Metric Abuse – aka Lying with Statistics

I’m behind the curve a bit here, but I’ve seen and heard a bunch of
people making really sleazy arguments about the current financial stimulus
package working its way through congress, and those arguments are a perfect
example of one of the classic ways of abusing statistics. I keep mentioning metric errors – this is another kind of metric error. The difference between this and some of the other examples that I’ve shown is that this is deliberately dishonest – that is, instead of accidentally using the wrong metric to get a wrong answer, in this case, we’ve got someone deliberately taking one metric, and pretending that it’s an entirely different metric in order to produce a desired result.

As I said, this case involves the current financial stimulus package that’s working its way through congress. I want to put politics aside here: when it comes to things like this financial stimulus, there’s plenty of room for disagreement.
Economic crises like the one we’re dealing with right now are really uncharted territory – they’re very rare, and the ones that we have records of have each had enough unique properties that we don’t have a very good collection of evidence
to use to draw solid conclusions about recoveries from them work. This isn’t like
physics, where we tend to have tons and tons of data from repeatable experiments; we’re looking at a realm where there are a lot of reasonable theories, and there isn’t enough evidence to say, conclusively, which (if any) of them is correct. There are multiple good-faith arguments that propose vastly different ways of trying
to dig us out of this disastrous hole that we’re currently stuck in.

Of course, it’s also possible to argue in bad faith, by
creating phony arguments. And that’s the subject of this post: a bad-faith
argument that presents real statistics in misleading ways.

When you see statistics used in arguments, the number one thing that you
should always be careful about is to make sure that you understand exactly what the statistic means. One of the classic ways of misleading people with statistics is take take a valid statistic measuring some quantity, and present it as if it represented an entirely different quantity.

The current example of this comes from a variety of people who’ve been citing the work of a conservative anti-Keynesian writer named Amity Shlaes. Ms.
Shlaes has put forward an argument that the New Deal was a failure, and that it
didn’t really reduce unemployment. The full argument is in her book,
The Forgotten Man: A New History of the Great Depression; a short version of the
particular claim that I’m objecting to can be found in a Wall Street Journal Op-Ed
article here.

Now, in school, we’ve all seen the history of the Great Depression. We’ve all
heard about how the New Deal put so many people to work, doing all sorts of things – building roads, bridges, buildings, aircraft carriers, etc. All told, the WPA (the organization that the government used to fund all of the New Deal work programs) employed something over three million people. So how did employing three
million people not help improve the unemployment situation? It seems ridiculous on the face of it – did employing three million people somehow destroy more than three million jobs, without anyone noticing that?

The answer is that when Ms. Shlaes talks about unemployment, she’s not really talking about unemployment. She’s careful to not really cite unemployment rates. What she does is cite the number of permanent jobs.

The WPA programs in the New Deal were designed as a deliberate stop-gap measure. That is, there were millions of unemployed people in America, who weren’t able to afford the basic necessities, like food and housing. The point of the WPA was to get those people working, immediately, doing something to earn a wage, so that they could feed their families, pay their rent, etc. It wasn’t to give them permanent jobs working for the government, but to give them paying jobs until
the economy recovered enough that private employers would start hiring again.

So in other words, Ms. Shlaes deceptively picks a statistic that excludes the jobs created by the New Deal, and then uses that to argue that the New Deal didn’t create any jobs. If you actually state that properly, it becomes transparently ridiculous: “If you exclude all of the jobs created by the New Deal, then the New Deal didn’t create any jobs”.

You can make the argument that they way the government tried to help the
economy delayed that recovery. I don’t find the argument particularly convincing, but that’s probably as much my philosophical bias as anything else. There are some strong arguments that the Roosevelt economic stimulus programs were the right thing; and there are some strong arguments that it was the wrong thing. Both arguments are, largely, speculative – there just isn’t enough data about this kind of situation to be able to really know which argument is right – both arguments follow logically from their assumptions, but we don’t know which set of assumptions is right in this situation.

