Political Methodology Committee on Concepts and Methods Working Paper Series

Transcription

1 Political Methodology Committee on Concepts and Methods Working Paper Series 10 December 2006 The Study of Critical Junctures in Historical Institutionalism Giovanni Capoccia & R. Daniel Kelemen University of Oxford C&M The Committee on Concepts and Methods IPSA International Political Science Association CIDE Teaching and Research in the Social Sciences

2 Editor Andreas Schedler (CIDE, Mexico City) Editorial Board José Antonio Cheibub, University of Illinois at Urbana-Champaign David Collier, University of California, Berkeley Michael Coppedge, University of Notre Dame John Gerring, Boston University George J. Graham, Vanderbilt University Russell Hardin, New York University Evelyne Huber, University of North Carolina at Chapel Hill James Johnson, University of Rochester Gary King, Harvard University Bernhard Kittel, University of Amsterdam James Mahoney, Brown University Gerardo L. Munck, University of Southern California, Los Angeles Guillermo O Donnell, University of Notre Dame Frederic C. Schaffer, Massachusetts Institute of Technology Ian Shapiro, Yale University Kathleen Thelen, Northwestern University The C&M working paper series are published by the Committee on Concepts and Methods (C&M), the Research Committee No. 1 of the International Political Science Association (IPSA), hosted at CIDE in Mexico City. C&M working papers are meant to share work in progress in a timely way before formal publication. Authors bear full responsibility for the content of their contributions. All rights reserved. The Committee on Concepts and Methods (C&M) promotes conceptual and methodological discussion in political science. It provides a forum of debate between methodological schools who otherwise tend to conduct their deliberations on separate tables. It publishes two series of working papers: Political Concepts and Political Methodology. Political Concepts contains work of excellence on political concepts and political language. It seeks to include innovative contributions to concept analysis, language usage, concept operationalization, and measurement. Political Methodology contains work of excellence on methods and methodology in the study of politics. It invites innovative work on fundamental questions of research design, the construction and evaluation of empirical evidence, theory building and theory testing. The series welcomes, and hopes to foster, contributions that cut across conventional methodological divides, as between quantitative and qualitative methods, or between interpretative and observational approaches. Submissions. All papers are subject to review by either a member of the Editorial Board or an external reviewer. Only English-language papers can be admitted. Authors interested in including their work in the C&M Series may seek initial endorsement by one editorial board member. Alternatively, they may send their paper to workingpapers@conceptsmethods.org. The C&M webpage offers full access to past working papers. It also permits readers to comment on the papers.

3 The authors thank Melani Cammett, John Gerring, Stephen Hanson, Sara Hobolt, Jack Levy, Michael Rosen, Margaret Stevens, and participants and the audience of the APSA Panel Critical Junctures, Path Dependency and Process Tracing in the 2005 Annual Convention and the seminar on Comparative Political Economy in the Department of Politics and IR at Oxford University for their comments and suggestions. The usual disclaimer applies. 1

4 I. Introduction The concept of critical junctures is an essential building block of historical institutionalism. Many causal arguments in the historical institutionalist literature postulate a dual model of institutional development, with alternations between relatively long periods of path dependent institutional stability and reproduction, punctuated occasionally by brief phases of institutional flux referred to as critical junctures during which more dramatic change is possible. The causal logic behind such arguments emphasizes the enduring impact of choices made during critical junctures in history. These choices close off alternative options and lead to the establishment of institutions that generate self-reinforcing path-dependent processes. In Pierson s words, Junctures are critical because they place institutional arrangements on paths or trajectories, which are then very difficult to alter (Pierson 2004: 135: see also Mahoney 2001: 114). Path dependence is a crucial causal mechanism for historical institutionalists and critical junctures constitute the starting points for many path dependent processes. Despite the theoretical and practical importance of critical junctures as the genetic moments for institutional equilibria, analyses of path dependence often devote little attention to them, focusing instead on the reproductive phase launched after a path dependent process is initiated. The notable exceptions are the macro-historical analyses of the development of entire polities (e.g. Lipset and Rokkan 1967; Berins Collier and Collier 1991; Mahoney 2002). Authors in this tradition put the moments of institutional formation at the center of their analyses. More generally, however, applications of the path dependence approach to institutional analysis often bracket off the moment of institutional formation (variously referred to as critical juncture, turning point, crisis, unsettled time etc.) 1. Critical junctures are too often treated as bookends, or dei ex machinae, on otherwise carefully constructed stories of institutional development. On the whole, the literature offers relatively little 1 These synonyms are often used interchangeably, with no indication that they represent different types of fluidity phases. We will use the term critical juncture throughout our analysis. 2

