It is a familiar refrain in criticisms of positive economics, in particular positive neoclassical economics (which indeed has by far the lion’s share of work of this kind), that it relies too strongly on unrealistic theories. Now when people speak of unrealistic theories, what they tend to mean in practice is the reliance in positive neoclassical economics on modelling, and more specifically modelling on the basis of assumptions that are known to be false. More than any other aspect of neoclassical methodology, this has come in for much criticism both from within and without the discipline of economics. It has nonetheless also had its defenders – most famously Milton Friedman, to whom is ascribed the thesis that the unrealisticness of assumptions does not matter at all as long as the theory so developed has better predictive value than any other.(1) This has also often in the minds of the critics been associated with the mathematization of modelling and economics in general that has taken place in the second half of the 20th century, and which is often seen as masking the falsehood of the models and thereby the theories by hiding it behind mathematical formulae. Yet although I think neoclassical economics is by and large poor economics and much of these things are very worth criticizing, it is important to look more closely at these matters and to separate some of the different aspects of these methodological issues and the basis for criticizing them.
First of all, it is important to understand the defense for such approaches in economics. It will not to at all, as many of the critics from outside economics tend to do, to attack a model purely for the unreality of its assumptions alone. When economists use models assuming perfect competition, or perfect information, or the profit maximization of firms, this does not entail at all a commitment to a belief that such conditions actually hold in real life. This explains the irritated and contemptuous response neoclassical economists often have for the ‘vulgar’ criticisms of their methods: they do not need to be explained that in reality there isn’t actually perfect information, because they know this. The point of the model can be of various kinds, but at all times the point of the unrealism of the assumptions is to abstract from the aspects of reality that are deemed to be nonfunctional for the purposes of the model. One cannot sensibly criticize neoclassical economists for doing this, because all theory necessarily requires such assumptions: if one does not abstract from reality at all, one is left with all of reality, and one does not have a theory. As Friedman pointed out in his celebrated paper, an entirely ‘realistic’ theory of the wheat market would need to contain every possible information about every wheatstalk, every trader up to the color of their hair and their eyes, and so forth – an infinity of knowledge.(2) In natural science it is to some degree (although by no means always) possible to develop theory on the basis of direct and repeatable experimentation under constant conditions, but in economics the nature of the science does not permit this. As Karl Marx put it in the preface to Capital: “In the analysis of economic forms, moreover, neither microscopes nor chemical reagents are of use. The force of abstraction must replace both.”(3) The point of the unrealistic assumptions is therefore to allow them to elucidate forces or capacities in reality that actually prevail, but that are hidden behind exogenous counter-forces, interfering variables and so forth, and which cannot by experiment be shown due to the nature of social science. The point is emphatically not to claim the assumptions of the model world to apply to the real world, but of the forces in the model world to be analogous to those in the real world in the relevant aspects.
The second point of defense is one of parsimony. The ability to abstract from all real operating forces to those which are believed to be the controlling ones allows us to go from the ‘world of appearances’ to the world of abstraction in which we can limit ourselves to those causes only. Under these circumstances, it becomes possible (hopefully) to explain a great number of disparate such appearance phenomena with a small number of dominant underlying real causes, which then allows us to do much with little. This serves the purpose of moving towards further explanatory unification in science, which is an important, perhaps even defining, goal in social and natural science alike.
