13 ms·

Milton Friedman's Thermostat

123 points by cousin_it 14y ago

kevin_morrill 14y ago
Really interesting. Biggest complaint is the author takes a long time to explain the thermostat model. Here is the definition: Everybody knows that if you press down on the gas pedal the car goes faster, other things equal, right? And everybody knows that if a car is going uphill the car goes slower, other things equal, right? But suppose you were someone who didn't know those two things. And you were a passenger in a car watching the driver trying to keep a constant speed on a hilly road. You would see the gas pedal going up and down. You would see the car going downhill and uphill. But if the driver were skilled, and the car powerful enough, you would see the speed stay constant. So, if you were simply looking at this particular "data generating process", you could easily conclude: "Look! The position of the gas pedal has no effect on the speed!"; and "Look! Whether the car is going uphill or downhill has no effect on the speed!"; and "All you guys who think that gas pedals and hills affect speed are wrong!"
- 1337biz 14y ago
  I'm really looking forward to the comments on this one. Overall it is a variation on the "correlation is not causation" theme and that defining variables in a social setting is super tricky, if not to say impossible.
  - cousin_it 14y ago
    It seems to me that when people say "correlation is not causation", they typically mean you can't learn about a causal relationship by just looking at the correlation. The article makes the converse statement: you can't infer the absence of a causal relationship from the absence of correlation, or equivalently (contrapositive), even when you know there is a causal relationship, it won't necessarily lead to a visible correlation in the data.
    - endersshadow 14y ago
      Well, more to the point: The data analysis game changes entirely once you've introduced an intelligent actor (in this case the driver).
      - cousin_it 14y ago
        Feedback processes don't have to be intelligent, e.g. the thermostat mentioned in the title.
  - Symmetry 14y ago
    It's actually much more subtle than "correlation is not causation". More like "goal seeking behavior can remove all correlation from a causal relationship," which is very different since causation normally causes correlation, and normally is you don't see correlation you can be sure that there isn't any causation.
SeanLuke 14y ago
> And no, you can not get around this problem by doing a multivariate regression of speed on gas pedal and hill. That's because gas pedal and hill will be perfectly colinear. Um, what? For a constant speed, which is this person's example, this claim is completely false.
- drunkpotato 14y ago
  When the incline increases, the driver pushes the gas pedal harder. Inline up correlates with gas pedal up. It's a statement of relation between gas pedal and hill incline. The claim is correct.
pdeuchler 14y ago
I think the better question is why are (almost all) economists unaware of basic economic theory? Not to go off topic, but it's hard to take any "economist" talking head seriously these days. I'm currently re-reading The Road to Serfdom by F.A. Hayek (highly recommend) and it's scary to see that even 60 years ago we understood cycles and events that are still being ignored today. The sad fact remains that real economic policy is no longer in vogue in Washington or the L.A. TV sets. What is popular these days is hand wavy gestures that seem to please the most amount of people... the science be damned. Rather scary stuff.
- rvkennedy 14y ago
  I think you underestimate the influence of Hayek, Friedman et al, whose moralistic market fetish has become the dominant ideology of Washington, London et al. For a more evidence-based approach, I'll go with Keynes' General Theory, which works, but has few adherents among the major parties in the West.
  - loganfrederick 14y ago
    Provide evidence that Hayek and Friedman are the dominant ideology in Washington. If we wanted to try and accurately describe the economic policy by the government in Washington, you'd have to describe the monetary and fiscal policies. Both of these since 2007 have been Keynesian far more than the other two schools (Monetarist and Austrian). This hasn't always been the case (Friedman was much more in vogue from the late 70s to approximately late 90s as measured by policies put in place). Keynes General Theory could theoretically work as he describes in his book, but is unrealistic in practice because of the political dynamics in government.
    - yk 14y ago
      The idea that markets usually produce a optimal outcome is certainly austrian ( examples for this is the washington consensus). On the other hand, there are serveral proposals from 2008 - 09 on the size of the stimulus, eg [1], which argue that a stimulus should be something like twice the size of the actual stimulus. ( And additionally bailing out car companies is not exactly Keynesian, since aggregate demand is not about one specific industry.) [1]http://krugman.blogs.nytimes.com/2009/10/05/the-story-of-the-stimulus/ http://krugman.blogs.nytimes.com/2009/10/05/the-story-of-the...
nhebb 14y ago
It sounds like the problem boils down to the difference between understanding a scalar (speed) and a vector (velocity). So economics would need a way to represent economic forces similar to the way a free body diagram is used in engineering and physics. And in order to do that, you would need an economics coordinate system. And therein lies the problem of economists building models without known or absolute reference points. Economics != Physics, QED.
- cousin_it 14y ago
  I think I can make up a variant of the problem that doesn't boil down to scalar vs vector. Imagine the driver is still trying to keep a constant speed, but the road is now horizontal, and the car gets slowed down by mud instead of hills.
