22 Oct 2013

The Non-Falsifiable “Empirical” Frameworks of Paul Krugman and Scott Sumner

Federal Reserve, Krugman, Scott Sumner 26 Comments

[UPDATE below.]

The defenders of both gentlemen will no doubt frame the matters differently, but my following summaries of four total blogs posts–two from Krugman and two from Sumner–are perfectly accurate.

KRUGMAN #1: In this post Krugman endorses (with a parenthetical caveat) Raj Chetty’s claim that empirical work shows that the extension of unemployment benefits has little effect on the unemployment rate. Krugman then comments, “But are such results actually being used to inform policy debate? Have conservative economists like Casey Mulligan said “OK, we were wrong to argue that extended unemployment benefits are the cause of high unemployment”? … You know the answer.” The overall point of the post is that yes, economics is an empirical science–which Krugman defines as being one where data are used to reject certain hypotheses–but certain economists are not themselves scientists, instead they “treat their field as a form of theology.”

KRUGMAN #2: The very next day, Krugman follows up this above post by writing:

[U]nless you at least try to think in terms of a broader model, all the empirical work in the world can’t answer some questions — and you can all too easily draw the wrong conclusions. Take Chetty’s very example, the effects of unemployment insurance. He reports evidence that extended benefits have only a small effect on the time people spend searching for work. But suppose the result went the other way; would that say that UI was hurting employment? Not necessarily, and I’d say not at all: right now the economy is constrained not by a lack of willing workers but by a lack of demand, so that making workers more choosy about accepting jobs would, to a first approximation, have no effect at all on overall employment. [Bold added.]

==> So, just to summarize, Casey Mulligan isn’t a scientist, but instead a theologian, for refusing to accept the empirical results about unemployment benefits. Yet Krugman elaborates that if the results of that experiment had turned out exactly opposite, Mulligan would still be wrong and Krugman would still be right.

* * *

SUMNER #1: In this post, Scott Sumner reports that John Taylor accused Alan Greenspan of fueling the housing bubble by holding interest rates too low. Greenspan totally denies it. Sumner concludes that Taylor is totally wrong in early phases but somewhere from 0% to 10% right in the year 2006.

SUMNER #2: The very next day, Sumner follows up on the above post by arguing that the Fed caused the housing bubble by keeping interest rates too low in 2006.

==> On this one more than with Krugman, Sumner’s defenders will object and say I’m misconstruing the argument. But I repeat: What I have written above is a perfectly correct description of what happened. That should be disturbing to Keynesians and Market Monetarists, but my falsifiable prediction is that it won’t be.

UPDATE: I should clarify that what Krugman did, in general, isn’t necessarily scandalous. For example, if Casey is arguing that OJ was guilty, while Paul claims that OJ is innocent, they would have to react to evidence about OJ’s alibi in different ways. If store surveillance camera has a video on the other side of town at the time the murders occurred, and OJ shows up on the video, then Paul could say, “Oh come on Casey, just admit you’re wrong.” In contrast, if OJ doesn’t show up on the video, then Paul can still reasonably say, “I still maintain that I’m right.” But (a) this particular example of unemployment isn’t quite like that, since Mulligan could conceivably embed his story in a macro GE effect like this guy says, and (b) Krugman pulls heads-I-win-tails-you-lose all the time, so I just thought this was hilarious when he deploys it in a post complaining that nobody follows the evidence.

Also, here’s Jim Manzi from the comments:

The UI “natural experiment” is not an experiment at all. The set of states that took the action were not randomly selected. It is plausible that those states that extended UI varied systematically form those that did not. For example, they may have had different prospective expectations of future unemployment, or may have had different balance of power between interest groups and on an on in ways that could be related to future employment. Inevitably, this results in the requirement for some kind of econometric adjustment (or more naively, no attempt to correct for bias) that means this is more econometric modeling, and not a controlled experiment or even a reasonable approximation to a controlled experiment. It’s especially bad, since N = 50, and generally n = somewhere between 5 and 20. I’ve gone into this in painful detail for the abortion-crime natural experiment, among others.

