it's still a good joke. Some college grammarian is rattling on about the finer points of negatives, says double negatives are troublesome but ok, and some languages allow triple negatives, but none allow double positives.
Wise ass in the back row says, "Yeah, right."
“I’m far from not an expert on the subject”
Most likely true, regardless of subject, for any person distant from Dr. Ed.
Professor Volohk,
That's actually expected, and to be consistent with previous results, the CDC has to put out the data that way. I don't blame the CDC at all for doing what is consistent with previous data and previous results. Perhaps you can blame the NYT for hyping it too much. But not the CDC.
Is it rationale to expect 2020's death rates for 2021-2099 and beyond? Not necessarily. But to be consistent with all the previous data, that's actually what needs to happen. The CDC can't "guess" what future death rates are for 2021 and 2022. That would just be making up data. It has to use the existing data that it has.
Now what you would expect is in 2022 for life expectancy to bounce back up to the 2019 numbers (and indeed, I do). But again, for consistency of data and analysis, you simply cannot just put in 2019's numbers. It is inaccurate.
AL,
I tend to agree with you. While the CDC estimate reflects a methodological error. The NYT's reporting (and its lede) was dishonest.
I wouldn't even say it's a methodological issue. It is defined as how life expectancy is calculated.
The error here is in misinterpretation of the data.
Sure, but all kinds of well-defined statistical tests are extremely misleading or subject to misinterpretation. For instance, p-values prove far less than most believe and are notoriously easy to manipulate. Why bother using a statistic that carries so many problems in the first place? If you insist on doing so, put out all relevant statistics to get a fuller picture of what the data actually imply. You can't claim innocence for misinterpretation when you put out a statistic that is extremely likely to be misinterpreted.
This isn't a statistical test. It's a statistic. These are different.
It takes hard numbers, and makes a calculation. deaths, at certain ages, and extrapolates a hypothetical cohort born that year for those death rates.
There are other statistics like this. For example, before and during WWII, the US fertility rate greatly dropped. This doesn't mean that women were suddenly "less fertile"
Thanks for this and your other comments—really enlightening. I should have consulted them before posting my comment, as they address the points I raised.
But, I think at least one question remains. Presumably Dr. (or should I say "kiddo" if he hasn't delivered any babies?) Bach knows better than most all the details you gave about CDC's methodologies. So why did he pointedly fail to mention any of that in his article? It seems like a glaring omission, whether deliberate or not. And it obviously undermines a lot of what he was arguing.
Bach mentions this pretty clearly.
"Don’t blame the method. It’s a standard one that over time has been a highly useful way of understanding how our efforts in public health have succeeded or fallen short. Because it is a projection, it can (and should) serve as an early warning of how people in our society will do in the future if we do nothing different from today."
He follows in a manner that takes issue with the fact that CDC published a statistic they should have known was problematic and would be misinterpreted by the public:
"But in this case, the CDC should assume, as do we all, that Covid-19 will cause an increase in mortality for only a brief period relative to the span of a normal lifetime. If you assume the Covid-19 risk of 2020 carries forward unabated, you will overstate the life expectancy declines it causes. It’s not like I am the first person to notice this problem. Researchers have regularly demonstrated that life expectancy projections are overly sensitive to evanescent events like pandemics and wars, resulting in considerably overestimated declines.
And yet the CDC published a result that, if anything, would convey to the public an exaggerated toll that Covid-19 took on longevity in 2020. That’s a problem."
The issue is, the CDC always publishes period life expectancy. Always. Every single year.
The real issue is, the NYT reporting.
Yup, Armchair's right, DJK's wrong.
The Bach article seems contradictory and/or internally inconsistent. First, he argues that CDC was mistaken, or even worse, "relied on an assumption it had to know was wrong." So that's basically accusing it of deliberate misconduct. But then, as DJK quoted, he concedes that CDC's methodology isn't to blame.
Second, he shifts his argument to complaining that—while correct—the report's "finding [] bears no relation to any realistic scenario." But that's about the same as saying that CDC should have produced some other type of report that he'd personally like better, even though, as Armchair pointed out, CDC never signed up for that and has been very clear about that fact.
Third and last, he shifts gears again, now saying that it's CDC's fault that NYT and other media outlets are illiterate when it comes to statistics and research methodologies. So even though nothing's actually incorrect with the report, CDC should retract it anyway just because NYT et al. misinterpreted it. Like Armchair said, that's hardly CDC's problem.
