Agents and Structures: Two Views of Preferences, Two Views of Institutions

Transcription

1 International Studies Quarterly (1998) 42, Agents and Structures: Two Views of Preferences, Two Views of Institutions WILLIAM ROBERTS CLARK Princeton University AND Georgia Institute of Technology Two analytically distinct approaches to the study of domestic politics have been referred to as the new institutionalism. The fundamental difference between the two brands of institutionalism can be seen in the way they handle the relationship between agents and structures. Structurebased approaches to institutions give ontological primacy to structures and view agents as being constituted by them. Agency-centered approaches view human agents as ontologically primitive and view institutions as structures that are created by goal-maximizing individuals. The two approaches are compared, with special attention given to the way they treat the preferences that actors hold. I argue that contrary to arguments made by many structure-based theorists, the agency-centered approach is capable of contributing to discussions regarding the sources of actor preferences. A limited-information model of the strategic interaction between workers and capitalists is used to demonstrate ways in which the agency-centered approach can begin to make preferences endogenous. Over the last two decades, two distinct brands of theorizing have both been dubbed the new institutionalism. While there are some characteristics shared by all of the new institutionalists, I argue that the two broad approaches to institutional analysis can be distinguished by the unit of analysis they make ontologically primitive (Wendt, 1987:337; Dessler, 1989). In one version of new institutionalism institutional structures are seen as either constraints on, or the products of, individual choices; in the other, actors are seen as constituted by social structures which are, in turn, constituted by the actions of these actors (Krasner, 1988:73; see also Krasner, 1982). I will refer to the former group as agency-centered and the latter as structure-based institutionalists. 1 Author s note: This article has had an unusually long gestation period. Consequently, it is difficult to acknowledge all who have contributed to its development, and many I do acknowledge may not recognize the final product. Nevertheless, I thank Robert Bates, Paul Bonk, Kenneth Finegold, Colin Flint, Barbara Geddes, Ken Gilmore, Mark Hallerberg, Robert Kaufman, Jack Levy, David Reeths, Michael Shafer, Steve Vallas, Alexander Wendt, Richard Wilson, Scott Wolfel and participants at the Georgia Tech Political Economy Brown-Bag and the Rutgers University Political Economy Seminar, and the editors and anonymous reviewers of this journal. I owe special thanks to William J. Long and Mark Yellin for many insightful discussions on topics related to this paper. 1 I am indebted to Nancy Bermeo for these terms which capture a crucial difference between the schools in a language I believe both schools would accept. My use of the terms agency-centered and structure-based are roughly equivalent to Krasner s (1988) utilitarian and institutionalist perspectives, except I would argue that Krasner is incorrect in 1998 International Studies Association. Published by Blackwell Publishers, 350 Main Street, Malden, MA 02148, USA, and 108 Cowley Road, Oxford OX4 1JF, UK.

2 246 Agents and Structures The most visible difference between agency-centered and structure-based approaches to institutions is methodological. Agency-centered institutionalists apply the deductive, often formal, modeling techniques associated with economics to political institutions. 2 In contrast, structure-based institutionalists typically apply the tools of historical-sociology and/or traditional political science to the study of state structures and their effects on policy outcomes. 3 Since these methodological differences are tendencies and not unavoidable consequences of the theoretical positions adopted by these schools, there is considerable opportunity for the creative combination of approaches. 4 Nevertheless, deeper epistemological and ontological differences will continue to thwart the productive integration of these approaches until they are brought to light. This paper is meant as a contribution to this process, which promises to be a protracted one. The perspective adopted is meant to be ecumenical, but not noncommittal it is a defense of the agency-centered approach that seeks to incorporate and learn from structure-based criticisms. My primary argument is that the gulf between structure-based and agencycentered institutionalisms has been exacerbated by a semantic misunderstanding that has led to an exaggeration of the conceptual differences that separate the way these schools handle preferences. Structure-based institutionalists have emphasized the ways in which institutions help define the interests actors hold, while agencycentered approaches have focused on the ways in which actors behave given an exogenously determined set of goals. While there has been a tendency in the agency-centered literature to treat preferences as unproblematic, assertions by structure-based institutionalists that the choice-theoretic technology typically employed by agency-centered theorists requires them to hold preferences constant and therefore removes the formation of preferences from the purview of their studies are overdrawn. I will argue in this paper that the agency-centered approach to institutions, particularly when it makes use of dynamic, limited-information game theory, is capable of shedding considerable insight on the recursive relationship between agent and structure, and that it is capable of addressing many of the concerns of structure-based institutionalists and structurationist theorists. 5 Furthermore, since structurationist approaches have been closely associated with scientific realism, I will also attempt to highlight the many tenets of scientific realism that are frequently realized in the practice of agency-centered theorists. 6 implying that the substantive positions of a narrow band of theorists who maintain that institutional change is frictionless and tends toward efficient equilibria are a natural product of the choice-theoretic approach. One of North s (1981, 1990) main goals, for example, is to account for the persistence of inefficient property rights. 2 Examples are Shepsle, 1986, 1989, Weingast and Marshall, 1988, North, 1981, 1990, North and Weingast, 1989, Geddes, 1991, Knight, 1992, Levi, 1988, and Calvert, 1995a, 1995b. 3 Representative works include Skocpol, 1985, Skocpol and Finegold, 1982, Finegold and Skocpol, 1984, Krasner, 1988, Stepan, 1978, Skowronek, 1982, and contributors to Evans, Skocpol, Rueschemeyer, et al., 1985, and Steinmo, Thelen, and Longstreth, There are many fine critical reviews of the new institutionalisms : see Levi, 1988, appendix, 1996a, Kiser and Hechter, 1991, Chong, 1994, Hall and Taylor, 1994, Pontusson, 1995, Cammack, 1992, and Geddes, Keohane (1988) identifies two distinct approaches to international institutions that are similar to the agency-centered and structure-based approaches to domestic institutions. 4 See Bates and Weingast, 1995, Laitin and Fearon, 1995, and Levi, 1996b. 5 I use the terms structure-based or structuralists to describe analysts who treat structures as ontologically primitive and structurationists to describe analysts who recommend avoiding ontological reductionism that is, giving primacy to either agents or structures. 6 The realist philosophy of science, or scientific realism, has been posed as an epistemological alternative to empiricism. Its defining features are that ontology (what exists) is independent of epistemology (what we know), that abductive inferences are valid, and, therefore, that entities can exist even if they cannot be observed. See Aronson, 1984, Harre, 1986, Bhashkar, 1979, and Aronson, Harre, and Way, Theorists taking an explicitly scientific realist approach to international politics (Wendt, 1987; Dessler, 1989) have also stressed the approach s emphasis on explanation as the determination of causal processes rather than the discovery of covering laws, or constant conjunctions. Grafstein (1992) argues for a realist approach to the study of institutions.