26 Responses to “The Non-Falsifiable “Empirical” Frameworks of Paul Krugman and Scott Sumner”

  1. valueprax says:

    Sorry, I am going to go a tangential direction here: alright, so if we know empirically that unemployment benefits don’t have a cost in terms of higher unemployment, shouldn’t we be looking for the cost they DO have in terms of something because no one here is trying to make a free lunch argument?

    In other words, the tradeoff isn’t higher unemployment (according to the study), but there still has to be a tradeoff of some sort, so wouldn’t an honest debate about the issue revolve around identifying what is being given up as an opportunity cost and then debating whether this is a good thing to give up or not in return for unemployment benefits?

    Instead, the participants have become focused on whether or not they’re lying about the outcome of the given experiment. Odd.

    • Gene Callahan says:

      Clearly, they have to cost the people paying the taxes to pay them the opportunity cost of what they gave up.

  2. Wonks Anonymous says:

    I don’t know if I’d say you’re misconstruing Sumner’s argument, though you might be taking it too seriously. He said he was being contrarian and riffing off the argument of one of his commenters. Also, it looks like an example of him changing his mind (to the extent that you take it seriously). He previously denied that the Fed was responsible since so many supposed “bubbles” took place when NGDP growth wasn’t particularly high, assuming the correlation was just zero. Now he’s considering the possibility that the correlation is just the opposite of what many folks assume it to be.

    But speaking of UI, here’s an argument that the effect is through job creation rather than search time:

    • Major_Freedom says:

      It’s pretty obvious why a market monetarist would jump for joy at the theory that the housing bubble was caused by too low inflation.

      Think of a Marxist who heard the theory that central ownership of the means of production really can have a price system for the means of production, and hence the planners really can rationally allocate resources.

      It would feel like a ton of bricks was lifted off one’s shoulders. In fact, that’s probably why the original poster made that comment in the first place.

    • Bob Murphy says:

      Fair enough Wonks Anonymous, you’re right that Scott seemed to be framing it as “I’m refining my views here.” But even so, it was more like, “I’m now giving a more comprehensive explanation instead of the partial truths I offered previously.” He’s not admitting that he was wrong when he said Taylor was incorrect for saying low interest rates caused the housing bubble.

  3. Major_Freedom says:

    Krugman abuses the logical structure of empiricism when he attacks his ideological enemies.

    Empiricism, strictly speaking, makes no apodictic conclusions about the real world (except of course itself, which contradicts, but let’s leave that aside for the moment). It only ever makes claims that certain relationships between variables have or have not yet been observed.

    Empiricism is grounded on skepticism. In other words, even if a theory of a relationship between A and B was not yet observed, and is hence “falsified”, it retains for itself the possibility that the relationship does in fact exist, it’s just that other, previously unaccounted for variables may have corrupted the possibility of B being observed as expected. Empiricism cannot deny that a new experiment, with new control variables, can potentially reveal the relationship between A and B such that B does occur as expected.

    This is how an empiricist, according to empiricism, must operate.

    This is what Krugman did in the second post Murphy cited. Krugman considered the empirical result of UI and unemployment going “the other way”, the bad way, the troublesome way, and like the cool empiricist mind he utilized, he said that it won’t prove apodictically that UI worsens the unemployment situation. No, because a previously unaccounted for variable, “demand”, can be the control variable, such that when controlled for, the test of the relationship between UI and unemployment goes the other way after all, in Krugman’s ideological favor.

    But when Krugman addresses Chetty’s results, which again the empiricist cannot treat as apodictically true, Krugman calls any skeptic of this result, which should in fact be him if he is going to remain a consistent empiricist, a “theologist”.

    Just let that hypocrisy sink in. The correct empiricist approach to Chetty’s results, indeed ANY empirical result, which is to be skeptical of the result, and to question it, and to consider previously unforeseen variables heretofore not controlled for, all of which just so happens to disagree with Krugman’s ideology, are theologians, whereas Krugman is being “scientific”…for violating the scientific foundation of empiricism when it suits his ideology.

