A student among the econ: Seeing through mathemagics
from Asad Zaman
In my article on “Education of an Economist”, I have explained how I gradually came to realize that all I had learnt during my Ph.D. training at Stanford University was false. I would like to make this more specific and concrete, by providing some examples. A leading example is a paper on Power and Taxes which I studied as a graduate student. But, let me start from the beginning.
Leijonhufvud, in his classic “Life Among the Econ” explains how the priestly caste of the Math-Econ is the most admired and revered among the Econ. With a math degree from MIT, I was eager to join the ranks of the Mathematical Economists, the most prestigious among the alternatives facing me. But a few experiences soured me on this idea. I ended up with a degree in Econometrics, which also makes heavy use of mathematics, under the illusion that this may be closer to reality. I learned much later that all the theory I studied at Stanford pertains to a fantasy world created by economists, which has no relationship to the real world. read more
Leave a Reply
—– look inside —– $5.94 / $20.00
—– look inside —– $4.90 / $8.00
—– look inside —– $15.99
—– look inside —– $5.99 / 12.99
—– look inside —– $5.93 / $12.99
—– look inside —– $4.97 / $9.90
—— Ugarteche, Puyana and Madi ——
Gerson Lima / Maria Alejandra Madi
Edward Fullbrook and Jamie Morgan
————— Michael Hudson ————–
Maria Alejandra Madi / Jack Reardon
————- Edward Fullbrook ————-
—————— Steve Keen —————–
————— Richard Smith —————
————– Gustavo Marques————
– Victor Beker and Beniamino Moro –
————– Lars Pålsson Syll ————-
—————– Stuart Birks —————-
Edward Fullbrook and Jamie Morgan
———— Armando Ochangco ———-
Shimshon Bichler / Jonathan Nitzan
————— Mauro Gallegati ————–
————— Herman Daly —————-
————— Asad Zaman —————
—————– C. T. Kurien —————
————— Robert Locke —————-
Real-World Economics Review Blog 
Asad, I read your blogs and article with interest. I also did my PhD at Stanford in economics in a joint program with the business school (that they discontinued after 3 of us graduated). When I finished in 1975 I was already doubting most of what I was taught. John Gurley brought a fresh leftist breath of air to a neoclassical fantasy (although they wouldn’t let him teach graduate classes) I remember spending only one class — led by economist Bill Moffat — on second best theory. I was blown away — with one imperfection everything I was taught in micro went out the window. I then asked Alan MacAdams in another class (visiting from Cornell University) what this implied since we had nothing close to perfect competition. He replied we had “workable competition” but when I asked what was that and did it have any of the efficiency benefits of perfect competition I was met with silence.
Even then I was surprised that studying economic history was optional, not required. And, even then, econometrics seemed false. To get unbiased coefficients you had a few completely unrealistic assumptions that, when violated, left you without a clue as to the real impact of a variable. And the parallel with the few completely unrealistic assumptions of perfect competition needed to get true market prices was obvious. The neoclassical fantasy was absurd from the beginning and check out my attempt to show that econometrics is truly just the emperor’s not-so-new clothes.
Click to access Klees74.pdf
Thanks for your comments. I looked at your regression paper, and I have similar ideas. I have put it this way in one of my papers: with ANY (relevant) data set, and ANY empirical hypothesis regarding the data, one can find a regression which will confirm the hypothesis. ALSO, one can find a regression which contradicts the hypothesis. Regression is just fraud by numbers. But, I have been working on approaches which can yield useful results — but we start by acknowledging that the data only provide hints. Real work must be done by “shoe leather” – one can never make conclusions about the real world using data alone, but one can be guided to useful research questions by looking at data.
AGREED!
Vague, vague. What do you mean by shoe leather? And if you walk down one path how do you know you would have come to the same conclusion if you had walked down another? The objections raised to statistical analysis, correctly done, apply to any empirical research whatever. All “scientific” generalisations are conjectures that survived the last test and we cannot “know” they will always be right.
Not only are all methods fallible but people can cheat with any method. But a single data set can be consistent, inconsistent or inconclusive about a given hypothesis. It cannot be all three at once. Accusations of fraud can be levelled at practitioners but not at a method. Your slightly overheated language should alert you to the fact that you are not thinking dispassionately.
Crosstabs are data. Regression coefficients are not. They are the result of torturing the very real crosstab data until “Nature” confesses, ha ha. And Gerald, “dispassionate” is overrated.
Why was the single data set comprising the eastern coastline of South America and the western coastline of Africa used to vehemently and passionately deny, then affirm, Continental Drift theory? What good is a method that uses the same dataset to disprove (with strong emotions!), then prove, a theory?
Freedman (1997:113) writes: “For nearly a century, investigators in the social sciences have used regression models to deduce cause-and-effect relationships from patterns of association. … . In my view,
this enterprise has not been successful. See: Freedman, D.A. (1997). From Association to Causation via Regression.In Causality in Crisis?ed.
V. McKim and S. Turner.University of Notre Dame Press, South Bend, (with discussion) 113-82. Reprinted in Advances in Applied Mathematics, 18, 59-110.
For shoe leather terminology, and what it means, see:
Statistical Models and Shoe Leather Author(s): David A. Freedman
Source: Sociological Methodology, Vol. 21 (1991), pp. 291-313 https://psychology.okstate.edu/faculty/jgrice/psyc5314/Freedman_1991A.pdf
For an academic article (not a blog post) which provides sharp and detailed arguments, in addition to those given above, see: Zaman, Asad, Methodological Mistakes and Econometric Consequences (September 2, 2012). International Econometric Review, Vol 4. Issue 2, p. 99-122, September 2012, Available at SSRN: https://ssrn.com/abstract=2140372
On something we can agree. You cannot deduce cause and effect relationships from regressions and no sensible analyst should imagine you can. But you cannot damn a technique because it has been misused – because any tool can be misused.
Regression can be used for two purposes. One is to identify persistent correlations that tell you nothing about cause and effect but might be useful in providing leading indicators for forecasting. (Such work is no good for policy or any what-if analysis). Two it can be used to test a pre-existing causal hypothesis. It cannot prove such a hypothesis, of course, but it can reveal when the data are inconsistent with the hypothesis. This will seldom be a knock-out blow; defenders of the hypothesis will object to the specification and suggest alternatives or point to some influence that has not been controlled. Fair enough, you try the alternative specification, proxy the uncontrolled variable and test again. That’s how empirical study advances. We are not confined to linear specifications. The complexity of a model is limited only by the amount of data available for testing and the imagination of the theorist.
We have been running regressions for over a century. Over this entire period, can you give ONE example of just one regression which added to our knowledge ? – (beyond the pairwise correlations we can do without regression)
Exactly!
PS I am familiar with the Freedman article to which you refer. He is more cautious in his critique of method, as opposed to practice, than you are. And there is a lot to be said about his points anyway. This may not be the place, though, for a lengthy commentary.