This does, however, raise some real questions which can be fruitfully used as points of criticism. The main point is the necessity of understanding the nature of the predictive value that these models are supposed to have and that they are supposed to be judged by. As is well-known by now in philosophy of science, the actual factual outcomes which would constitute the empirical test for theories are underdetermined by theory: for any given factual case there is always more than one theory that can account for it. In practice it is therefore not easily possible to apply a simple methodology that ignores whatever the assumptions may be and builds a predictive model, and then compares it with rival predictive models for the purposes of empirical testing. There will never be agreement about to what extent reality matches with given such theories, as is shown by the long-term persistence of strong theoretical divisions, whether between Marxists and neoclassical economists or more narrowly between Keyenesians and monetarists. It is almost impossible to determine a priori therefore at what level one should declare the problem to occur if the facts do not match the predictions of a model: maybe the data are wrongly gathered, or maybe the data are polluted by third variables affecting them, or maybe the model was wrongly constructed, or maybe the theory is incorrect. Which of these one thinks the most likely one in any given case tends to depend strongly in practice on the political and theoretical implications it would have and the degree to which they fit one’s preconceived idea of how the world works. As Friedman states: “Observed facts are necessarily finite in number; possible hypotheses infinite. If there is one hypothesis that is consistent with the available evidence, there are always an infinite number that are.”(4)
Understanding this fact also allows us to understand the different uses to which modelling in neoclassical economics can be put (and not only in neoclassical economics). Because one cannot go directly from a model to reality to check whether the predictions hold true, the models function, as Uskali Mäki and others have pointed out, as a model world. They are worlds in which the assumptions made hold true – worlds in which perfect information exists, profit is maximized, or whatever. This allows them to be judged on their theoretical virtues as a way of distinguishing one hypothesis about the known facts from another, in light of the above mentioned underdetermination.(5) These theoretical virtues, again, would be primarily those of parsimony and of elegance of explanation, as well as the manner in which it allows hitherto complicated matters to be formalized for ease of use. In such a way, this formalization can allow us to grapple better with complicated questions by distilling them into clearly defined elements when they would otherwise remain muddled. As Levins and Lewontin defend mathematical formalization in science:
Mathematics is used mostly in modeling in order to predict the outcomes of systems of equations. But is also has another use: educating the intuition so that the obscure becomes obvious. When we abstract from the reality of interest to create mathematical objects, we do this because some questions that would seem intractable can now be grasped immediately.
The role of assumptions in models can therefore be manifold, as manifold as the purpose of the models themselves. Models in economics will then namely be seen as having two major possible functions (there are also some minor ones which I will not go into in detail): either they function as purely theoretical model worlds, and the purpose of the exercise is to elucidate some aspect hitherto unacknowledged about this model world, thereby improving the precision of and knowledge about a given positive economic theory, or a model attempts to directly make a claim about real causes that operate in reality, by either making a predictive claim on the basis of the model’s abstraction of reality or by making a claim about the reality of the causal factor identified in the model. These two different approaches are both common in neoclassical economics and are not always properly differentiated by the economists themselves. Yet they should attract accordingly different criticisms on their merits. Uskali Mäki again has described very well what the difference is between the ‘substitute systems’, as he calls the former type, and the ‘surrogate systems’, as he calls the latter type:
One kind of criticism attacks styles of inquiry that treat a model as a substitute system only, not even intending it as a means for gaining access to the real world. The alleged problem is that there is no attempt. The other kind of criticism acknowledges a model being treated as a surrogate system, but blames it for failing in accessing the social world. The alleged problem is that there is a failed attempt. The history of economics exhibits both kinds of criticism.
In order to criticize neoclassical economics effectively, one must separate these two types of criticism. The unrealisticness of assumptions is warranted insofar as a given model’s identification of causes can, in the terminology of Mäki, be said still to resemble the real causes operating in the world; it is an empirical question whether this is the case or not, and therefore one that is very liable to the underdetermination problem identified above. However, there are also cases in which the unrealisticness is always unwarranted. There are several such cases. The most important one is the case in which either the model is only used to elucidate other models or assumptions in positive neoclassical economics, without making the attempt to connect it with the real world in any way, except highly indirectly – on the basis of the principle that most of neoclassical economics can be accepted as known and true, and therefore elucidating theoretical aspects of neoclassical economics’ assumptions is helpful. Here the model is then subject to the ‘so what?’ critique: even if it is true that a given model has certain properties as a thought experiment, it is still necessary to justify empirically the connection of the thought experiment to reality. However, there is obviously room for maneouvre here: as Mäki also points out, what appears from a critic’s point of view as a substitute system can from the point of view of a practitioner appear as a surrogate system, it just happens that the model discussed in a given paper is only remotely connected with the eventual application to reality. Again, there will be disagreement on the point of this being true or not, and the more remote the application to reality is, the greater the room for challenge. Very often in economic papers the models discussed make no hypothesis about reality at all, or when they do, the hypothesis makes an immediate and unjustified leap from the theoretical properties of the model to the theoretical properties of reality – here is an excellent terrain for criticizing neoclassical economics, which seems especially prone to these errors.
A slightly less significant but also relevant point of criticism for modelling in this context is the use of assumptions in models for the purposes of tractability, including presentation for pedagogical purposes and the like. Here, it is of the utmost importance that such assumptions when unrealistic are as trivial and as irrelevant to the actual point of contention as possible. As Mäki has pointed out but perhaps not emphasized strongly enough, contrary to the habitual practice of many neoclassical economists, any unrealistic assumptions made for the purposes of tractability must not have any ontological implications. As his example goes:
A few decades ago economists lacked the mathematical tools for dealing with increasing returns and monopolistic competition in a general equilibrium framework. This violated the ontological convictions of many economists working on development issues: these economists conceived of (major parts of) the economy as being governed by positive feedback mechanisms and market imperfections. In case a conflict between ontology and tractability is resolved in favor of tractability while suppressing ontology, the obvious suspicion is that the models that ensue are (or are to be) treated as substitute systems only.