    Brutal. Just brutal.

    • valueprax says:


      Oh, I guess my point wasn’t as tangential as I thought, because I think what you’re saying here is what I was saying– if this variable doesn’t have an effect, shouldn’t we be earnestly searching for the one that does?

      Seeing as how no one is believing in a free lunch here. Nope, we wouldn’t make that kind of basic mistake. And we know all about Say’s Law, you need not remind us of that!

      • Major_Freedom says:

        “shouldn’t we be earnestly searching for the one that does?”

        Let’s say while studying history you do find explanatory independent variables whose values do seem to gel with the dependent variable in question.

        What is the foundation for concluding that these relationships will be the same in the future, such that Krugman is justified in calling for a particular government activity tailored to those past relationships?

        Regarding Say’s Law: It may make people’s eyes roll by being repeated so often, but I can tell you that it is one of the most misunderstood, misstated, and elusive economic concepts there are. The correct explanation is worth repeating however many times the law is brought up, either directly or as a useful tool of understanding (and critiquing) a particular type of argument.

        I’ll break my own recommendation and won’t get into the details of it here, but I must say that I think the best way of explaining Say’s Law is by utilizing Bastiat’s framework of “Seen versus Unseen.” Without it, the debate will tend to gravitate towards only what is seen, and when that happens, it is so incredibly easy to conclude that a general overproduction is possible, indeed common. Not only that, but the belief in a general overproduction will seem so plausible, that Say’s Law itself will be viewed as not only wrong, not only ridiculous, but an ideological weapon of deceit designed to protect fat cat capitalist class interests, against the good and pure social democratic forces.

    • Gene Callahan says:

      You give a good analysis of empiricism as a philosophical doctrine. But I think Krugman just means “I pay attention to facts.” (Not that he’s doing that consistently either, but I don’t think he means “empiricism” a la Bertrand Russell.

      • Silas Barta says:

        Gene_Callahan … just complimented … Major_Freedom … on his philosophical analysis.

        For the rest of the day, I’m going to be on the lookout for swine aviation and weather reports of cold fronts coming through Hell.

        • Major_Freedom says:

          I am always in full support of anyone who attempts a drastic change in their lives, assuming of course they remain peaceful, to in whatever way improve their lives (not that I am saying agreeing with me necessarily implies someone will have a better life).

          It is extremely difficult to do, and something I won’t discourage.

        • Gene Callahan says:

          Silas, I calls it as I sees it, and should major freedoms analysis be good, I call it good.

  4. Dan says:

    Krugman is such a tool.

  5. RPLong says:

    You can’t really nail Sumner on stuff like this, because with him the whole name of the game is framing. If you say something about where the Fed sets interest rates, then Sumner just tells you, “Never reason from a price change!” So the Fed sets the rate with a policy objective in mind, but you can’t talk about it because that’s “reasoning from a price change.”

    I mean, it’s pretty obvious that this is just rhetorical sleight-of-hand, isn’t it?

    • RPLong says:

      In fact, the Objectivist in me wants to suggest that the fallacy being committed is “context-dropping.” Taylor is talking about what the Fed’s interest rate policy is. Sumner drops the fact that the Fed has an interest rate policy as the context of the discussion, and instead starts talking about what a market-determined interest rate means in terms of money demand.

      But, no. That’s not right. The whole point of Fed policy is to influence money demand. So why can’t we talk about it? Why is that no longer the context?

      “Blank out.”

      • Major_Freedom says:

        I love reading Objectivists. Really. I can’t tell you how often I pine for the phrases “Blank Out” and “Context Dropping”. Good show.

  6. Jim Manzi says:


    The UI “natural experiment” is not an experiment at all. The set of states that took the action were not randomly selected. It is plausible that those states that extended UI varied systematically form those that did not. For example, they may have had different prospective expectations of future unemployment, or may have had different balance of power between interest groups and on an on in ways that could be related to future employment. Inevitably, this results in the requirement for some kind of econometric adjustment (or more naively, no attempt to correct for bias) that means this is more econometric modeling, and not a controlled experiment or even a reasonable approximation to a controlled experiment. It’s especially bad, since N = 50, and generally n = somewhere between 5 and 20. I’ve gone into this in painful detail for the abortion-crime natural experiment, among others.

