Tyler Muir suggested this lovely catchphrase, which should stand next to “Correlation does not imply causation” in our menagerie of econometric sayings.
That's a "deepity." It's trivially true insofar as the typical Pearson correlation coefficient only captures linear relationship. At a deeper level it is false insofar Pearson correlation is not the only game in town.
How can we know what we do not know? Are there constants akin to the constants of physics and chemistry to be measured when it comes to human choice and action? What if there are no constants? What if there aren't even variables that have small, stable variance when it comes to human behavior?
I am neither economist nor philosopher, but this sounds like just a variation on the philosophical concept of 'necessary and sufficient' or the idea of INUS (Insufficient but Necessary part of an Unnecessary but Sufficient condition) - used in science, as well. I think it underscores for me that it seems economists really only have retrospective correlation for their theories and impose explanations on them - that is they force theories to fit data. Or, put another way, they don't really have a fundamental apriori first principals theory of human behavior which can predict economic behavior.
Lucas wanted to regain control. So instead of saying the key in economics was making decisions under conditions of uncertainty, he flipped it. He focused on an (1-uncertainty), or more specifically, risk... that portion of the unknown that we think we might know something about as we look to the future. < Use the Force Luc-as>. Now what we know might be only a small portion of what might happen. But once we begin to focus on expectations and say they drive people's behavior, you are in a very different world. and economists are so enthralled to use expectations even if the expectations' ability to predict is near zero...is it significant? To me this is the great failing of Lucas and of expectations as used in economics. Who is really driven by expectations? I mean, except for some unusual circumstances? Most likely if you have expectations, you are more prepared to act once 'X' happens than than that you do 'y' now because you expect "X" in the future.
I disagree. I think people make many decisions based on expectations. There were some studies on AIDS in the early 2000s that showed people made different decisions regarding risky sex based upon future expectation of life span. People in Africa engaged in risky behavior because lifespans weren't as long generally, and because there wasn't a lot of hope for the future regarding income/standard of living. In the US, decisions were different because the expectations of the future were different. Might have been an Emily Oster study but don't quote me on that. Read about it in the old Becker-Posner blog.
Not a new world at all... High 't' statistics and low R-bar Square. Welcome to my world.
Drug Cos love to talk of the power of statins to reduce the risk of heart attack. The risk can be reduced by 50% or more! with statins. from (not the real numbers...sorry) less than 0.5% to less than 0.025%. Small risk... big (relative) reduction. but people with high - very high- cholesterol levels can and do live a long time. And people with low cholesterol still get heart attacks. And if we take out the TIAs, the impact of statins on reducing 'heart attacks' disappears since they have NO IMPACT on all csause mortality exceept for men under 50 who already have had a heart attack. So, all this is just parsing the statistics. Drug Co' are MASTERS at it. Economtists are pikers by comparion.
But we do get carried away with all this. Do do you want to workshop R-squared or statistical significance? Well, presumably both together, a dual mandate. Good luck.
I'm mostly a fan of the "causal revolution" in economics, but I appreciate your critiques. There is IMO a tradeoff between A) not-so-important questions that we can answer well and B) very important questions for which are answers are ultimately more speculative. The world in which you and I grew up probably focused too much on B questions, but the present world probably focuses too much on the A's.
Housing is a clean case where “causation does not imply variation” shows up in the wild.
Mortgage rate increases do causally raise the discount rate. But once a large share of owners are sitting on very low fixed coupons, the supply elasticity of existing homes collapses. At that point, the clearing margin switches from price to quantity - turnover and composition (new vs. existing) - rather than prices adjusting.
So the causal channel (rates to discount rate) is intact, but the variance from that shock shows up almost entirely in listings and turnover instead of prices. Builders with the ability to buy down rates can still clear at the payment-capacity boundary, while existing owners simply don't sell.
One place this tends to get missed in the asset-pricing literature is that the empirical focus is usually on price response to shocks, not which margin is doing the adjusting. If the margin itself has changed (from price to quantity), then searching for the shock’s imprint on price will naturally make the effect look small or "puzzling". It’s not that the causal effect is weak, it’s that the observed variation is happening on a margin they’re not measuring.
Isn't this why we use Linear Algebra? We take all the variables we think might influence a certain outcome and we get an equation that gives the weights of each variable that maps the history and we can see which ones have a large influence and small influence?
I think, it has been decades since I took that class
This paper might fit into why we don't update our priors as much as expected. I apply it to estimates returns on investment, but it could apply pretty well to significant coefficients in a new and unexpected context too.