  - Evbn 14y ago
    That's exactly the same. Vector refers to directionality, not number of dimensions.
    - pseut 14y ago
      But the only thing varying is a scalar, is the point. The fact that you can write it as a vector is irrelevant.
- ajuc 14y ago
  Why? The analogy works just as well for temperature of water in a boiler, and the thermostat that modifies heating power inversely to the current temperature of the water. Both are scalars in that example, and the basic problem remains.
  - nhebb 14y ago
    You have a point there. But going with your example, the ideal gas law (PV = nRT) still models a system where all inputs are known. While not a perfect model, it's much better for predicting the performance of a thermodynamic system than most economic models are at predicting economic systems.
rayiner 14y ago
The level of discourse in policy-oriented economics papers is shockingly low. My pet policy issue is spectrum management (allocation of radio frequency ranges). Its a subject that lends itself to "neat" economic arguments. Ronald Coase wrote an instrumental paper in 1959 suggesting we should propertize spectrum and allow trading, which would lead to efficient allocations. It saddens me how many policy papers I see from "think tanks" that completely ignore the second part of that paper, as well as Coase's earlier work: the efficient allocation assumes that trading spectrum has negligible transaction costs. This works for say big TV stations in a city, not so much for say allocating spectrum for personal wireless devices. In my research for a seminar paper, I came across people who really should know better who ignored crucial features of the problem domain, ignored transaction costs, ignored considerations of lock-in, etc. I'm tremendously skeptical of economics-based policy papers. A biologist works on a paper for years, and concludes that this kind of gene expression might be linked to this kind of nutrient deficiency. An economist at a "think tank" works on a paper for a fraction of that time and gives advice on how to structure a tens of billions of $ industry...
- 1337biz 14y ago
  The basic problem is here in my perspective, that Think Tanks almost always subscribe to a specific line of thought and exist to promote this cause. While on the other hand less political motivated researchers are too much in love with constructing models, that nobody outside of their peer group understands and often come with questionable value for real life implications.
  - rayiner 14y ago
    Possibly. I read some absolute drek as well in a U of Chicago symposium on the subject... Re: think tanks, isn't it something that even in supposedly less rigorous fields like, oh English literature, such upfront results-orientation would be considered the badge of unserious scholarship?
    - stdbrouw 14y ago
      > Re: think tanks, isn't it something that even in supposedly less rigorous fields like, oh English literature, such upfront results-orientation would be considered the badge of unserious scholarship? Nowhere near as fishy as how think tanks work, but how scientists and academics think about hypotheses is actually very complex and you'd be surprised at how often unserious scholarship is tolerated. E.g. it's not uncommon to first write a paper and then tag on a hypothesis that just happens to agree with your conclusion, to change the hypothesis when your own research proves it to be wrong (it's hard to get research published about things that didn't work), or to try everything you can to prove a hypothesis, rather than to try and falsify it.
eavc 14y ago
>And no, you can not get around this problem by doing a multivariate regression of speed on gas pedal and hill. I'm no expert on statistics, but this seems like a pretty basic scenario for a statistical analysis to provide insight into. These two phenomena would be perfectly correlated in both the time of appearance and the degree. You wouldn't be able to derive anything about the effect of hills or gas pedals on speed, but you'd be given a powerful clue as to what's going on. The second there's even a tiny imbalance between the two, you're given the relationship to speed.
- simonster 14y ago
  Well, in this contrived scenario, you can't tell what would happen if the driver _didn't_ push down on the gas pedal while going up the hill. The relationship between the gas pedal and the speed of the car is entirely obscured. However, in a real world system, the chance that there is zero imbalance is also nearly infinitesimal. You could take the most skilled driver in the world and you would still be able to pull out the relationship between gas pedal and speed out of his driving pattern. Similarly, with the right variables and enough data, you could pull a relationship like the one described here out of economic data. (As a side note, I think the way in which this article is written is pretty obnoxious. The first 500 words are dedicated to the author congratulating himself for being aware of an idea that he hasn't even explained yet.)
- ratsbane 14y ago
  Exactly. This whole article could have been distilled down to http://en.wikipedia.org/wiki/Multicollinearity http://en.wikipedia.org/wiki/Multicollinearity (but I still thought it was a good read)
klochner 14y ago
