" In economics, we’ve taken on a model of argument in which data, regressions and modelling (which are then summarised on a scatterplot) are the standard of proof. "
When you come from physics or applied math, the idea that a regression deserves the word "model" is a bit strange. As I wrote in Models.Behaving.Badly,
"Had Newton been a social scientist performing statistical regressions in psychology, economics or finance, he would likely have proposed the gravitational inverse square law as a power of 2.5 +- 0.31."
I’m going to take this as an opportunity to voice semi-controversial opinions and so give mine: the vast majority of economists could be replaced with video games designers and we’d genuinely get better results. Seriously - watch this one 30 minute video on the Sims and see 100 years of economic modelling being blown out the water by some early 2000s meme game. https://youtu.be/9gf2MT-IOsg?si=U-cejmYeSitjLJ5a
Economics looks a lot better when you view it as an explanatory science rather than a predictive science. But explanatory sciences are soft, and predictive sciences are hard. There are some people who prefer hardness to intellectual intercourse.
Is economic geography the right target? You might just as well say econometrics, mightn’t you? Aren’t you just talking about the ability of computers to generate something that looks like a line from scattergrams? My first school statistics textbook in the late 1960s (authored by Loveday if memory serves me rightly) had in chapter one the notion of spurious correlation illustrated by the suicide rate in the uk plotted against the density of lampposts. No idea how that looks now.
Econometrics is just applied maths (that's why it's so awful), and probably deserves a separate bee, but in econgeog, it is really the theoretical models (of agglomeration, creative class, etc) that get on my nerves, because they produce such platitudinal and general results that you can more or less flick paint from a toothbrush at the page, draw a line through it and call it empirical evidence.
I think econgeog may also have meant something different before the days of computer generated multivariate “correlation”. But there is certainly a mathematical analysis of drawing lines through toothbrush splatter (if only I could remember what it is) - it’s why you can always find leylines if you want to look for them. Quite like econometrics really.
I feel that you are pointing at a specific instance of a more general thing, roughly speaking "treating a residual too seriously". Economics produces these like locusts because of its preferred techniques, as I think you implied (looking at you, "total factor productivity"). But the same thing often happens outside economics; with the "placebo effect", for example.
Perhaps; but the really important thing is distinguishing the residuals that matter from the ones that don’t, isn’t it? Just as any model-based analysis should begin with “how well does this model really address the question”. I’m sure you know the joke about the professionals cast away on a desert island with a cache of tinned food. The economist’s solution is “let’s assume a tin opener”.
I think you are being unfair to the placebo effect here. It does make sense to classify it as a residual in some sense because it impacts RCTs by changing both the control and the study arm with "equal" likelihood. But its also a bonafide phenomenon in the sense that there appears to be a genetic basis for who is more or less susceptible to placebo effect (mediated by neurotransmitters like serotonin or dopamine receptors). I suppose the real similarity between econometrics and placebo is that they both display a sort of reflexivity: systems where interpretations of properties of the system can change the state of the system.
The interesting thing is that psychology massively bought into "p=0.05 and you have a Fact", and ended up having to abandon enormous tracts of knowledge as nonsense, and then econogeog did the same damn thing <em>in the knowledge that had happened</em>.
Isn't the basic issue here that becuase we cannot do randomized control trials at the national-level? We are reduced to national-level scatter plots and maybe--if we think the broad conditions on the right hand side of the imagined regression have been stable--an even more questionable time series. Scatter plot analysis and economic geography have obvious limitations, but it does seem like one of the short list of tools available to comparative analysis of policy, economic systems, and other elemnts of social organization.
I, for one, shamelessly love the scatter plot showing a relatioship between a country's corruption index performance an the number of parking tickets issued by the NYPD to its UN mission. It's not a full theory, but it's not a coincidence.
" In economics, we’ve taken on a model of argument in which data, regressions and modelling (which are then summarised on a scatterplot) are the standard of proof. "
That's called "physics envy," me lad. It's a less dangerous disease than one might think. People know better in their real-life conduct. The MIT men still work for the Harvard men. Or to update: the IIT-Kanpur men still work for the Harvard women.
For reasons I can't fully justify, I turn off immediately when I read the word "city" in any context like this. In particular, I've always regarded the "global city" phenomenon as another word for cronyism. As with many of my opinions, I seem to be in the minority on this https://crookedtimber.org/2004/02/24/cities-and-cronyism/
It will always be with us. Because city sizes follow a power law [citation needed], the sample size for the largest cities will always be too small, and those are the ones the authors wanted to talk about, right?