3 WILLIAM ROBERTS CLARK 247 I will begin with a brief discussion of the ways in which structure-based and agency-centered institutionalists typically handle preferences. I will then examine the structure-based critique of exogenous preferences and provide an illustration of the way agency-centered theorists can begin to make preferences endogenous. Agents, Structures, and Preferences in Institutional Analysis While there is considerable subtlety and variety within the approach, all structurebased institutionalists maintain that the preferences actors hold are deeply embedded in the structural environments in which they operate. Some structure-based analysts, most notably those associated with the statist approach, go so far as to assign agency to structures, while others make the weaker claim that institutions shape the preferences that actors hold. Structure-based institutionalism was born, in part, of a critique of pluralist and marxist approaches that asserted that the state is not a mere arena on which the game of politics is played, a cash-register that aggregates the preferences of societal actors. 7 Crude versions of this argument seem to be based on the assumption that if a particular outcome cannot be traced back to the preferences of at least one societal group, it must serve the interests of the state, or some faction within it. This, often implicit, functionalist assertion can be seen as problematic from several perspectives. Most fundamentally it commits the logical fallacy of affirming the consequent. 8 In addition, it reveals an approach that fails to take the strategic nature of politics seriously. 9 Political outcomes do not necessarily reflect the preferences of any one actor, or group of actors, rather they are a product of many actors pursuing their interests. It is important to take this insight one step further: political actions do not necessarily reveal preferences over outcomes either; at best they reveal subjective beliefs about what actors consider the best response to the expected behavior of other relevant actors. Finally, the public choice literature tells us that collective decision procedures (those based on successive pair-wise comparisons, for example) typically lead to outcomes which do not reflect either individual or group preferences (Arrow, 1951:2). 10 The notion that one must choose between viewing the state as a mere arena of social conflict or as an actor in its own right sets up a false dichotomy. In between these explanatory alternatives is an approach that views institutions as transforming, or shaping political actions, rather than summing them. Bates (1987:10) captures this perspective succinctly: Rarely do we find a direct and obvious relation between economic changes and political outcomes. Rather, we find the effects refracted, 7 Almond (1988) points out that the statists have misrepresented, or ignored, more sophisticated forms of pluralism, such as the work of Schattschneider (1942), Wilson (1973), and Lowi (1964, 1969). 8 This is a logically invalid argument that commits the fallacy of affirming the consequent (McGaw and Watson, 1976:34). The logical structure of the argument is as follows: State actors enact policies that benefit themselves or their constituents. Policy x did not benefit any societal constituency. The state enacted policy x. State actors benefit from policy x. This conclusion would be valid if the antecedent read state actors enact only those policies that benefit themselves or their constituencies. If, however, it is understood that the policy process can lead to outcomes that do not satisfy the antecedent above, then the conclusion is invalid, since it can be false even when all of the premises are true. 9 The strategic nature of politics is both a key insight of and the analytical domain of game theory (see Ordeshook, 1992:ch. 1). 10 It should be clear that this logic can be generalized beyond the context of voting to any situation in which actors must choose between a set of greater than two alternatives that are addressed in a pair-wise manner.

