3 ms·

The problem here is that hunger and war are woolly concepts.[1] But I wouldn't say causation is a myth at this level. You can reasonably ask whether hunger caus

by dash2 6y ago

The problem here is that hunger and war are woolly concepts.[1] But I wouldn't say causation is a myth at this level. You can reasonably ask whether hunger causes war and run IV regressions, instrumenting hunger with e.g. rainfall, which have a reasonable chance at testing that hypothesis.[2] [1] More thoughts about this at: https://wyclif.substack.com/p/on-social-science-about-complex-stuff https://wyclif.substack.com/p/on-social-science-about-comple... [2] Like the literature coming from http://data.nber.org/ens/feldstein/ENSA%20Sources/Geospatial/Rainfall%20Africa/_Paper__Economic_Shocks_and_Civil_Conflict.pdf http://data.nber.org/ens/feldstein/ENSA%20Sources/Geospatial...

__MatrixMan__ 6y ago
I'm gonna check out those links but I just wanted to clarify that I think causation is a myth at all levels, even the hardest of sciences. It's just that it's a more useful myth in those places.
- dash2 6y ago
  So what's your objection to the standard Pearl/Rubin idea: "X causes Y means that if X, and nothing else, varies, then Y varies"?
  - __MatrixMan__ 6y ago
    When was the last time you encountered a situation where only one thing was varying? I don't have a problem with the idea on it's own. It's a great way to understand all kinds of things. But data to support explanations of that sort is hard to come by. What I object to is the belief that all explanations that can in principle be put in terms of cause, must in practice be put in terms of cause--otherwise they're useless (which, I think, motivated a lot of unnecessary handwaving in the bbc article, and probably in the ones that it references too). Researchers should contribute useful data so that others can reference it to make decisions. That data becomes less useful when they have incentives to skew things so that they can meet the awkwardly high bar of "supports a causal argument" just to be worthy of publication. It ought to be ok to say: "these things go together. Here's the evidence. It might not be possible to untangle the causal web between them, but if you try one I bet you'll see the other," and leave it at that. Not to say that untangling the causal web isn't worth doing, but sometimes it's just not in the cards, and demanding that we search for it in the cards is like going hungry because all of the carrots left in the grocery store are strangely shaped.
    - dash2 6y ago
      Ok, I’m sympathetic to that POV. I’m grateful for the credibility revolution in econometrics, but it clearly can become an obsession. Prediction on its own can be worthwhile. I wouldn’t say causation is a myth - maybe rather that it’s overvalued.

by dash2 6y ago

__MatrixMan__ 6y ago

I'm gonna check out those links but I just wanted to clarify that I think causation is a myth at all levels, even the hardest of sciences. It's just that it's a more useful myth in those places.

dash2 6y ago
So what's your objection to the standard Pearl/Rubin idea: "X causes Y means that if X, and nothing else, varies, then Y varies"?
- __MatrixMan__ 6y ago
  When was the last time you encountered a situation where only one thing was varying? I don't have a problem with the idea on it's own. It's a great way to understand all kinds of things. But data to support explanations of that sort is hard to come by. What I object to is the belief that all explanations that can in principle be put in terms of cause, must in practice be put in terms of cause--otherwise they're useless (which, I think, motivated a lot of unnecessary handwaving in the bbc article, and probably in the ones that it references too). Researchers should contribute useful data so that others can reference it to make decisions. That data becomes less useful when they have incentives to skew things so that they can meet the awkwardly high bar of "supports a causal argument" just to be worthy of publication. It ought to be ok to say: "these things go together. Here's the evidence. It might not be possible to untangle the causal web between them, but if you try one I bet you'll see the other," and leave it at that. Not to say that untangling the causal web isn't worth doing, but sometimes it's just not in the cards, and demanding that we search for it in the cards is like going hungry because all of the carrots left in the grocery store are strangely shaped.
  - dash2 6y ago
    Ok, I’m sympathetic to that POV. I’m grateful for the credibility revolution in econometrics, but it clearly can become an obsession. Prediction on its own can be worthwhile. I wouldn’t say causation is a myth - maybe rather that it’s overvalued.