``Managerial Conservatism and Rational Information Acquisition, '' Journal of Economics and Management Strategy (Spring 1992), 1(1): 175-202. Conservative managerial behavior can be rational and profit- maximizing. If the valuation of innovations contains white noise and the status quo would be preferred to random innovation, then any innovation that does not appear to be substantially better than the status quo should be rejected. The more successful the firm, the higher the threshold for accepting innovation should be, and the greater the conservative bias. Other things equal, more successful firms will spend less on research, adopt fewer innovations, and be less likely to advance the industry 's best practice. http://rasmusen.org/published/Rasmusen_92JEMS.conservatism.pdf
Great example! "More seriously, regress wages on education, but “control for” industry. The R2 goes up, we explain much more variation of wages (sort of where this post wants to go, but not this way). But the point of education is to let you move from the burger flipping industry to investment banking, so controlling for industry destroys the causal interpretation of the coefficient."
How about calling it “Cochrane’s Razor”. We care about causation only when it causes enough variation to worry about. We sharpen Occam’s Razor with econometrics to find out.
I think we should call it econometric folly. There is statistical significance, so econometricians will salivate over it, but it has little impact on anything.
Reminds me of “clinically significant” - sure a cancer drug can be proven to reduce mortality. But if it’s an extra month? And you want to charge $100k for it?
at $100 per mo human life with an expectancy of 72 years is valued at $86 million. But of course these are the cheap, low quality of life years you're buying...
Expect the BIG hits for Geriatrics to come on stream...
Intubate! intubate Dance to the music!
Ball of infusion delusion?
Or I didn't shoot the sherrif..But I sure shot thru your inheritance.
Not an econometrician, but my intuition is that adding fixed effects is not to take the means out of the specific subgroups. Rather, it is trying to take into account at least some of the impact of un-controlled variables (say, to the first moment). I could be wrong, but in fact, for any hyperplane (i.e. linear model), if you have enough variation locally, you can estimate it well (so, the means of the x's varying across subgroups is not necessarily a problem by itself).
I chatted with some people in my little economist's corner, and, indeed, everyone has a different opinion on this issue. So, I could well be wrong.
The hardest one is conventional wisdom doesn’t make something true…and it’s only going to get worse with LLMs. The phony opioid epidemic is the most deadly example as fentanyl has almost nothing to do with prescription opioids being overprescribed. Btw, why did the fentanyl crisis become so much worse than anything anyone predicted as more laws were passed to restrict opioid access??
Correlation does not work in the presence of nonlinearity.
-Nassim Taleb
That's a "deepity." It's trivially true insofar as the typical Pearson correlation coefficient only captures linear relationship. At a deeper level it is false insofar Pearson correlation is not the only game in town.
How can we know what we do not know? Are there constants akin to the constants of physics and chemistry to be measured when it comes to human choice and action? What if there are no constants? What if there aren't even variables that have small, stable variance when it comes to human behavior?
Ah!! and an Inspector Clouseau fan! ..and I do not know what I do not know...
😂
I am neither economist nor philosopher, but this sounds like just a variation on the philosophical concept of 'necessary and sufficient' or the idea of INUS (Insufficient but Necessary part of an Unnecessary but Sufficient condition) - used in science, as well. I think it underscores for me that it seems economists really only have retrospective correlation for their theories and impose explanations on them - that is they force theories to fit data. Or, put another way, they don't really have a fundamental apriori first principals theory of human behavior which can predict economic behavior.
Necessary vs. sufficient, and there exists vs. for all, are cousins.
When I saw the title, I expected to see a mention of the Lucas critique.
Lucas wanted to regain control. So instead of saying the key in economics was making decisions under conditions of uncertainty, he flipped it. He focused on an (1-uncertainty), or more specifically, risk... that portion of the unknown that we think we might know something about as we look to the future. < Use the Force Luc-as>. Now what we know might be only a small portion of what might happen. But once we begin to focus on expectations and say they drive people's behavior, you are in a very different world. and economists are so enthralled to use expectations even if the expectations' ability to predict is near zero...is it significant? To me this is the great failing of Lucas and of expectations as used in economics. Who is really driven by expectations? I mean, except for some unusual circumstances? Most likely if you have expectations, you are more prepared to act once 'X' happens than than that you do 'y' now because you expect "X" in the future.
I disagree. I think people make many decisions based on expectations. There were some studies on AIDS in the early 2000s that showed people made different decisions regarding risky sex based upon future expectation of life span. People in Africa engaged in risky behavior because lifespans weren't as long generally, and because there wasn't a lot of hope for the future regarding income/standard of living. In the US, decisions were different because the expectations of the future were different. Might have been an Emily Oster study but don't quote me on that. Read about it in the old Becker-Posner blog.
Not a new world at all... High 't' statistics and low R-bar Square. Welcome to my world.
Drug Cos love to talk of the power of statins to reduce the risk of heart attack. The risk can be reduced by 50% or more! with statins. from (not the real numbers...sorry) less than 0.5% to less than 0.025%. Small risk... big (relative) reduction. but people with high - very high- cholesterol levels can and do live a long time. And people with low cholesterol still get heart attacks. And if we take out the TIAs, the impact of statins on reducing 'heart attacks' disappears since they have NO IMPACT on all csause mortality exceept for men under 50 who already have had a heart attack. So, all this is just parsing the statistics. Drug Co' are MASTERS at it. Economtists are pikers by comparion.