4 248 Agents and Structures as it were, through the society s institutional endowment. (Emphasis added) 11 Unlike an approach to institutions that emphasizes the way they refract social conflicts, the structure-based institutionalist attempt to assign the state independent causal weight typically leads them to anthropomorphize it. That is, they seek to demonstrate that the state may formulate and pursue goals (Skocpol, 1985:9), and in the process transform the state into an actor in its own right (March and Olsen, 1984:738 (referring to institutions); Stepan, 1978; Krasner, 1984, 1988; Haggard and Moon, 1983). There is nothing inherently wrong with attributing human agency to collectivities: the field of microeconomics has been quite successful in treating firms (collections of individuals with diverse needs, goals, and desires) as if they were unitary actors. 12 The usefulness of the unitary actor assumption does depend, however, on the goal of the study. The structure-based institutionalists use of the unitary assumption, for example, interferes with their stated goal of developing an explanation of political behavior that takes institutions seriously. When one speaks about the state as an actor with interests, the effect is to black box all that goes on within state institutions. This effect is quite clearly understood by international relations theorists who have engaged in debates that have centered on the need to maintain the unitary action assumption for explanations among states (Waltz, 1979) versus the need to develop an understanding of what occurs within the policy apparatus within states (Allison, 1971). What has emerged quite clearly from this debate is that if one maintains the assumption of a state interest, one is prohibited from speaking about the behavior of actors within the state. To explain the behavior of actors within the state, we must disaggregate the state, and look at organizations within it, the individuals who inhabit them, and the institutional framework that structures their behavior. This task cannot be completed if institutions and the state or even state actors are confounded. Some structure-based institutionalists do attempt to distinguish between actors and institutions, but they continue to argue that the behavior of actors is in the end reducible to the institutions from which they emerge. These analysts (Krasner, 1988; Smith, 1992) argue that choice-theoretic approaches address institutions only as forms of constraints on actors, and not as independent forces that constitute the actors themselves. According to this version of the structure-based argument institutions not only shape the conduct of actors, but also create them. If this is the case, individual humans can be treated as epiphenomenal institutions both create them and determine their behavior. Because agency-centered theorists tend to think of institutions as sets of rules that shape the behavior of actors (Shepsle, 1989), they are better able than structurebased analysts to clearly delineate between agent and structure. Actors are agents because they make choices in their attempt to achieve their goals. Institutions are structures because they constrain the behavior of actors. Collections of actors can more profitably be thought of as organizations than institutions (North, 1990). When actors within an organization have sufficiently homogenous preferences, as the firm under neo-classical assumptions, the organization can be treated as an actor. If, however, we refer to the collectivity as an institution, and speak of it as possessing interests, agent and structure are confounded in such a way as to impede our analysis: it becomes difficult to separate the actors from the set of rules 11 See also Gourevitch, 1989:89, which argues along the same lines. 12 And its limitations within the field of economics has been demonstrated by the new economic organization literature; for a review see Moe, 1985.

5 WILLIAM ROBERTS CLARK 249 that structure their behavior. In game-theoretic terms, institutions specify the rules of the game. They delineate the range of possible actions, and the payoffs that result from the action combinations of the players (be they individuals or organizations). While the structure-based institutionalists have attempted to establish the interests of states, or state actors, through inductive processes, the agency-centered institutionalists typically employ either deductive or abductive reasoning to establish the preferences that actors hold. The standard view of the agency-centered approach maintains that analysts deduce the choices of institutional actors from axiomatic assumptions about human behavior. 13 Most generally, this involves the rational choice assumption that humans engage in goal-oriented behavior and attempt to maximize utility; they try to do the best they can (Frohlich and Oppenheimer, 1978:11; Frieden, 1991). There is considerable disagreement over what constitutes rational action (Simon, 1955) and the extent to which individuals consider the goals of others in their actions (Keohane, 1984:ch. 7; Margolis, 1982), but the diverse versions of rational choice are all consistent with this weak formulation. While rational choice theorists all share the assumption of goal-oriented behavior, exactly what that goal is varies according to the empirical situation under study. Actors need not be driven by pecuniary interests the value they maximize is not necessarily economic. In other words, while all rational choice theorists agree that actors seek to maximize their interests, what that interest is remains the object of theorizing. Establishing the preferences of actors is therefore a key and problematic task for rational choice theorists. This task is accomplished in neo-classical microeconomics through the identification of a functional prerequisite for firms. Firms must achieve a long-run rate of return on investment if they are to continue to survive. Decision makers within the firm may have other goals, such as serving society by building a better mouse trap, improving working conditions for their employees, or enhancing their own careers, but they are unlikely to be able to advance these goals in the absence of long-run profitability. Agency-centered institutionalists often apply a similar logic to establish the preferences of political actors. Elected officials cannot pursue any goals if they are not in office; as a result, political survivability will be the over-riding concern in their goal formation (Downs, 1957; Weingast and Marshall, 1988; Ames, 1987). Similarly, although rulers of early modern states were at times faced with contradictory incentives, they sought first and foremost to maximize net revenues (North, 1981; Levi, 1988) which were a functional prerequisite for maintaining their monopoly on the use of force. Less recognized is the reliance of agency-centered theorists on abductive (or retroductive ) reasoning a form of reasoning that scientific realists argue distinguishes the practice of scientists from the prescriptions of empiricist epistemology (Hanson, 1969; Wendt, 1987). An abductive inference is an inference made about unobserved phenomena or processes on the basis of observation. 14 For example, if I awake to find puddles in my driveway and infer that it has rained during the night, I have made an abductive inference. Choicetheoretic scholars make abductive inferences when they begin their analysis with an observation and then ask, What preferences and beliefs must actors have held to produce such behavior in equilibrium? This technique is obviously problematic 13 Citing Bueno de Mesquita, 1985, and Wendt, 1987:351, n. 36, implies that rational choice theory s reliance on deductive reasoning places it in the empiricist, rather than realist, epistemological camp. I will argue to the contrary below. 14 The key distinction between scientific realists and empiricists involves the ontological status of unobservables. The former treats them as real, while the latter may permit unobservables into their analytical frameworks, but treats them in a purely instrumental fashion.

