Home > Uncategorized > On causality and econometrics

On causality and econometrics

from Lars Syll

causal-inference-in-statistics-233x165The point is that a superficial analysis, which only looks at the numbers, without attempting to assess the underlying causal structures, cannot lead to a satisfactory data analysis … We must go out into the real world and look at the structural details of how events occur … The idea that the numbers by themselves can provide us with causal information is false. It is also false that a meaningful analysis of data can be done without taking any stand on the real-world causal mechanism … These issues are of extreme important with reference to Big Data and Machine Learning. Machines cannot expend shoe leather, and enormous amounts of data cannot provide us knowledge of the causal mechanisms in a mechanical way. However, a small amount of knowledge of real-world structures used as causal input can lead to substantial payoffs in terms of meaningful data analysis. The problem with current econometric techniques is that they do not have any scope for input of causal information – the language of econometrics does not have the vocabulary required to talk about causal concepts.

Asad Zaman / WEA Pedagogy

What Asad Zaman tells us in his splendid set of lectures is that causality in social sciences can never solely be a question of statistical inference. Causality entails more than predictability, and to really in depth explain social phenomena require theory. Analysis of variation — the foundation of all econometrics — can never in itself reveal how these variations are brought about. First, when we are able to tie actions, processes or structures to the statistical relations detected, can we say that we are getting at relevant explanations of causation.

5cd674ec7348d0620e102a79a71f0063Most facts have many different, possible, alternative explanations, but we want to find the best of all contrastive (since all real explanation takes place relative to a set of alternatives) explanations. So which is the best explanation? Many scientists, influenced by statistical reasoning, think that the likeliest explanation is the best explanation. But the likelihood of x is not in itself a strong argument for thinking it explains y. I would rather argue that what makes one explanation better than another are things like aiming for and finding powerful, deep, causal, features and mechanisms that we have warranted and justified reasons to believe in. Statistical — especially the variety based on a Bayesian epistemology — reasoning generally has no room for these kinds of explanatory considerations. The only thing that matters is the probabilistic relation between evidence and hypothesis. That is also one of the main reasons I find abduction — inference to the best explanation — a better description and account of what constitute actual scientific reasoning and inferences.

Some statisticians and data scientists think that algorithmic formalisms somehow give them access to causality. That is, however, simply not true. Assuming ‘convenient’ things like faithfulness or stability is not to give proofs. It’s to assume what has to be proven. Deductive-axiomatic methods used in statistics do no produce evidence for causal inferences. The real causality we are searching for is the one existing in the real world around us. If there is no warranted connection between axiomatically derived theorems and the real-world, well, then we haven’t really obtained the causation we are looking for.

  1. Frank Salter
    January 27, 2020 at 2:09 pm

    The underlying problem with econometric understanding is that the measurements are treated as if they are simply numbers. They are not. All quantities are products of the value and the unit of measurement — Q = {Q}[Q] where Q is a quantity, {Q} is the numerical value in terms of the physical unit [Q]. The acceptable mathematical manipulations on quantities are set out by the quantity calculus.

    The level of complexity of economic relationships needs to be dealt with by dimensional analysis. One of the founding papers is “ON PHYSICALLY SIMILAR SYSTEMS; ILLUSTRATIONS OF THE USE OF DIMENSIONAL EQUATIONS.” by E. Buckingham in Physical Review. 1914. It is freely available at URL: https://journals.aps.org/pr/abstract/10.1103/PhysRev.4.345

    Without understanding the vast simplification which these techniques provide in the physical sciences, economists will never progress.

  2. ghholtham
    January 27, 2020 at 4:33 pm

    “Some statisticians and data scientists think that algorithmic formalisms somehow give them access to causality”
    I’ll take your word for it Lars but I have never met any. The procedure is: you frame a causal hypothesis and test it on the data. Your premises can be simplifications but cannot be flagrantly counterfactual (unless you are Milton Friedman) and the implications of your formulation have to be corroborated by the data. Then your causal hypothesis is retained until disproved. It is true that econometricians talk of “Granger causality” but that is just a test of causal ordering to distinguish between theories when two variables are related. “Correlation is not causation” is something every statistician is taught in their very first class. Data do not generate hypotheses; people do. They can claim corroboration from data but not proof.