Am I the only one who read Friedman's paper? [1] The answer to OP's question [2] - "why are economists unaware of milton friedman's thermostat" - is because the OP is misinterpreting Friedman's paper, and therefore has misattributed this nonexistent "theory". Friedman just said that the Fed acts like a thermostat, and in the linked paper, claimed that the period subsequent to 1985 had increased price stability owing to improvements in their understanding of inflation and it's relation to the quantity of money, hence their "better thermostat". AFAICT, the OP has made up this "Friedman's Thermostat" insofar as it relates to economic forecasting, and now he wants to know why no one has heard of it. As an aside, "exception that proves the rule" is an almost universally misapplied phrase, that should only be used if you're really really sure you know what it means (kind of like "begs the question"). [1] http://online.wsj.com/article/0,,SB106125694925954100,00.html http://online.wsj.com/article/0,,SB106125694925954100,00.htm... [2] OP = Nick Rowe - http://worthwhile.typepad.com/worthwhile_canadian_initi/about-nick-rowe.html http://worthwhile.typepad.com/worthwhile_canadian_initi/abou...
- taylorwc 14y ago
  Agreed. An analogy made by Friedman in a WSJ op-ed != established theory.
- Symmetry 14y ago
  Right, if you find yourself asking "Why has nobody else noticed this" you should always consider that the answer is "because it isn't there." But on the other hand, the idea that intentional behavior can remove correlation from causal relationships is important and deserves more thought than people give it. And since economics has something of a history of naming things for the wrong person, calling this "Friedman's Thermostat" and spreading it around might be a decent idea after all.
- j2labs 14y ago
  The delivery is so emotionally charged too. It took several paragraphs of "I can't believe no one gets this!" before we even got to Milton's theory.
  - deleted 14y ago
eykanal 14y ago
OK, I think I'm missing something. To quote the article: > And no, you can not get around this problem by doing a multivariate regression of speed on gas pedal and hill. That's because gas pedal and hill will be perfectly colinear. This is one of the reasons why you check for multicollinearity [1] when performing multivariate linear regressions; aside from introducing significant instability into the model, if your predictors are correlated, then you're essentially measuring the same thing twice. The stated problem is actually a good example as to why you should avoid having highly correlated predictors in a multivariate model; by blindly pursuing the regression despite the inputs being correlated, we miss out on the fact that there is a relationship present (i.e., the dependent variable (speed) is actually a factor of both of the independent variables (pedal height and hill slope)). [1]: http://en.wikipedia.org/wiki/Multicollinearity http://en.wikipedia.org/wiki/Multicollinearity
- Irregardless 14y ago
  Sounds like the nail in the coffin on this topic. It always sucks when you accuse an entire profession of overlooking a fundamental principle, only to find out you've committed the same error in a more egregious fashion.
- cynicalkane 14y ago
  You aren't missing anything. The author's article isn't about the optimal ways to do statistics, it's a criticism of the informal economic discussion that dominates policy, blogs, and pop-economics--and, according to the post, an unfortunate number of actual econ papers--that says things like "economic plan X doesn't work because there hasn't been a change". The criticism generalizes to any statistical methods, though. You always need to think about the underlying model before doing any kind of statistics in any science.
  - chimeracoder 14y ago
    Yes, but I wouldn't say that "almost all" economists don't understand the principle (as in the title). I would be more likely to say that "almost all" reporters don't understand basic economics & statistics.
mathattack 14y ago
A similar problem happened at Harvard Business School. In the mid 80s they found no correlation between GMAT scores and success at the school, so they stopped requiring applicants to submit scores. This was a thermostat-unaware view. The real answer was that they were giving it the right amount of weight. If they were overweighting the GMAT, unqualified people with high GMATs would get in. If they were not underweighting it, the opposite. Approximately 10 years later they did a correlation of student GMATs (which weren't used for admissions purposes) and school success, and found that there was indeed now a correlation. The lesson was that if you're properly managing a factor, you won't see it in the outcome. Or in the thermostat example, if you're properly managing the thermostat, you won't see the changes.
vavoida 14y ago
find that explanation shorter & faster, although polsci http://themonkeycage.org/blog/2012/07/31/milton-friedmans-thermostat/ http://themonkeycage.org/blog/2012/07/31/milton-friedmans-th...
tomrod 14y ago
In addition to klochner's comment, this is why economists rely on exogenous variation. Any applied econometrician worth his or her salt would not say "the gas pedal has no effect", but that the data doesn't rule out the hypothesis that it does.
dbecker 14y ago
I am an econometrician... exactly the sub-population that the author faults most for not knowing about Friedman's thermostat. We haven't heard of the thermostat because it does not introduce new ideas to us. If the data includes hill height, this is an example of multicollinearity. If your data does not include hill height, this is an example of endogeneity. Both concepts are taught in considerable depth in an undergraduates first course in econometrics. It's a nice examples, but the author's claim that these phenomena are outside are awareness shows a striking ignorance of basic econometrics.
- ScottBurson 14y ago
  Okay, fine, but then what do you think of the Casey Mulligan column that the author links to (third paragraph)? It certainly seems at first glance to contain instances of the error the author is describing. (I am not an economist and have no position on this; just curious.)
  - dbecker 14y ago