6 250 Agents and Structures from a falsificationist perspective unless new predictions are deduced from the abductively inferred preferences and beliefs and tested on observations other than those that sparked the study. 15 While the establishment of institutional actors through deductive or abductive means is problematic, it offers a key advantage over the inductive approach because it allows us to (a) posit what actors goals will be, (b) deduce strategic alternatives available to them, and (c) use this information to formulate falsifiable predictions or explanations (Snidal, 1986:40). This allows us to avoid what Levi refers to as the myth that interest equals action (Levi, 1996a:3). This brief discussion of the ways in which structure-based and agency-centered institutionalists characterize the motivations of political actors suggests that these schools have found different solutions to the agent-structure problem. While both approaches include agents and structures in their analysis, they differ markedly in which they treat as ontologically primitive. The agency-centered approach shares rational choice s commitment to methodological individualism in the sense that an explanation is not complete until the outcomes explained are traced back to the behavior of individuals (Elster, 1989:158). Thus, while agency-centered theorists do not dismiss the explanatory import of structural factors, they maintain an a priori assumption that those structures are, themselves, to be viewed as the product of human agency. Contrast this view with the structural approach, which stresses that social structures have an existence that is independent of the behavior of the individuals who inhabit them. The structural approach draws on a more sociological tradition that can be traced back to Durkheim s assertion that collective life did not arise from individual life; on the contrary, it is the latter that emerged from the former (Durkheim, 1984:220 1; see also Mead, 1934:7). Thus, agency-centered and structure-based theorists both include agents and structures in their analysis, but they differ radically in their answers to what Calvert (1995a:218; see also Udehn, 1996:172) refers to as the chicken and the egg question If institutional structures are the product of human choices, how can they be said to influence those choices? Recently, some social theorists have argued that the stalemate between agencycentered and structure-based approaches to the social sciences is based on a false dichotomy, or the fallacy of the dualism of structure (Wendt, 1987; Giddens, 1979, 1984). These analysts claim that this dualism is transcended in structuration theory by the theorem of the duality of structure, which maintains that the constitution of agents and structures are not two independently given sets of phenomena ; and that the structural properties of social systems are both medium and outcome of practices they recursively organize (Giddens, 1984:25). Thus, correct answers to the chicken and the egg problem will not arise until we understand the recursive relationship between structure and agent because [s]tructure is not external to individuals.... It is in a certain sense more internal than exterior to their activities (Giddens, 1984:25). I wish to argue that because it requires analysts to make their assumptions about the substantive entities and configurations posited by a theory explicit, game theory is a useful ontological tool that can further our understanding of the recursive relationship between agent and structure in a manner that is consistent with the central assumptions of the agency-centered approach. Thus game theory is compatible, though by no means uniquely so, with the theoretical goals of both structure-based institutionalists and structurationist social theorists. Current 15 Green and Shapiro (1994) argue choice theory has shaky empirical foundations because choice theorists seldom follow through on this. See contributions to Friedman, 1996, for counter claims. For an excellent introduction to theory construction using retroductive logic see Lave and March, 1975.

