[Elster, Jon (1983), Reply to comments (on Marxism, functionalism and game theory), Theory and Society 12:111-120]
Reply to comments
(On the article "Marxism, functionalism and game theory")
[start of page 111]
It is both gratifying and somewhat confusing to be the object of comments as acute and diverse as those made in the July issue of Theory and Society on my "Marxism, Functionalism and Game Theory." To some extent the discussants demolish each other's objections. Thus Roemer and (more ambiguously) Van Parijs support my view of functional explanation against Cohen, while Giddens defends it against Berger and Offe. Yet this does not relieve me of the task of replying more directly to the critics. Rather than deal with the comments one by one, I shall organize my reply under five general headings: methodological individualism, functionalism, structuralism, game theory, and Marxism.
This view is criticized both by Berger and Offe and by Giddens. The former do so in a very explicit statement that microeconomics and macroeconomics "can and must be methodologically isolated from each other." The example is an interesting and historically important one. Schumpeter - who invented the term methodological individualism1 - was always skeptical about Keynesian economics because
it keeps analysis on the surface of things and prevents it from penetrating into the industriaI processes below, which are what really matters. It invites a mechanistic and formalistic treatment of a few isolated contour lines and attributes to aggregates a life of their own and a causal significance that they do not possess.2I believe this observation remains sound, and so do the many economists who for some decades have been engaged in providing microfoundations for the aggregate relations of macroeconomics.3 Because Berger and Offe actually cite an "example for such nonreducible laws," viz., the "neo-Ricardian formulation of an inverse relation between the rate of profit and the real wages," I may perhaps be allowed to show how it can be reduced to microeconomic terms.
[end of page 111, start of page 112]
Consider a simple model of an economy that produces only one final (consumption) good with the help of one intermediate or capital good.4 To make one unit of the capital good one needs a11 units of the capital good and a12 units of labor. To make one unit of the consumption good one needs a12 units of capital and a12 units of labor. We set the price of the capital good equal to 1 by convention. The real wage (in units of the consumption good) is w, the rate of profit is r, and the price of the consumption good is p. We assume that the capitalists do not calculate profits on the advance for wages, only on constant capital. We can then lay down the following equilibrium conditions:
a11 (1 + r) + a12 * w * p = 1
a12 (1 + r) + a12 * w * p = p
Each equation says that capital + profit on capital + wages must equal price. The underlying postulate of an equal rate of profit in the consumption sector and the capital sector follows from the microeconomic postulate of profit maximization and the assumption of unrestrained competition, and can be justified in no other way. Given these equations, one can derive r as an explicit function of w, and obtain the downward-sloping curve cited by Berger and Offe. Much more sophisticated treatments are of course possible.5 They all rest on the same basic idea: given the technical production data, one can specify in microeconomic terms the equilibrium conditions from which the macrorelations can be derived.6
I find Gidden's strictures on methodological individualism hard to grasp, and his own alternative virtually impenetrable as presented both in his comment and, more elaborately, in the version offered in his "Agency and Structure."7 So let me latch on to the most tangible part of his argument, viz., that the
structural properties of linguistic . . . systems cannot be expressed as qualities or descriptions of the conduct of either individual or collective agents. Syntactical rules, for example, are not attributes of individual speakers, speech acts, or of texts. They are instantiated in, and reproduced through, speech and writing, but that is something different.I agree that language is the most plausible-looking example of a supra-individual entity instantiating itself in individual behavior. Yet the very fact of (structural) linguistic change shows the need to anchor these rules firmly in individual usage. It is the strain and conflict of rules in individual usage that set up a pressure for change, and relative stability is similarly explained by the (temporary) attenuation of such strain. The unfortunate legacy of Saussure is to set up a methodological dichotomy between synchronic and diachronic linguistics, with the concomitant view that the synchronic struc-
[end of page 112, start of page 113]
ture somehow has primacy over individual usage. Here, as in other cases, the dichotomy may be useful for heuristic purposes, but will not bear a great theoretical weight. Consider as an analogy the economic distinction between shifts along the production function and shifts of the function.8 This distinction has proved useful in many empirical studies, although in the very process of applying technical knowledge new insight is acquired so that, strictly speaking, it does not make sense. For purposes of short- or medium-term analysis it is often possible to treat a slowly moving frontier as being at rest, but one must not then go on to hypostatize it as an immutable "structure."