5 methodological guidance to those who would employ the concept of critical junctures: this article seeks to redress this imbalance. 2 We address two questions one conceptual and one empirical concerning the use of critical junctures in historical institutionalist explanations. First, how should one define a critical juncture? While in principle critical junctures can be invoked in order to interpret all sorts of developmental processes, ranging from evolutionary biology, to macro-history, to organizational decision-making processes, to individual life histories, we focus on institutional analysis. In institutional analysis, critical junctures are characterised by a situation in which for a relatively short period the structural (i.e. economic, cultural, ideological, organizational) influences on political action are substantially relaxed, with two main consequences. First, the range of plausible choices open to powerful political actors expands substantially. Second, the consequences of their decisions for the outcome of interest are potentially much more momentous. Contingency, in other words, becomes paramount. We offer a definition of critical junctures that aims to promote a more rigorous and hopefully more fruitful use of the concept in institutional analysis. Second, how should one analyze critical junctures? What is the best approach to analyze these moments when the freedom of political actors and impact of their decisions is heightened? What methods should be used? What kind of reasoning? What are the strengths and weaknesses of these methods? We argue that because heightened contingency is a core characteristic of a critical juncture, counterfactual reasoning and narrative process tracing must be explicitly employed to analyze them. In this context, we also address specific issues relevant to both cross-sectional and longitudinal comparisons of critical junctures. We provide criteria for adjudicating between rival claims concerning the causal impact of critical junctures. In sum, while the theoretical apparatus of critical junctures remains a heuristic for historical research (Ebbinghaus 2005, 25) and therefore defies fine-grained measurement, the concepts on which it is built and the methods used need not be ambiguous or ad hoc. The article proceeds as follows. In the first section, we review prominent applications of the concept of critical junctures and its synonyms in political science and highlight the limitations of the existing literature. Second, we present our 2 This dearth of conceptual instruments to define, study and compare critical junctures is even more striking if compared with the rich conceptual apparatus available to scholars to analyze path-dependent processes themselves -e.g. increasing returns, lock-in, sequencing etc. (Page 2006). 3

6 definition of critical junctures and highlight its consequences for the application of the concept to empirical analysis. Next, we discuss methodological issues connected to the study of critical junctures. We then provide empirical illustrations of our theoretical points. Finally, we conclude by summarizing our argument. II. Critical Junctures and Institutional Formation in the Literature The dualist conception of political and institutional development, based on an alternation between moments of fluidity and rapid change and longer phases of relative stability and institutional reproduction has a venerable pedigree in social sciences and political history. An early example can be found in Polanyi, who speaks of critical periods and connecting stretches of time (1944, 4). Of the many subfields in the social sciences in which work has been done on path dependent institutional development, critical junctures have received the most sustained attention in macro-historical analyses of the development of entire regimes or regions. For example, Seymour Martin Lipset and Stein Rokkan were pioneers in this regard, tracing the roots of the origins of Western European party systems to three crucial junctures in the history of each nation (Lipset and Rokkan 1967, 37-38). 3 While Lipset and Rokkan say that the variety of empirical party systems can be reduced to a set of ordered consequences of decisions and developments at critical junctures, their analysis is couched in a largely structural language, leaving little place for more fine-grained analysis of political decision-making and meaningful choices (ibid., 44). The concept of critical junctures is also central in Berins Collier and Collier s study of the political development of eight Latin American countries. According to their definition, a critical juncture is a period of significant change, which typically occurs in distinct ways in different countries (or other units of analysis) and which is hypothesized to produce distinct legacies (Berins Collier and Collier 1991, 29). In his comparative study of the political development of Central America, Mahoney (2002) uses a similar approach and defines critical junctures as choice point[s] when a particular option is adopted among two or more alternatives, defined by antecedent historical conditions. Mahoney emphasizes the connection between critical junctures and path-dependent processes, explaining that once a particular option is selected [in 3 Further examples of a similar approach are Moore (1966) and Luebbert (1991). 4

7 a critical juncture], it becomes progressively more difficult to return to the initial point when multiple alternatives were still available. (Mahoney 2000: 513; 2001: 113). More explicitly than Berins Collier and Collier, Mahoney emphasizes the importance of agency and meaningful choice: in many cases, critical junctures are moments of relative structural indeterminism when wilful actors shape outcomes in a more voluntaristic fashion than normal circumstances permit these choices demonstrate the power of agency by revealing how long-term development patterns can hinge on distant actor decisions of the past (Mahoney 2002: 8; see also Katznelson 2003: ). Gourevitch s (1986) study of the domestic political impact of international economic crises shares this focus on individual actors and their decisions. Gourevitch defines crises as the open moments when system creating choices are made and argues that in moments of crisis politicians, their goals and their decisions are what matters most (1986, 239). We share the emphasis on the power of agency during critical junctures, but even these important contributions do not address several key issues. In particular, conceptual shortcomings stand out in four crucial areas: power asymmetries, time horizons, units of analysis, and near misses. First, much of the existing literature draws on analogies that obscure the role of power asymmetries during critical junctures. Berins Collier and Collier, as well as Mahoney, see their studies as examples of a more general approach to the analysis of institutional development, in which critical junctures give rise to path dependent processes, and explicitly link their work to research on path dependence in institutional economics. Berins Collier and Collier assert that underlying their study is a common understanding of change that is a cornerstone of comparative-historical research on development. It suggests what Paul A. David has called a path dependent pattern of change, in that outcomes during a crucial transition establish distinct trajectories within which, as he has engagingly put it, one damn thing follows another (Berins Collier and Collier 1991: 27). Similarly, Mahoney illustrates his definition of critical juncture by referring to the economics literature on competing technologies (Mahoney 2002: 7). Yet the application of analogies, ubiquitous in this literature, with processes in economics or physics in which a series of small events lead a system to tip one way or the other may conceal the impact of power asymmetries in politics: even in periods of extreme fluidity, the decisions of some actors are freer and in particular more influential than those of others 5