    That's a great question, and I'm afraid I'm not going to do it justice. But the place where I think you see the thermostat concern is in the paragraph: A 1983 study by Lars Peter Hansen of the University of Chicago and Kenneth Singleton of Stanford showed that short-term rates on Treasury bills and short-term returns on stocks traded on the New York Stock Exchange had very little correlation with consumer spending. Many empirical studies have confirmed this sort of result (this comparison of inflation-adjusted Treasury bill returns and business sector profitability is a recent example). This summarizes "many empirical studies" in a single sentence. Unfortunately, it doesn't include citations. There are a number of ways to address the "thermostat" problem. Each potential solution only works in specific circumstances. I'd hope the underlying research adequately addresses this concern, but it's hard to say without seeing the research. In general, summarizing "many" 40 page papers in a paragraph is inherently difficult, and Mulligan seems to be especially vague here. If the claims here have any credibility, it would only be through an "appeal to authority," which most of us won't find very compelling. So to answer your question, I'm unimpressed with the Mulligan column... in part because we have no indication what problems there are in the underlying analysis.
DannyBee 14y ago
As a meta-comment, this article dances around and only after an entire page of text, does it get to the actual idea.
danielweber 14y ago
Why does it 11 paragraphs repeating himself about how important this idea is? I thought this was some set-up for a joke in which he never tells us about the thermostat.
pseut 14y ago
I have no idea if the author describes "friedman's thermostat" correctly, but to say that Economists are unaware of the phenomenon he describes is beyond absurd (disclosure: I'm an Economist!). There's another way to find the relationship between the gas pedal, slope, and speed that the author did not mention (but is somewhat related to some of his suggestions): write down a model of the driver that assumes he or she uses the gas pedal to try to keep speed constant: E_{t-1} speed(gas_t, slope_t, parameters) = const (eq1) where E_{t-1} is the conditional expectation of the period-t term, given the information available in period t-1. Maybe speed should be nonparametric, but that will introduce some new problems in estimation; maybe you know enough about physics to write down a parametric formula for speed; whatever, the exact details of speed(.,.) are kind of beside the point for the author's argument. Now, speed(gas_t, slope_t, parameters) - E_{t-1} speed(gas_t, slope_t, parameters) (eq2) is a martingale difference sequence when "parameters" is set to their true value, so the sequence equals zero in expectation for all t, assuming the model is true, and that can be the basis for estimation through, say, Generalized Method of Moments. Because combining (eq1) and (eq2) gives us E speed(gas_t, slope_t, parameters) = const, so you can estimate the parameters as \hat parameters = argmin (average_over_t speed(gas_t, slope_t, parameters) - const)^2 Everything I laid out is an extreme simplification of the DOMINANT STRATEGY in applied macro (with, in all likelihood, some errors due to sloppiness). Now, for any realistic economy, it's going to be hard as hell to write down a sensible formula for "speed(.,.,.)" and since there aren't that many years since WWII (which is kind of seen as the beginning of the "modern" economy) there isn't a lot of data to estimate the model, but neither of those issues have anything to do with the author's "critique" and are basically the direction of almost all research in macro.
- ignostic 14y ago
  At least the author admits that a lot of people know more about this stuff than he does. It's tough to make big assertions when you're relatively naive.
- Nick_Rowe 14y ago
  pseut: "so you can estimate the parameters as \hat parameters = argmin (average_over_t speed(gas_t, slope_t, parameters) - const)^2" No you can't. Think about it. 1. If the driver has as much information as the econometrician, setting both parameters on slope and gas equal to 0 will fit the data equally well. 2. If the econometrician observes slope and gas and speed with error (as will almost always be the case), then GMM will estimate both parameters on slope and gas as zero (if those 3 errors are independent of each other), even if the true parameters aren't. 3. If the driver observes gas or slope with error, and the econometrician doesn't, then the econometrician can indeed actually estimate the parameters. How likely is 3? Only a stupidly arrogant econometrician would assume a priori that he knows better how to drive the car than the guy actually driving it. (OK, maybe the econometrician has final revised data, and the driver has only real time data, and the final revised data is better than the real time data. But even then the driver will be observing other indicators that the econometrician doesn't have data on, so the econometrician will interpret the driver's response to those omitted variables as "gas pedal shocks", and will screw up the estimation royally. Yep, all that Sims VAR stuff is wrong. Here, read this. http://worthwhile.typepad.com/worthwhile_canadian_initi/2011/10/james-hamilton-on-christopher-sims-and-identifying-monetary-policy-shocks.html http://worthwhile.typepad.com/worthwhile_canadian_initi/2011... You have just proved my point: economists don't understand Milton Friedman's thermostat. Forget your fancy stuff. Just STOP AND THINK about what you are really doing when you try to estimate parameters.
ignostic 14y ago
I thought he was never going to get to the point. This entire post should be about 1/5th the size.
wissler 14y ago
This is a subset of David Hume's analysis of causality, see his "An Enquiry Concerning Human Understanding."
baddox 14y ago
There are exactly 500 words before a single bit of substance in this article.

