Lockdowns Saved Trillions! …. Say Models

Joe Nocera — read his Lockdowns Haven’t Proved They’re Worth the Havoc — sent us an article from the Washington Post which discusses how much money the lockdowns saved us.

Economists at the University of Wyoming estimated the economic benefits from lives saved by efforts to “flatten the curve” outweighed the projected massive hit to the nation’s economy by a staggering $5.2 trillion. Another study by two University of Chicago economists estimated the savings from social distancing could be so huge, “it is difficult to think of any intervention with such large potential benefits to American citizens.”

In other words, the economists are saying, “the cure” doesn’t come at a cost at all when factoring in the economic value of the lives saved.

5.2 trillion!

The article goes on to discuss “Value of a Statistical Life” and other obscure topics we can ignore.

The studies referred to are (Wyoming) “The benefits and costs of using social distancing to flatten the curve for COVID-19” by Thunstrom and others, to appear in the Journal of Benefit-Cost Analysis, and (Chicago) “Does Social Distancing Matter?” by Greenstone and somebody else, a preprint at the Becker Friedman Institute.

The Chicago paper came out in March 2020. Here’s the pertinent part of the Abstract (my emphasis):

This paper develops and implements a method to monetize the impact of moderate social distancing on deaths from COVID-19. Using the Ferguson et al. (2020) simulation model of COVID-19’s spread and mortality impacts in the United States, we project that 3-4 months of moderate distancing beginning in late March 2020 would save 1.7 million lives by October 1. Of the lives saved, 630,000 are due to avoided overwhelming of hospital intensive care units. Using the projected age-specific reductions in death and age-varying estimates of the United States Government’s value of a statistical life, we find that the mortality benefits of social distancing are about $8 trillion or $60,000 per US household.

8 trillion!

This is hilariously wrong. The Ferguson model was long ago busted. It predicted, as the Abstract said, a “savings” of 1.7 million lives. Somehow, if we roamed the streets free, the coronavirus was going to be worse that the Spanish flu. It is a bad model, a buggy model, a dumb model.

What we have here is not uncommon in academia: models of models. This is only two levels deep, but in theory (!) you can stack these one upon another until you reach…tenure. Or a big grant, which contains juicy overhead for your Dean.

How many times have I told you that models, all models, only say what they’re told to say?

This isn’t rhetorical. Answer the question.

These Chicago people told a model to say every life saved by lockdowns was worth this-and-such, and a second model was told to say this many lives will be saved. The joint model, then, was told to say 8 trillion! would be gained by locking down.

Does it seem to you the economy is 8 trillion! richer, now that we’re coming to the end of the lockdowns? Or does it seem instead that the gibbering panic cost money, not gained it?

The Wyoming paper was even bolder. It did all its own modeling. I might have miscounted, but it looks like that gained the US 5.2 trillion! with four equations. Here they are (with no attempt to make pretty or explain):

This is at least a clear example of a model saying exactly what it was told to say. Here in those equations (if I didn’t miss any) are what it was told to say.

It said 5.2 trillion!

And the reporter at the Washington Post, following his own or his editor’s model, only said what he was supposed to say.

To support this site and its wholly independent host using credit card or PayPal (in any amount) click here

27 Thoughts

  1. I immediately laughed out loud when I saw the Wapo headline yesterday. Couldn’t help myself.

    Double digit unemployment, literally thousands of businesses closed and not generating revenue, gasoline demand 40% below normal, whole office buildings empty, the travel and hospitality industries limping along… and these chuckleheads say that the lockdown saved trillions of dollars.

    Here’s an idea: let’s keep everything locked down permanently and we’ll MAKE money.

    I suspect Bastiat probably would’ve laughed at it too.

  2. Michael Dowd: Gates was apparently going for 90% reduction, him being in the remaining 10%, of course. I wonder if 10%, at least half of which are the rich, spoiled, evil elite, can grow food, make clothes, etc to keep themselves, the elites, going. I know!!! Let’s make a model!!!!

    That’s it. We are living in the Matrix and someone flushed all the red pills.

  3. This isn’t the right target for criticism. Making some statement of the costs of Option A versus the costs of Option B is the only way to decide between them.