8 Second, the literature provides very little guidance on how to deal with time horizons in historical institutional arguments that involve critical junctures. 4 Many authors understand critical junctures not as transitory periods, but as historical moments that may last for many years. However, the literature has not fully explored the logical and theoretical consequences of this approach. We draw attention to potential pitfalls concerning both the relative and absolute duration of critical junctures. First, the duration of the critical juncture must be brief relative to the duration of the path dependent process that it initiates. Treating an entire decade as a critical juncture with respect to an outcome observed a century later might be sensible. However, it would clearly not be sensible to consider a decade-long period a critical juncture with respect to an outcome observed only one year later. Second, the absolute duration of a critical juncture has an impact on the possibility for actors to act more freely and for the consequences of their actions to have a larger impact than in normal times: the longer the juncture, the higher the probability that political decisions will be constrained by some re-emerging structural constraint. Third, rigorous arguments based on the notion of critical junctures must specify precisely the unit of analysis with respect to which the juncture is argued to be critical. Often, scholars identify relatively brief periods of momentous political, social or economic upheaval and assert that these are critical junctures in a general sense (e.g. Ebbinghaus 2005, 16). However, even during periods of massive social and political upheaval, certain institutions may remain unaffected. Likewise, during periods when a political domestic or international regime as a whole is stable, particular institutions may face critical junctures. Even where various institutions are inter-connected, the occurrence of a critical juncture for one institution need not constitute a critical juncture with respect to all of its counterparts. Finally, critical junctures are too often equated with moments of change (e.g. Berins Collier and Collier 1991; Abbott 2001). However, as counterintuitive as it may seem, change is not a necessary element of a critical juncture. A situation of heightened contingency may lead to change, but it may as well give rise to reequilibration, that is, aborted change. Most researchers, following Pierson s (2000) advice to go back and look, trace the roots of institutional change back to the origin of a path. We certainly do not disagree with this approach, but contend that, if used 4 A notable exception to this rule is Pierson (2004), whose work we discuss below. 6

9 exclusively, it overlooks the fact that some critical junctures may result in reequilibration of an institution. To sum up, especially when we move away from the specific use of the concept in macro-historical analyses and survey the literature on the development of other kinds of institutions, 5 we notice that the existing literature on path dependence rarely gives critical junctures the attention that they deserve. Most often the concept is simply referred to, and there is little effort to define it clearly. The emphasis of the literature is often on the mechanisms of reproduction underpinning path dependence, rather than on the genetic phase of the critical juncture itself (e.g. Pierson 2000), while attempts to clarify and operationalize the concept (e.g. Ebbinghaus 2005, Hogan 2005) have been generally unable to eliminate conceptual confusion. Taken as a whole, the scholarship in which the concept is used or referred to lacks conceptual consistency. III. Theory Definition In the context of the study of path dependent phenomena, we define critical junctures as relatively short periods of time during which there is a substantially heightened probability that agents choices will affect the outcome of interest. By relatively short periods of time, we mean that the duration of the juncture must be brief relative to the duration of the causal process or outcome it purports to affect. By 'substantially heightened probability', we mean that the probability that agents choices will affect the outcome of interest must be high relative to that probability before and after the juncture. This definition captures both the notion that, for a brief phase, agents face a broader than normal range of feasible options and that their choices from among these options are likely to have a significant impact on subsequent outcomes. Furthermore, by emphasizing that the probability that actors choices will affect outcomes decreases after the critical juncture, this definition suggests that their choices during the critical juncture trigger a path dependent process. In fact, the resort to the concept of critical 5 The concept of critical juncture has been applied to national welfare policies (e.g. Gal and Bargal 2002; Ebbinghaus 2005), the EU budgetary policy (Laffan 2000), labor unions (e.g. Hogan 2005),, regulations of competition in product markets and banking (Djelic and Quack 2005), foreign policy (Ikenberry 1989), the modern state (Krasner 1984), the causes of war (Levy and Goertz 2005), and more. 7