Functionalism in my article is conceived us functional explanation. Cohen, Roemer, and Giddens understand it similarly, the former defending it and the latter two accepting my criticism of it. Berger and Offe and Van Parijs choose to talk about something different. Berger and Offe shift the emphasis from functional explanation to functional equivalence, while Van Parijs prefers to talk about "satisficing explanation" rather than functional explanation. I shall deal with these nonstandard views in turn, and then address myself to Cohen's defense of the standard view.
I am baffled by Berger and Offe's account of functionalism. How does one explain an institution by listing others that could do the same job? Perhaps some insight is gained when one perceives that institution A might have the same function (i.e., the same beneficial consequences for something or other) as the actual institution B, but not in the sense of explaining the emergence or persistence of the latter. If anything, identifying functional equivalents to B would dim the prospect of successfully explaining it in the functional manner.9 Moreover, Berger and Offe then go on to make their brand of functionalism virtually vacuous, by turning it into the analysis of the intended or unintended, positive or negative, consequences of the institution in question. Here the institution is the explanans, not the explanandum. Clearly, to trace the open or hidden consequences of an institution is often a valuable task. Somewhat less clearly, it may also be useful to point to alternative institutions that might have brought about the same consequences. Neither of these tasks, however, is of any help in providing an explanation of the institution in question. Because my article is concerned exclusively with this explanatory issue, the Berger and Offe argument is a red herring - calling "functionalism" a mode of argument only superficially related to what I discuss.
Van Parijs raises a number of points in his comment. First, he objects to my argument against the Markov-process approach to social change. I agree
[end of page 113, start of page 114]
that the argument is too brief to carry conviction and that, even if valid, it does not show the difference between biology and social science to be more than a matter of degree. Elsewhere my views on this topic are spelled out more fully.10 Part of the problem is that natural selection in biology is the object of a precise and powerful theory, while the Markov-approach remains a mere metaphor, 11 with one exception only: the series of simulation studies conducted by Richard Nelson and Sidney Winter, showing that in a population of firms subject to technical change there will be at all times a large proportion of nonoptimizers. In their words, "changes in the 'best' techniques known by firms and in the external environment of product demand and factor supply conditions may be sufficiently rapid relative to the speed of adjustment of the overall system that a wide range of behavior can survive at any time."12 Van Parijs' observation that "endogenous pressures to change provide social systems with a way of speeding up adaptation that is not available to biological systems" is correct, but irrelevant, because the process that leads to increased adaptation also makes for more rapid change of the system to which the firms have to adapt. For each adapting firm, the environment consists largely of other adapting firms.
Second, Van Parijs tries to answer my objection (or one of my objections) to functionalist explanation by changing the nature of the beast. As in his recent book,13 he argues that institutions may be explained by the absence of destabilizing consequences rather than by the presence of optimal ones. The difficulty with this suggestion is that it rescues functionalism by diluting its explanatory power. Below the critical level of crisis-inducement there may be many "satisfactory" institutions, and non-functional explanation must be invoked to explain why one rather than the other is observed. Third, Van Parijs attempts to refute my claim that "positive long-term consequences could never dominate negative short-term effects in the absence of an intentional actor" by giving a biological counterexample. If the example - parental altruism - is redescribed so as to make the gene rather than the organism the unit of selection,14 it is seen to be perfectly consistent with my claim. It would provide an objection only if the gene, to increase its frequency in the population, had to undergo an initial decrease.
Van Parijs objects mainly to my view that there are few non-intentional mechanisms in the social sciences that can support a functional explanation, but agrees, on the other hand, with my view that without specification of a mechanism such explanations cannot be sustained. Cohen objects to the latter, more fundamental claim. More precisely, he tries to rebut two of my arguments against his view that functional explanation can dispense with knowledge of mechanisms. I accept his rebuttal of my second argument,
[end of page 114, start of page 115]
concerning the purported ambiguity concerning time. I do not believe, however, that he succeeds in refuting my first argument. I shall not only uphold it, but add, another argument that, together with the first, corresponds to two well-known objections to Hempel's "covering-law" model of scientific explanation. These turn on the problem of epiphenomena - i.e., the confusion of causation with correlation - and the problem of preemption - i.e., the confusion of causation with necessitation. Analogously, we may have a well-confirmed "consequence law" à la Cohen of the form "If (if A, then B), then A," and yet this does not provide an explanation of specific instances of A if both A and its tendency to produce B are effects of a common cause C. This is the first argument of my paper. Cohen tries to rebut it by referring to "tests which, when appropriate results are forthcoming, render the hypothesis that there exists such a C implausible." I believe, however, that such tests are always inconclusive in the absence of a priori assumptions, i.e., in the absence of a mechanism.15 Also - corresponding to the notion of causal preemption - we may have a well-confirmed consequence law that has no explanatory power in a specific instance because some other mechanism preempted the operation of the mechanism underlying the law. (Recall that Cohen insists there must exist some mechanism sustaining a valid consequence law; he denies only that we must have knowledge of it to provide a satisfactory explanation.)
