The case against rational choice

Green and Shapiro’s Pathologies had a tremendous impact because it claimed to be judging RC theory by the standards its practitioners had set for themselves. Traditionally, one way in which RC theorists have been able to side-step arguments about inaccurate assumptions is by invoking Friedman’s instrumentalist argument that theories should be judged in terms of their predictive success rather than the realism of their assumptions.Footnote 22 Green and Shapiro seemed to demonstrate that this predictive record was nevertheless quite poor. Their argument can be summarized as follows: (i) The value of any theory ought to be measured in terms of its empirical achievements; (ii) in the case of RC theory, these achievements are extremely limited; (iii) applications of RC theory exhibit a number of methodological pathologies, including the search for confirming rather than falsifying evidence, post hoc theory development and arbitrary domain restriction; and (iv) the prevalence of these pathologies can be accounted for in terms of the theory’s method-driven rather than puzzle-driven approach.

One of the initial lines of defence here was attack. In a series of essays, subsequently published as The Rational Choice Controversy, a number of practitioners argued that Green and Shapiro had fallen foul of their own methodological standards and unfairly searched for areas of work, such as the paradox of voting, where RC theory was already known to be empirically weak, and ignored areas of work, like legislative politics and agenda-setting, where it was strong.Footnote 23 Shepsle also argued that, whatever its limitations, there was no credible rival to RC – his first law of wing-walking being that you should not let go until you have something else to hold on to.Footnote 24 To which Green and Shapiro rather bitingly noted that this law ‘might be easier to take seriously if one could develop a degree of confidence that the aircraft in question were in fact airborne’.Footnote 25

If the initial tone of the debate was quite hostile, it may nevertheless be that, in the longer term, Pathologies encouraged RC theorists to invest more time in empirical testing. Sanchez-Cuenca argues that the ‘last ten years or so’ have ‘witnessed an impressive integration of formal modelling and empirical testing’: with Boix’s Democracy and Redistribution, Epstein and O’Halloran’s Delegating Powers, Huber and Shipan’s Deliberate Discretion and Tsebelis’s Veto Players being singled out for praise.Footnote 26 In a recent paper with Hampsher-Monk, I have calculated that, in accordance with Sanchez-Cuenca’s contention, the proportion of articles published within the American Political Science Review and the American Journal of Political Science employing RC theory and including some form of empirical testing has increased. Between 1984 and 1993, only 52 per cent of the RC articles in these journals included empirical tests. Between 1994, when Pathologies was published, and 2006 this proportion increased to 70 per cent.Footnote 27 Critics might still argue that too much RC theory remains untested, or that, when tested, it only generates ‘banal’ results.Footnote 28 But such reservations should not be allowed to detract from the extent of the changes which have occurred.

The renewed emphasis upon empirical testing may be welcome but it is unlikely to lead to any mass conversions to the RC faith. The key issue for critics has always been about the assumptions RC makes. The argument here is straightforward. RC experimental evidence has conclusively shown that people have incomplete and sometimes intransitive preferences; that they are vulnerable to framing, anchoring and priming effects; that they miscalculate risk; that they excessively discount the future; and that they frequently display a willingness to co-operate with others that is difficult to reconcile with the assumption of self-interest.Footnote 29 In short, the criticism here is that far from being perfectly rational and zealously self-interested, people are boundedly rational and have bounded will-power and self-interest.Footnote 30

I do not have the space available here to offer a thorough review of this work. But, as a token example, consider research on the ultimatum game. Here, two players must decide how to divide a sum of money which has been given to them. The first makes a proposal about how to divide the money and the second can either accept or reject this offer. If the second player rejects it, then neither player receives anything. If the second player accepts the proposed division, the money is split accordingly. The game is only played once and is usually played anonymously. RC theory predicts that the first player will propose giving themselves nearly all of the money, leaving the second player with a choice between getting almost nothing and getting nothing. But when the ultimatum game was first run with cash-motivated subjects, Guth observed that the modal proposal was to split the money equally.Footnote 31 This result has since been replicated hundreds of times, with one review article reporting that the average offer made to the second person is between 30 and 40 per cent: a finding a number of critics refer to in their commentaries upon RC.Footnote 32