10 junctures is only significant in the context of path dependent arguments, where normally the choices of political actors tend to be constrained by external circumstances. As such, the critical juncture constitutes a qualitatively different situation than the normal historical development of the institutional setting of interest. Finally, our definition sets the stage for addressing the four shortcomings in the literature identified above concerning units of analysis, time horizons, near misses and power asymmetries. Identification of a unit of analysis When confronted with the question of whether an event, or series of events, constitutes a critical juncture, the first question one must ask is a critical juncture in the development of what? In other words, the analysis must be anchored in a unit of analysis. The dual model of development with alternations between critical junctures and phases of stability and reproduction has been used in a variety of settings, ranging from an individual s life course (e.g. Abbott 2001), to the analyses of natural evolution based on the punctuated equilibrium model (e.g. Gould 2002) to a variety of subfields of physics (Ball 2004). In the broad tent of political science, this general dual approach can be applied to a wide variety of units of analysis, such as the psychological history of a leader, trends in public opinion, the development of national identity or the outbreak of wars. In historical institutionalism in particular, however, the units of analysis are typically some form of institutional setting in which actors decisions are constrained in phases of equilibrium and freer in phases of change. Such institutions may range from a single organization (e.g. a political party, a union, a corporation), to the structured interaction between organizations (e.g. a party system, relationships between branches of government), to public policies, to a political regime as a whole (e.g. Greif and Laitin 2004: 640). The key point here is that an historical moment that constitutes a critical juncture with respect to one institution may not constitute a critical juncture with respect to another. In some accounts, periods of crisis for an encompassing institutional setting (i.e. a regime as a whole) are simply assumed to constitute critical junctures with respect to all connected, subsidiary institutions. However, even when political systems as a whole face unsettled times (Swidler 1986; Katznelson 2003), many institutions may remain unaffected: as Thelen and Streeck note, there often is considerable continuity through and in spite of historical break points (Streeck and 8

11 Thelen 2005, 8-9). Likewise, during settled times, particular institutions may face critical junctures. Even where there are clear connections between institutions, what constitutes a critical juncture for one, may leave others largely unaffected. For example, a period identified as a critical juncture with respect to a country s party system, may not be a critical juncture with respect to its overall regime type or system of interest intermediation. While relevant events happening at one of these levels of analysis may influence the others, analytically it is important to keep them separate, and to refer the critical juncture clearly to the development of a specific unit. The identification of the levels of change and stability, and of the significance of events for each level, is a matter of theoretical choice (Thelen 2003: 213; see also Shermer 1995: 71). Time horizons and alternative models of institutional change Most historical institutionalists treat critical junctures not as instantaneous events, but as moments in a more metaphorical sense, which may actually last for a number of years. If a critical juncture is not a discrete event, but rather an accumulation of related events during a relatively compressed period, at what point does it become more accurate to speak of gradual evolution than rapid change? A gradualist view of institutional evolution, which discounts the role of critical junctures and punctuated equilibrium, suggests that choice points come with great regularity and that the accumulation of choices generates outcomes. 6 Pierson s (2004) categorization of social science accounts in terms of their time horizons is helpful to clarify when it is, and when it is not, appropriate to use explanations that invoke critical junctures. As Pierson explains, social science accounts may involve causes that have either short-term or long-term time horizons, and they may be designed to explain outcomes that have either short-term or longterm time horizons. Table 1 Time horizons in different social science accounts Time horizon of the outcome Short Long 6 In the literature on evolution, from which much of the social science literature on institutional evolution draws inspiration, scholars who reject the notion of punctuated equilibrium emphasize precisely this point (Dawkins 1996). 9

12 Time horizon of the cause (Source: Pierson 2004: 92) Short Long I III Thresholds; Causal Chains II Cumulative Effects IV Cumulative Causes As discussed in more detail below, a critical juncture should have a duration that is short relative to either the duration of the outcome it purports to explain and/or the period between the end of the critical juncture and the eventual outcome. 7 Thus, causal accounts that involve a short term cause with long term effects (upper right quadrant) are the strongest candidates for a critical juncture explanation. Short term causes with short term outcomes (upper left quadrant), may also involve critical junctures. If an explanation relies on a cause that has a relatively long time-horizon (bottom row cells), then it is highly unlikely that a critical juncture will be a significant part of the cause. In accounts that involve long term, cumulative causes there may be a tipping point a point at which the cumulative cause finally passes a threshold and leads to a rapid change in the outcome, but a tipping point is not a critical juncture. It may be the case that actions taken on the verge of the tipping point might have forestalled it. However, causal accounts that involve cumulative causes suggest that the probability of a particular outcome increases steadily over time, and thus would not present compressed moments in which an agent s decisions are particularly likely to affect outcomes. 8 Change and negative cases Many scholars define critical junctures on the basis of their outcome, namely change (e.g. Berins Collier and Collier 1991, 29-30; Abbott 2001; Gal and Bargal 2002, 437; Hogan 2005). Tempting as it may be to equate critical junctures and change, this view is not tenable with the emphasis on structural fluidity and heightened contingency which are the defining traits of critical junctures. Contingency implies that wideranging change is possible and even likely, but also that re-equilibration is not 7 For encompassing typologies of models of institutional change, see Streeck and Thelen (2005); Pierson (2004: 134 ff.); Thelen (1999, 2003, 2004). 8 Pierson categorizes 'causal chain' explanations as 'long term cause - short term outcome' accounts (lower left quadrant of his table). We contend that insofar as such accounts ground their explanation at a critical juncture, it is more plausible to treat that (short term) critical juncture as the cause and the causal chain that it unleashes as part of a long-term outcome. 10