In my article I discuss and reject structuralism, i.e., the view that the constraints on action typically are so strong as to make rational choice within them irrelevant. Both Giddens and Berger and Offe take me to task for this objection. Cohen also makes a related, although different point.
Giddens disagrees with my view that the "action of individuals can then only be conceptualized as occurring in whatever space is left over from the operation of such constraints." He argues that as a consequence of this view I "find great difficulty in recognizing the point stressed by Berger and Offe," viz., that the "game starts only after the actors have been constituted, and their order of preferences has been formed as a result of processes that cannot themselves be considered as being part of the game." Not only have I no difficulty in recognizing this point; I explicitly recognize it in note 46 of my article. Moreover, I have written two books arguing at some length against the dualism of "choice versus structure." On the standard, dualistic view the structural constraints embody an element of necessity, the preference-guided choice an element of freedom. One nonstandard objection is that people are sometimes free to choose their constraints, as in the story of Ulysses who
[end of page 115, start of page 116]
bound himself to the mast to resist the Sirens' temptations. 16 Another is that the preferences are sometimes shaped by the constraints, as in the story of the fox and the sour grapes. 17 I admit, however, that in my article I largely follow the standard view, mainly because one cannot introduce too many complexities at a time. In any case, I fail to see how the fact that preferences are endogenously shaped by the social system can serve as an objection to the view that they determine what occurs in the space left over from the operation of the constraints. Preferences are not rock-bottom, but non-rock-bottom explanations may also be valid.
I believe, in fact, that Giddens conflates two objections to the rational-choice approach: that preferences are endogenous, and that (in any case) they have no alternatives on which to operate. Berger and Offe address themselves more directly to the latter issue. After an opaque reference to Althusser, they state that it is only when.
class action is not perfectly determined by objective structures that game theory can be put to use. This usefulness does not, however, indicate that, in explaining class struggle, game theory is superior to such alternatives as those that focus on structural constraints.With this I wholly agree. I am concerned to refute the view that constraints are everything, not to argue that they are nothing. The "relative autonomy of the state," for instance, can be understood as the scope for state action that is left after the operation of such constraints as a minimal capitalist profit and, perhaps, a minimal amount of social welfare.
According to Cohen, " Marxism is fundamentally concerned not with behavior, but with the forces and relations constraining it and directing it." It is concerned not with the mechanics of the class struggle, but with its long-term outcome; or, in Roemer's phrase, not with the disequilibrium processes that loom so large in history, but with the equilibrium states. Whatever the exegetical and substantive correctness of this view, it clearly is not structuralist in the general sense defined above. It does not deny that the behavior of economic agents has motivated choice as its proximate cause, but it adds that the motives themselves are to be explained by their tendency to favor the development of the productive forces. Thus the quoted statement may be simply rephrased, more generally, as saying that explanation in terms of motives and preferences is never rock-bottom. As I hope I have made clear by now, I also subscribe to this view. Yet the task of providing an endogenous explanation of preferences (and beliefs) is so difficult that for many purposes we must simply take them as given and proceed on that basis to explain behavior.
[end of page 116, start of page 117]
Of the discussants only Giddens enters into the details of the game-theoretic part of my article. Though disagreeing about its centrality to Marxism, Roemer and Cohen concur that game theory can be a helpful aid to understanding class struggle. Giddens seems less enthusiastic. He objects both to the abstract framework of game theory and to the triviality of the insights to be gained from it. I agree that the assumptions of the theory tend to be excessively abstract, but I also think he puts his finger on the least important problem in citing the rationality assumption as the main culprit. I myself think the real difficulty lies elsewhere, viz., in the implausible strong information requirements needed to generate behavior in accordance with the solution to the game. The theory of games with incomplete information is largely in its infancy, but may lead to a more realistic picture when fully developed.