I'd think it'll bump up above 2019 since many of those dying now would have died in 2022. I hope nothing increases the rate of death in 2022 to prove me wrong, though.
Again, to re-emphasize this point, here is a chart of US life expectancy at birth from 1860.
We see a large drop in life expectancy in 1865 from 1860. This does not mean that a child born in 1865 was actually likely to die sooner than in 1860. What this reflects is the large death rate in the US Civil War, and a whole bunch of 20 and 30 year old men dying (who would've been born in 1840).
Again, you see the same dip from 1915 to 1920. This reflects the Spanish flu (and WWI). It doesn't mean that babies born in 1915 would all magically live a year longer than those in 1920.
https://www.statista.com/statistics/1040079/life-expectancy-united-states-all-time/
I'm a little more charitable than some: my take is that the CDC did the same old calculations they do year after year. It's a perfectly valid thing to do. And then innumerate media gave a misleading interpretation.
If there is something to criticize about the CDC, one could argue they should have realized that reporters might not understand what they were saying and conjure up a misleading spin, and could have added a for-the-masses explanation to the report.
The real error here is with the NYT reporting.
"It’s a perfectly valid thing to do. "
No, it is not perfectly valid. It is perfectly lazy and poor science. And you repeat the same procedure without even a hint of the inherent flaw is poor science that could be called misconduct in the cancel culture.
It's not lazy or poor. It's the standard for how it is done, using the existing data you have at the time.
If you fail to interpret what the data really means though, that the real issue.
Standards are not always good, and even if good, are not always the right thing to follow. Blind obeisance to standards is how we got crap like "97%: and the "science is settled". If it's settled, it ain't science; and if it's science, it ain't settled. And if your blind obedience causes you to translate "ain't" in your head, then you are doing it wrong.
You shouldn't follow them blindly, but you shouldn't change how you calculate, say, the GDP or CPI without a lot of thought either. People depend on them meaning the same thing every year.
Sometimes you'll see asterisks in a financial statement saying 'this years revenue reflects a one time writeoff resulting from our purchase of Acme Corp' or something - and maybe the CDC should have done that. But you just don't change the definitions of your metric willy nilly. And in the CDC's defense, the underlying report doesn't look like it is targeted at a general lay audience.
standard =/= correct
The so-called standard is a lazy reporter's excuse.
It would not be accepted in a scientific paper. Typically such "standards are roundly criticized by the peer reviewers and would at least result in a paragraph to explain the implied (in the NYT) conclusion. It is an incorrect methodology used to support a politicized conclusion.
Misleading does not stop being misleading because other people do it.
Don Nico, the politicized conclusions seem to be all yours. You don't like that someone creating a statistic by a customary method delivered a result which embarrasses your political preference. So you say they used the statistic wrong. Factor your logic and it comes out as you asserting that any statistic which embarrasses your political preference is wrong.
"embarrasses your political preference"
It doesn't embarrass anyone's political preference. It embarrasses the reporters who reported on it in a confusing way.
No, Stephen, I have nothing to say about politics except that the NYT's lede was a political one.
A late friend of mine was a highly respected theoretical physicist. He liked to start an argument with, "Let's assume a spherical chicken..." State his premise and proceed with a rigorous analysis. ,Was the analysis correct. Formally, yes. Did the conclusion of that analysis have anything to do with fried chicken? No.
There conclusion is not embarrass my politics. You're dead wrong about that. What is wrong is applying the analysis to the spherical chicken, and the NYT claiming that the analysis describes real chicken. Do you see the difference?
The distinction is one which I apply almost every day as the editor-in-chief of two world-class scientific journals. I has naught to do with left, right, up or down politics.
The distinction is one which I apply almost every day as the editor-in-chief of two world-class scientific journals.
Let me guess, Nature, and New England Journal of Medicine?
The NYT said nothing about chicken, spherical, metaphorical, or otherwise. It used a standard statistic in a standard way. In general, that is a safer and wiser approach—an approach more amenable to reader follow-up and fact checking—than a reporter trying to invent on the fly an alternative statistic which has no publicly available standard for comparison. Inevitably, such an approach creates a novelty which hasn't been vetted for any purpose at all. That is the method vulnerable to admitting smuggled political bias into the story. Using the standard statistic in the standard way is the way you take the story straight down the middle.