13 excluded. If an institution enters a phase of fluidity, in which several options are possible, this does not exclude the possibility that the outcome of the critical juncture will involve the restoration of the pre-critical juncture status quo. Hence, change is not a necessary element of a critical juncture. If change was possible and plausible, considered and ultimately rejected, in a situation of high uncertainty, then there is no reason to discard these cases as non-critical junctures. Moreover, ignoring the near misses of history would actually deprive scholars of important and interesting negative cases with regard to the outcomes they seek to explain. Capoccia (2005), for example, analyzes several cases in which governing elites of European democracies, during the interwar years, took key decisions that played an important role in avoiding the breakdown of democracy. By extending his analysis to both crises that resulted in regime breakdown, and others that resulted in re-equilibration of democracy, he achieves more leverage to identify the key actors, events, decisions and their interconnections, than he could have with an analysis limited to cases of breakdown. Power asymmetries and key actors Although most of the literature on path dependence in comparative politics purports to draw inspiration from institutional economics (e.g. Arthur 1994; David 2000; North 1990), typical accounts of path dependence in institutional economics do not resort to the dual conception of institutional development typical of the use of this framework in political science and sociology. Rather, they refer to situations in which a host of small events sets in motion a process characterised by increasing returns. For example, Arthur explains: Suppose in a certain island cars are introduced, all at more or less the same time. Drivers may choose between the right- and left-hand sides of the road. Each side possesses increasing returns: as a higher proportion of drivers chooses one side, the payoff to choosing that side rapidly rises. Casual thought tells us that we would observe a good deal of randomness to the proportions initially driving on each side, but that, if one side by chance got sufficiently ahead, other drivers would fall in on this side, so that eventually all cars would drive on (would allocate themselves to) the same side of the road In such a situation the actual outcome would likely be decided by a host of small events outside our knowledge drivers reactions, dogs running into the road, the timing or positioning of traffic lights (Arthur 1994: 14). 11

14 While this passage describes the properties of lock-in (or inflexibility) and non-ergodicity (or path dependence) in a vivid fashion, it is apparent that the genetic moment of this non-reversible process is conceptualized differently than 'critical junctures' normally are in historical institutionalism. The phase of fluidity (e.g. when there is still uncertainty as to which side of the road one should drive) is conceptualized as a series of microscopic small choices by actors, which largely for random and unpredictable reasons eventually tip one way or the other, leading the system into a trapping region (David 2000: 10). The typical implications of critical junctures in political analyses are unlikely to respond to that model. In theory, we could certainly conceptualize political change as the result of a series of random small events, which then tip one way or the other, but this would conceal a key dimension of politics: power. What selects an institutional equilibrium during a phase of institutional fluidity are decisions of some political actors, who find themselves in a situation of power, and are therefore freer and more influential than others. For this reason, political science analyses of critical junctures most often focus not on random small events, but instead on decisions by particularly influential actors leaders, bureaucrats, judges and examine how they steer outcomes toward a new equilibrium. 9 Indeed, if we return to the question of traffic flows, it is striking that in actuality, decisions of political elites, rather than the consequences of a host of microscopic decisions by drivers, often determine the direction of traffic. 10 IV. Methods On the basis of the above, in essence we can consider the analysis of critical junctures as the analysis of decision-making under conditions of uncertainty. The methods adopted should therefore reconstruct, in a systematic and rigorous fashion, each step of the decision-making process, identify which decisions were most influential, what 9 We do not mean to suggest that accidental concatenations of unrelated, contingent, events (so-called Cournot effects Pierson 2004, 57) cannot play an important role in influencing the outcome of a critical juncture. What counts as contingent and unrelated, however, depends on the theoretical framework adopted, while focusing on the range of options available to actors offers a more consistent basis to analyze contingency in critical junctures. Events can obviously play an important role in influencing such decisions and their consequences. 10 Even a cursory survey of the history of the direction of traffic in Europe shows that the establishment of a standard had very little to do with drivers decisions and much to do with political decisions. For example, in France, after some years of fluidity and uncertainty as which side of the road carts and other vehicles should drive, an official keep-right rule was introduced in Paris in Later, Napoleon's conquests spread the new French standard of driving on the right to Belgium, the Netherlands, Luxembourg, Switzerland, Germany, Poland, Russia and many parts of Spain and Italy. 12