But to argue that unemployment was not reduced by hiring three million people? That’s idiocy. And Ms. Shlaes (and most of the people citing her) know it. In fact, she basically admits it herself, but handwaves her way past it: “To be sure, Michael Darby of UCLA has argued that make-work jobs should be counted. Even so, his chart shows that from 1931 to 1940, New Deal joblessness ranges as high as 16% (1934) but never gets below 9%. Nine percent or above is hardly a jobless target to which the Obama administration would aspire.”

Read that carefully. She’s admitting that WPA programs reduced
unemployment by nearly half. (And even that’s using skewed figures. Different ways of estimating unemployment during the Depression range as high as 25%.) But even in
the midst of her argument about how the New Deal didn’t decrease unemployment,
she’s admitting that it reduced unemployment quite dramatically.

As I said, this is a typical way of using statistics in a misleading way. Pick a statistic that measures quantity A, and use it as if it measures quantity B. You can see arguments like this all over the place.

To cite another example, this time from the other side of the political spectrum: when criticizing the Bush administration’s fiscal policies, you
constantly hear people talking about the Clinton surplus. They tell you that
under Bill Clinton’s fiscal policies, the federal government’s budget went
from operating at a huge deficit to a huge surplus.

The problem is, there was no surplus. There was never a real surplus
under President Clinton. It’s once again a game of switching metrics.

The government collects takes to pay for government programs. There are
multiple kinds of taxes – personal income taxes, corporate taxes, capital gains
taxes, inheritance taxes, social security taxes, etc. Some of the money collected
as those taxes goes into a general funds pool – the basic pool from which the
government operates. Some of those taxes go into special funds, which are supposed to be used only for a specific purpose. An example of the latter kind of tax
is social security, which is only supposed to be used to pay social security
benefits. Any excess in social security is supposed to be put away for
future retirees.

When people talk about the Clinton surplus, what they mean is that by
the end of Clinton’s term in office, the total collection of funds from all taxes during a fiscal year exceeded the total funds spent by the federal government during that fiscal year. Sounds great, right?

Only the special purpose taxes, like social security, were included in the tax collections. But in order to use social security funds for the general
budget, the goverment has to borrow money from social security. It issues
government bonds – the exact same bonds that make up the federal deficit. But instead of selling those bonds to private purchasers, it sells them to itself. But it’s debt, nonetheless.

When someone talks about the surplus, they’re playing a misleading
game by using invalid metrics. But by citing one quantity (total government income including money borrowed from social security), and pretending that it represented a different quantity (total government income from general taxes), they can dishonestly claim to have balanced the federal budget, and produced a surplus.

This tactic – using the wrong measure – is incredibly widespread, because it’s
incredibly effective. It’s easy to use to deceive people, because most people won’t pay attention to the exact definition used in the statistic – they’ll focus on the number. And if someone objects that the statistic is wrong, it’s easy to twist the
argument into a debate about the number, rather than about the meaning – which, in turn, makes it look to people observing the argument, like there’s a legitimate debate.

The lesson here should be clear. Always, always make sure that you understand
exactly what a statistic really measures. When someone makes an argument
based on statistics, make sure that the statistics they cite really do measure what
the arguer claims they measure.

0 thoughts on “Metric Abuse – aka Lying with Statistics

  1. Infopractical

    The regular story is at least as deceptive. Nobody ever talks about the jobs lost due to taxation. I’ve read numerous economists on this issue who agree with Ms. Shlaes’ reliance on permanent jobs as the better metric.
    To borrow money from the future to create temporary jobs is just a generational shift in employment. There are plenty of arguments that it decreases permanent jobs in the long term on top of temporary jobs. This means that the number of temporary jobs is likely to be a wash (even Krugman agrees with this point).
    You should stick to computer science. This is clearly not a topic on which you are particularly well read. This argument is decades old, and the tide of sentiment has shifted farther and farther from the use of temporary jobs as any kind of a positive metric. Only the people who wave the magic Keynesian multiplier wand to its greatest overstated factor believe in it at all, while it is well agreed that it comes with substantial liabilities on the money borrowed.