7 WILLIAM ROBERTS CLARK 251 practice 16 in game theory states that the following aspects of a situation need to be included in a model to yield explanations of political behavior: 1. a statement of who the actors are 2. a statement of the actions available to the actors 3. a statement of the order in which events take place 4. a statement about the expected consequences of the various possible action/ strategy combinations 5. a statement regarding what information is known to the actors when they act 6. a statement regarding the beliefs actors hold about the structure of the game and the beliefs and preferences of the other actors 7. a statement of the actors preferences over the outcomes produced by the various possible action/strategy combinations Factors one through five can be thought of as structural factors. They may be the product of individuals decisions, but they are not part of what determines the essence of particular actors. While factors 6 and 7 are clearly related to the intrinsic properties of individual actors, closer consideration reveals that the division between agent and structure in game theory is not nearly so clear cut. Rather than address the recursive relationship between agent and structure that game theory posits in the abstract, I will attempt to demonstrate insights that game theory might shed on this issue in the context of a comparison of the ways agency-centered and structure-based scholars tackle the problems of institutional analysis. The Structure-Based Critique of Exogenous Preferences The most powerful and consistent critique of the agency-centered approach to institutions is that its assumption of fixed and exogenous preferences makes it impossible to address the question of preference formation (Thelen and Steinmo, 1992; Steinmo, 1989; Weir, 1989). The logic of this critique typically advances in the following manner: a. Actors preferences change during the course of many interesting and important political conflicts, and/or are endogenous to particular institutional arrangements that are prone to change. b. Agency-centered theory assumes actors preferences are fixed and exogenous. c. Agency-centered theory is, therefore, ill-equipped to address many interesting and important political conflicts. There are at least three possible responses to this argument, each worthy of some attention. First, it can be argued that, since much agency-centered theory is aimed at explaining or predicting behavior based on a given set of preferences, this is not an immanent critique (Stigler and Becker, 1977). Second, it can be argued that while changes in preferences are rare, changes in strategies (easily handled by the agency-centered technology) are much more common. Finally, while difficult, it is possible to use the tools employed by agency-centered theorists to gain insights into the process of preference formation. Note that these responses are not unrelated; the second response suggests that the first response does not leave choice theorists as dependent as they may, at first glance, appear. The third defense suggests that choice theorists are not compelled to accept the first. I will address each response in turn. 16 Current practice is exemplified in textbooks aimed at teaching applied game theory. See, for example, Morrow, 1994, or Rasmussen, As Kuhn (1970) has suggested, textbooks are a revealing place to view the habits of mind of a research community.

8 252 Agents and Structures As noted, the first response is that the attack on fixed preferences is not an immanent critique. Since rational choice theory s central claim is that it provides a technology that allows the analyst to use assumptions about preferences (and information and available actions) to predict behavior, this line of argument amounts to criticizing the approach for being unable to accomplish a task it had not attempted. Thus, one could use this criticism to argue that rational choice theory does not answer important questions, but not to argue that it does not answer the questions it says it does. Many rational choice theorists are content to allow other disciplines and approaches to explore where preferences come from and to use their findings as inputs in their own studies. To some extent, it is the reticence of some agency-centered theorists to address the question of preference formation that has emboldened critics to charge that the approach is incapable of doing so. The second defense is to demonstrate that many of the changes in preferences observed by the agency-centered approach s critics are better understood as changes in strategies, which rational choice theory may be uniquely qualified to explain. If this can be done, the conclusion potentially drawn from the first response that rational choice institutionalism is true, but trivial can be avoided. I will argue that much of what passes for debate over the applicability of rational choice models in institutional analysis is driven by semantic confusion over the meaning of terms such as preferences, interests, strategies, et cetera. Attention to the ontological implications of these terms can help to clarify their meaning. When rational choice theorists use the term preferences they are referring to a dispositional trait of actors specifically, the way an actor values alternative outcomes of the decision process being modeled. Preferences (along with beliefs) help define who an actor is. Strategies (and the actions they involve), in contrast, define what an actor can do; they are the actions that, taken together, produce outcomes. Strategies are means, preferences are ends. Actors prefer ends, but they choose means. Most rational choice models hold statements about who an actor is (what the actor likes and believes) constant in order to produce statements about what that actor will do in particular strategic environments. They typically do so through the use of comparative statistics in which the preferences of actors are held constant and aspects of their strategic environment (such as institutions) are allowed to vary. The appropriate method (decision theory or game theory) is then used to see if the change in strategic environment is expected to lead to a change in behavior. Rational choice models do not typically employ the concept of interests, in part, perhaps, because the term s ontological status is often unclear. On the one hand, interests may be another word for preferences (e.g., the actor s position in the economy determined her interests ), but more often, the term seems to refer to a strategic choice (e.g., given the institutional environment in which she operated, it was in her interest to do x ). Much of the strength of the argument about the limitations imposed on the usefulness of rational choice theory by the assumption of fixed and exogenous preferences disappears when the conceptual confusion surrounding the terms preferences, interests, and strategies is clarified. A few examples may demonstrate this point. While the assertion that institutions influence the preferences actors hold is a common theme among historical institutionalists (Steinmo et al., 1992), many of the examples they marshal in support of this hypothesis are more usefully interpreted as examples of institutions influencing the strategies actors choose. Hall, for example, argues that a less fragmented union movement might have been inclined to pursue neocorporatist solutions longer than the British unions did and it might have been able to implement them more effectively (Hall, 1992:107). There is considerable evidence to support this counterfactual claim, but the assertion that in this case we can see how the institutions devised to represent the interests of a social group, such as the working class, themselves affect the definition and