The status of this experimental evidence is contested. Economists have argued that when choices are relatively simple, the stakes are relatively high and situations are repeated, and participants in experiments routinely act in ways that are consistent with the standard assumptions of RC theory.Footnote 33 Others have expressed more general concerns about drawing strong conclusions from evidence that is often limited. Santos is typical in suggesting that experiments which can tell us something about how people behave in one carefully circumscribed and artificially created situation cannot necessarily tell us a great deal about how people generally reason and make decisions.Footnote 34

Yet the accumulated experimental evidence about people’s behaviour is difficult to dismiss out of hand or to regard as being caused by individual idiosyncrasies which will cancel each other out when the behaviour of enough people is examined. Furthermore, the rise and rise of historical institutionalism provides political scientists with a methodological alternative to RC which makes very different assumptions about agency. Historical institutionalism tends to a view that agents are embedded within, and their agency constrained by, institutions; that agents’ preferences are formed endogenously within institutions (rather than being exogenously fixed); and that embedded agents lack the cognitive resources always to identify the utility-maximizing strategy. This is why, within the new institutionalism, clear distinctions have traditionally been drawn between RC and historical institutionalism.Footnote 35

The story here is, however, not a straightforward one. In recent years, there is some evidence of what Thelen calls ‘border crossing’ and a narrowing of the differences between these approaches.Footnote 36 Three examples will suffice. Scharpf accepts that institutions can shape norms and behaviour and that, as a result, ‘we will not gain much mileage from assumptions of self-interested action per se’.Footnote 37 His actor-centred institutionalism nevertheless seeks to show how game theory can be used to explain the results of strategic interaction and so reconcile approaches ‘conventionally treated as being mutually exclusive’.Footnote 38 In a study of the development of vocational training regimes in Germany, Britain, the United States and Japan, Thelen argues that the question ‘Why do institutions take the form they do?’ is best answered using historical rather than RC institutionalism.Footnote 39 Yet from this starting point, she nevertheless shows how the evolution of training regimes can be understood as the product of, amongst other things, self-interested bargaining between unions, employers and the state. Her general argument, that institutional reproduction and institutional change have to be closely studied together and are in some important ways quite closely linked, is presented as a challenge to existing RC and historical institutionalist literature. Finally, Rothstein argues that the problem of explaining how some countries escape from ‘social traps’ in which efficient co-operation for common purposes can come about only if people trust that most other people will also choose to co-operate ‘cannot be handled solely by either the rationalist or the culturalist approach’.Footnote 40 Within his overall argument, an important role is played by the notion of a ‘collective memory’, which can be manipulated by strategic actors in order to enhance levels of overall levels of trust but ‘cannot be reduced to simple utility maximization’.Footnote 41

If the boundaries between historical institutionalism and RC theory have become blurred, those between RC and a more recently minted constructivist institutionalism remain frozen. Constructivist institutionalism, as it has been expounded by, principally, Hay, Blyth and Schmidt, holds that both RC and historical institutionalism are ‘overly structuralist’ and do not grant ‘purposeful actors a proper role’.Footnote 42 Indeed, it is the belief that historical institutionalism has ceded too much ground to RC, which appears to have provided a part of the motive for the development of a constructivist alternative.Footnote 43 Central to this position is a conviction that it is the ‘ideas that actors hold about the context in which they find themselves rather than the context itself which informs the way in which actors behave’.Footnote 44 This focus upon ideas distinguishes constructivist institutionalism not only from historical institutionalism – where institutions are given explanatory primacy – but also from RC – where interests derived from an actor’s structural location are the prime mover.