    Reply
  2. Mark C. Chu-Carroll

    Re Infopractical:
    I’ll just point out that you don’t actually address anything that I said in my post.
    Ms. Shlaes claims that the people working for the WPA weren’t employed, and that therefore, the WPA didn’t reduce unemployment. That’s a *ridiculous* argument. As I said, there are good faith arguments on both sides for whether the New Deal type spending was the best way to try to fix the Depression. But to claim that it didn’t affect unemployment – because you’ve defined employed people as being unemployed – is just lying.
    It’s got nothing to do with whether you like Keynesian spending. It’s got nothing to do with whether you like the kinds of jobs that the WPA-style programs created. To me, it’s just a matter of simple honesty: to say that a person is unemployed, despite the fact that they’re working and earning a wage, with which they’re managing to pay their rent, feed their family, etc. – that’s just dishonest.
    I’ll also note that people like Ms. Shlaes don’t use the same metric when they count *current* unemployment. When they cite modern unemployment figures, people in temporary employment are counted as *employed*. That’s a rather strong piece of evidence for the dishonesty argument – why is it, that when they want to report on how the economy fared over the last 8 years, they cite the *lower* unemployment figure – the same figure that they at other times argue is invalid?
    I’d also like to see a citation of just where Krugman agreed with Shlaes about this?

    Reply
  3. Sili

    It would seem to me that by this metric *anyone* employed by *any* branch of Government would have to be counted as unemployed.
    Doctors, firemen, police, the entire Civil Service, &c &c.

    I’ve read numerous economists on this issue who agree with Ms. Shlaes’ reliance on permanent jobs as the better metric.

    I would say that the better metric is how many people have to starve, leave their homes and die from easily cured deceases?

    Reply
  4. Mark C. Chu-Carroll

    Re #3:
    To be fair, that’s not really correct. Ms. Shlaes doesn’t exclude WPA workers from employment rolls because they were working for the government; she excludes them because they were working in what were explicitly designed to be temporary jobs. Professional firemen, police, etc., are working in permanent jobs. Their employer happens to be the government. But the main argument that Shlaes makes isn’t that government employees don’t count; it’s that temporary employees don’t count.

    Reply
  5. wtanksleyjr

    Good article. I disagree with your main premise, but only because you claim too much.
    You said, “Read that carefully. She’s admitting that WPA programs reduced unemployment by nearly half.” But she did not admit that, at least not in the quote — she merely quoted two numbers, the higher of which was twice the lower. She didn’t claim they were in any chronological order, much less did she claim that the New Deal _caused_ employment to change from one number to another.
    Although I’m sympathetic, I don’t buy what she’s saying… It’s possible that WPA prolonged the Depression, but it’s not clear to me how we’d collect any evidence to actually show that. There are economic models which would explain such an effect, but there are also economic models that explain the opposite effect. The only fact I’ve seen cited that would support her claims is that Canada followed less Keynesian policies and suffered a shorter Depression (until ’33 rather than ’39). BUT, even if you believe that “less Keynsian” allegation, Canada’s situation was very different; in particular, they suffered no bank failures at all (according to Wikipedia).
    Economics is hard. I don’t think you’re justified, therefore, in calling her an idiot. Yes, she doesn’t (and can’t) assign cause and effect; but you are in the same boat, and you admit it, and yet you go on to claim that the WPA certainly created 300 million jobs. Yes, it created 300 million tasks filled by people — but how many jobs did it (and the other New Deal programs) destroy? How many of those 300 million people were hired from true unemployment, and how many actually left a viable job in order to start a more attractive one? Worse yet, how many of the WPA jobs ended artificially before the Depression was over, therefore creating temporary unemployment? It doesn’t seem possible to end a WPA-type program without laying off the “temporary” workers — won’t that create an economic disturbance? The New Deal was able to finally end the programs because an even larger jobs program, WWII, came on the scene. If we try to replicate the WPA, we’re taking a very real and very untested risk that we’ll be able to STOP without destroying our economy.
    These are all important and unanswered questions — and yet they must be answered if we are to accept the claims that this stimulus package must be passed right now in order to avoid otherwise-certain total disaster.
    -Wm

