Colonial origins, causal claims, the baggage left behind in the overhead bin via the absence of a structural model, settler-colonist mortality, modern pro-prosperity “institutions”, and structural...
On Jojosi knappers and the Gesher Benot Yaʿaqov hominins it's interesting ... 'we are different because of our economic function,' Sounds good but... bees have hives, Pukekoes (NZ swamp hen) have tribes and I often observe them having close relations during feeding to paradise ducks.. so lots of smarts and planned behaviour, even in blowflies.
I have never trusted papers using Instrumental Variables unless the instrumenting equation (your overhead compartment) is presented clearly and discussed properly.
Yes indeed. I was lucky enough to have been a short-time roommate of John Bound in my younger days, the Master Knower of the statistical pitfalls of "weak instrument" and downstream "file drawer" problems. The Cauchy Distribution cannot be reasoned with, negotiated with, or banished by incantation.
> **John Quiggin**: I have never trusted papers using Instrumental Variables unless the instrumenting equation (your overhead compartment) is presented clearly and discussed properly.
======
======
**DELONG’S GRASPING REALITY: Trying to make my readers—and myself—smarter. I think I am a go-to source to understand things economic in the past and in the present. Too online since 1995. Subscribe! <https://graspingreality.substack.com/subscribe>. Currently featuring**:
Another Aussie economist told me this in grad school (I'm paraphrasing). "All econ data is basically crap, so your instrument will be too. If you're estimating a demand curve, and it slopes downward, it's a demand curve."
So I go to comment on an excellent, new article on the Fermi Paradox, but instead I see comments from mid April on a completely different topic. Is there a bug in the comments section?
I was quite prepared to express my agreement with your views about structural models. But then I reflected that America is the world's most prosperous large country and its governing institutions are currently riding the hot rails to hell. Now I am not so confident.
Very interesting. I did not know that Hicks changed his views.
Though I have to say the list of great names in econ who are skeptical of the formal models of neoclassical is long. Marshall, who is often listed as the founder of anglophone neo-classical, was very clear about this. I would argue that in this he was simply following Mill. Marshall did not accept the break with classical economics and was not a pure marginalist. Keynes was simply following in his footsteps. The Austrians are, of course, skeptical.
On the other side of the Atlantic, Veblen spent his whole career deconstructing neoclassical and founding institutionalism along the way. And then both Viner (a Marshallian) and Knight are very skeptical also. Not surprising in Knight's case because he is a theorist of Knightian uncertainty like Keynes. And yet, all these thinkers seem to have made little impact on post-war economics.
Very much enjoyed the post re Hicks and, for what it is worth, very much agree. Seems that we often need simplicities for some tasks, but it is extremely important to understand the underlying nuances. As a rule, we tend toward tangible solutions because hand-waving, even if communicating something important, is a bit discomforting.
Well yes, AI gets certain obvious facts wrong and that's embarrassing. But AI learns from its errors. If you discuss "War and Peace" with it six months from now it will do much better.
You have to recognize what the LLM was trained on to recognize biases of the time that generated the text.
For example, if you searched the card catalog of pictures in a Southern historical library, and asked to to show you a picture of a "farmer" you would get a picture of a white farmer. You have to search the term "Negro Farmer" to see a black farmer. (Learned this at a Digital Public Library convention.
What I like to do is ask ChatGpt to rewrite a poem in the style of Elizabethan English...so the poem "Mary had a little lamb" comes out as "Mary begot a lamb..." Two meanings of the word "had" Also asked it to rewrite in the style of Ernest Hemingway, with the result: "Mary had a lamb. It was white. It followed her. Even to school."
1. I agree with Mr Koop 2. It seems like most of the "prosperous" countries in the study were English speaking....is adoption of English Common Law a better indicator of institutions & prosparity ?
As preamble: I'm just a lay person interested in economics. I'm likely to have missed what others have seen.
Throwing caution to the wind then...
Glancing through the arguments I'm not seeing what strikes me as a substantial missing element to the analysis - the incredibly cheap "cost" of asset purchase.
I suspect settler mortality was lowest where Native Americans had thrived but, by virtue of European disease vectors, had died out.
Adjusting for this exogenous factor seems to me a useful exercise in this analysis.
Since I'm not an economist I'm sure this analysis is found in research on the effects of Black Death depopulation on growth I just haven't read it yet.
On that note, I'm wondering about the growth effects of other more directly genocidal expansions, like, say, the Akkadians.
And then a potential counter example, the Sea Peoples ... maybe one can depopulate too much and destroy trade routes.
I might be wrong, but doesn't IV reduce attenuation bias in the case of classical measurement error even if the instrument isn't a " better measure of what matters"? I'm thinking of the classic intergenerational mobility application, were you use a second measure of the father's occupational status as an instrument for the first (e.g. Zach Ward's fantastic 2023 AER).
I think you're right though to want the structural model! AJR is indeed super fun to teach, brings up lots of important questions.