    Yes, the model will tell us whatever we tell it, but that’s actually a pretty big step forward. It’s much better to argue about a the inputs to a model — or even the model itself — than about the choice between A and B. There are subject-matter experts who can answer (sort of) technical questions like, “If we put everyone on house arrest, how many people will die?” That question they can take a stab at. They cannot answer questions like, “Should we put everyone on house arrest?”

    The execution is bad. The projections are highly suspect. The uncertainty in the inputs swamps the precision of the calculation. But that shouldn’t be a criticism of the basic idea of cost/benefit calculation.

  4. Hi SSGT Briggs,

    Just found your site and it’s fantastic. So you might be interested this:

    Effectiveness of workplace social distancing measures in reducing influenza transmission: a systematic review

    “The economic impact of the next influenza pandemic in the United States, in the absence of vaccination and other mitigation measures, has been estimated to be $71 to $166 billion”

    Get that? We went from “hey this might cost you ~$120 billion if you don’t social distance” to “we saved you $5.2 trillion by forcing social distancing!”

    Oh it gets worse. Of all the literature on social distancing, only 15 studies made the cut. 3 were epidemiological. Here’s the quality of those studies:

    “The overall risk of bias in the epidemiological studies was serious in two studies and critical in one study (Table 2). All three studies had moderate or serious risk of bias in the confounding domain, and two studies had moderate risk of bias in the outcome measurement domain. In addition, because the outcomes used in these studies were surrogates for influenza illness, the evidence was indirect [41].”

    Of the remaining 12, they were models.

    “we did not assess the quality of the modeling studies”


    So what did this review of the available literature on social distancing in workplaces find out? Well, you could possibly reduce an non-vaccinated flu pandemic by 23%, with the following caveats:

    “The modeling studies reported that percentage reduction in cumulative influenza attack rate in the general population declined with higher R0 values”

    Any R0 over 1.9, social distancing starts losing value

    “No studies provided empirical information regarding the impact of workplace social distancing measures on changing workplace contact rates. Third, the studies included did not report the effects of workplace social distancing on two of our primary outcomes of interest (lost workdays, harms). ”

    Get that? No studies had a baseline on what workplace contact rates actually look like. No studies incorporated lost workdays or harms into the analysis of the effectiveness of reducing attack rates.

    Are these people kidding us?

  5. Darin – Briggs was not downplaying any idea of cost/benefit calculation. The article was devoted to critiquing the concept of “garbage in, garbage out” mode of model predictions.

  6. Darin,

    A cost benefit analysis is only morally justified when it is you evaluating your cost against your benefit. Even then, you stand a chance of being wrong unfortunately.

    In an authoritarian or societal analysis, you don’t get to choose the experts (they were mostly wrong), you don’t get to choose the model (it gave a different result each time it was run), you don’t get to curate the model inputs (false positives and misclassifications), nor to you get to question the model outputs. Invariably, it will mean great cost (to you) and great benefits to others (the experts).

    In a rational society individuals would take prudent precautions and learn as they go. We would not hunker down everyone at the first prediction of doom. People need to protect the sick, but continue working to ensure the well being of the rest, not just for today, but for the future. We would learn as we go and make adjustments to our models.

    Last week, the CDC adjusted the fatality rates, making this no worse than a normal flu season. No learning, however. The emotional politicians and absurd media continue driving panic. No getting back to work even though we are in a better place now than when this started. Meanwhile they continue to raise our debt by millions of millions of dollars and are set to unleash an army of contact tracers.

    Hey, let’s try another cost benefit analysis. Horse hockey!

  7. pK, fine. Individual liberty and all that. Now, back to reality.

    Governments *are going* to make decisions about lockdowns. Your choices are:

    a) an explicit cost/benefit model that attempts to show at least whether benefits (“to whomsoever they accrue”) outweigh costs, or

    b) the seat of your governor’s pants.

    Is this really a hard choice? Would we have been better off if instead of forecasting 2 million fatalities using a dumb model, Imperial College had forecast “a s***-load of fatalities” using some other means? At least the model can be scrutinized.

    I agree that the models were “wrong.” I agree that there were plenty of shenanigans, even with the models. But perfection was not one of the options we were offered.

    Pointing out that it’s hard to calculate how many lives were saved by lockdowns (more than zero?) using a model is an argument easily turned on itself. Would things have been better *without* a model? Your answer depends on your inputs, i.e., your model says what you tell it to.