Developments within economics as well as political science have conspired to emphasize the limitations of RC assumptions and the availability of alternative approaches. We might start here by noting that there is a measure of nonchalance in Mueller’s earlier definition of rational choice as the application of the methods of economics to politics, in so far as he is actually equating economics with the methods of neo-classical economics. Such has been the grip of the neo-classical approach upon economics that this difference might once have seemed slight. This is, however, no longer the case. Davis argues that ‘there is now considerable evidence that economics is undergoing a significant change in which a collection of new research programs all at odds in important respects with standard neoclassical economics is increasingly dominating the economics research frontier’.Footnote 45

As examples of this new research, Davis cites game theory, evolutionary economics, behavioural economics, experimental economics and neuro-economics. With the notable exception of game theory, which has formed a staple of RC analysis, these research programmes, all originating in disciplines outside economics, constitute a challenge not only to neo-classical economics but to its political science offspring.

This is perhaps most obviously the case with behaviourism. Within economics, behaviourism has acquired a considerable cachet.Footnote 46 Within political science, behaviouralist arguments about bounded rationality, bounded willpower and bounded self-interest once provided the foundations for research on voting and party identification and incremental decision making within organizations.Footnote 47 Once sidelined by RC, evidence for a resurgence in behaviourist-inspired research can be seen in Baumgartner and Jones’s work on agenda setting and punctuated equilibrium, Sunstein and Thaler’s work on ‘nudging’ and the architecture of choice, and Jonathan Bendor’s work on turnout, party competition and bureaucratic decision making.Footnote 48 Behaviourism, like historical institutionalism, now provides a credible rival to RC theory.

Experimental economics does not constitute a rival approach to RC in the same way as behaviourism. Indeed, experiments can be used to test RC predictions. But experiments of the sort devised by Donald Green in the United States and by Peter John in Britain need not make any particular assumptions about how people reason and so provide an inductive alternative to a deductive RC method.Footnote 49 Neuro-economics poses a different kind of challenge by bringing into question something which has previously been seen as a key source of RC theory’s advantage: that ‘rational action is its own explanation’.Footnote 50 Neuro-economics is a reductionist programme which promises to explain rational action by identifying the neural mechanisms by virtue of which people assess rewards, calculate future consequences, compare options, negotiate deals, perceive threats and modify their preferences.Footnote 51 Neuro-economics has not, as yet, generated significant applications within political science. Still, it is not hard to see how existing work within economics might be easily extended to political decision making.

Defending rational choice assumptions

In the introduction to Pathologies, Green and Shapiro state that they do not object to the use of RC theory per se; only to the attempt to use it to explain any and every political phenomenon. Confronted with experimental evidence about how people behave, one alternative argument, therefore, available to RC theorists is to maintain that explanations which assume decision-theoretic rationality and self-interest will prove effective when people are rational and self-interested, even if there are many cases in which the assumptions will not prove valid and in which RC explanations consequently will be inappropriate. A ‘partial universalism’ of this sort carries the qualified blessing of Green and Shapiro. They suggest that RC models are likely to prove effective in those situations where: (i) the stakes are high and actors are self-conscious optimizers; (ii) preferences are well-ordered and relatively fixed; (iii) actors are presented with a clear range of options and little opportunity for strategic opportunity; (iv) strategic uncertainty is minimal; and (v) actors have the opportunity to learn from feedback.Footnote 52 Hay complains that this kind of partial universalism implies a ‘profound ontological unevenness and inconsistency – some parts of the social universe are simple, some complex’.Footnote 53 But by Hay’s usual standards, this is fairly unconvincing. Why cannot certain parts of the world be simpler than others?

Partial universalism concedes considerable ground to RC’s critics in so far as it would seem to put a large part of politics off-limits. If, for example, Brennan is right about the logic of choice and the prevalence of expressive rationality in the electoral arena, then it may be that RC theory cannot tell us anything useful about voting.Footnote 54 George Orwell once remarked, the quickest way of ending a war is to lose it and partial universalism looks like a partial surrender. But, in so far as even the most intense of critics is likely to concede that political actors are often self-interested and capable of making informed choices – if only because in some situations there is a norm encouraging such behaviour, it also creates the grounds for an academic division of labour and the possibility of a potentially peaceful coexistence.