    • Bob Murphy says:

      But Jim, that guy Chetty teaches at Harvard. He wouldn’t tell readers of the NYT with confidence something that is wrong…

  7. Silas Barta says:

    What about Bryan_Caplan’s (easily-strawmanned) Bayesian approach on this? “Seeing X would force me to lower my confidence in believe Y, but not to abandon it?”

    • Bob Murphy says:

      That would be fine. When Krugman says, “I wish Mulligan would lower his priors” I will stop blogging about him.

  8. adamastor says:

    I believe there is an obvious objection to this: Mulligan asserted a strong causation from UI to unemployment increases (or non-decreases) and the cited study found only a very slight correlation, which is of course no proof, but a strong evidence against the initial assertion..

    The inverse claim is in no way equivalent, no matter how strong the results, as it would merely show correlation between the two, which tells you nothing without further study.

    None of this means Krugman is right and the others are wrong about the initial assertion, as the study is probably severely lacking as a “proper” controlled experiment, but I don’t think it demonstrates any logical inconsistency on Krugman’s part (which was Bob Murphy initial point)

  9. Gamble says:

    I am going to try and start something new, see if takes flight.

    Since government creates everything, ( ” You didn’t build that”) then they should stop taxing us and just take their cut immediately.

    Yes, they simply should take it off the top of their production rather than first giving us what they created and then taking some back in the form of taxation.

    Hopefully you all realize this will be impossible for them to do since they create nothing however a simple idea like this has to destroy Krugmans and every other statist perpetual garbage mouth.

    I can see Rand Paul or Ted Cruz presenting a bill like this.

  10. kebko says:

    I’m the one that wrote the piece Scott Sumner referenced in Sumner #2 above. If you accept Sumner’s priors, there is no contradiction. You are simply refusing to accept his priors and then blaming your subsequent lack of understanding on him.

    It’s like if you claimed that disease was caused by evil spirits. Then you see Scott say (1) that germs cause disease and then later say (2) that he knew that a witch that didn’t wash her hands would spread disease. And then, you respond, “See! He contradicted himself! He said that witches cause disease!”

    • Bob Murphy says:


      The irony here is that it’s *you* who are refusing to leave your own (and Scott’s) worldview.

      The “Taylor Rule” is couched in terms of interest rates being set as a function of price inflation and the output gap, right? Here, look at Wikipedia.

      So, if Taylor is saying Greenspan screwed up and fueled a bubble, the *cause* Taylor lists is “nominal interest rates that are too low.”

      Scott says he’s wrong, because Scott read “low nominal interest rates” as “Taylor means loose money, and I know money was fine because NGDP growth wasn’t too bad.”

      Then later Scott says, “Actually, the housing bubble was caused by nominal interest rates being too low.”

      So in your analogy, it would be if Scott came back and said yes the disease was caused by evil spirits, but they were actually trying to help and were just really dumb. (or something)

      • kebko says:

        The post on 2000’s home prices is referring to low long term real rates combined with a low inflation premium on the long end of the curve, so the causal factor we are talking about there is not the contemporaneous Fed Funds rate or Taylor rule rate at the short end of the yield curve.

        • Bob Murphy says:

          Sumner wrote:

          As a result, real estate “bubbles” are more likely to occur during periods when nominal interest rates are relatively low and average people find it easier to qualify for mortgage loans. I initially missed this point because I focused too much on the Fisher effect and not enough on the strange practice in America of structuring mortgage payments in nominal terms.

          Yeah, I’m sure Kevin you can come up with 65 caveats–I predicted guys like you would, in the OP–but c’mon, Scott is quite clearly saying that housing bubbles are caused by low nominal interest rates.

Leave a Reply