  8. The faulty analysis here is you don’t make cost-benefit analysis after the fact. At that point anyone can say it would have cost anything.

    Were there any estimates of the cost of not shutting down during a pandemic? Oh wait… what do you know… we do have estimates! From the proponents of social distancing themselves!
    Effectiveness of workplace social distancing measures in reducing influenza transmission: a systematic review

    “The economic impact of the next influenza pandemic in the United States, in the absence of vaccination and other mitigation measures, has been estimated to be $71 to $166 billion [2].”

    Got that? The next pandemic could cost us ~$120 billion. So pray tell me, how did we get from there to “actually no the pandemic would have been many times more expensive and because we shut everyone down, it saved $5.2 trillion?”

    Thanks in advance.

  9. Are models ever really scrutinized? I don’t mean on blogs like this one, I mean on an official policy level.

    That is, are there any recent examples of:

    -A policy decision being made based on a model.
    -That model turns out to be catastrophically wrong
    -The people who made the policy decision criticize the model and admit that it led them to make bad choices.

    and ideallly

    -A new model is created which fixes the flaws of the old model and is actually accurate.

    The closest I’ve seen is for the incorrect models to get quietly get swept under the rug and replaced by new models which make predictions that justify the exact same policies that the old ones did. The only difference in the new models is that they are updated to match new information.

    As far as I can see the “self-correcting science” idea for models is a fantasy, it doesn’t happen in the real world.

    And yes, having a catastrophically inaccurate model is worse than having none at all. There are two reasons for this. First, if a model is bad enough it can lead to worse predictions than random guessing. Second and more importantly the existence of a model gives false certainty to policy makers and the public.

  10. The funny thing is we can actually determine the number of lives lost through Planned Parenthood. By the same calculation, the cost of this organization continuing must be horrendous.

  11. David Burns
    “The faulty analysis here is you don’t make cost-benefit analysis after the fact. ”

    Why the hell not?! Knowing whether you made a good decision or a bad one is a good tool for improving future decisions. I say, do cost/benefit analysis at every turn!

    “At that point anyone can say it would have cost anything.”

    I mean… sure. But writing those assumptions down (i.e., a model) is a one way to curb the urge to cook the books.

    My point remains: your gripe is with execution, not strategy. The problem isn’t models per se (or even post hoc modeling). It’s *bad* modeling.

  12. We don’t need no stinking models next time around. Whatever we did in 1968 … just do that. Adjusted for population, we lost many more Americans than from Rock Star Virus. We fought a war. We sent men around the Moon. We weathered executions. We played baseball. And the economy roared along. And though I was just eleven, I think I would have remembered pervasive mask-shaming.

  13. “And yes, having a catastrophically inaccurate model is worse than having none at all. There are two reasons for this. First, if a model is bad enough it can lead to worse predictions than random guessing. Second and more importantly the existence of a model gives false certainty to policy makers and the public.”

    Having no model is not an option. You WILL make decisions, and those decisions imply something about the model you are using. Your choice is between being explicit about your model or not.

    I do agree accountability and transparency are lacking. But the models — such as they are — help with that problem. When my governor says he’s relying on such-and-such model for his decisions (which mine has), that’s better than him not saying it.

    Sure, the second- and third- iterations of the models have been bad, too. Maybe they haven’t converged as quickly as you’d like. But, again, how does saying, “We have no idea. Let’s just lock down some more!” help?

  14. Darin,
    There is no distinction between bad and good modeling. Seriously, even in the paper I posted above. Read it. They filtered out all the crappiest research and came up with 15 studies worth examining. They had 3 epidemiological studies that they called biased. Of the remaining 12 modeling studies, they said:

    “First, most of the included studies were based on modeling and few were in actual settings… Second, we did not assess the quality of the modeling studies.”

    Just like how no one assesses the quality of the modeling studies out today. So there’s no good or bad models. No one, not even the supporters of social distancing even assesses whether they are good or bad, or what that would look like.

    We don’t find out how bad they are until they release their code, I guess. Heh.

  15. My problem isn’t that models “aren’t converging fast enough.” My problem is that they aren’t converging at all, nor are they intended to do so. The purpose of models is to push predecided policy decisions, not to inform policy decisions. As such there is no motivation for models to reflect reality and so they do not “converge” towards accuracy.