123 points by cousin_it 14y ago

Really interesting. Biggest complaint is the author takes a long time to explain the thermostat model. Here is the definition: Everybody knows that if you press down on the gas pedal the car goes faster, other things equal, right? And everybody knows that if a car is going uphill the car goes slower, other things equal, right? But suppose you were someone who didn't know those two things. And you were a passenger in a car watching the driver trying to keep a constant speed on a hilly road. You would see the gas pedal going up and down. You would see the car going downhill and uphill. But if the driver were skilled, and the car powerful enough, you would see the speed stay constant. So, if you were simply looking at this particular "data generating process", you could easily conclude: "Look! The position of the gas pedal has no effect on the speed!"; and "Look! Whether the car is going uphill or downhill has no effect on the speed!"; and "All you guys who think that gas pedals and hills affect speed are wrong!"

1337biz 14y ago
I'm really looking forward to the comments on this one. Overall it is a variation on the "correlation is not causation" theme and that defining variables in a social setting is super tricky, if not to say impossible.
- cousin_it 14y ago
  It seems to me that when people say "correlation is not causation", they typically mean you can't learn about a causal relationship by just looking at the correlation. The article makes the converse statement: you can't infer the absence of a causal relationship from the absence of correlation, or equivalently (contrapositive), even when you know there is a causal relationship, it won't necessarily lead to a visible correlation in the data.
  - endersshadow 14y ago
    Well, more to the point: The data analysis game changes entirely once you've introduced an intelligent actor (in this case the driver).
    - cousin_it 14y ago
      Feedback processes don't have to be intelligent, e.g. the thermostat mentioned in the title.
- Symmetry 14y ago
  It's actually much more subtle than "correlation is not causation". More like "goal seeking behavior can remove all correlation from a causal relationship," which is very different since causation normally causes correlation, and normally is you don't see correlation you can be sure that there isn't any causation.

SeanLuke 14y ago

> And no, you can not get around this problem by doing a multivariate regression of speed on gas pedal and hill. That's because gas pedal and hill will be perfectly colinear. Um, what? For a constant speed, which is this person's example, this claim is completely false.

drunkpotato 14y ago
When the incline increases, the driver pushes the gas pedal harder. Inline up correlates with gas pedal up. It's a statement of relation between gas pedal and hill incline. The claim is correct.

pdeuchler 14y ago

I think the better question is why are (almost all) economists unaware of basic economic theory? Not to go off topic, but it's hard to take any "economist" talking head seriously these days. I'm currently re-reading The Road to Serfdom by F.A. Hayek (highly recommend) and it's scary to see that even 60 years ago we understood cycles and events that are still being ignored today. The sad fact remains that real economic policy is no longer in vogue in Washington or the L.A. TV sets. What is popular these days is hand wavy gestures that seem to please the most amount of people... the science be damned. Rather scary stuff.

rvkennedy 14y ago
I think you underestimate the influence of Hayek, Friedman et al, whose moralistic market fetish has become the dominant ideology of Washington, London et al. For a more evidence-based approach, I'll go with Keynes' General Theory, which works, but has few adherents among the major parties in the West.
- loganfrederick 14y ago
  Provide evidence that Hayek and Friedman are the dominant ideology in Washington. If we wanted to try and accurately describe the economic policy by the government in Washington, you'd have to describe the monetary and fiscal policies. Both of these since 2007 have been Keynesian far more than the other two schools (Monetarist and Austrian). This hasn't always been the case (Friedman was much more in vogue from the late 70s to approximately late 90s as measured by policies put in place). Keynes General Theory could theoretically work as he describes in his book, but is unrealistic in practice because of the political dynamics in government.
  - yk 14y ago
    The idea that markets usually produce a optimal outcome is certainly austrian ( examples for this is the washington consensus). On the other hand, there are serveral proposals from 2008 - 09 on the size of the stimulus, eg [1], which argue that a stimulus should be something like twice the size of the actual stimulus. ( And additionally bailing out car companies is not exactly Keynesian, since aggregate demand is not about one specific industry.) [1]http://krugman.blogs.nytimes.com/2009/10/05/the-story-of-the-stimulus/ http://krugman.blogs.nytimes.com/2009/10/05/the-story-of-the...

nhebb 14y ago

It sounds like the problem boils down to the difference between understanding a scalar (speed) and a vector (velocity). So economics would need a way to represent economic forces similar to the way a free body diagram is used in engineering and physics. And in order to do that, you would need an economics coordinate system. And therein lies the problem of economists building models without known or absolute reference points. Economics != Physics, QED.