There is, however, another option available to practitioners. Confronted with evidence that people are not really infallible, self-interested decision-makers, RC theorists can argue that their position has been unfairly caricatured and that they had never said that they were. In her 1997 Presidential Address to the American Political Science Association, Elinor Ostrom reviewed the evidence relating to people’s decision-making capacities and motives. She concluded not that RC theory ought to be abandoned but instead that it ought to be revised so as to allow for the possibility, within its basic assumptions, of error, learning and the existence of ingrained reciprocity norms.Footnote 55 Over the last decade or so, a growing number of practitioners seem to have heeded Ostrom’s advice. As a consequence, Quackenbush, Lovett and Ferejohn specifically deny that RC theorists are wedded to the assumption of self-interest.Footnote 56 On such a reading, RC is perhaps best understood as a ‘research tradition’ rather than a ‘research programme’: one constituted by a shared set of epistemic and methodological norms of investigation rather than a ‘hard core’ of assumptions.Footnote 57

Although she might well baulk at the label, Ostrom’s own work on Governing the Commons might serve as an exemplar of such a revised and improved RC approach, combining, as it does, an account of how calculative but boundedly rational and self-interested actors struggle with and sometimes manage to overcome collective action problems.Footnote 58 Citations might also be awarded here to Dennis Chong’s Collective Action and the Civil Rights Movement and Rational Lives: Norms and Values in Politics and Society, Brennan and Hamlin’s Democratic Devices and Desires, or Brennan’s other work on economizing on virtue or esteem, Regina Smyth’s Candidate Strategies and Electoral Competition and Jonas Tallberg’s Leadership and Negotiation in the European Union.Footnote 59 Finally, and with perhaps the greatest accompanying fanfare, Analytical Narratives promised to develop a ‘middle ground between ideographic and nomothetic reasoning’ characterized by the combination of ‘thin’ game-theoretic forms of reasoning with ‘thick’ descriptions ‘capturing the influence of history, the importance of uncertainty, and the capacity of people to manipulate and strategize, as well as the limits of their ability to do so’.Footnote 60

In the course of a case study of central banking, Bell recognizes the existence of what he calls RC2 models, which update and more richly specify the context and behaviour of rational actors and allow for the possibility of endogenous formation of preferences.Footnote 61 And he accepts that these models represent a significant improvement upon what he calls RC1 models, in which preferences are assumed to be fixed, motives selfish and reasoning flawless. But Bell is far from satisfied. He calls upon RC theorists to abandon their method and ‘move formally in the direction implied’ in their recent work by embracing historical institutionalism.Footnote 62 This is not a proposal for more ‘border crossing’. It is a call for RC to abandon its sovereign status. RC theorists are likely to remain unimpressed. One argument available to them here is to suggest that there is a trade-off between explanatory breadth and depth and that RC can be used to generate more widely applicable explanations of types of events, which can then be modified to fit the details of specific token cases.Footnote 63 But putting aside the details of this dispute, there is surely a more general point to be made. In terms of the assumptions they make about individuals’ decision-making capacities and motives, there is, as Bell recognizes, a great deal of difference between RC1 and RC2.

What is rational choice explaining?

In this part of the article, I want to shift the focus away from what it is that RC assumes to what it is that it explains. I have so far followed the ‘received interpretation’ that RC is a form of intentional explanation which accounts for people’s actions in terms of their mental states.Footnote 64 Von Wright drew a distinction that is still useful here between:

Von Wright goes on to suggest that, in the social sciences, explanations are frequently ‘quasi-causal’.Footnote 66 The explanation of a particular event entails identifying its ‘contributory causes’ with these causes being connected to the event not via a covering law but by a set of singular statements which constitute the premises of practical inferences.Footnote 67 Before this quasi-causal explanation can take place, however, it is first necessary to interpret an action which will mean attributing intentions to agents. RC explanations remain distinctive and controversial in so far as they assume not only that agents’ actions can be explained in terms of their beliefs and desires but that their actions were the best possible actions they could have taken given their beliefs and that their beliefs were the best possible beliefs given the information available.Footnote 68 Critics will, of course, baulk at the optimality constraint here, but the point is that this conventional account of what it is that RC theorists are doing is one that other political scientists will at least recognize. There are, however, alternative accounts available of what it is that RC explains: alternatives that, in turn, can be used as a defence against the charge that RC employs inadequate assumptions.