    Asad says: “The problem with current econometric techniques is that they do not have any scope for input of causal information – the language of econometrics does not have the vocabulary required to talk about causal concepts.” The mystery here is what sort of “causal information” he is talking about. In framing a hypothesis a theorist or statistician draws on intuition, knowledge of the world, imagination or whatever as well as observed empirical regularities. Is that what Asad means? Econometric techniques have nothing to do with framing hypotheses. They are indispensable for testing hypotheses. You can’t test the deepest causal theory except through its implications for observable phenomena , i.e. data. Econometricians who try to extract causal theories from data are said in the trade to be “fishing” and the term is normally pejorative.
    It would be helpful if Lars and Asad pointed to specific examples of what they are objecting to; then they could not be accused of attacking straw men.

  3. ghholtham
    January 27, 2020 at 4:56 pm

    May I repeat my question or challenge to Asad (and Lars too since he seems to agree with Asad): if I look at the practical procedures of a Bhaskarian scientist and a Kantian scientist, as he terms them, could I tell which was which? And how could I?

    • January 28, 2020 at 11:06 pm

      A pity Asad and Lars have not taken up this challenge, but I’ll have a go, since I have read some of both Kant and Bhaskar and are sympathetic to their positions given their contexts. Both start from disagreement with Hume, whose position Gerald doesn’t seem to realise he is taking when seeing theorists framing hypotheses with intuition etc. This despite reaching the same conclusion as myself: “Econometric techniques have nothing to do with framing hypotheses. They are indispensable for testing hypotheses.” One frames testable hypotheses by seeing what one can “measure”.

      Hume argues that one cannot see causality, as in Newton’s gravitational forces. All one can see is the results of measurements, and how one can even see he gets round by different people (“scientists” agreeing the the location of points on a graph (i.e. correlation). If one goes along with this, deduction has to proceed from correlated sets of measurements and there is no room for causality in the framing of the deductive argument (the logic of the time), only the uncertain result of what Hume called “induction”. Hume’s goal was social science. Before quantum theory and logical positivism, most physicists stuck with Newton’s forces.

      Kant, only a few years after Hume, was “roused from his slumbers” by discovering this. As I put it, he distinguished concepts (interpretive language) from both percepts (phenomena) and reality (forms of energy). His German, even in translation, is difficult to follow, but A D Lindsay’s “The Philosophy of Immanuel Kant” (The People’s Books, Jack and Nelson, 1919) is remarkably lucid and to the point, so I quote this (from p.47).

      “The doctrine it implies is very easy to misunderstand, partly because idealism is generally used in a very different sense than that in which Kant used it, partly because Kant’s statement of the distinction between things in themselves and phenomena depended on a view of knowledge [that of Hume] which he was very much concerned to refute, but with which we are not now familiar. If we are to understand Kant’s philosophy, we must know what he means by idealism, and wherein his idealism differs from that of his predecessors.

      “The word idealism is, naturally, contrasted with realism. It suggests an assertion that something is not real, but only an idea. … Something like this had been held by Kant’s predecessors. For the fundamental principle on which [they] had been agreed, and which is sometimes [mistakenly] called Cartesian, and sometimes called subjective idealism, is that the mind somehow knows itself and its own actions and states with more directness and certainty than it knows external objects. The doctrine is commonly based upon a confused view of sense perception.