cousin_it 14y ago
I think I can make up a variant of the problem that doesn't boil down to scalar vs vector. Imagine the driver is still trying to keep a constant speed, but the road is now horizontal, and the car gets slowed down by mud instead of hills.
- Evbn 14y ago
  That's exactly the same. Vector refers to directionality, not number of dimensions.
  - pseut 14y ago
    But the only thing varying is a scalar, is the point. The fact that you can write it as a vector is irrelevant.
ajuc 14y ago
Why? The analogy works just as well for temperature of water in a boiler, and the thermostat that modifies heating power inversely to the current temperature of the water. Both are scalars in that example, and the basic problem remains.
- nhebb 14y ago
  You have a point there. But going with your example, the ideal gas law (PV = nRT) still models a system where all inputs are known. While not a perfect model, it's much better for predicting the performance of a thermodynamic system than most economic models are at predicting economic systems.

rayiner 14y ago

The level of discourse in policy-oriented economics papers is shockingly low. My pet policy issue is spectrum management (allocation of radio frequency ranges). Its a subject that lends itself to "neat" economic arguments. Ronald Coase wrote an instrumental paper in 1959 suggesting we should propertize spectrum and allow trading, which would lead to efficient allocations. It saddens me how many policy papers I see from "think tanks" that completely ignore the second part of that paper, as well as Coase's earlier work: the efficient allocation assumes that trading spectrum has negligible transaction costs. This works for say big TV stations in a city, not so much for say allocating spectrum for personal wireless devices. In my research for a seminar paper, I came across people who really should know better who ignored crucial features of the problem domain, ignored transaction costs, ignored considerations of lock-in, etc. I'm tremendously skeptical of economics-based policy papers. A biologist works on a paper for years, and concludes that this kind of gene expression might be linked to this kind of nutrient deficiency. An economist at a "think tank" works on a paper for a fraction of that time and gives advice on how to structure a tens of billions of $ industry...

1337biz 14y ago
The basic problem is here in my perspective, that Think Tanks almost always subscribe to a specific line of thought and exist to promote this cause. While on the other hand less political motivated researchers are too much in love with constructing models, that nobody outside of their peer group understands and often come with questionable value for real life implications.
- rayiner 14y ago
  Possibly. I read some absolute drek as well in a U of Chicago symposium on the subject... Re: think tanks, isn't it something that even in supposedly less rigorous fields like, oh English literature, such upfront results-orientation would be considered the badge of unserious scholarship?
  - stdbrouw 14y ago
    > Re: think tanks, isn't it something that even in supposedly less rigorous fields like, oh English literature, such upfront results-orientation would be considered the badge of unserious scholarship? Nowhere near as fishy as how think tanks work, but how scientists and academics think about hypotheses is actually very complex and you'd be surprised at how often unserious scholarship is tolerated. E.g. it's not uncommon to first write a paper and then tag on a hypothesis that just happens to agree with your conclusion, to change the hypothesis when your own research proves it to be wrong (it's hard to get research published about things that didn't work), or to try everything you can to prove a hypothesis, rather than to try and falsify it.

eavc 14y ago

>And no, you can not get around this problem by doing a multivariate regression of speed on gas pedal and hill. I'm no expert on statistics, but this seems like a pretty basic scenario for a statistical analysis to provide insight into. These two phenomena would be perfectly correlated in both the time of appearance and the degree. You wouldn't be able to derive anything about the effect of hills or gas pedals on speed, but you'd be given a powerful clue as to what's going on. The second there's even a tiny imbalance between the two, you're given the relationship to speed.

simonster 14y ago
Well, in this contrived scenario, you can't tell what would happen if the driver _didn't_ push down on the gas pedal while going up the hill. The relationship between the gas pedal and the speed of the car is entirely obscured. However, in a real world system, the chance that there is zero imbalance is also nearly infinitesimal. You could take the most skilled driver in the world and you would still be able to pull out the relationship between gas pedal and speed out of his driving pattern. Similarly, with the right variables and enough data, you could pull a relationship like the one described here out of economic data. (As a side note, I think the way in which this article is written is pretty obnoxious. The first 500 words are dedicated to the author congratulating himself for being aware of an idea that he hasn't even explained yet.)
ratsbane 14y ago
Exactly. This whole article could have been distilled down to http://en.wikipedia.org/wiki/Multicollinearity http://en.wikipedia.org/wiki/Multicollinearity (but I still thought it was a good read)

klochner 14y ago

Am I the only one who read Friedman's paper? [1] The answer to OP's question [2] - "why are economists unaware of milton friedman's thermostat" - is because the OP is misinterpreting Friedman's paper, and therefore has misattributed this nonexistent "theory". Friedman just said that the Fed acts like a thermostat, and in the linked paper, claimed that the period subsequent to 1985 had increased price stability owing to improvements in their understanding of inflation and it's relation to the quantity of money, hence their "better thermostat". AFAICT, the OP has made up this "Friedman's Thermostat" insofar as it relates to economic forecasting, and now he wants to know why no one has heard of it. As an aside, "exception that proves the rule" is an almost universally misapplied phrase, that should only be used if you're really really sure you know what it means (kind of like "begs the question"). [1] http://online.wsj.com/article/0,,SB106125694925954100,00.html http://online.wsj.com/article/0,,SB106125694925954100,00.htm... [2] OP = Nick Rowe - http://worthwhile.typepad.com/worthwhile_canadian_initi/about-nick-rowe.html http://worthwhile.typepad.com/worthwhile_canadian_initi/abou...