One particularly hard-line response is that RC theory does not attempt to provide psychologically realistic explanations of why people act in certain ways and that criticisms about the accuracy of its assumptions are, therefore, misplaced. RC theory operates with a revealed preference view of utility which absolves it from the need to make any claims about mental states. Choices reveal preferences, and rather than thinking about people trying (and often failing) to maximize their utility, the utility function is simply a formal representation of those choices. Binmore suggests:

Nevertheless, the revealed preference view cannot escape making some assumptions about agents, their preferences and their rationality. In order to use information about how people have behaved in one situation to predict how they will behave in another, it must be assumed that agents’ preferences are continuous, complete, reflexive and transitive. This, it might be argued, amounts to a substantive set of claims about what goes on in people’s brains, which have been empirically tested and frequently found wanting.Footnote 70

Phillip Pettit offers another account of what it is that RC explains. He accepts that most of the time most people are, in non-market settings, content to let their actions be dictated by habits or cultural norms rather than by calculations of self-interest.Footnote 71 People are not, by and large, ‘rational fools’.Footnote 72 They do favours for friends, vote when called upon to vote and give honest directions to strangers. But, for most people, there are natural limits to such behaviour. If, in a particular situation, acting out of habit or norm is likely to significantly threaten self-interest, then ‘alarm bells will ring’ and people will come to consider their own interests more carefully and, as a result, act in a more self-interested manner.Footnote 73 On this account, self-interest normally plays little or no actual role in motivating action. But it is nevertheless ‘virtually present in deliberation, for there are alarms which are ready to ring at any point where the agent’s interests get to be possibly compromised and those alarms will call up self-regard and give it a more or less controlling deliberative presence’.Footnote 74

Once an alarm bell has rung and an agent has been motivated to act in a self-interested manner, there is obviously no difficulty in seeing how self-interest might enter into the explanation of their actions and why, therefore, RC theory might prove valuable. But how can the virtual presence of self-interest explain anything in those cases where self-interest did not actually lead an agent to behave in the way they did? Pettit argues that the virtual presence of self-interest may explain the resilience of a particular course of action. Imagine a person who acts in a particular way simply because there is an established cultural norm encouraging them to do so but that they would continue to act in this way even if an alarm bell was rung and they started to act purely on the basis of their self-interest. The fact that they would have continued to behave in the same way, even if something that did not happen had happened, cannot help us to explain their actions; yet it can help us to explain the resilience of that action. The fact that RC theory does not present us with an accurate account of how people actually reason does not mean that it is unable to explain the resilience of particular patterns of behaviour.

A third argument here is that RC should be judged not in terms of the realism of its assumptions or the extent of its empirical successes but as a heuristic ‘aid to thought’.Footnote 75 One version of this argument can be traced back at least as far as Popper’s defence of the ‘zero method’. Popper argued that it made pragmatic sense to assume perfect rationality and self-interest in the social sciences in order to generate benchmark predictions which could be adjusted in the light of empirical evidence.Footnote 76 On such a reading, RC theory serves as a cognitive short-cut: a way of coping with complexity whilst generating falsifiable predictions.Footnote 77 In reincarnating Popper’s argument, Schotter suggests that RC is ‘strong but wrong’, and that there is an unavoidable trade-off between a theory’s capacity to make predictions and the accuracy of those predictions.Footnote 78 The source of RC theory’s comparative advantage clearly lies in its capacity to generate clear-cut predictions – that people will defect in a one-shot prisoner’s dilemma; that people will propose unequal divisions in a dictator game; and so on. Experimental evidence shows that these predictions are frequently inaccurate. But these predictions, although false, may nevertheless be ‘very instructive’ in so far as they can be used to inform subsequent research.Footnote 79 RC ‘does not provide us with the truth about human behaviour but rather with a very compelling platform with which we can seek the truth’.Footnote 80

