Games Real Actors Play

Actor-centered Institutionalism In Policy Research

Fritz W Scharpf

• Publisher: Routledge
• Year: 2018
• ISBN:
  9780429979903
• Edition: 1
• Topic: Politique et relations internationales
• Subtopic: Politique

Games Real Actors Play provides a persuasive argument for the use of basic concepts of game theory in understanding public policy conflicts. Fritz Scharpf criticizes public choice theory as too narrow in its examination of actor motives and discursive democracy as too blind to the institutional incentives of political parties. With the nonspecialist in mind, the author presents a coherent actor-centered model of institutional rational choice that integrates a wide variety of theoretical contributions, such as game theory, negotiation theory, transaction cost economics, international relations, and democratic theory.Games Real Actors Play offers a framework for linking positive theory to the normative issues that necessarily arise in policy research and employs many cross-national examples, including a comparative use of game theory to understand the differing reactions of Great Britain, Sweden, Austria, and the Federal Republic of Germany to the economic stagflation of the 1970s.

1
Policy Research in the Face of Complexity

In order to be pragmatically useful, the findings of interaction-oriented policy research should not only be case-specific and post hoc, in the sense in which that is true of historical research, but they should also allow lessons drawn from one case to be applied to others and, ideally, to produce lawlike generalizations with empirical validity. In the social sciences, however, this ideal is generally difficult to realize, and in interaction-oriented policy research it is nearly impossible.

Intentional Action: Boundedly Rational and Socially Constructed

The reason is straightforward: Policy is produced by human actors who are not merely driven by natural impulses or by the compulsion of external factors. Instead, public policies are the outcomes—under external constraints—of intentional action. Intentions, however, are subjective phenomena. They depend on the perceptions and preferences of the individuals involved. People act not on the basis of objective reality but on the basis of perceived reality and of assumed cause-and-effect relationships operating in the world they perceive. And people act not only on the basis of objective needs but also on the basis of preferences reflecting their subjectively defined interests and valuations and their normative convictions of how it is right or good or appropriate to act under the circumstances. Intentional action, in other words, cannot be described and explained without reference to the subjective "meaning" that this action has for the actor in question.

For social science research this condition creates an obvious problem, since we cannot directly observe subjective phenomena but always depend on what is at best secondhand information. Moreover, to say that intentions are subjective also suggests the possibility that they may be idiosyncratic, varying from one individual to another and from one time and place to another. If this were all that we could count on, a social science that is searching for lawlike regularities and for theory-based explanations and predictions would be not merely difficult but impossible. All we could aspire to do would be to describe what happened in historical narratives and perhaps to search for ad hoc explanations based on information about individual motives and worldviews of the actors involved that we might infer from such unreliable sources as personal interviews, memoirs, and contemporary documents. Since there would be no way in which we could apply lessons from one case to another, our work would also lack pragmatic usefulness.

If, nevertheless, the social sciences do claim to discover regularities of human action that allow not only interesting descriptions of past events but also theory-based explanations that are potentially useful for practical purposes, that claim presupposes the existence of mechanisms that, in some way, are able to structure, or to standardize, the individual perceptions and preferences that we are likely to encounter in empirical research. In fact the social sciences have come to rely on two such mechanisms, both of which depend at bottom on an evolutionary argument.

At one end of the social science spectrum are mainstream economics and those variants of political science and sociology that have become committed to the rational-actor approach. In the evolution of the human species, so it is assumed, there must have been a premium on accurate perceptions of the environment and on behavior that would increase the survival chances of the individual and its progeny. As a consequence, rational self-interested action is thought to have become genetically fixed as a universal characteristic. Among the social sciences proper, neoclassical economics depends most completely on this working hypothesis. Economic actors are assumed to be exclusively motivated by economic self-interest—which is interpreted to imply the maximization of profits for firms and the maximization of wealth for households. On the cognitive side the corresponding assumption is that actors will perceive the economic environment in the same way that it is perceived by the scientific observer and that their computational capacities are on the whole adequate to the task of selecting courses of action that will in fact maximize their self-interest. When these assumptions are granted, choices will be determined by external conditions—namely, by the available opportunities for investment and consumption, by their relative prices, and by the actors' own budget constraints. Since data on these conditions are, at least in principle, accessible to the researcher, neoclassical economics claims to explain and predict the behavior of economic subjects on the basis of general laws combined with objectively available information.

For the world of competitive market economies, the assumptions of neoclassical economics may indeed approximately describe the intentions of economic actors—and as the intensity of competition increases, their empirical plausibility increases as well (Latsis 1972). Moreover, since empirical economic research does not usually try to explain or predict individual choices but rather is interested in the aggregate effects of large numbers of individual choices, random deviations from the assumed central tendency do not much matter, and economics is, on the whole, reasonably successful in explaining and predicting—under ceteris paribus conditions—the responses of capital owners, firms, workers, and consumers to changes in the relative prices of capital, raw materials and energy, labor, and goods and services.