I will have to think about this... I think I should have said "less correlated with the noise in the instrument"...
With Y, X, and Z, the IV estimate of the effect of Y on X is the OLS coefficient of Y regressed on Z divided by the coefficient of X on Z. But what we are looking for in the attenuation-bias case is the regression of Y on the "true" components of X, call them W, with X = W + ε...
It has never been clear to me why it is optimal to use the second measure of status as an instrument for the first, rather than think that they are both noisy predictors of the true status variable and constructing, explicitly, the optimal weighted measure...
I think the trick is that both the coefficient of Y on Z and of X on Z have attenuation bias, so the ratio is an improvement.
(The validity assumption is the error in Z is uncorrelated with the error in X, and both errors are classical. So since Z is the independent variable in the two OLS regressions, both are attenuated by the same factor, σ^2_z/(σ^2_z+σ^2_e) where Z = W+e).
touché... what I should have said is that there is prosperity which is determined by institutions and other stuff, systematic and random; true institutions and measured "institutions", true settler-climate and measured settler climate, and that all of these interact via some true stuctural model that has:
* true climate affecting original settler institutions
* true climate affecting other factors that feed into modern-day prosperity
* true prosperity-affecting institutions and other institutions, together adding up to measured contemporary institutions
* the causal channel from institutions today to prosperity today
* the reverse-causation channel
And you really need to get properly Mediæval Judea Pearl on this with boxes and circles and lines and arrows to talk about it coherently...
AR use Ambos Nogales as a natural experiment that shows the power of institutions. The problem with this is that they don't have any data. The Mexican statistical agency, INEGI, does not estimate aggregate output for sub-state regions (municipios, like US counties). However, it is not difficult to estimate given labor employed by sector and sectoral shares of state income. When you do the estimate, you find that the difference between Nogales Sonora and Nogales Arizona is only about 40 percent of the difference between the US and Mexico, in terms of per capita GDP at PPP. I think AR & J's work is brilliant, but not the whole story.
On Jojosi knappers and the Gesher Benot Yaʿaqov hominins it's interesting ... 'we are different because of our economic function,' Sounds good but... bees have hives, Pukekoes (NZ swamp hen) have tribes and I often observe them having close relations during feeding to paradise ducks.. so lots of smarts and planned behaviour, even in blowflies.
I have never trusted papers using Instrumental Variables unless the instrumenting equation (your overhead compartment) is presented clearly and discussed properly.
Yes indeed. I was lucky enough to have been a short-time roommate of John Bound in my younger days, the Master Knower of the statistical pitfalls of "weak instrument" and downstream "file drawer" problems. The Cauchy Distribution cannot be reasoned with, negotiated with, or banished by incantation.
> **John Quiggin**: I have never trusted papers using Instrumental Variables unless the instrumenting equation (your overhead compartment) is presented clearly and discussed properly.
======
======
**DELONG’S GRASPING REALITY: Trying to make my readers—and myself—smarter. I think I am a go-to source to understand things economic in the past and in the present. Too online since 1995. Subscribe! <https://graspingreality.substack.com/subscribe>. Currently featuring**:
* The Fourteen-Lion Parade <https://braddelong.substack.com/p/fourteen-lion-parade>
* Lessons for Debt Control from Clinton's Success in the 1990s <https://braddelong.substack.com/p/lessons-for-debt-control-from-clintons>
* "Star Trek: The Wrath of Khan" as Franchise Hinge <https://braddelong.substack.com/p/2025-12-24-star-trek-the-wrath-of>
<https://braddelong.substack.com/p/donald-trump-is-too-old-to-be-president>
* Time for a "Land Acknowledgement”! <https://braddelong.substack.com/p/time-for-a-land-acknowledgement>
* The Spearpoint Right Now of Human Technological Engineering <https://braddelong.substack.com/p/the-spearpoint-right-now-of-human>
Another Aussie economist told me this in grad school (I'm paraphrasing). "All econ data is basically crap, so your instrument will be too. If you're estimating a demand curve, and it slopes downward, it's a demand curve."
So I go to comment on an excellent, new article on the Fermi Paradox, but instead I see comments from mid April on a completely different topic. Is there a bug in the comments section?
I was quite prepared to express my agreement with your views about structural models. But then I reflected that America is the world's most prosperous large country and its governing institutions are currently riding the hot rails to hell. Now I am not so confident.
Does anyone perceive an ancestor, although non-consensual, in the bot farms built by cyber criminals, decades ago?
Very interesting. I did not know that Hicks changed his views.
Though I have to say the list of great names in econ who are skeptical of the formal models of neoclassical is long. Marshall, who is often listed as the founder of anglophone neo-classical, was very clear about this. I would argue that in this he was simply following Mill. Marshall did not accept the break with classical economics and was not a pure marginalist. Keynes was simply following in his footsteps. The Austrians are, of course, skeptical.