" In economics, we’ve taken on a model of argument in which data, regressions and modelling (which are then summarised on a scatterplot) are the standard of proof. "
When you come from physics or applied math, the idea that a regression deserves the word "model" is a bit strange. As I wrote in Models.Behaving.Badly,
"Had Newton been a social scientist performing statistical regressions in psychology, economics or finance, he would likely have proposed the gravitational inverse square law as a power of 2.5 +- 0.31."
I’m going to take this as an opportunity to voice semi-controversial opinions and so give mine: the vast majority of economists could be replaced with video games designers and we’d genuinely get better results. Seriously - watch this one 30 minute video on the Sims and see 100 years of economic modelling being blown out the water by some early 2000s meme game. https://youtu.be/9gf2MT-IOsg?si=U-cejmYeSitjLJ5a
Economics looks a lot better when you view it as an explanatory science rather than a predictive science. But explanatory sciences are soft, and predictive sciences are hard. There are some people who prefer hardness to intellectual intercourse.
Is economic geography the right target? You might just as well say econometrics, mightn’t you? Aren’t you just talking about the ability of computers to generate something that looks like a line from scattergrams? My first school statistics textbook in the late 1960s (authored by Loveday if memory serves me rightly) had in chapter one the notion of spurious correlation illustrated by the suicide rate in the uk plotted against the density of lampposts. No idea how that looks now.
Econometrics is just applied maths (that's why it's so awful), and probably deserves a separate bee, but in econgeog, it is really the theoretical models (of agglomeration, creative class, etc) that get on my nerves, because they produce such platitudinal and general results that you can more or less flick paint from a toothbrush at the page, draw a line through it and call it empirical evidence.
I think econgeog may also have meant something different before the days of computer generated multivariate “correlation”. But there is certainly a mathematical analysis of drawing lines through toothbrush splatter (if only I could remember what it is) - it’s why you can always find leylines if you want to look for them. Quite like econometrics really.
I feel that you are pointing at a specific instance of a more general thing, roughly speaking "treating a residual too seriously". Economics produces these like locusts because of its preferred techniques, as I think you implied (looking at you, "total factor productivity"). But the same thing often happens outside economics; with the "placebo effect", for example.
Perhaps; but the really important thing is distinguishing the residuals that matter from the ones that don’t, isn’t it? Just as any model-based analysis should begin with “how well does this model really address the question”. I’m sure you know the joke about the professionals cast away on a desert island with a cache of tinned food. The economist’s solution is “let’s assume a tin opener”.
I think you are being unfair to the placebo effect here. It does make sense to classify it as a residual in some sense because it impacts RCTs by changing both the control and the study arm with "equal" likelihood. But its also a bonafide phenomenon in the sense that there appears to be a genetic basis for who is more or less susceptible to placebo effect (mediated by neurotransmitters like serotonin or dopamine receptors). I suppose the real similarity between econometrics and placebo is that they both display a sort of reflexivity: systems where interpretations of properties of the system can change the state of the system.
The interesting thing is that psychology massively bought into "p=0.05 and you have a Fact", and ended up having to abandon enormous tracts of knowledge as nonsense, and then econogeog did the same damn thing <em>in the knowledge that had happened</em>.
html tags are verboten? html tags are verboten.
Isn't the basic issue here that becuase we cannot do randomized control trials at the national-level? We are reduced to national-level scatter plots and maybe--if we think the broad conditions on the right hand side of the imagined regression have been stable--an even more questionable time series. Scatter plot analysis and economic geography have obvious limitations, but it does seem like one of the short list of tools available to comparative analysis of policy, economic systems, and other elemnts of social organization.
I, for one, shamelessly love the scatter plot showing a relatioship between a country's corruption index performance an the number of parking tickets issued by the NYPD to its UN mission. It's not a full theory, but it's not a coincidence.
" In economics, we’ve taken on a model of argument in which data, regressions and modelling (which are then summarised on a scatterplot) are the standard of proof. "
That's called "physics envy," me lad. It's a less dangerous disease than one might think. People know better in their real-life conduct. The MIT men still work for the Harvard men. Or to update: the IIT-Kanpur men still work for the Harvard women.
For reasons I can't fully justify, I turn off immediately when I read the word "city" in any context like this. In particular, I've always regarded the "global city" phenomenon as another word for cronyism. As with many of my opinions, I seem to be in the minority on this https://crookedtimber.org/2004/02/24/cities-and-cronyism/
It will always be with us. Because city sizes follow a power law [citation needed], the sample size for the largest cities will always be too small, and those are the ones the authors wanted to talk about, right?
1. Scanty data, unfalsifiability
2. ????
3. Profit!