At the other end of the social science spectrum, cultural anthropology and mainstream sociology also make a claim to evolutionary foundations. But rather than the survival value of rational self-interested action in natural environments, they emphasize the extreme complexity and uncertainty of an environment constituted by other human actors whose subjective worldviews and preferences cannot be directly observed, as well as the enormous difficulties that individuals would have to overcome in communicating with each other and in coordinating their actions in social encounters. Human societies could only have evolved, so it is argued, because these difficulties are overcome through a "social construction of reality" (Berger/Luckmann 1966) that assures the convergence of cognitive orientations and through social norms and institutionalized rules that shape and constrain the motivations or preferences of all participants in social interaction. Culture and institutions, in other words, are necessary preconditions of human interaction. They allow individuals to find some sense in their otherwise chaotic worlds and to anticipate to some extent the otherwise unpredictable—and hence threatening—intentions of others with whom they must interact. Most important for our purposes, they also create the behavioral regularities that can then be discovered by social science research and used in theory-based social science explanations.

These two paradigms are usually presented in opposition to each other. In fact, however, they are not mutually exclusive. On the one hand, even if the underlying assumptions of culturalist approaches were granted, it would not follow that human action can be explained exclusively by reference to culturally "taken-for-granted" beliefs and institutionalized rules of "appropriate behavior." Human actors are not merely acting out culturally defined "scripts," nor are they rule-following automata—they are intelligent and they have views of their own and interests and preferences of their own, which sometimes bring them to evade or to violate the norms and rules that they are supposed to follow.

But on the other hand, neither is it realistic to think of human actors as always being omniscient and single-minded self-interest maximizers who will rationally exploit all opportunities for individual gain regardless of the norms and rules that are violated. Human knowledge is limited and human rationality is bounded, and hence much human action is based not on the immediate cognition of real-world data and causal laws but on culturally shaped and socially constructed beliefs about the real world. At the same time, most human action will occur in social and organizational roles with clearly structured responsibilities and competencies and with assigned resources that can be used for specific purposes only. In these culturally and institutionally defined roles, pure self-interest will not explain much beyond the choice of assuming, or refusing to assume, certain roles. But once a role has been assumed, action within that role is practically impossible to explain without reference to cultural and social definitions of that role and to the institutionalized rules associated with its proper performance.

Thus, while the rational-actor paradigm may capture the basic driving force of social interaction, its information content with regard to the operative intentions of human actors outside of the economic field is close to zero—unless we are able to resort to institution-specific information for the specification of actor capabilities, cognitions, and preferences. This is the gist of the framework of actor-centered institutionalism that will be presented in Chapter 2.

Many Variables and Few Cases

From the point of view of generalizability, however, we are still far from home. Institutional definitions of capabilities, cognitions, and preferences are by no means universal in their substantive content. Cultural history and cultural anthropology have informed us about the enormous variability of what is culturally "taken for granted" from one place to another and from one time to another, and we also know from legal history, constitutional history, comparative law, and comparative government how much institutions do in fact vary in time and place. So if behavior is shaped by institutions, the behavioral regularities we can expect are also likely to vary with time and place. Hence the best that we could hope to discover is not the universal theories that are the aim of the natural sciences but, as the late James S. Coleman (1964, 516–519) put it, "sometimes true theories"—providing explanations, that is, that hold only under specific institutional conditions. In order to assess the domain of such explanations, we therefore must vary the institutional context in comparative studies.

If we do so, however, we confront the fact that the institutional factors that will plausibly affect policy outcomes can only be described in a multidimensional property space. For instance, even if we limit comparison to highly developed Western societies and democratic political systems, and even if we consider only institutions at the national level, the institutional settings that are known to affect policy processes can be described as being either unitary or federal, parliamentary or presidential, having two-or multiparty systems in which interactions are competitive or consociational, and with pluralist or neocorporatist systems of interest intermediation. In comparative political science research, these variables are assumed to be of general policy relevance, whereas others—for example, the autonomy or dependence of central banks or the existence of insurance-based or tax-financed health care systems—may need to be considered only in particular policy contexts. Worse yet, interaction effects among the characteristics listed are likely to be important: In the case discussed in the Introduction, it was clear that the beneficial effects of neocorporatist institutional arrangements could be undermined by the existence of an independent central bank. Similarly, federalism in a two-party system will generate effects that differ from federalism in a multiparty system (Scharpf 1995). Hence even the first-mentioned five dichotomies will amount to 2⁵, or 32, different institutional constellations that, for all we know, may differ significantly in their impact on public policy.

But that is not all. The effect of institutions on public policy is also likely to be modified by changes in the external policy environment. Thus while neocorporatist concertation was successful in avoiding both inflation and unemployment under the stagflation conditions of the 1970s, the same institutional factors lost most of their effect on economic policy outcomes in the economic environment of globalized capital markets in the following decade (see Appendix 1). Similarly, in the benevolent economic environment of the postwar decades, a wide variety of welfare-state institutions have been equally successful in providing social security at acceptable cost. In the economic environment of the 1990s, however, Continental welfare states relying primarily on payroll taxes to finance transfer payments seem to be in greater difficulty than Scandinavian welfare states that are financed from general tax revenues and that emphasize services rather than transfer payments (Esping-Andersen 1990).