taylorwc 14y ago
Agreed. An analogy made by Friedman in a WSJ op-ed != established theory.
Symmetry 14y ago
Right, if you find yourself asking "Why has nobody else noticed this" you should always consider that the answer is "because it isn't there." But on the other hand, the idea that intentional behavior can remove correlation from causal relationships is important and deserves more thought than people give it. And since economics has something of a history of naming things for the wrong person, calling this "Friedman's Thermostat" and spreading it around might be a decent idea after all.
j2labs 14y ago
The delivery is so emotionally charged too. It took several paragraphs of "I can't believe no one gets this!" before we even got to Milton's theory.
- deleted 14y ago

eykanal 14y ago

OK, I think I'm missing something. To quote the article: > And no, you can not get around this problem by doing a multivariate regression of speed on gas pedal and hill. That's because gas pedal and hill will be perfectly colinear. This is one of the reasons why you check for multicollinearity [1] when performing multivariate linear regressions; aside from introducing significant instability into the model, if your predictors are correlated, then you're essentially measuring the same thing twice. The stated problem is actually a good example as to why you should avoid having highly correlated predictors in a multivariate model; by blindly pursuing the regression despite the inputs being correlated, we miss out on the fact that there is a relationship present (i.e., the dependent variable (speed) is actually a factor of both of the independent variables (pedal height and hill slope)). [1]: http://en.wikipedia.org/wiki/Multicollinearity http://en.wikipedia.org/wiki/Multicollinearity

Irregardless 14y ago
Sounds like the nail in the coffin on this topic. It always sucks when you accuse an entire profession of overlooking a fundamental principle, only to find out you've committed the same error in a more egregious fashion.
cynicalkane 14y ago
You aren't missing anything. The author's article isn't about the optimal ways to do statistics, it's a criticism of the informal economic discussion that dominates policy, blogs, and pop-economics--and, according to the post, an unfortunate number of actual econ papers--that says things like "economic plan X doesn't work because there hasn't been a change". The criticism generalizes to any statistical methods, though. You always need to think about the underlying model before doing any kind of statistics in any science.
- chimeracoder 14y ago
  Yes, but I wouldn't say that "almost all" economists don't understand the principle (as in the title). I would be more likely to say that "almost all" reporters don't understand basic economics & statistics.

mathattack 14y ago

A similar problem happened at Harvard Business School. In the mid 80s they found no correlation between GMAT scores and success at the school, so they stopped requiring applicants to submit scores. This was a thermostat-unaware view. The real answer was that they were giving it the right amount of weight. If they were overweighting the GMAT, unqualified people with high GMATs would get in. If they were not underweighting it, the opposite. Approximately 10 years later they did a correlation of student GMATs (which weren't used for admissions purposes) and school success, and found that there was indeed now a correlation. The lesson was that if you're properly managing a factor, you won't see it in the outcome. Or in the thermostat example, if you're properly managing the thermostat, you won't see the changes.

vavoida 14y ago

find that explanation shorter & faster, although polsci http://themonkeycage.org/blog/2012/07/31/milton-friedmans-thermostat/ http://themonkeycage.org/blog/2012/07/31/milton-friedmans-th...

tomrod 14y ago

In addition to klochner's comment, this is why economists rely on exogenous variation. Any applied econometrician worth his or her salt would not say "the gas pedal has no effect", but that the data doesn't rule out the hypothesis that it does.

dbecker 14y ago

I am an econometrician... exactly the sub-population that the author faults most for not knowing about Friedman's thermostat. We haven't heard of the thermostat because it does not introduce new ideas to us. If the data includes hill height, this is an example of multicollinearity. If your data does not include hill height, this is an example of endogeneity. Both concepts are taught in considerable depth in an undergraduates first course in econometrics. It's a nice examples, but the author's claim that these phenomena are outside are awareness shows a striking ignorance of basic econometrics.