9 WILLIAM ROBERTS CLARK 253 expression of those interests (Hall, 1992:107) can have controversial implications for the use of the rational choice approach to the study of institutions. If the term interests is meant to capture the actors preferences over outcomes, Hall s assertion can be used to support the argument that the practice of treating preferences as exogenous may create problems for choice-theoretic institutional analysis by creating an endogeneity problem: institutions devised to represent interests (actors preferences) themselves affect the definition of those interests (preferences). If in contrast, the term interests is meant to refer to strategic choices of the actors, Hall s conclusion becomes: institutions devised to represent interests (actors choices) themselves affect the definition of those interests (choices). Since it is not clear what it would mean to represent an actor s strategy choice there is prima facie cause to understand Hall s first use of the term interest to mean actors preferences. While it is logically possible to assert that institutions devised to represent actors preferences themselves affect the definition of those preferences, Hall s counterfactual assertion does not support this interpretation. Hall argues that a difference in institutions might have led, not to a difference in the disposition, but to a difference in the behavior of the actors. Under alternative institutional arrangements they would, we are told, have been inclined not to value this or that outcome differently, but to pursue neo-corporatist solutions longer and implement them more effectively. Thus, Hall s analysis of the British case leads to the conclusion that is quite comprehensible from a choice-theoretic framework institutions devised to represent actors preferences themselves affect the strategies actors will choose. 17 As the above example demonstrates, the use of a single term to refer to both the disposition and the behavior of actors invites confusion. Conceptual confusion between dispositions and behaviors (preferences and strategies) occurs frequently in the structure-based literature. Similar instances occur in Weir s (1989) discussion of Labour s abandonment of nationalization in favor of demand management and Rothstein s discussion (1992:39) of the effect of labor-market institutions on workers preferences to join or not join unions. In the latter case, Rothstein, while attempting to explain cross-national differences in the propensity of workers to join unions, infers from Skocpol s (1988) assertion of a correlation between labor-market institutions and union densities that government labor-market institutions change workers preferences whether or not to join unions and concludes that we cannot understand national differences in unionization solely by using some sort of rationalchoice or game-theoretic approach, because we need to know how preferences were established in the first place. There are at least two reasons to challenge this conclusion. First, one cannot infer changes in preferences from a correlation between institutional structures and behaviors unless one has examined the strategic environment in which choices are made and eliminated other explanations for the change in behavior without committing the inferential errors discussed above. Second, if we view the decision to join a union as a strategic choice rather than an outcome over which actors have preferences, the research question goes from being something that many have argued is outside the domain of game theory to being an example of precisely the form of human behavior game theory was developed to explain. As these examples indicate, the fact that structure-based and agency-centered institutionalists often mean different things by the word preferences has discouraged a potentially fruitful dialogue on the sources of preferences from taking place. 17 This passage is less a criticism of Hall as it is a criticism of the implications drawn by two of the volume s editors (Thelen and Steinmo, 1992). There is, in general, no tension between Hall s work and the agency-centered approach discussed in this paper. There is, for example, little difference between his understanding of institutions (Hall, 1986) and the approach of Douglas North (1981, 1990). Thelen and Steinmo consider the former one of the historicalinterpretive institutionalists and the latter a rational choice institutionalist.

10 254 Agents and Structures Instead, structure-based institutionalists have reached the conclusion that agencycentered approaches cannot contribute to our understanding of this problem and many agency-centered institutionalists have been content to let someone else deal with this difficult question. If, however, the preferences that are changing in these important cases can be conceived as strategies, then agency-centered theorists do have a great deal to contribute to the understanding of these substantive problems. Furthermore, even if what we are observing is a change in preferences, advances in limited information game theory might allow agency-centered theorists to contribute to the understanding of this more difficult set of questions. Either way, little progress is likely until some agreement is reached on what is meant by terms such as preferences and strategies. Toward an Agency-Centered Approach to Endogenous Preferences Part of the confusion surrounding preferences and strategies is endemic to the agency-centered approach. There exists no universal model of social behavior and no general theory of preference formation. Instead, what exists is a sea of more or less contextually specific models about different aspects of human behavior that include a set of statements about the goals of the actors involved. As noted above, what these models share is the assumption that the behavior to be explained is goal oriented; assumptions about the goals of the actors are typically model specific. Within the context of each model, it is assumed that actors have complete and transitive preferences over a set of outcomes and that they choose strategies in an attempt to obtain the most preferred outcome possible. Strategies are thought of as being entirely instrumental in that actors should have no independent predilection for one set of strategies over another (Frieden, 1997:6). Confusion surrounding preferences and strategies may arise because while outcomes are typically thought of as the terminal nodes of a game or decision tree, in fact, there is typically nothing terminal about them. Actors are not typically thought to cease to exist at the end of the game or decision problem. They collect their payoffs cash their checks, file for bankruptcy, enact policies, get removed from office, et cetera. Since the actors do not cease to exist at the terminal node, the payoffs they receive can be thought of as the values the players expect to receive, in equilibrium, from games that begin at those terminal nodes. As a consequence, their payoffs can easily be thought of in instrumental terms that is, as means toward the satisfaction of some more fundamental goal. Most people have policy preferences, for example, not because they place intrinsic value on the enactment of particular policies, but because they value the expected effects of policies. An industrialist may lobby for trade protection because she expects some more fundamental goal to be better met under protection than under import competition. Note that in this framework preferences have an ambiguous status they are the end of strategy choice, but are also the means of satisfying some deeper, more fundamental goal. While this ambiguity invites confusion, 18 it also provides a mechanism for explaining changes in preferences. This section explores this ambiguity further by examining the opportunity to explain preferences by examining the role they play in satisfying deeper goals that actors have while at the same time emphasizing the importance of keeping strategies and preferences analytically distinct. 19 As will, I hope, become clear, beliefs play an important role in this process. Actors prefer a particular outcome because they believe it will best satisfy some deeper goal. I explore these issues in the context of a simple 18 See Frieden, 1997, for a clear discussion of many of the potential pitfalls. 19 The approach adopted here is similar to Sen s (1977) discussion of meta-rankings in that it posits a hierarchy of preferences.