      ” … For if we know, and must eternally know, only ideas inside our head, why should we ever imagine that there an outside world exists? … This idealism Kant is careful to refute, and he points out that there is no evidence for its fundamental proposition that we know our mind more directly than we know objects. We are only conscious of ourselves in knowing something not ourselves. … But it is a fact, and one that has got to be explained, that in judgement we go beyond what is present in our minds, and that in so anticipating what we shall experience, we assume that certain principles hold of all that has or may be present”. His examples include not just causality but space and time. In effect he is saying you cannot measure without a consistent measuring instrument.

      Bhaskar, following Popper, Kuhn and Lakatos in our era of quantum dynamics and communication theory, focussed not on deduction (the logic of unchanging states) but on dialectic, the (perhaps half-baked) evolutionary logic of Hegel rather than the materialism of Marx. His “A Realist Theory of Science” (1975, Verso) begins:

      “[A systematic realist account of science] must provide a comprehensive alternative to the positivism which since the time of Hume has fashioned our image of science”. On p.118: “The fact that closed systems are a presupposition of the actualist account of science is reflected in (a) in the absence of a theory of their establishment and (b) in the absence of a clear contrast between pure and applied phases of scientific activity”. Kant is not indexed in this book; to see how Bhaskar’s views related to his one needs to look in “Dialectic: The Pulse of Freedom” (1993, Verso).

      P.239-40: “[I]nsofar as all transcendental arguments turn on agency (including Kant’s original one, once one rejects an impossible empiricist account of experience), all transcendental argument must be seen to presuppose the category of absense. Even more simply, a sentence without absences,pauses or spaces would be unintelligible. Thus absence is a condition of any intelligibility at all.”

      p.359: “Kant involuted structure, but the synthetic a priori [invented before it could be used] could not do anything to discriminate between the transfinity of possible causal laws consistent with the empirical data”. [See discussion of ‘retrodictability” in the “Realist Theory”].

      p.365: “An Achilles Heel critique seeks to show that it is precisely where a position seems strongets that it is actually most week. Thus Kant cannot sustain an intelligible concept of freedom, or Hegel of historicity”. [Arguably he can].

      p.393. This in particular is what you won’t find in Kantian science: “4D = Fourth Dimension. Unified by the category of transformative praxis or agency. In the human [or Algol68 computing] sphere it is implicit in the other three [1M, 2E, 3L]”. C.f. data, variables and interpretations in computing.

  4. Yoshinori Shiozawa
    January 28, 2020 at 3:02 am

    If we argue about causality in economics, we should read Michael Joffe’s article Causal theories, models and evidence in economics.

    In my impression, it is useless to argue causality between macroeconomic variables (variables as indicators of macro state of an economy). They may be connected by causal chains but only through a long (and probably too long) chains of causalities and it is not suitable to judge the existence of causality between them. Economics must change tradition of arguing causality between macroeconomic variables and seek to find more direct causal relations. This requires the drastic change of of our research program. It seems that Lars and Asad are still trapped in an old tradition.

  5. ghholtham
    January 28, 2020 at 3:13 pm

    Well, Shiozawa-san is correct. There is no unconditional causality among macroeconomic variables because they are aggregates embedded in a complex system of inter-relationships. A change in X may normally precipitate a change in Y but there will be circumstances where all the other variables in the system conspire so that it is not so. That is why relationships in macroeconomics, like that between income and consumption or profit and investment are perceptible but highly stochastic and variable, The long chains of causality exist. The point is the more links there are in the chain the more likely is it to be broken along the way from time to time.
    If we descend to a more basic level and look for causal relationships at a lower level of aggregation we cannot just flip those up to the macro level as the new-classical school attempts to do. if the lower level theory is realistic its macro implications can only be explored by computer simulation. Just as geologists can solve simple models with strains and stresses from basic physics but when trying to understand and predict real-world earthquakes have to resort to computer simulation too. Complex systems cannot be “solved” algebraically but they can be mimicked/.