    Reply
  6. Matt McKnight

    Mr. Carroll, I wanted to point out a few errors in your text.
    The WSJ editorial says:
    “New Deal spending provided jobs but did not get the country back to where it was before.”
    You say:
    “…Ms. Shlaes deceptively picks a statistic that excludes the jobs created by the New Deal, and then uses that to argue that the New Deal didn’t create any jobs. “
    This would appear to be a complete contradiction of what she said, making much of your post a mere strawman attack. It seems useful to distinguish between jobs that are being created because there aren’t enough jobs and other normal jobs, to see if you are still need to spend money on creating jobs.
    You say: “She’s admitting that WPA programs reduced unemployment by nearly half. “
    No where did she imply the causation that you suggest, and your use of the word “admitting” would seem to indicate that you think this is true. You say the WPA employed “something over 3M people”. For this to be nearly half, total unemployment would have had to have dropped from around 7M to around 4M. However, the unemployment numbers in 1933 were around 11-12M depending on who is counting.
    It’s important to note that today’s unemployment statistics include people on workfare.

    Reply
  7. Repton

    I read an interesting blog post on unemployment rates in the United States recently. The up-shot: the US Government defines unemployment in a variety of ways. In particular: the official unemployment rate _does not_ include “discouraged workers”, which is people who do not have jobs but are not looking for work because they don’t think there is any work to get (it also excludes “marginally attached” workers, which is all other unemployed people who aren’t looking for work).

    Reply
  8. Matthew Platte

    They call it “dog-whistle” for several reasons. We heard (and heard about) many such calls during the recent election campaign.
    Another reason the term applies is related to the persistence of the 27%. You know how your neighbor’s dog can bark day and night with seldom an interruption? That’s also a characteristic of the dittoheads.
    As has already been mentioned, we are presently in a Paul Krugman/Joe Scarboring news cycle where government jobs aren’t really jobs. And the dittoheads are out there – and in here making comments on your blog – in force, peddling their Big Lie of the Day.
    It’s worth noting, however, that they’re only 27%. Possibly less; maybe more but nowhere near a majority.
    It’s unpleasant but necessary to point out their mendaciousness as often as one can.

    Reply
  9. William Wallace

    She actually wrote:

    AMITY SHLAESwrote:
    If he proposes FDR-style recovery programs, then it is useful to establish whether those original programs actually brought recovery. The answer is, they didn’t. New Deal spending provided jobs but did not get the country back to where it was before.

    I am not sure how that is dishonest.

    It seems ridiculous on the face of it – did employing three million people somehow destroy more than three million jobs, without anyone noticing that?

    That is part of the problem.
    Consider an analogy.
    Dyke leaks. Glorious leader has everybody bring caulk and start caulking the leaks. This stops a lot of water, but the dyke breaks elsewhere anyway.
    Did the glorious leader help by doing what he could, or did he hurt by diverting resources?

    Reply
  10. mystyk

    “There are three kinds of lies: lies, damned lies, and statistics.” — Mark Twain
    Of course, the real problem is not that numbers lie, but that the people using them lie. I keep thinking about how the term “unemployment” as officially reported has changed over the years to remove a good chunk of those who are, well, unemployed. Also, the entire concept of “core” inflation, which one economist described as “inflation without the inflation.”