Some writers about baseball may disagree that Babe Ruth was the greatest ever. They may think jiggering statistics proves it was someone else. You can do that. But the resulting article gets very long, and ultimately the subject changes. It becomes all about the new statistical method. That isn't a reasonable demand to make of writers and editors delivering day-to-day news coverage.
It is the mathematical definition of how 'life expectancy' is defined. It's no different than saying 'the arithmetic mean is the sum of the observations divided by the number of the observations'. That is just true.
If one was reporting on the average net worth of residents of a small town year after year, you still calculate the mean the same way even if the number swings wildly because Bill Gates moves there.
If a clueless reporter doesn't understand how that works and reports 'everyone in Podunk is now a millionaire!' because the mean moved from $50k to $50M, that's an error by the reporter, not the statistician.
It's the definition of how a particular estimate of life expectancy is calculated. It is disingenuous to not present all relevant estimates.
In your example, a competent statistician would also present other statistical information. Simply comparing the median to the mean tells you that something is afoot.
You're not asking for science, though. You're asking for an actuarial prediction. That's a fine endeavor for an underwriter, but the CDC did what a public health agency should do.
Then lazy or sensationalist reporters presented the CDC's report in a way that robbed it of nuance and meaning.
Michael,
Acturarial predictions are a science. There are established and experimentally tested methods. It is not the same as an exercise in mathematics. The tables of numbers are meaningless absent the assumptions of the computational/population model.
The CDC produces tables based on population models and assumptions in exactly the same way (maybe with less or maybe with more precision than a corporate actuarial department.
The technical appendix to the VSRR no 10 describes the calculational technique. The text explains several reasons why the resulting table of survival may be incorrect in either direction. Reading the report over, the authors did NOT claim that the changes were due to the coronavirus infections. They were based on preliminary death rates.
I think the CDC could have issued a disclaimer with the data indicating that while this is a standard way of modeling life expectancy it suffers from magnifying short-term events, and that we should anticipate life expectancy ticking back up to pre-pandemic levels in a year or two
It's not "modelling" life expectancy. It's a definition, based on the current year's death rates.
Kevin is exactly right. I don't see why you insist that you cannot understand the greater utility of telling the entire truth.
What "entire truth"?
Do you want to redefine a statistical measure?
You really want to be dense about this topic.
You define and use such statistical quantities for a real world purpose.
If your definition confounds that purpose at a given time, reporting it without comment as if it does is faulty science.
Don Nico, try not to be thunderstruck. The NYT is not the arbiter of scientific truth. It does not purport to practice science. It practices journalism, and does what it can to be expert at that. Which is part of the reason it is the great newspaper it is.
You seem to see yourself as a practitioner of science. Fair enough. That is a different calling. Based on what I have seen from you about what journalism ought to be, I suggest you may be better at science than you are at journalism. Maybe scientific journals are the right place for you.
The entire truth is saying want things are and why the calculation may not be accurate. The authors gave some reasons why they may not be accurate, but they did not say that the some of the assumptions of the calculation may be faulty. That would be the whole truth.
You're hung up on a definition of a mathematical procedure and are not making any contact with apply a tool to the real world.
I am sure that you are unable to understand that.
But I do appreciate that the conversation did not degenerate into name-calling as so many discussions do in this venue.
The majority of the loss life expectance, like 2/3, came from untreated cancer and heart disease. That loss was caused by the Democrat Governor lockdown to take down Trump. These traitors allowed infected young people to travel to nursing homes to provide intimate care to moribund patients. That served the purpose of reducing cost for the Medicaid payers. The cheapest patient is the deceased patient. But, they shut down outpatient clinics for diagnosis and outpatient management of the majority of the most frequent causes of death, heart disease and cancer.
That quack Democrat Governor lockdown is the biggest mass murder in US history.
Deaths of despair caused by the lockdown was a major contributor, suicides, murders, and overdoses, all shot up 40% in 2020, compared to 2019.
Doctors, like lawyers, offer worst case scenarios. Most worst case scenarios never come to fruition, though sometimes they do. But it's up to the client/patient to hear what the worst case scenario is and then make an informed decision about how much risk to assume. So even granting that the CDC numbers were too big, I'm not sure, based just on that, that the CDC necessarily did anything wrong.