  6. January 29, 2020 at 3:26 am

    As I commented on Prof. Asad Zaman’s serial posts, econometrics is surely a method, referential, effective but far from perfection. If we economists pursue perfection, econometrics will be not satisfactory to us; but, if we pursue only usefulness, econometrics will be valuable, among many other methods. Hence, over-criticizing econometrics just reflects the underlying preference of perfectionism of critics, which has been committed by mainstream for long times. Mainstream economics explains some aspects of the world, thus it is impossible to be denied totally. What need to do today is to revise and remedy it, so as to build a new synthetic economics, with plural methods, plural data (than quantitative data), plural relationships, a whole framework, directly toward the real world. Thanks! https://goingdigital2019.weaconferences.net/papers/how-could-the-cognitive-revolution-happen-to-economics-an-introduction-to-the-algorithm-framework-theory/

  7. gerald holtham
    January 29, 2020 at 11:51 am

    I agree with BinLi’s remarks here, though I don’t know much about algorithm-framework theory.

    Thanks to Dave Taylor for his exegesis but I am still in a bit of a fog. We can step past idealism or phenomenalism and agree there is a real world. As Quine observed, language is an inter-subjective activity and all its concepts are based on “things glimpsed not glimpses”. I take the old-fashioned Popperian view that induction is a psychological process that cannot confer scientific legitimacy on general statements. That leaves me with his schema of conjecture and refutation and the provisional nature of all scientific knowledge. The strain of agnosticism in that does have a Humean flavour but we are in a different place. This discussion seems to be about the status or nature of the conjecture. People can’t help but think in terms of causation but at the end of the day our notions can only be tested by the behaviour of observable phenomena and their temporal ordering. So my question remains: whether I believe I am exploring deep real structures of causation or whether I think I am just trying to construct a model of necessary and sufficient conditions to predict phenomena what would I do differently? In empirical work it seems to me I’d do much the same thing in either case. I might tell different stories about them but from the same starting point I’d settle on the same equations. Attempted refutation follows the same process in either case. Is that wrong?

    Incidentally, Hume may be old hat but he wrote the most beautiful English prose, discussing tough concepts with great clarity and a touch of humour. You can read him for pleasure. The contrast with filthy sentences like: “Kant involuted structure, but the synthetic a priori [invented before it could be used] could not do anything to discriminate between the transfinity of possible causal laws consistent with the empirical data”. is stark. I’m a Philistine I suppose but when a writer makes you work that hard to figure out what he means, I can’t help but think he should have worked harder himself to make himself clear. One can only hope obscurity rests in the expression, not in the thought itself.

    • January 29, 2020 at 2:17 pm

      Gerald, I have a visual rather than linguistic mind, so I see meanings and am fairly insensitive to wording. I will totally agree Lindsay is much more lucid than either Kant or Bhaskar, but he has the advantage of already being familiar with their work, where they are having to find words to express things no one has previously given words to. This is like Lindsay offering fish for dinner and the others having to invite us to come fishing, hoping we will catch the fish for ourselves. In the “filthy sentence” you object to, blame me for the bracketed explanation.

      I didn’t like what Hume was saying, but found Kant and Bhaskar agreeing with what I had worked out for myself. Popper conflated Hume’s “induction” (econometrics) with intuition (which is a psychological process), and failed to see what Bhaskar read into Kant: that one can only refute what has already been conjectured, not what is absent. In my clarification of Bhaskar’s phases of science, “induction” functions entirely logically as the statistical quality control procedure governing acceptance of the results of applied science.

      The answer to your question was given you as the difference between pure and applied science. Pure science is seeking necessary and conditions by retroduction to the underlying structures directing causes. (C.f. Peirce’s “abduction”: abstraction from the evidence). Applied science takes unproven abstract conditions for granted, builds models enabling them to deduce consequences and seeks either to sell these or (with Popper) to disprove them.

      “Philistine: a person of material outlook indifferent to culture”. I wouldn’t call you a Philistine, but you do seem to have difficulty thinking outside the box of your own culture, where new understanding is by definition not yet commonplace. You seek lucidity, but a great teacher argued that students in pursuit of new understanding need to be made to think for themselves.