11 WILLIAM ROBERTS CLARK 255 model of the process of class compromise in Western European democracies. It should be stressed that the point of this modeling exercise is to clarify the agencycentered approach s position on the difference between preferences and strategies. Infinitely more sophisticated, descriptively accurate, and fertile models are available elsewhere. Interested readers might start with Przeworski, 1985, which the current discussion leans heavily upon. Assume, for the moment, that individuals (in a particular concrete historical setting, such as late twentieth century capitalism) have a first-order preference for wealth maximization. Assume also, for the purposes of illustration, that two groups of individuals (workers and capitalists) can be usefully represented by the actions of their leaders (the head of a union confederation and the head of an employers association; hereafter workers and capitalists ). Workers maximize expected real wages which are a function of real wages and the national level of employment, while capitalists maximize profits which are a function of investment, the capital-output ratio, and the real-wage level. 20 First, assume that the workers have a choice between exercising wage restraint (R) and pressing for real-wage increases via increased labor militancy (M); but they are uncertain which action would help them best attain their goals (the highest possible expected real wages). By way of example, suppose that the primary source of their uncertainty revolves around the response of capitalists to wage claims. Capitalists can respond to the actions of workers with either continued investment (I) or disinvestment (D). Figure 1 illustrates this highly stylized depiction of the strategic interaction between capital and labor. Note that at this stage in the analysis wage restraint (R), worker militancy (M), investment (I), and disinvestment (D) are actions that are the object of choice, and as such FIG. 1. Wage Restraint Investment Game 20 In this discussion we have assumed that workers satisfaction of wealth maximization is sufficiently linked to increased real wages and employment that they can both be treated as unproblematic tastes. Note that there is nothing about these assumptions that renders them incomprehensible from the standpoint of rational choice theory. Increases in real wages and employment are clearly instrumental, and it is to a large extent an empirical question whether the leaders of labor confederations pursue these goals as if they were pursuing their own wealth maximization. The theory of the firm and principal agent analysis was developed precisely to provide a mechanism for opening up such black boxes.

12 256 Agents and Structures are strategies that are available to each of the actors. In contrast, increased wages, increased employment, et cetera are outcomes of a process that is jointly determined by the choices of workers and capitalists, but which cannot be fully controlled by either. Since workers are assumed to have stable and known preferences regarding wages and employment, it is possible to construct their preference ordering over outcomes. Let us assume that workers would prefer increases in employment and wages (Outcome 3) to all other outcomes and would prefer all outcomes to declines in employment in the absence of offsetting wage increases (Outcome 2). Finally, assume that workers prefer growth in employment (Outcome 1), even in the absence of wage gains, to increased wages when accompanied by decreased employment (Outcome 4). 21 This statement of workers complete preference ordering (O3 > O1 > O4 > O2) over outcomes does not tell us anything, however, about the workers likely choice of strategy because this choice of strategy maps to outcomes only when we and they have a set of expectations about the likely responses of capitalists to worker behavior. 22 For simplicity, we will assume that workers believe that capitalists are either conditional investors or unconditional disinvestors but are uncertain as to which type of capitalists they are dealing with. 23 Both conditional investors and unconditional disinvestors prefer O4 to O3, but the former prefers O1 to O2. 24 Given their belief that the capitalists are one or the other of the two types posited above, workers can begin to think about the possible consequences of alternative choices of strategy. Specifically, if capitalists are conditional investors (top panel of Figure 2), a decision to exercise wage restraint amounts to a choice between O1 and O4. Since we know from the preference ordering discussed above that workers prefer O1 to O4, we can conclude that workers with the assumed preference ordering will exercise restraint if they believe that capitalists are conditional investors (paths followed when players follow subgame perfect equilibrium strategies are bold). Note that, if the capitalists response is fully and unambiguously anticipated, choosing wage restraint is the strategic equivalent of choosing O1, while choosing no restraint is the strategic equivalent of choosing O4. In contrast, if workers believe that capitalists are unconditional disinvestors (bottom panel of Figure 2), choosing no restraint is still the equivalent of choosing O4, but choosing restraint is now the strategic equivalent of choosing O2. This simple depiction of capital-labor relations can shed some light on metatheoretical debates concerning the limitations of agency-centered approaches to institutions. First, note that workers preferences are identical in the two games, but the strategies they are expected to choose in equilibrium are different. This suggests that analysts who infer changing preferences from changes in behavior, or crossnational differences in preferences from cross-national differences in behavior, may be committing what psychologists refer to as the fundamental attribution error (Ross, 1977; Miller, 1989; Miller, Ashton, and Mishal, 1990) or what Frieden refers to as a sin of commission (Frieden, 1997:21). The problem is that there are a number of logically possible reasons a labor confederation would pursue a strategy of increased militancy at time t but wage restraint at time t+1, and changing 21 Presumably because current restraint increases profits that can be reinvested in ways that increase worker productivity and wages in the long run. 22 Similarly, it is impossible at this stage of the analysis (i.e., in the absence of propositions about capital s preferences and therefore equilibrium behavior) to make meaningful inferences about workers preferences based on observations of their behavior. 23 There are four logically possible strategies that capitalists can pursue. For brevity, only two are addressed here; this amounts to an assumption that workers place a probability of zero on the chance that capitalists will adopt one of these other strategies. Note that from this standpoint games of perfect information can be thought of as simplified presentations of special cases of games of imperfect information. 24 We will put off, for the moment, a discussion of why capitalists might have one or another set of preferences.