15 options were available and viable to the actors who took them, and clarify both their impact and how they connect to other important decisions. Several methods are particularly apt to the task at hand: process-tracing (George and McKeown 1985; George and Bennett 2005), systematic process analysis (Hall 2003), analytic narratives (Bates et al 1998; Levi 1999 and 2002), and in general any form of structured, theory-guided narrative. A further key element of the analysis of critical junctures is the analysis of contingency: during critical junctures decisions are taken in situation of high uncertainty and unpredictability, given the relaxation of the normal structural and institutional constraints on action. Taking contingency into account requires researchers to analyze what happened in the context of what could have happened (Berlin 1974, 176), a task for which we have to complement our hindsight perspective (useful to identify moments of change) with a foresight one, which allows reconstructing not only what the consequences of actual decisions were, but what consequences other, viable choices plausibly would have had. 11 This move takes us to into the realm of counterfactual analysis. Below we elaborate on these two methodological tools, counterfactual analysis and theory-guided narrative Counterfactual analysis Leading historians and philosophers have long dismissed analysis based on counterfactuals as virtual history. (e.g. Carr 1986). Recently, counterfactuals have been given back their due role in the social sciences. 12 Several contributions have highlighted the important role that the construction of counterfactuals can play in increasing the number of cases in small-n research (e.g. Fearon 1991), in constructing causal arguments (e.g. King et al. 1994), and in assessing the causal impact of specific factors on historical outcomes (e.g. Immergut 2005; Lebow 2000a and 2000b; Sen 2002; Tetlock and Belkin 1996). This literature has elaborated logical and methodological rules to assess the plausibility of a counterfactual argument, differentiating counterfactuals with good heuristic value from those that belong to the thought-provoking but insufficiently rigorous realm of virtual history. While putting forward a whole roster of criteria (including clarity and logical consistency), this literature suggests that to be plausible, counterfactuals must respect 11 A foresight perspective is obviously necessary to identify correctly those critical junctures that did not in the end lead to change. 12 The same trend is observable in history and historiography (e.g. Bulhof 1999). 13

16 first of all the criterion of theoretical consistency. Mahoney explains that analysts should focus on a counterfactual antecedent that was actually available during a critical juncture period, and that, according to theory, should have been adopted (Mahoney 2000, 513; see also Mahoney and Goertz 2004). Historical consistency is also crucial. Also known as the minimal-rewrite rule, this criterion constrains counterfactual speculations in several ways: for example, by considering only policy options that were available, considered, and narrowly defeated by the relevant actors; or by ruling out counterfactuals in which the antecedent and the consequent are separated by such wide stretches of time that it is implausible that all other things would remain equal (Tetlock and Belkin 1996, 23-24). Several authors have highlighted the connection between the analysis of contingency during critical junctures and counterfactual thought experiments (see e.g. Mahoney 2000, 513). The focus on the role of political actors and their decisions during critical junctures is amenable to plausible counterfactual thought experiments, which can be supported by the historical record (e.g. Lebow 2000a). 13 The difference between a factual and a counterfactual argument as regards the availability of historical sources should not be exaggerated: there may be as much historical evidence on decisions that were taken as on those that were considered, discussed, and ultimately discarded. Similarly, there may be enough evidence to produce informed speculations at least on the immediate consequences of the decisions that could have realistically been taken (e.g. Lebow 2000b). Theory-guided narrative The use of narrative is ubiquitous in historical accounts in the social sciences. While perhaps the most prominent examples of the use of historical narrative focus on long-term processes, the use of narrative is arguably even more important in the analysis of shorter phases such as critical junctures. To use Polanyi's expression, in the analysis of critical periods time expands and so must our analysis (Polanyi 1944, see also Turner 1999, 302). Recent theoretical and methodological literature in both history and the social sciences has shown large consensus that narrative does 13 As Turner points out, reducing history to impersonal causes gives rise to a deterministic version of the past that lends a spurious air of high probability to what happened and blots out the effects of contingency that spring from immediate circumstances and individual choices (1999, 305). Affected by this certainty of hindsight bias, structural explanations...fail to recognize the uncertainty under which actors operated and the possibility that they could have made different choices that might have led to different outcomes' (Lebow 2000a, 559; see also Sewell 1992, 2). 14