ScottBurson 14y ago
Okay, fine, but then what do you think of the Casey Mulligan column that the author links to (third paragraph)? It certainly seems at first glance to contain instances of the error the author is describing. (I am not an economist and have no position on this; just curious.)
- dbecker 14y ago
  That's a great question, and I'm afraid I'm not going to do it justice. But the place where I think you see the thermostat concern is in the paragraph: A 1983 study by Lars Peter Hansen of the University of Chicago and Kenneth Singleton of Stanford showed that short-term rates on Treasury bills and short-term returns on stocks traded on the New York Stock Exchange had very little correlation with consumer spending. Many empirical studies have confirmed this sort of result (this comparison of inflation-adjusted Treasury bill returns and business sector profitability is a recent example). This summarizes "many empirical studies" in a single sentence. Unfortunately, it doesn't include citations. There are a number of ways to address the "thermostat" problem. Each potential solution only works in specific circumstances. I'd hope the underlying research adequately addresses this concern, but it's hard to say without seeing the research. In general, summarizing "many" 40 page papers in a paragraph is inherently difficult, and Mulligan seems to be especially vague here. If the claims here have any credibility, it would only be through an "appeal to authority," which most of us won't find very compelling. So to answer your question, I'm unimpressed with the Mulligan column... in part because we have no indication what problems there are in the underlying analysis.

DannyBee 14y ago

As a meta-comment, this article dances around and only after an entire page of text, does it get to the actual idea.

danielweber 14y ago

Why does it 11 paragraphs repeating himself about how important this idea is? I thought this was some set-up for a joke in which he never tells us about the thermostat.

pseut 14y ago

I have no idea if the author describes "friedman's thermostat" correctly, but to say that Economists are unaware of the phenomenon he describes is beyond absurd (disclosure: I'm an Economist!). There's another way to find the relationship between the gas pedal, slope, and speed that the author did not mention (but is somewhat related to some of his suggestions): write down a model of the driver that assumes he or she uses the gas pedal to try to keep speed constant: E_{t-1} speed(gas_t, slope_t, parameters) = const (eq1) where E_{t-1} is the conditional expectation of the period-t term, given the information available in period t-1. Maybe speed should be nonparametric, but that will introduce some new problems in estimation; maybe you know enough about physics to write down a parametric formula for speed; whatever, the exact details of speed(.,.) are kind of beside the point for the author's argument. Now, speed(gas_t, slope_t, parameters) - E_{t-1} speed(gas_t, slope_t, parameters) (eq2) is a martingale difference sequence when "parameters" is set to their true value, so the sequence equals zero in expectation for all t, assuming the model is true, and that can be the basis for estimation through, say, Generalized Method of Moments. Because combining (eq1) and (eq2) gives us E speed(gas_t, slope_t, parameters) = const, so you can estimate the parameters as \hat parameters = argmin (average_over_t speed(gas_t, slope_t, parameters) - const)^2 Everything I laid out is an extreme simplification of the DOMINANT STRATEGY in applied macro (with, in all likelihood, some errors due to sloppiness). Now, for any realistic economy, it's going to be hard as hell to write down a sensible formula for "speed(.,.,.)" and since there aren't that many years since WWII (which is kind of seen as the beginning of the "modern" economy) there isn't a lot of data to estimate the model, but neither of those issues have anything to do with the author's "critique" and are basically the direction of almost all research in macro.

ignostic 14y ago
At least the author admits that a lot of people know more about this stuff than he does. It's tough to make big assertions when you're relatively naive.
Nick_Rowe 14y ago
pseut: "so you can estimate the parameters as \hat parameters = argmin (average_over_t speed(gas_t, slope_t, parameters) - const)^2" No you can't. Think about it. 1. If the driver has as much information as the econometrician, setting both parameters on slope and gas equal to 0 will fit the data equally well. 2. If the econometrician observes slope and gas and speed with error (as will almost always be the case), then GMM will estimate both parameters on slope and gas as zero (if those 3 errors are independent of each other), even if the true parameters aren't. 3. If the driver observes gas or slope with error, and the econometrician doesn't, then the econometrician can indeed actually estimate the parameters. How likely is 3? Only a stupidly arrogant econometrician would assume a priori that he knows better how to drive the car than the guy actually driving it. (OK, maybe the econometrician has final revised data, and the driver has only real time data, and the final revised data is better than the real time data. But even then the driver will be observing other indicators that the econometrician doesn't have data on, so the econometrician will interpret the driver's response to those omitted variables as "gas pedal shocks", and will screw up the estimation royally. Yep, all that Sims VAR stuff is wrong. Here, read this. http://worthwhile.typepad.com/worthwhile_canadian_initi/2011/10/james-hamilton-on-christopher-sims-and-identifying-monetary-policy-shocks.html http://worthwhile.typepad.com/worthwhile_canadian_initi/2011... You have just proved my point: economists don't understand Milton Friedman's thermostat. Forget your fancy stuff. Just STOP AND THINK about what you are really doing when you try to estimate parameters.

ignostic 14y ago

I thought he was never going to get to the point. This entire post should be about 1/5th the size.

wissler 14y ago

This is a subset of David Hume's analysis of causality, see his "An Enquiry Concerning Human Understanding."

baddox 14y ago

There are exactly 500 words before a single bit of substance in this article.