This connects to the point about theoretical virtues: while it is not right to criticize neoclassical economics for using models given their theoretical virtues, it is right to criticize them when the theoretical virtues alone are the reason for the models and the attendant unrealistic assumptions. The purpose of any assumptions must at all times be firmly structured with an eye to the real world.
To sum up: it will not do for critics of positive neoclassical economics to complain about the unrealisticness of assumptions in modelling alone. It is necessary to distinguish between different types of models, particularly whether their purpose is of description or prediction of causes in the real world by abstracting from all causes deemed negligible or whether their purpose is to create a thought experiment in order to elucidate aspects about positive economics to make them clearer, easier to teach, easier to model etc. In the former case, we must judge them by whether we think they stand the empirical test – this cannot be definitively resolved, and will remain a subject of contention. But Mäki has suggested that by reformulating the unrealistic assumptions in terms of positive theses about the negligibility of causes abstracted away, it becomes easier to find a way of testing them. He here distinguishes the weaker version (‘negligibility’), which implies that the model can justifiedly abstract from real existing causes because they are held to be negligible for the purposes of the model, and the stronger version (‘applicability’), which holds that the causes in the domain discussed actually are negligible.(9) Whether this is sufficient as a ground for criticism of positive neoclassical economics can be doubted, but distinguishing the two is important, and it does open the way for critiques to be more focused on making implicit assumptions in neoclassical models and papers explicit, as well as pointing out inconsistencies in the conclusions drawn by economists from the logical structure of their own models. This in turn is a fruitful base to contrast them with alternative economics approaches such as Marxism, which may need many fewer assumptions and be able to boast at least equal empirical plausibility. With regard to the latter case, the case of pure thought experiment, the unrealisticness of the assumptions can and should be attacked directly, because there is no plausible reason for permitting unrealism if it is no immediate help in understanding the world as the real object of study. Here, the ‘vulgar’ critique of neoclassical economics is more in the right place. As Mäki summarized it:
Both have a chance of being true, but do so in virtue of different kinds of fact. On the one hand, paraphrases in terms of negligibility and applicability make reference to, and may be true partly in virtue of, properties of the real target system of a model. Thanks to this, they can be used for justifying the original idealization. On the other hand, the metaclaims listed above are about the roles of idealizing assumptions in modeling practice, so are true or false in virtue of pragmatic facts only – facts about disciplinary practices of inquiry and education such as early-step, tractability, academic entry, pedagogic value. These claims may be very informative about important facts of academic practice, so may be justified as such factual claims. But (…) re-nomination in terms of early-step status, tractability, aesthetics, pedagogy, and academic entry may support nothing but substitute modeling if left unsupported by other arguments that do appeal to real-world facts. Negligibility and applicability assumptions do make appeal to the properties of real targets and are therefore able to support surrogate modeling.
1) Milton Friedman, “The Methodology of Positive Economics”, in: Essays in Positive Economics (Chicago, IL 1966), p. 3-16, 30-43. See also: http://www.ppge.ufrgs.br/giacomo/arquivos/eco02277/friedman-1966.pdf
2) Friedman, p. 32.
3) Karl Marx, “Preface to the First German Edition”, Capital Vol. I [Hamburg 1867]. http://www.marxists.org/archive/marx/works/1867-c1/p1.htm
4) Friedman, p. 9.
5) A.o. Uskali Mäki, “Unrealistic assumptions and unnecessary confusions: Rereading and rewriting F53 as a realist statement”, in: Mäki (ed.), The Methodology of Positive Economics. Reflections on the Milton Friedman Legacy (Cambridge 2009), p. 90-116. In this article Mäki also contests the necessity of reading Friedman as an anti-realist.
6) Richard Lewontin and Richard Levins, “Strategies of Abstraction”, in: Lewontin & Levins, Biology Under the Influence (New York, NY 2007), p. 166.
7) Uskali Mäki, “Realistic realism about unrealistic models”, in: Kincaid & Ross (eds.), A Handbook of the Philosophy of Economics (Oxford 2009), p. 68-98.
9) Uskali Mäki, “The truth of false idealizations in modeling”, in: Humphreys & Imbert (eds.), Representations, Models and Simulations (forthcoming).