    Reply
  11. Uncle Al

    (3,472,073)^7 + (4,627,011)^7 = (4,710,868)^7
    is hogwash. Do it in your head in ~30 sec but not in your computer.
    The STIMULUS is 100% process and no product. It is obscenely over-leveraged credit seeking to fill vast holes dug by obscenely over-leveraged credit. It opulently rewards the stewards of disaster without oversight while enslaving the victims with vicious surveilance. It is the perfect Washington War on Sanity.
    3971397; 7+1=8; 8426842; 8 NE 2

    Reply
  12. Pat

    @ Infopractical
    > Nobody ever talks about the jobs lost due to taxation.
    On the contrary, I hear a lot about the jobs lost due to taxation. It’s a very popular topic of discussion particularly on right-leaning talk radio and in the Opinion section of your local newspaper.
    Interestingly, however, I don’t see a lot of academic papers that support this claim. Can you provide some references? Most of the papers and books that I see with a cursory examination of the outstanding academic literature actually claim the opposite; that the argument that “tax cuts spur job creation” is not supported.
    > I’ve read numerous economists on this issue who
    > agree with Ms. Shlaes’ reliance on permanent
    > jobs as the better metric.
    Excellent. Citations, please?
    > To borrow money from the future to create
    > temporary jobs is just a generational shift in
    > employment.
    I agree that this is correct, but it is unfortunately incomplete. Creating temporary jobs using borrowed money can in fact be economically efficient if the temporary jobs produce infrastructure upgrades that increase future productivity or reduce future costs, enabling the future workers to produce more with the same amount of effort, which expands the tax base in the future.
    Just like a corporation can use debt to finance projects that will result in a long term growth payoff. For-profit corporations do this all the time, it is one of the central business strategies in a free-market economy. It is, in fact, one of the core symptoms of our current problem -> corporations are not able to acquire capital via debt right now, and thus they cannot finance projects that will result in a long term growth payoff.
    > There are plenty of arguments that it decreases
    > permanent jobs in the long term on top of
    > temporary jobs. This means that the number of
    > temporary jobs is likely to be a wash (even
    > Krugman agrees with this point).
    Again, I will grant this may be the case, but if this is so obvious it should be fairly easy for you to provide references supporting this claim.

    Reply
  13. Daithi

    Matt McKnight,
    I don’t know if you realize it but Mark’s last name is not Carroll. It is Chu-Carroll. You may not think this is a big deal, but I know that Mr. Chu-Carroll finds it annoying when people refer to him as Mr. Carroll. I know where he is coming from because my last name is Ivey and it bugs me to no end when people call me Mr. Ivry.

    Reply
  14. Jud

    To borrow money from the future to create temporary jobs is just a generational shift in employment.
    Unfortunately, this argument proves too much. By these lights, building the interstate highway system, public schools, etc., simply time-shifted unemployment.
    Moved from the public to the private sector, this principle would mean that borrowing money to buy cars would create future unemployment among auto workers. Sorry, just can’t see this making any sense as a general proposition.

    Reply
  15. Infopractical

    Pat, it’s interesting that you ask for citation from me, but not from Mark when he makes claims like, “Ms. Shlaes claims that the people working for the WPA weren’t employed, and that therefore, the WPA didn’t reduce unemployment.” which is a flat out fabrication. Nowhere in Ms. Shlaes book does she make this claim.
    What Mark is doing is smoke-and-mirrors for people who haven’t read this particular book. Mark is upset that Ms. Shlaes does not like the same metrics that he does. He wants immediate job creation and employment to be more linked than Ms. Shlaes intends in her discussion, where momentary employment numbers are reaggregated into a different kind of metric. What Mark doesn’t know, but anyone with a reasonable education in economics does, is that this is…extremely common and not in any way controversial.
    Mark is dramatically overstating his understanding of economics and winding up looking like a fool in order to grand stand to his own crowd.
    But you’re telling me that “Most of the papers and books that I see with a cursory examination of the outstanding academic literature actually claim the opposite; that the argument that “tax cuts spur job creation” is not supported.” when there is very clearly a raging debate that has been going on for decades. This tells me that you are intentionally isolating yourself from half the literature that you must also have gaping holes in your economics education. It’s sad that you stand up to comment.
    But as for citation of the point Ms. Shlaes made, you could actually look to…the citation she gives. Not that it’s hard to do that much leg work. Or will you simply take Mark’s word for it all?