My reading of Schotter’s argument is that RC theory serves as the methodological equivalent of a set of learner plates: a necessary but unrewarding stage, through which most people pass. Hindricks offers a slightly more positive assessment in the course of an essay defending the use of unrealistic assumptions within economics.Footnote 81 He argues that models ought to be judged not in terms of the realism of their assumptions but their ‘explanatory scope’. RC models resting on implausible assumptions are valuable if they can be used to deduce original and interesting conclusions, which subsequently encourage other researchers to investigate whether the same conclusions hold when these underlying assumptions are modified and presumably made more realistic. Consider Downs’s popularization of the median voter theorem.Footnote 82 If there are two parties, political space is one-dimensional, parties can move to and occupy any point in this space, parties are vote-maximizers, voting is positional rather than directional, voters’ preferences are fixed and there is perfect information, then Downs demonstrates that parties will converge upon the position of the median voter. These assumptions might be dismissed as being implausible. Yet the ‘explanatory scope’ of Downs’s work has been such that generations of theorists have used his basic model to delineate the circumstances in which party competition takes a centripetal form.Footnote 83

A slightly different way of phrasing the same argument is to say that RC theory performs a valuable service for political science by exposing the limitations of existing research and identifying new empirical puzzles. Consider Olson’s Logic of Collective Action and its prediction that unless the number of individuals in a group is quite small, or unless there is coercion or some other device to make individuals act in their common interest, rational self-interested individuals will not act to achieve their common or group interests.Footnote 84 As a prediction, this is wrong. Collective action problems are, after all, often overcome. But it might be argued that the real value of Olson’s work is to be found in the way in which it exposed the limitations of pluralist theory and encouraged political scientists of all persuasions to consider the circumstances in which people join groups. Similarly, whilst the paradox of voting has proved to be an empirical disaster zone for RC theory, it has stimulated research into the decision to vote. In this vein, Chong concludes that ‘like any good theory, rational choice theory raises interesting issues for study and charts a program and methodology of research’.Footnote 85

Finally, and whatever the obvious inadequacies of its assumptions, a final argument here is that RC theory is heuristically useful because it can tell us something about how the world would be if in the future people were perfectly rational and entirely self-interested. Just as engineers stress-test buildings and regulators (hopefully) stress-test banks, institutional designers can use the assumptions of RC theory to see whether institutions are capable of withstanding ‘knavish’ behaviour or can ‘economise on virtue’.Footnote 86 Alternatively, Hay suggests that RC theory can serve as a useful fable, warning people of the dangers of a world in which people are ‘rational fools’ and life is nasty and brutish.Footnote 87

The claim that RC theory stands as a ‘useful aid to thought’ and, more specifically, the claim that it serves the broader interests of political science by generating baseline predictions, setting research agendas and facilitating counterfactual analysis is not uncontroversial. Sceptics might argue that the predictions RC makes are so inaccurate as to be of no value; that designing institutions on the assumption that people are knaves simply encourages knavish behaviour; and that, with the notable exception of Tsebelis’s work on Veto Players, which has generated an active research programme within comparative politics, RC theory has become increasingly method-driven and uninteresting.Footnote 88 Yet the heuristic defence of RC theory nevertheless serves to narrow the gap with the rest of political science. There is a great difference between arguing that RC will be a field within economics and will encompass all of political science and suggesting that RC is the handmaiden of progress within political science.

One final argument in this part of the paper takes us back to Pathologies of Rational Choice. Green and Shapiro, it will be recalled, equate theoretical value with empirical potency: ‘A theory of politics has no payoff if its hypotheses do not survive empirical scrutiny’.Footnote 89 In doing so, they embraced an epistemological standard shared by many – but not all – RC theorists. Cox, for example, suggests that RC theory can be used to provide ‘more satisfying’ explanations, even when it does not add to our existing empirical knowledge.Footnote 90 How so? One argument here, which Cox appears to edge towards, is that explanation can involve showing how something fits into a more general pattern. Explanation, that is, can come from unification. Just as Newton showed how tides, projectiles and planetary motion could be explained using the same set of underlying principles, RC theory shows how it is possible to account for a diverse range of phenomena using a small number of core assumptions.Footnote 91 On such a reading, the most valuable works in the RC corpus are not those which exhibit an ‘impressive integration of formal modelling and empirical testing’ but those like, for example, Russell Hardin’s explanation of ethnic conflict in One for All, which apply the RC method to new areas of work.Footnote 92