  8. gerald holtham
    January 29, 2020 at 11:59 am

    By the way, Occam knew how to “discriminate between the transfinity of possible causal laws consistent with the empirical data”. Pick the simplest and don’t multiply entities unnecessarily. How would Bhaskar do it?

    • January 29, 2020 at 3:02 pm

      No,. Occam is (in Einstein’s phrase) keeping it “too simple” by leaving out language. Bhaskar here is on about what Asad is discussing in terms of Simpson’s paradox. Bhaskar’s simplest is to be found in a five-line diagram on p.69 of “The Possibility of Naturalism” (3rd edn, 1998, Routledge), which is Wheatstone’s Bridge without the meter linking the two potential dividers, i.e. the indicator of truth when the ratio of interest (the theory) is the same as the reference ratio (the facts).

  9. ghholtham
    January 29, 2020 at 6:35 pm

    Well, perhaps you are right. I have difficulty with the notion of meaning as distinct from wording. We are communicating with words after all. No doubt it is the British empiricist in me but I like my metaphysics to be as clean as possible. I accept that phenomenalism and logical positivism don’t work. We need a metaphysics but I can’t accept Occam was too simple, if you are allowed to be profuse with unobservable entities you can come up with an infinity (or is that transfinity) of nonsense. I do suspect Asad of multiplying unnecessary entities but I’m hoping someone can descend to my level and explain why that’s not so.
    Btw I’m sure Popper hadn’t failed to notice that you can’t refute what hasn’t been conjectured, What we haven’t imagined, we can’t know.

    • Craig
      January 29, 2020 at 8:26 pm

      “What we haven’t imagined, we can’t know.”

      Precisely. Paradigms are both temporal patterns and human mindsets. If one never shows an awareness of the need to analyse on the integrative level of the paradigm let alone attempt to do so outside of the current paradigm (which itself most people are unconscious of) then their awareness will remain blunted and blinded.

      Unconsciousness and unwillingness to look are the first and most essential barriers to overcome in both psychoanalysis and the temporal pattern/human mindset phenomenon AKA paradigms.

  10. ghholtham
    January 29, 2020 at 7:03 pm

    Sorry to go on but I find this stuff interesting. So here is another question: abduction or retroduction from phenomena can lead to different views of the underlying causal structures. The history of science is full of competing theories that were data-compatible at the time. Asad himself has remarked on that, So how can abduction be a logical process? I’ve heard of fuzzy logic but that’s ridiculous. There is no unique process of abduction. Attempted abductions can lead to different places. This cannot therefore be a method of quality control in pure science. Yes the theory has to be internally consistent and coherent and it will evidently be consistent with the experience that gave rise to it. But we gain confidence in it when it passes more empirical tests. That, coherence and parsimony are the quality controls.
    And when experience shows a theory applies only within limits and outside those it needs to be superseded what do we say about the original abduction? Evidently it proceeded from a partial data set so it didn’t lead to truth. But all our data sets are partial…
    I suppose I’m trapped in my cultural bubble – but aren’t we all?

  11. Ken Zimmerman
    February 1, 2020 at 3:33 pm

    Facts, as you note have many different, possible, alternative explanations, but we want to find the best of all contrastive (since all explanation takes place relative to a set of alternatives) explanations. But here is no means, and never has been for humans to pick the “best” explanation. Our approach to causation needs to change. Causation is not a thing. It is a story. A story that explains within the framework of a specific culture how events or actors relate to one another, and how both relate to human culture and society. For example, every society have a creation story that explains the origins of that society and its place in the “universe.” That’s causation, a story. A believable story we can share. But this “believable” story is still part of complex events. Complexity makes identifying causality in a predictive way impossible.