But we do get carried away with all this. Do do you want to workshop R-squared or statistical significance? Well, presumably both together, a dual mandate. Good luck.
Asking a question on these stats specifically: What's the conditional probability of a statin helping you live longer given that you have had a TIA?
I'm mostly a fan of the "causal revolution" in economics, but I appreciate your critiques. There is IMO a tradeoff between A) not-so-important questions that we can answer well and B) very important questions for which are answers are ultimately more speculative. The world in which you and I grew up probably focused too much on B questions, but the present world probably focuses too much on the A's.
I don't even criticize the focus on causal relations. Just don't over interpret them.
Housing is a clean case where “causation does not imply variation” shows up in the wild.
Mortgage rate increases do causally raise the discount rate. But once a large share of owners are sitting on very low fixed coupons, the supply elasticity of existing homes collapses. At that point, the clearing margin switches from price to quantity - turnover and composition (new vs. existing) - rather than prices adjusting.
So the causal channel (rates to discount rate) is intact, but the variance from that shock shows up almost entirely in listings and turnover instead of prices. Builders with the ability to buy down rates can still clear at the payment-capacity boundary, while existing owners simply don't sell.
One place this tends to get missed in the asset-pricing literature is that the empirical focus is usually on price response to shocks, not which margin is doing the adjusting. If the margin itself has changed (from price to quantity), then searching for the shock’s imprint on price will naturally make the effect look small or "puzzling". It’s not that the causal effect is weak, it’s that the observed variation is happening on a margin they’re not measuring.
Isn't this why we use Linear Algebra? We take all the variables we think might influence a certain outcome and we get an equation that gives the weights of each variable that maps the history and we can see which ones have a large influence and small influence?
I think, it has been decades since I took that class
This paper might fit into why we don't update our priors as much as expected. I apply it to estimates returns on investment, but it could apply pretty well to significant coefficients in a new and unexpected context too.
``Managerial Conservatism and Rational Information Acquisition, '' Journal of Economics and Management Strategy (Spring 1992), 1(1): 175-202. Conservative managerial behavior can be rational and profit- maximizing. If the valuation of innovations contains white noise and the status quo would be preferred to random innovation, then any innovation that does not appear to be substantially better than the status quo should be rejected. The more successful the firm, the higher the threshold for accepting innovation should be, and the greater the conservative bias. Other things equal, more successful firms will spend less on research, adopt fewer innovations, and be less likely to advance the industry 's best practice. http://rasmusen.org/published/Rasmusen_92JEMS.conservatism.pdf
Great example! "More seriously, regress wages on education, but “control for” industry. The R2 goes up, we explain much more variation of wages (sort of where this post wants to go, but not this way). But the point of education is to let you move from the burger flipping industry to investment banking, so controlling for industry destroys the causal interpretation of the coefficient."
How about calling it “Cochrane’s Razor”. We care about causation only when it causes enough variation to worry about. We sharpen Occam’s Razor with econometrics to find out.
I think we should call it econometric folly. There is statistical significance, so econometricians will salivate over it, but it has little impact on anything.
Reminds me of “clinically significant” - sure a cancer drug can be proven to reduce mortality. But if it’s an extra month? And you want to charge $100k for it?
at $100 per mo human life with an expectancy of 72 years is valued at $86 million. But of course these are the cheap, low quality of life years you're buying...
Expect the BIG hits for Geriatrics to come on stream...
Intubate! intubate Dance to the music!
Ball of infusion delusion?
Or I didn't shoot the sherrif..But I sure shot thru your inheritance.
This kind of echoes McCloskey's old complaints about statistical significance vs economic significance
Not an econometrician, but my intuition is that adding fixed effects is not to take the means out of the specific subgroups. Rather, it is trying to take into account at least some of the impact of un-controlled variables (say, to the first moment). I could be wrong, but in fact, for any hyperplane (i.e. linear model), if you have enough variation locally, you can estimate it well (so, the means of the x's varying across subgroups is not necessarily a problem by itself).
I chatted with some people in my little economist's corner, and, indeed, everyone has a different opinion on this issue. So, I could well be wrong.
The wage example is great but it is actually even worse than you make it out to be ("regress wages on education, but “control for” industry").
That's a typical example of a bad control which introduces a selection effect (see Angrist and Pischke page 64 I believe).
The hardest one is conventional wisdom doesn’t make something true…and it’s only going to get worse with LLMs. The phony opioid epidemic is the most deadly example as fentanyl has almost nothing to do with prescription opioids being overprescribed. Btw, why did the fentanyl crisis become so much worse than anything anyone predicted as more laws were passed to restrict opioid access??