The unificationist argument, or, more precisely, the argument that explanation through unification can be achieved independently of causal explanation, remains controversial. In Kinkaid’s discussion, he draws a distinction between a unification which consists in the provision of a common classificatory scheme or descriptive vocabulary, where no satisfactory scheme previously existed, and physical unification, where phenomena previously regarded as having quite different causes or explanations are shown to be the result of a common set of mechanisms or causal relationships.Footnote 93 The distinction is relevant because critics are likely to argue that whilst RC theory provides a common descriptive vocabulary and set of methods, it falls far short of physical unification, because – precisely as Pathologies showed – it has been unable to use a small number of assumptions to reveal common causal relationships. Furthermore, a critic might argue that, in the absence of this physical foundation, unification simply licenses a method-driven Flight from Reality in which, rather than explaining some event or process, effort is exclusively devoted to showing how that event or process might be reconciled with the use of the RC framework.Footnote 94

The method of isolation

I have already suggested that the ‘turn’ towards mechanism-centred explanation creates a point of common ground between RC and the rest of political science. Mechanisms cause things to happen in particular ways. But particular mechanisms only cause things to happen in predictable ways when left to operate in isolation. In the social sciences, prediction is extremely difficult because most outcomes are the results of the operation of a number of mechanisms. So how can we know what difference particular mechanisms make? One answer to this question involves the application of the ‘method of isolation’. Models constructed using the method of isolation take account of just some of the factors which are at work in the real world: the potential influence of other factors being ‘sealed-off’.Footnote 95 The question I address in the final section is this: can the use of the method of isolation justify the use of unrealistic assumptions within RC? If it does, then RC theorists can acknowledge that the ‘hard core’ assumptions they employ are inaccurate but nevertheless claim that the RC models which employ them can explain events and outcomes.

Within the Journal of Economic Methodology, the method of isolation has been argued to justify the use of unrealistic assumptions.Footnote 96 Consider Schelling’s famous ‘chequerboard’ model of racial discrimination, which was lauded as a classic example of mechanistic explanation.Footnote 97 Here, two sets of counters (dimes and pennies in Schelling’s initial presentation) are arranged randomly on a chequerboard, each cell of which adjoins eight other cells. Schelling postulates a condition to determine whether a coin is to remain in its initial place. For example, a rule can be introduced stipulating that a coin is to be moved when more than two-thirds of the coins on the adjoining cells are of a different type. Schelling shows that, over time, and almost regardless of the initial distribution, such a rule results in an outcome in which the two sets of coins occupy different parts of the board. The point is to show how even mildly segregationist preferences can, over time, lead to a quite rigid racial division within cities.

As a model of residential moving and racial segregation, the chequerboard model makes a number of unrealistic assumptions, including costless moving and the absence of any income differences between the two sets of coins. For this reason, Schelling’s model cannot be thought to constitute a complete explanation of the dynamics of racial segregation. Yet the use of these unrealistic assumptions might be defended on the grounds that they allow Schelling to ‘seal-off’ his model from the effects of other mechanisms in order to show how, when left undisturbed, the sorting mechanism operates.Footnote 98 In this way, Schelling’s model can be said to provide a ‘credible’ account of racial discrimination even though it employs unrealistic assumptions.Footnote 99

There is, however, a significant problem here. The method of isolation can justify the use of assumptions which are unrealistic in the sense of being partial. Schelling’s model ‘isolates away a host of elements’, including income differentials in order to focus upon the operation of the sorting mechanism.Footnote 100 But it might be argued that the core assumptions of RC theory are unrealistic in the very different sense of being simply false. People are not infallible and self-interested decision makers. The problem with RC theory comes, in other words, not with what is left out of the models but what is put in.