Worst case scenarios by professionals are fraud in rent seeking, to pump up fees. They violate professionalism, and are unethical.
Actually it's probably unethical not to tell your clients about worst case scenarios. That way, if the worst case actually does happen, the client was warned.
It's also probably unethical to not wrap enough context around a worst-case scenario (minor details like roughly how likely you think it is for the worst-case scenario to occur) to help your client make a measured judgment about how much resources they should expend to try to avoid that worst-case scenario.
The Table in the report is not wrong per se. Not commenting on its misleading interpretation is less than good science.
If you've had a particularly bad week, then you welcome at least those 5 days being lopped off your life.
This really needs to be re-emphasized, repeatedly. It's a fundamental misunderstanding
"Life Expectancy" as reported by the CDC is "Period life expectancy" which is a statistical measure, not a model. It is defined the mean length of life of a hypothetical cohort assumed to be exposed, from birth through death, to the mortality rates observed at a given year. It is an official statistic. Not a model.
Other official statistics include
1. Population
2. Total fertility rate
3. Employment rate
4. Gross Domestic Product
and of course...
5. Life expectancy.
"which is a statistical measure, not a model."
It is a measure defined within a model. It is as usual as is the underlying model.
... is as USEFUL... (not usual)
Flu deaths dropped. One suspects intentional fraud, for the Medicare bounty for COVID. Pay for something, you get more. Trump was played by the
Democrat, Deep State traitors advising him. The majority of deaths were in moribund people in nursing homes, ridding Medicaid of those costs. Most COVID death certifications are with COVID, not from COVID. Many are lies. Guy shot in the head was heard to cough. Make $13000 for adding COVID to the death certificate.
For disclosure, I greatly benefited from the COVID lockdown. My income broke records, and my occupational dream of 20 years, ending travel, came true. When I sued and lost to stop the lockdown, it was 100% against personal interest. I also greatly benefited from lawyer perfidy and judge idiocy, before the lockdown. Some legal training paid for itself many times over in cancelling debt and in tax matters. Stop asking me, where did a lawyer hurt you? All my arguments are against interest, and morally superior.
Maybe these people can help this subject.
Reductions in 2020 US life expectancy due to COVID-19 and the disproportionate impact on the Black and Latino populations
Theresa Andrasfay and Noreen Goldman
https://www.pnas.org/content/118/5/e2014746118
Abstract:
"COVID-19 has generated a huge mortality toll in the United States, with a disproportionate number of deaths occurring among the Black and Latino populations. Measures of life expectancy quantify these disparities in an easily interpretable way. We project that COVID-19 will reduce US life expectancy in 2020 by 1.13 y. Estimated reductions for the Black and Latino populations are 3 to 4 times that for Whites. Consequently, COVID-19 is expected to reverse over 10 y of progress made in closing the Black−White gap in life expectancy and reduce the previous Latino mortality advantage by over 70%. Some reduction in life expectancy may persist beyond 2020 because of continued COVID-19 mortality and long-term health, social, and economic impacts of the pandemic."
Sorry Eugene, your 400K is close, but the number of people who died because of COVID (as opposed to the number recorded as infected) must be calculated using excess deaths, no other measure can answer that question -- raw deaths is particularly useless due to methodology differences across the innumerable institutions that report deaths.
If you google CDC Excess Deaths and download the CSV marked "National and State Estimates of Excess Deaths," CDC seems to think that (so far) since Feb 2020 America has seen about 680,000 excess deaths (sum on column H for the rows with dates since Feb 2020)
As you can see, excess deaths returned to near-normal in March, but there is a lingering tail. The future is uncertain, but that should give you the most reasonable estimate of the total impact of COVID possible today.
So using your math, we find on average Americans lost around a week, which is probably about right. There will be lingering effects for some survivors, but note excess hospitalizations also track excess deaths pretty closely.
Note the COVID technological tail has implications too... with the plethora of mRNA vaccines suddenly emerging for things like malaria that have never been possible before we may be on the verge of large gains globally. And new platforms like peptide vaccines that could radically lower costs are getting serious looks.
Cato and the Reason Foundation are going to find that they have to deplatform Mr. Volokh if he keeps posting stuff like this.