    Inaccurate models are not even thrown out. The Imperial College model is about as busted as any model can be. It’s clear that the assumptions about the disease put into the model made it far more deadly than the reality shows, and the code used is ludicrously bad. But it is still referenced when it is politically convenient. The point of the model was never to reflect reality. The point was twofold: First, to justify a lockdown. Second, to allow politicians to say that they saved millions of lives.-

    Darin, you’re living in a fantasy land where the people who make models have good intentions and follow good practices.

    And it’s not even as though models help with “transparency.” Politicians will still make decisions based on whatever the hell they feel like even if their decisions are not justified by the model they are supposedly using. For example before he recently backed down, the governor of Minnesota had intended to restrict religious gatherings to only 10 people for the long term even as restrictions lessened on businesses and other gatherings. Despite him constantly touting that he was only following the models, there was nothing in his models that justified specifically restricting religious gatherings, it was just something he felt like doing. So while he said that he was following the University of Minnesota models in his decisions, in reality he only did that when it was convenient for him to do so.

  16. At the moment? I’m just suggesting that we don’t give cover to inaccurate models by pretending that they were well intentioned or that they provided anything of value.

    In the long run? Tarring and feathering the people responsible for models like the Imperial College model might be a start.

  17. Few in the establishment are ever punished (Denny Hastert was a rare exception) and Tiny Tony Fauic (PBUH) and the Doctor with no medical license, Brix, will not be punished or shamed for their wildly wrong predictions but will be the subjects of endless encomiums by the establishment media.

    Prolly most remember where they were on 911, can those same men remember any Intelligence Agency Employee being shamed or punished for failing to identify and prevent that catastrophic failure?

  18. “Does it seem to you the economy is 8 trillion! richer, now that we’re coming to the end of the lockdowns?”

    The answer being “compared to what?” and the “what” that authors of the paper are talking about is an “economy” which they model to be 8 trillion poorer than it is now.

    Funny one-liner, but better to be accurate. And “the economy” is reification 😉

  19. Taking a quick look at the mortality data, 75% of the lives saved will be retirees (over 65’s) who are net receivers from the economy due to Pensions (Social Security to you yankee doodles) and state funded medical costs, not contributors..

    So this valuation would go to the “not even wrong” bin…

  20. “At the moment? I’m just suggesting that we don’t give cover to inaccurate models by pretending that they were well intentioned or that they provided anything of value.

    “In the long run? Tarring and feathering the people responsible for models like the Imperial College model might be a start.”

    I agree, Rudolph. These are problems of execution, not modeling per se, though. In my view, our host has been a little loose with terminology, and the commenters have run with it. Mr. Briggs would not (I assume) suggest not analyzing and interpreting available data to support decision-making (i.e., modeling). He is mocking

    1. the bad *execution* of models and bad behavior of modelers — unfounded assumptions, over-confidence, hiding assumptions, over-complex code, commenting beyond field of expertise — and

    2. the bad *interpretation* of models and other errors by political decision-makers, which is, of course, caused by the first error.

    (Naturally, I don’t speak for Mr. Briggs, so you should take my interpretation of his comments for what they’re worth.)

  21. I’ve been hoping that someone else would ask this question, in the selfish hope that I wouldn’t need to subject myself to (possibly justifiable) criticisms of being cold-blooded and unsympathetic. Many of you will not like it. I don’t like it. I have grandchildren, so trust when me when I claim that I fully understand that many of those in the group I discuss so callously below have grand kids who love them dearly. But I think that the dynamic should be mentioned in context.
    All of the cost discussions seem to have contained the implicit assumption that all of the lives that were (or will be) lost, actually or potentially, to COVID-19 are of statistically equal economic value. We know that the disease disproportionately kills the elderly and/or already seriously ill. In my area, it appears that the largest group infected, and the vast majority of the dead, were institutionalized, mostly in elderly care facilities. If that is the way of it elsewhere, what happens if you apply that differential to the cost models? In other words, what if nearly all of the dead were individuals who were currently economically non-productive, and whose continued maintenance was heavily subsidized by the economically productive? What would that do to any objective model of the net cost of COVD19 over an 18- 24 month period? What would it do to an honest projection of economic values and costs going forward, given the removal of a significant economic burden? Obviously that would (will?) be a huge negative for the long term care industry and its suppliers and employees. But overall?

Leave a Reply

Your email address will not be published. Required fields are marked *