13 WILLIAM ROBERTS CLARK 257 FIG. 2. Equilibrium Behavior in Wage Restraint Investment Game for Two Types of Conditional Investors preferences is only one of them. As the example here demonstrates, it is entirely possible that workers preferences have been stable, but their beliefs about capitalists preferences (and therefore, equilibrium behavior) have changed. Alternatively, it is possible that workers possessed no uncertainty about capitalists preferences, but that the change in workers behavior is the result of a change in capitalists preferences. Which of these possible explanations is the case is essentially an empirical matter, and, while agency-centered theorists are reluctant to infer changes in preferences from changes in behavior, there is nothing in the approach that rules

14 258 Agents and Structures out changes in preferences. The agency-centered approach s position regarding changing preferences is neither an empirical statement nor a logical deduction from unquestionable axioms. It is, instead, a pragmatic, methodological rule of thumb that maintains that theoretical approaches that search for explanations for changes in behavior should examine the structural environment in which actors are embedded before inferring that changes in preferences are the cause of observed changes in behavior (Stigler and Becker, 1977). 25 Since agency-centered theory does not rule out the possibility of changing preferences, it must come to grips with the charge that it is incapable of explaining changes in preferences or accept the charge that it cannot contribute to an important source of human behavior. To address this question, it is crucial to clarify further the distinction between preferences and strategies. As noted above, when choice theorists assert that preferences must be held constant, they are speaking exclusively about preferences over outcomes. That is, whatever is chosen by the analyst and/or treated by the actors as the unquestioned goal of their behavior in that context must be held constant if we are to be able to deduce propositions about the recursive relationship between beliefs and actions. With few exceptions, the choice of what should be treated as the outcomes over which actors have preferences is not determined by empirical reality, but rather is a function of the research question at hand. If, as in the above example, we are attempting to explain the conditions under which the leaders of labor confederations will exercise wage restraint, it is important to treat wage restraint as an action or strategy choice as a means to some other end not as an outcome over which workers have preferences. If we treated it as an outcome that actors had preferences over, our explanation would be tautological workers choose restraint when they prefer it to militancy. Thus, if we were to characterize the research problem addressed in the wage-restraint game above as What explains the union confederation in country X s preference for restraint over labor militancy? and we accepted the notion that rational choice theory can t explain preferences, we might wrongly infer that rational choice theory can t contribute to our understanding of the strategic interaction between capital and labor. If, instead, we posed the research question as What explains the union confederation in country X s choice of wage restraint over labor militancy? we could then provide an answer to this question in terms of dispositional traits of actors (the preferences workers hold over alternative macroeconomic outcomes); the structure of the game (workers moved first and capitalists moved last; capitalists knew workers choice when they decided but the reverse is not true); the beliefs that actors held about the motivations and intentions of other actors (as captured in their preferences over outcomes); and the way in which strategy combinations mapped to macroeconomic outcomes (i.e., the implicit macroeconomic model that the actors believe governs their world). 26 Thus, with a few possible exceptions, the decision to rule a question out of bounds for agency-centered approaches is a decision made by the analyst, not a consequence of the subject being studied. If we conceive of the behavior in question as the product of a choice made by agents rather than a reflection of the attributes of agents, we render the question amenable to choice-theoretic analysis. 27 There are possible 25 In this light, Wendt s (1994) suggestion that rationalist and structurationist approaches possess rival hypotheses regarding the stability of preferences that can and should be directly tested is a curiously empiricist position. Agency-centered theorists tend to treat preferences as unobservables and recommend evaluating the veracity of assumptions about preferences by comparing a model s implications with observed experience. 26 Note that the contribution of game theory is not in identifying the correct model of a particular phenomenon, but rather in using models to evaluate the internal logic of competing explanations and to generate testable propositions about the world. 27 Even when actors are guided by cognitive scripts, using rules of thumb, or engaging in habit-driven behavior, remembering that they retain the capacity to choose aids in understanding (a) the conditions under which they might