Giddens also argues that some of the conclusions derived from game-theoretic analysis are reached in a needlessly roundabout way. He cites two examples. The first is the game-theoretic analysis of the "second-best" problem, leading among other things to the conclusion that unilateral activism may be harmful rather than helpful to the class struggle. I agree with Giddens that this is a well-known idea. Yet the value of game theory, in my view, is well displayed by this example, because it (1) enables us to distinguish in a precise way this form of conditional solidarity from both unconditional altruism and the Prisoner's Dilemma and (2) allows us to perceive the structural identity of this predicament and others that might at first glance appear unrelated, such as the dilemma of unilateral disarmament. Game theory in this case is valuable, even invaluable, as a tool for conceptual analysis, even if it does not generate counterintuitive conclusions.
In the other example cited by Giddens, however, I believe that game theory also provides new insights. This is the Lancaster model of capitalism as a differential game between workers and capitalists. Giddens misstates my reasons for appreciating this model. It is useful not just because it embodies the idea that workers are exploited through being excluded from investment decisions, but also because it allows us to derive surprising conclusions from that well-known premise. It is indeed a puzzle why workers in countries where they have the political power and strong trade unions should nevertheless allow themselves to be exploited by the capitalists. The Lancaster model shows that this is a consequence of the distribution of economic power under capitalism. Crudely put, it is not in the interest of the workers to kill the goose that lays the golden eggs of economic growth. The model says more than this, however. It points to the importance for the workers of not letting the goose
[end of page 117, start of page 118]
retain all the eggs for itself. And, finally, it points to a symmetrical dilemma for the capitalist class. Although the details of the Lancaster solution may not be robust, this analysis of capitalism as interlocking dilemmas allows us to understand at least in a qualitative way the balance of class struggle. I do agree, of course, that these qualifications - that game theory only allows for a "conceptual" and "qualitative" understanding - point to a limitation. I admit as much toward the end of my article, and I do not mind repeating it now. I also agree with Giddens that the proof of the pudding is in the eating. If he is bored with some of my examples, so be it. Could I make a plea, however, not to dismiss social theories because they only tell us in fancy language what we already know? Could it not be the case, rather, that they give us grounds for believing what we believe?
Cohen and Giddens both object to my account of exploitation, on the grounds that it is un-Marxist to refer to Dahrendorf and to substitute power for wealth as the crux of the class struggle. I agree that this does constitute a departure from Marx, but a less important one than, say, the theory of social stratification. Marx defined classes through a relation of economic interaction, viz., the exploitation of one class by another. I submit that it is a much smaller departure from classical Marxism if we use political interaction or domination, than if we use economic comparisons, to distinguish classes from one another. 18 In any case, I fail to see why subscribing to Dahrendorf's views here also commits me to his denial that a "radical socialist transformation of 'post-capitalist' societies is feasible for the future," as Giddens suggests.
Cohen makes several powerful objections to my view, most of which I accept. First, he argues that "Elster misidentifies the illusion that survives after the marginalist one has been dissolved." On the basis of his further comments, I would now reformulate my view as follows. The workers are well aware that the means of production are the product of past labor, and yet accept the present capitalist possession as legitimate, because the earlier generation of workers produced them with the help of means of production legitimately possessed by the earlier generation of capitalists. On this view alienation and exploitation reinforce each other in a steady-state process that has been well described as follows:
Attention is focused not on past labor but on the present value of the embodiment of past labor, and its current productiveness can be taken to provide a justification for the attribu-Second, Cohen objects to calling the lack of power over investment decisions exploitation. I wonder, however, whether he is consistent here. He first argues that, generally speaking, exploitation means taking unfair advantage of someone - a characterization I fully accept. He then goes on to say, however, that the "exploitation of the worker lies in the appropriation [of surplus value], not in the subsequent disposal over what has been appropriated." But how can the capitalist take unfair advantage of the worker if he does not use the surplus for his own benefit - i.e., at least partly, for his own consumption? This may be largely a scholastic issue, however. On two substantive points I agree with Cohen. Lack of power over investment decisions, although unjust, is not exploitation. Also, even if exploitation as I now understand it - i.e., as distributive economic injustice created by voluntary exchange20 - may be small in aggregate terms, this does not mean it is unimportant. As Cohen rightly stresses, even a small minority of rich non- workers may deprive the workers of their sense of dignity. The unequal distribution of economic power is at the heart of capitalism. It is a bad thing in itself, whether or not it leads to exploitation of the workers, i.e., whether or not the whole surplus is ploughed back into working-class consumption. Also, the amount of injustice created by the surplus retained for capitalist consumption is not proportional to the size of that consumption in aggregate terms.