For comparative policy research, this means that the potential number of different constellations of situational and institutional factors will be extremely large—so large, in fact, that it is rather unlikely that exactly the same factor combination will appear in many empirical cases. In the natural sciences this difficulty would typically be overcome through experimental designs that permit the isolation and systematic variation of a single factor—which we can rarely hope to do in policy research. The closest equivalent to experimental designs in empirical research are comparative studies using carefully matched cases selected according to the logic of the "most-similar systems" or the "most-different systems" design (Przeworski/Teune 1970). If the cases differ (or agree) only in one variable or in a very limited set of variables, it may indeed be possible to derive causal inferences with a good deal of confidence.

The "most-similar systems" design was in fact used in the example study discussed in the Introduction: All four countries, Austria, Britain, Sweden, and West Germany, were hit by the same external oil-price shock in 1973–1974; all were in fairly good economic shape in 1973; all had governments that acted from a Keynesian worldview and that had a clear political preference for maintaining full employment; and all had relatively strong and generally "cooperative" trade unions. Since these factors could be "held constant," it was then possible to identify the influence of just two sets of institutional variables—union organization and central-bank independence—on the policy choices that were in fact adopted. Hence quasi-experimental designs may indeed work in policy research. However, two caveats are in order.

First, as the full-length study amply demonstrates, the four countries did differ in a great many other respects that I have not mentioned here, and the actual courses taken were also influenced by historical "accidents" that could not be represented in a parsimonious theoretical model (Scharpf 1991a). Hence the effectiveness of the quasi-experimental design depends on the level of detail at which explanations are being sought. Second, and more important, the research design used here depends on exceptional circumstances that policy researchers cannot count upon. In the general case, comparative designs are much more likely to encounter cases that differ not only in a few institutional variables but also in external conditions, actor identities and capabilities, actor perceptions, and actor preferences. Under such conditions, "most-similar" and "most-different" systems designs will not reduce variance sufficiently to facilitate quasi-experimental solutions (King/Keohane/Verba 1994, 199–206).

By the same token, however, the usual social science methods of inductive theory development and statistical theory testing will also run into difficulties here. Even if we disregard the logical objections to inductive generalization (Wilier/Wilier 1973; John 1980), we have little opportunity to discover "empirical regularities" by observing large numbers of similar cases and we have even less opportunity to subject hypotheses generated through inductive generalization to statistical tests using data sets that are different from the original observations. Given the number of potentially relevant independent variables, we will usually not have the requisite number of cases to perform statistical tests, even if the number of observations is inflated by combining cross-sectional and longitudinal data in "pooled time series."¹

Gary King, Robert Keohane, and Sidney Verba (1994) have, it is true, identified a range of useful and imaginative research strategies that could be employed to ease or to overcome the small-sample, or small-N, problem by generating additional observations at different levels or in different segments of complex cases. Unfortunately, however, this important work only helps to highlight the more fundamental difficulty that we encounter when trying to follow the methodological canons of empirically validated causal inference. It is best summed up in their discussion of a study searching for explanations of interstate cooperation in high-tech weapons development. Since only three cases could be studied, whereas there were seven potentially effective independent variables, the research design was judged to be indeterminate: It could not determine which of the hypotheses, if any, was true. Assuming that a sufficient number of additional case studies could not be carried out, the best advice that the authors can provide is "to refocus the study on the effects of particular explanatory variables across a range of state action rather than on the causes of a particular set of effects, such as success in joint projects" (King/Keohane/Verba 1994, 120).

More generally, King and colleagues have a consistent preference for designs searching for the effects of a particular explanatory variable rather than for the causes of a particular empirical outcome; in fact, all their methodological recommendations for coping with the small-NT problem have this "forward-looking" character. Everything else being equal, this certainly is a highly plausible methodological preference. When one is looking forward from a particular independent variable to its potential effects, h...

Table of contents

• Cover
• Half Title
• Title
• Copyright
• Dedication
• Contents
• List of Tables and Figures
• Acknowledgments
• Introduction
• 1 Policy Research in the Face of Complexity
• 2 Actor-Centered Institutionalism
• 3 Actors
• 4 Actor Constellations
• 5 Unilateral Action in Anarchic Fields and Minimal Institutions
• 6 Negotiated Agreements
• 7 Decisions by Majority Vote
• 8 Hierarchical Direction
• 9 Varieties of the Negotiating State
• Appendix 1: A Game-Theoretical Interpretation of Inflation and Unemployment in Western Europe
• Appendix 2: Efficient Self-Coordination in Policy Networks—A Simulation Study, with Matthias Mohr
• References
• About the Book and Author
• Index