So, at this point, it would seem that the method of isolation needs after all to be run in conjunction with an alternative account of what it is that RC is seeking to explain: taking us back to the arguments of the previous section. One such alternative here can be found in Ferejohn and Satz’s argument that RC theory constitutes a form of structural explanation in which it is variations in the constraints on actors’ choice sets and not variations in their motives and decision-making capacities which carry the explanatory burden.Footnote 101 Consider the example of a RC theorist who wants to explain historic variations in fertility rates. Ferejohn argues that the theorist will want to start by assuming that agents are rational and that they seek to maximize their wealth, but these assumptions do not themselves explain anything. Their role is instead to provide a fixed backdrop against which it can be more clearly seen how parametric shifts in constraints – presumably including such things as changes in inheritance laws and the introduction of labour-saving technologies – affect behaviour.

In analysing the operation of structural, that is aggregative, mechanisms like free riding, sorting, structure-induced equilibrium, cycling and agenda setting, RC theorists need not argue that people really are perfectly rational and self-interested or that variations in decision-making capacities and motives associated with the operation of individual-level mechanisms like weakness of will, framing and anchoring are of no explanatory significance. The RC theorist can simply argue that they are practising a division of labour and ‘sealing-off’ the influence of these other factors in a manner consistent with the method of isolation.

As I have presented it here, the structural argument implies a certain degree of modesty. RC theory offers incomplete explanations of political processes and outcomes which focus upon the operation of aggregative mechanisms. I argued before that partial universalism creates the possibility of a peaceful coexistence between RC and other branches of political science. The same applies here. RC theorists who commit to the structural defence need not deny the potential value of other explanatory approaches which analyse the operation of either individual-level mechanisms, like weakness of will, or even aggregative mechanisms, like group think, which assume sub-optimal decision making. But in judging the value of the resulting explanations, a key question suggests itself. Is the operation of the aggregative mechanisms identified by RC theorists using the assumption that individuals are rational and self-interested robust to variations in decision-making capacities and motives? If the answer to this question is ‘Yes’, then these aggregative mechanisms will continue to operate when people are boundedly rational and boundedly selfish, RC explanations can be widely applied and the assumption that people are perfectly rational and self-interested justified on the purely pragmatic grounds that it makes it easier to construct models. If the answer to this question is ‘No’, then the status of the initial assumptions does matter as aggregative mechanisms will only operate when people are perfectly rational and self-interested and RC models will only have a limited explanatory reach. Indeed, at this point, the structural argument simply becomes a variant of partial universalism.

The correct answer to this question about the reach of RC structural explanations will, I suspect, vary depending upon the particular mechanisms under consideration. Consider, once again, free riding. Looking back at Olson’s original argument, he formulates the collective action problem on the specific assumption that individuals are self-interested.Footnote 102 But Chong argues that the free-riding mechanism still operates in those situations in which participants are motivated by ‘extra-rational’ considerations.

Equally, whilst Downs’s initial argument rested upon the assumption that politicians are vote-maximizers who ‘act solely in order to attain the income, prestige and power which come from being in office’, the same competitive mechanism pushing parties towards the electoral centre has been argued to operate when politicians are policy seekers or motivated purely by ideological considerations.Footnote 104 There are, however, other mechanisms whose operation would seem to be parasitic upon the assumption of self-interest. The levels of free riding within a single play of the prisoner’s dilemma, as one variation of the collective action problem, would, for example, seem to be significantly affected by actors’ underlying motives.Footnote 105 Here, and as in the ultimatum game, the evidence suggests that people consistently behave contrary to the standard expectations of RC theory.

A common complaint about RC theory is that it is unnecessarily formal and that this formalism is used to hide its dependence upon implausible assumptions. Yet, in this context, it might equally well be objected that it is the informality of RC theory which often poses as a significant problem. Too many RC models that point to the operation of aggregative mechanisms fail to clearly specify the range of circumstances in which deductions from the model remain valid.Footnote 106