    The central question is why did humans create, invent causality? It was invented as a way for humans to do what they do best, make up a story. In this case a story that explains to them why some event or action occurred as it did, when it did. And as with all things humans invent, causality is contextual and relational. It’s existence and form depends on the historical circumstances of its creation. Its time and place of creation and its relationships with other actors and events. A few examples in the area of wealth. These are complex and extensive stories. I’ll just touch the surface. First, according to the ‘Foundation for Economic Education,’ “The free market is in the interests of the poor. The alternative, the transfer system, is not humane. It locks people into poverty by destroying opportunities and prohibiting many kinds of productive work. The welfare state encourages dependence, rather than self-reliance. The welfare recipient is deluded into believing there is such a thing as a free lunch. In a free market it is necessary to give in order to receive; in the welfare state, the scheme is to try to live at the expense of others.” Thomas Sowell, a well-known economist sort of goes anthropological in his book, Wealth, Poverty and Politics: An International Perspective. The central argument of this book is that the members of one group can learn from the members of another, provided they don’t allow envy and resentment to poison their worldview. And thus, everyone becomes wealthier. Perhaps his approach is more psychiatric than anthropological. Finally, according to Inequality.org, race plays a large role in who is wealthy and who is not, and by how much. With being white a major factor in having wealth and the amount one has.

    • Craig
      February 1, 2020 at 7:30 pm

      “Our approach to causation needs to change. Causation is not a thing. It is a story. A story that explains within the framework of a specific culture how events or actors relate to one another, and how both relate to human culture and society.”

      Yes, parables/stories are the tool of wisdom which is precisely what we need in economics and in the entirety of our lives.

      “A believable story we can share. But this “believable” story is still part of complex events. Complexity makes identifying causality in a predictive way impossible.”

      Time and space will always make PERFECT predictability impossible, but the wisdom of grace as in the integration of the discern-able truths in opposites is pretty damned predictive, and undoes any obsessive need for perfection. Perhaps god or the computer programmer who created the cosmos set it up this way….so we wouldn’t become so titanic-ally bored by perfect predictability.

      • Ken Zimmerman
        February 12, 2020 at 12:28 pm

        Craig, since humans make-up the causes, they can change them. The important part played by these stories is to help humans prepare for the inevitable uncertainty faced in all actions and events. Our earliest Sapiens ancestors were hunters and gatherers. How they chose to organize these activities reflects the many uncertainties of hunting and gathering. Economists (and others from religious followers to physical scientists) depict humans as now the masters of these uncertainties. Which is, of course nonsense creating dangerous consequences and a large number of sociopathic members of society.

      • Craig
        February 12, 2020 at 6:43 pm

        I’m not disputing your anthropological observations….only your tendency to utilize an anthropology ONLY point of view. It’s just that it (still) neglects to integrate the cognitive aspect of the phenomenon and study known as paradigm perception/wisdom.

        In other words anthropology is the abstract scientific half of paradigm perception/wisdom because it focuses on the culture/pattern, but like every other scientific/social scientific study it still clings to the patina of the present monopolistically demanding Science Only paradigm which (still) blunts if not robs it of potential insight.

        We need to Integrate…and keep on integrating…because the deepest truth of the cosmos is that it is in a dynamic, interactive, integrative state of grace…with just enough uncertainty enforced perhaps by time itself….to make it interesting.

        The irony is that with the NATURAL cognitive enhancing disciplines of wisdom applied it not only remains interesting it’s potentially joyously so.

      • Ken Zimmerman
        February 13, 2020 at 1:20 pm

        You are correct. Anthropologists cannot escape their location in time and place, and the relationships involved in that placement more so than can any other human. They have the same limitations as all other humans. But they attempt to at least glimpse outward to gain a bigger and broader perspective by studying humans (homo Sapiens) in all their parts and relationships over time. It’s a big and always uncertain and incomplete job. This is the total of the anthropology point of view. You’ll have to judge for yourself if it’s helpful.

  1. No trackbacks yet.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.