    Reply
  16. trrll

    But as for citation of the point Ms. Shlaes made, you could actually look to…the citation she gives. Not that it’s hard to do that much leg work. Or will you simply take Mark’s word for it all?

    I don’t know about others, but when somebody is asked for citations, and replies with a 5 paragraph screed that provides no citations, but attacks others for not providing citations, and ends by telling readers to go look it up, I tend to presume that the reason no citations are provided is that the writer has none, and that what he is claiming is most likely false.

    Reply
  17. Pat

    @ infopractical
    > Pat, it’s interesting that you ask for citation
    > from me, but not from Mark when he makes claims
    That’s only interesting if you’re trying to drum up the appearance of a controversy. Mark’s argument may or may not hold water; but I wasn’t commenting on Mark’s argument, I was commenting on yours. That doesn’t mean that I agree with Mark, just that I find yours to be lacking in rigor. If you want to attack Mark’s argument as lacking supporting evidence, feel free to do so… but once you start introducing your own propositions, you have to be able to defend them. And you are clearly introducing your own propositions.
    > But you’re telling me that “Most of the papers
    > and books that I see with a cursory examination
    > of the outstanding academic literature actually
    > claim the opposite; that the argument that “tax
    > cuts spur job creation” is not supported.” when
    > there is very clearly a raging debate that has
    > been going on for decades. This tells me that
    > you are intentionally isolating yourself from
    > half the literature that you must also have
    > gaping holes in your economics education.
    How does a self-described “cursory examination” in any way imply “intentional isolation”? If anything, it’s precisely the opposite. I’m fessing up to several weaknesses… namely, that I don’t consider myself an expert in economics and that I’m not extremely well read in the field.
    However, I do have access to two universities’ online libraries and their electronic journals, and that’s what I see with a cursory examination. This is admittedly very weak sauce, as there is little in the way of in-depth analysis of the results, but this is where you have an opportunity. Show me that my very cursory examination has major problems by providing me something to look at.
    Put another way, if I’m looking around a farm and I see something that looks like a duck and walks like a duck, I’m not going to bother to see if it quacks like a duck before I start to assume that it’s a duck. This is perfectly reasonable.
    If you know that it crows like a rooster, pointing that out to me makes me re-evaluate my assumption. Telling me that I’m lazy or ignorant because I haven’t sat around and waited to see if it quacks like a duck isn’t very compelling.
    Particularly if you *are* an expert, you should have a nice big block of citations at the ready. See, this is how one becomes an expert, by reading other experts. This is why there are software packages devoted entirely to tracking citations for people who do research. Asking for citations isn’t obfuscating the argument, it’s asking you to show that you actually have something behind your argument other than your opinion. If you do, you may actually convince me that you’re correct… which should be the point of the exercise, no?
    > Or will you simply take Mark’s word for it all?
    I don’t take Mark’s word on the economics, since he’s not an economist. But I’ve been reading this blog for a while, and the total volume and quality of the posts here is sufficient evidence to establish Mark as a general expert on mathematics, in spite of his Ph.D. in Computer Science (geek joke there, for the humor-challenged).
    Anyone who’s read a sufficient volume of math (particularly statistics) knows that Mark’s point regarding statistical metric “bait-n-switch”es is a common problem. When someone well versed in the mathematics of statistics critiques an argument (be it in economics or another social science or a population study in biology or whatever other field), it doesn’t take advanced knowledge of the other field to accept the critique as having some teeth… even if the argument is commonly accepted in the other field. Bad math is bad math, and it’s incredibly common in all fields.
    You defend a critical analysis of a statistical argument by providing additional evidence that the analysis has justifiable basis, not by claiming that the the mathematician doesn’t understand the field that he’s critiquing.

    Reply

Leave a Reply