The comments can be ordered on a linear scale according to the amount of disagreement with my position. Roemer has virtually no quarrels with my position, which is why I have virtually nothing to say on his comment. Van Parijs is also fairly close to my view, even if we disagree about the nature and importance of the non-intentional mechanisms that can sustain functional explanations. With Cohen the disagreement goes deeper, because he argues that we can dispense with knowledge of the mechanism altogether, though I accept his criticism of my analysis of exploitation. Next on the scale, still further come Berger and Offe. We share an interest in the same problems, such as the nature of the capitalist state and the problem of collective action, but we use quite different conceptual tools to handle them. I feel very far from Giddens's position, to the extent that it is at all intelligible to me. Social theory in his hand becomes extremely abstract, without acquiring the precision for which one is sometimes prepared to pay a high price in terms of level of abstraction. Though, at the most general level, I symphatize with his
[end of page 119, start of page 120]
objections to the dualism of choice versus structure, my agreement gives way to puzzlement when I try to understand how his views could make a difference for the working social scientist.
1. See F. Machlup, "Schumpeter's Economic Methodology," in Schumpeter: Social Scientist, ed. S. E. Harris, (Harvard Universily Press, 1951), 100. Schumpeter also made the important distinction between methodological and political individualism, failure to respect which has led many left-wing writers to embrace some variety of methodological collectivism.
2. J. Schumpeter, Business Cycles (McGraw-Hill, 1939), 100. For details, see my Explaining Technical Change (Cambridge University Press, 1982), ch. 5.
3. See for instance E. R. Weintraub, Microfoundations: The Compatibility of Microeconomics and Macroeconomics (Cambridge University Press, 1979), and J. Roemer, Analytical Foundations of Marxian Economic Theory (Cambridge University Press, 1981).
4. This is the kind of model that was at the center of the "capital controversy" some years ago. See for instance G. C. Harcourt, Cambridge Controversies in the Theory of Capital (Cambridge University Press, 1973).
5. See for instance Roemer, Analytical Foundations.
6. I discuss the explanatory structure of this and similar examples in Explaining Technical Change, ch. 1.
7. Anthony Giddens, Central Problems in Social Theory (Macmillan, 1979).
8. Elster, Explaining Technical Change, ch. 6.
9. See Cohen, Karl Marx's Theory of History: A Defence (Oxford University Press, 1978), 274 ff.
10. Elstor, Explaining Technical Change, ch, 2.
11. This statement holds for the analysis of changing systems of interrelated variables. Markov-chains have been used with some success in the study of local change, e.g., in the analysis of social mobility.
12. R. Nelson and S. Winter, "Factor Price Changes and Factor Substitution in an Evolutionary Model," Bell Journal of Economics 6 (1975), 472.
13. P. Van Parijs, Evolutionary Explanation in the Social Sciences (Rowman and Littlefield, 1981), Section 52.
14. R. Dawkins, The Extended Phenotype (W. H. Freeman, 1982).
15. See H. Simon, "Spurious Correlations: A Causal Model," in Causal Models in the Social Sciences, ed. H. Blalock (Macmillan, 1971).
16. See my Ulysses and the Sirens (Cambridge University Press, 1979).
17. See my Sour Crapes, forthcoming in 1983 from Cambridge University Press.
18. See my Logic and Society (Wiley, 1978), 20 ff.
19. M. Nuti, "Capitalism, Socialism and Steady Growth," Economic Journal 80 (1970), 56.
20. See my "Exploitation and the Theory of Justice," Nomos (forthcoming).
[end of page 120]
[Elster, Jon (1983), Reply to comments (on Marxism, functionalism and game theory), Theory and Society 12:111-120]