17 not necessarily equate with undisciplined or ad hoc story-telling. In political science, and in particular in comparative analyses, narratives are often adequately structured by explicit theoretical models (e.g. Büthe 2002, 483). A theoretical model provides criteria to simplify reality and thus drives the construction of the narrative, which focuses on the aspects considered salient by the theory itself: in the case of critical junctures, these will be the main actors, their goals, their preferences, their decisions, and the events that most directly influenced them. Whether a formal or a non-formal approach is used, however, the very nature of critical junctures requires two things. First, the narratives should specify not only the decisions and actions that were taken, but also those that were considered and ultimately rejected, thus making explicit the close-call counterfactuals that render the critical juncture critical. Second, the narrative should reconstruct the actual consequences of the decisions taken, and (as much as the available data allow) the likely consequences of those that could plausibly have been taken but were not. Below we briefly discuss the advantages and disadvantages of each approach. The application of game-theoretical tools to the analysis critical junctures has several advantages. The rigorous specification and formalization of actors payoffs, their available moves and strategies has the potential to make very explicit the alternatives facing the actors and their consequences. Furthermore, a formal approach may spell out at least some of the close-call counterfactuals that could have led to alternative equilibria. The analysis of off-the-path behavior and sub-game perfect equilibria (Weingast 1996) is particularly powerful in this regard. 14 One potential disadvantage of a formal approach is that the need for manageable game-theoretical models may require an excessively impoverished account of complex situations. This becomes more problematic in cases of cross-sectional comparison of critical junctures in different units, where it is normally more difficult to apply a single model to more than one case (e.g. Ferejohn 1991; Greif 1998; Capoccia 1999). A further potential disadvantage is that the specification of the game being played may not be 14 In his critique of the analytic narratives approach, Carpenter (2000) maintains that making explicit the counterfactuals by applying formal methods to historical accounts is of no added value, since it only specifies some of the possible (potentially infinite) counterfactuals. Carpenter misses the distinction between plausible and non-plausible counterfactuals. While game-theoretical modelling might not make explicit all the plausible counterfactuals, it can clarify the most plausible ones and exclude others as non-plausible. 15

18 satisfactory, if it does not demonstrate empirically that the hypothesized causal mechanisms were actually at play (George and Bennett 2005, 208). 15 Non-formal narrative accounts of institutional crises have a long tradition (e.g. Linz and Stepan 1978). Explicit reference to process-tracing (George and McKeown 1985; George and Bennett 2005, ) is increasingly common in such scholarship (e.g. Capoccia 2005). Tracing the process that led from a situation in which several options were open to a new equilibrium based on the choice of one of them is a flexible enough methodology: it can easily be applied to different units of analysis, can account for the paths not taken, and can offer a stylized but compelling reconstruction of the key decisions and choices that produced the final outcome. 16 The potential disadvantages of a non-formal approach mirror, to some extent, the advantages of formal approaches: while the narrative can be more respectful of the historical record, it may include too much detail, thereby sacrificing parsimony or elegance. V. Comparing Critical Junctures Multiple narratives and cross-sectional comparisons Critical junctures, like any other concept, can be compared along several dimensions of variation and can be part of different kinds of comparative research designs (e.g. Sartori 1991). The typical research design used in comparative arguments involving critical junctures is based on a theoretical framework that identifies similar historical processes in different units (e.g. countries) that involve critical junctures in which the same kind of actors act in a similar strategic environment and face similar challenges. Variation normally comes from the contingent outcomes of decisions and strategic interactions during critical junctures. In this context, comparing processes of development that involve critical junctures (or producing multiple narratives -- Büthe 2002) presents several advantages. First, what is a counterfactual argument in one unit may actually be a factual argument in another. In other words, if critical junctures take place in similar units and in similar conditions, then different decisions of the same actors can give 15 This is one of the main points in critiques of Analytic Narratives -see e.g. Elster (2000), or the symposium in Social Science History (4/2000). 16 It should be noted that process tracing is not incompatible with the use of formal methods and rational choices analysis (George and Bennett 2005, ). 16

19 rise to different outcomes, allowing variation and increasing the overall leverage of the analysis. Second, this facilitates the identification of negative cases, that is, junctures that, although presenting the same characteristics of structural fluidity and actors prominence, do not actually give rise to sweeping change. Third, comparing similar junctures (possibly with different outcomes) helps focus on the important actors, moments, and choices, while omitting less relevant contextual details. This kind of design is typical of both the classical macro-historical work on critical junctures mentioned above and of more recent comparative work. The earlier scholarship, however, is largely couched in terms of analysis of whole periods rather than specific decisions. Recent comparative studies by Kalyvas (1996), on the emergence of Christian Democracy in Europe, and by Capoccia (2005), on democratic crises in inter-war Europe, instead place the decision-making process at the center of the analysis. Both studies the first adopting a formal approach and the second a nonformal one, compare different cases in similar contexts, focus on the same set of actors and decisions in each case, and achieve variation on the dependent variable by including cases of change and non-change as results of similar moments of fluidity. A possible caveat for this research design emerges if the comparison involves similar junctures happening in different spatial units at different points in time. This can have an impact on the independence of at least some the cases being compared, as a consequence of political learning (Bermeo 1992). If the junctures are similar in some important trait (e.g. crisis of a democratic regime due to the rise of Fascist parties), but they happen at different points in time, it is likely that the actors involved in one of them in a certain unit will know about the outcomes of the others happened earlier, and adjust their behavior accordingly (Büthe 2002). In this case the research will have to account for the influence of earlier junctures on later ones. Critical-ness and longitudinal comparisons Longitudinal comparisons, that is comparisons of two or more critical junctures argued to explain an outcome in the same unit of analysis, present different problems. Rival explanations may identify different historical moments that they claim are critical junctures with respect to an outcome. Two scholars supporting rival claims may present convincing causal arguments, counterfactuals and evidence, and 17