On the other side of the Atlantic, Veblen spent his whole career deconstructing neoclassical and founding institutionalism along the way. And then both Viner (a Marshallian) and Knight are very skeptical also. Not surprising in Knight's case because he is a theorist of Knightian uncertainty like Keynes. And yet, all these thinkers seem to have made little impact on post-war economics.
Very much enjoyed the post re Hicks and, for what it is worth, very much agree. Seems that we often need simplicities for some tasks, but it is extremely important to understand the underlying nuances. As a rule, we tend toward tangible solutions because hand-waving, even if communicating something important, is a bit discomforting.
So this is the reason I am still waiting for my trickle down check?
Well yes, AI gets certain obvious facts wrong and that's embarrassing. But AI learns from its errors. If you discuss "War and Peace" with it six months from now it will do much better.
You have to recognize what the LLM was trained on to recognize biases of the time that generated the text.
For example, if you searched the card catalog of pictures in a Southern historical library, and asked to to show you a picture of a "farmer" you would get a picture of a white farmer. You have to search the term "Negro Farmer" to see a black farmer. (Learned this at a Digital Public Library convention.
What I like to do is ask ChatGpt to rewrite a poem in the style of Elizabethan English...so the poem "Mary had a little lamb" comes out as "Mary begot a lamb..." Two meanings of the word "had" Also asked it to rewrite in the style of Ernest Hemingway, with the result: "Mary had a lamb. It was white. It followed her. Even to school."
1. I agree with Mr Koop 2. It seems like most of the "prosperous" countries in the study were English speaking....is adoption of English Common Law a better indicator of institutions & prosparity ?
This is what my sometime roommate Andrei Shleifer thinks, strongly. You tangle with him at your intellectual peril..
As preamble: I'm just a lay person interested in economics. I'm likely to have missed what others have seen.
Throwing caution to the wind then...
Glancing through the arguments I'm not seeing what strikes me as a substantial missing element to the analysis - the incredibly cheap "cost" of asset purchase.
I suspect settler mortality was lowest where Native Americans had thrived but, by virtue of European disease vectors, had died out.
Adjusting for this exogenous factor seems to me a useful exercise in this analysis.
Since I'm not an economist I'm sure this analysis is found in research on the effects of Black Death depopulation on growth I just haven't read it yet.
On that note, I'm wondering about the growth effects of other more directly genocidal expansions, like, say, the Akkadians.
And then a potential counter example, the Sea Peoples ... maybe one can depopulate too much and destroy trade routes.
Now back to work.
I might be wrong, but doesn't IV reduce attenuation bias in the case of classical measurement error even if the instrument isn't a " better measure of what matters"? I'm thinking of the classic intergenerational mobility application, were you use a second measure of the father's occupational status as an instrument for the first (e.g. Zach Ward's fantastic 2023 AER).
I think you're right though to want the structural model! AJR is indeed super fun to teach, brings up lots of important questions.
I will have to think about this... I think I should have said "less correlated with the noise in the instrument"...
With Y, X, and Z, the IV estimate of the effect of Y on X is the OLS coefficient of Y regressed on Z divided by the coefficient of X on Z. But what we are looking for in the attenuation-bias case is the regression of Y on the "true" components of X, call them W, with X = W + ε...
It has never been clear to me why it is optimal to use the second measure of status as an instrument for the first, rather than think that they are both noisy predictors of the true status variable and constructing, explicitly, the optimal weighted measure...
I think the trick is that both the coefficient of Y on Z and of X on Z have attenuation bias, so the ratio is an improvement.
(The validity assumption is the error in Z is uncorrelated with the error in X, and both errors are classical. So since Z is the independent variable in the two OLS regressions, both are attenuated by the same factor, σ^2_z/(σ^2_z+σ^2_e) where Z = W+e).
touché... what I should have said is that there is prosperity which is determined by institutions and other stuff, systematic and random; true institutions and measured "institutions", true settler-climate and measured settler climate, and that all of these interact via some true stuctural model that has:
* true climate affecting original settler institutions
* true climate affecting other factors that feed into modern-day prosperity
* true prosperity-affecting institutions and other institutions, together adding up to measured contemporary institutions
* the causal channel from institutions today to prosperity today
* the reverse-causation channel
And you really need to get properly Mediæval Judea Pearl on this with boxes and circles and lines and arrows to talk about it coherently...
Amen to that! And to “get properly Mediæval Judea Pearl on” something is a phrase I now want to employ somewhere.
The second bullet point is the main issue I have with the IV. But hey, as you said, the paper starts some great conversations.
AR use Ambos Nogales as a natural experiment that shows the power of institutions. The problem with this is that they don't have any data. The Mexican statistical agency, INEGI, does not estimate aggregate output for sub-state regions (municipios, like US counties). However, it is not difficult to estimate given labor employed by sector and sectoral shares of state income. When you do the estimate, you find that the difference between Nogales Sonora and Nogales Arizona is only about 40 percent of the difference between the US and Mexico, in terms of per capita GDP at PPP. I think AR & J's work is brilliant, but not the whole story.