20 ultimately academic audiences may be convinced that both junctures are critical in their own way. 17 How can we assess such rival arguments? Assessing just how critical a critical juncture is requires an operationalization of critical-ness. In our view, critical-ness is measured by two factors, which we call probability jump and temporal leverage. The probability jump measures the change of probability of the outcome of interest which is connected with the juncture, and has two components: first, the change (increase) in probability of the outcome of interest in relation to its probability at the lowest point either before or during the critical juncture; 18 second, how close the probability of the outcome of interest post-critical juncture is to 1. The greater the probability of an outcome at the conclusion of a critical juncture relative to its probability at the lowest point during the critical juncture, and the closer that post-critical juncture probability is to 1, the more critical the juncture. The temporal leverage is the ratio of the duration of the impact of the critical juncture to the duration of the juncture itself. The higher the ratio, the more critical the juncture. In a formalization, these factors can be combined, weighted and discounted in several ways to form a single measure of critical-ness. Below we suggest one possible strategy: Y = Outcome of Interest P y = Lowest probability of outcome Y before or during critical juncture P y' = Probability of outcome Y after critical juncture T x = Duration of critical juncture T x' = Duration between end of critical juncture and outcome Y CJ y = Critical-ness score, with respect to outcome Y CJ y = ( P y ' (1 P P y ' y ) ) * Ln T T x ' x 17 See for example Mahoney s (2002) remarks on Yashar s (1997) interpretation of regime development in Costa Rica and Guatemala. Lynch s work on the development of pension systems in Italy and the Netherlands includes the analysis of two different critical junctures in each country (Lynch 2002). 18 One must consider the lowest probability of the outcome before or during the critical juncture rather than simply before it because, as explained above, not all critical junctures result in change: considering only the probability before and after the juncture would lead analysts to ignore the criticalness of critical junctures that result in re-equilibration. 18

21 For any CJ y >0, as CJ y increases, the critical-ness of a critical juncture with respect to Y increases. 19 Our measure of temporal leverage captures the notion that a critical juncture is more critical, the briefer it is relative to the duration of the causal process or outcome it purports to affect (i.e. where values of T x' /T x are high). 20 Our conceptualization of the probability jump combines, on the one hand, how much the probability of the outcome increased between its low point during the critical juncture when several paths of development were equally open and plausible and its status after the critical juncture (P y' -P y ); and on the other, how close the probability of the outcome of interest was to 1 after the critical juncture is concluded [1/(1- P y' )]. 21 The measure is designed to gauge the overall impact that the choices made during the critical juncture had on the likelihood of the outcome in question. The point here is not to suggest that historians and political scientists should calculate critical-ness scores for their arguments. 22 Rather, the model is simply a heuristic device that can make us more conscious of what we already are doing implicitly in rhetorical battles between rival historical narratives. For example, the model clarifies that a critical juncture should be shorter than the path-dependent process that it initiates. The model also captures why decisions taken years before an outcome, which raised the probability of that outcome considerably (say from.1 to.6), could be deemed more critical than decisions taken hours before the outcome, which raised its probability from.9 to.95. We discuss practical applications of our measure of critical-ness later in the article. Apart from being a tool to compare rival arguments, this operationalization of critical-ness has the potential to contribute to a more rigorous use of the concept of critical junctures itself. First of all, it can help scholars to concentrate on those junctures that are most critical. Table 2 conveys the same intuition of the formula above by broadly summarizing the variation in critical-ness scores that the model 19 If CJ y 0, then CJ was not a critical juncture with respect to outcome Y. 20 We take the natural log of (T x' /T x ) in order to discount the effect of time. Otherwise, critical junctures occurring in the distant past would produce the highest critical-ness measures,, even if they had a very modest impact on the probability of the outcome. 21 If the outcome of interest happens immediately after the end of the critical juncture, the probability jump fraction would have a denominator of 0, which would give an irrational number. However, in arguments built on increasing returns and path dependence, the probability of the outcome of interest at the beginning of the path cannot equal one, as this would deny the very logic of self-reinforcing mechanisms. 22 While historical arguments relied on assessments of the likelihood of various outcomes, it is obviously problematic to assign precise probabilities to predictions in historical explanations (e.g. Weber 1949, 183). 19