Program in Comparative-Historical Social Science (CHSS)

Transcription

1 Program in Comparative-Historical Social Science (CHSS) Northwestern University Buffett Center for International and Comparative Studies Working Papers Academic Year Working Paper 3 March 2007 CONTEXT AND CAUSAL HETEROGENEITY IN HISTORICAL ANALYSIS * Tulia G. Falleti** Assistant Professor Department of Political Science University of Pennsylvania falleti@sas.upenn.edu Julia Lynch Assistant Professor Department of Political Science University of Pennsylvania jflynch@sas.upenn.edu * Authors are listed alphabetically and both contributed equally to the elaboration of this paper. We are greatly indebted to Bear Braumoeller, Steven Hanson, Evelyne Huber, and Ian Lustick for their extensive and very helpful comments on earlier versions of this paper. This paper was presented at the 2006 Annual Meeting of the American Political Science Association (in Philadelphia, PA, August 31 - September 4, 2006) and at the Comparative Historical Analysis Study Group, Center for European Studies, Harvard University, November of **Ph.D., Northwestern University Department of Political Science, 2003.

2 About the Program in Comparative Historical Social Science (CHSS): Co-sponsored by the Departments of Political Science and Sociology, the Program in Comparative-Historical Social Science (CHSS) provides an institutional setting for faculty collaboration and graduate student training in comparative historical research. Students in the program complete their Ph.D. in either political science or sociology, but also receive a certificate from the University for expertise in the interdisciplinary area of CHSS. The program provides students with a common coursework structure integrated with their departmental curricula; resources for student research, including travel abroad; interdisciplinary venues at which to present work in progress and receive feedback; and opportunities for collaborative research. For additional information consult the program website at The CHSS Working Paper series will offer CHSS students and faculty a venue to have their work in progress formally available to scholars within and outside of Northwestern University. The series will be directed by Professors Edward Gibson and James Mahoney. It will feature papers focused on comparative and/or historical work, very broadly defined. Students and faculty who would like to have their papers considered for this series should contact Gibson at egibson@northwestern.edu and Mahoney at james-mahoney@northwestern.edu Working Papers: Academic Year No. 1/07 Edward Gibson, Associate Professor, Political Science Department Julieta Suarez-Cao, Ph.D. Candidate, Political Science Department "Competition and Power in Federalized Party Systems." January 2007 No. 2/07 Arthur Stinchcombe, Professor, Sociology Department "Competition and Power in Federalized Party Systems." March 2007 No. 3/07 Tulia G. Falleti, Assistant Professor, Political Science Department, University of Pennsylvania Julia Lynch, Assistant Professor, Political Science Department, University of Pennsylvania "Context and Causal Heterogeneity in Historical Analysis." March 2007

3 Abstract: Political scientists largely agree on the usefulness of causal mechanisms to open the black box that connects inputs and outcomes. Yet often overlooked in discussions of causal mechanisms is their particular relationship to context and the impact that context has on explanation. In this paper, we define causal mechanisms as portable concepts that can travel from one context to another. In turn, we define context as the relevant aspects of a setting (analytical, temporal, spatial, or institutional) in which an array of initial conditions leads to an outcome of a defined scope and meaning via a specified causal mechanism or a set of causal mechanisms. Drawing from these definitions, we study the ways in which causal mechanisms and layered contexts interact, discussing the implications for causal explanation and theory-building of different approaches to periodization in historical analysis. Keywords: Causal mechanism, context, critical juncture, historical institutionalism

4 Truth does not have to be timeless. 1 Ignoring context can be highly misleading, as can leaving the notion of context too vague. 2 In the years since the publication in 1983 of Jon Elster s Explaining Technical Change, a stimulating discussion of causal mechanisms has taken place in the social sciences. While there is little consensus on the appropriate definition of the concept 3, we see (and in the first part of this paper advocate) some convergence upon a notion of social mechanisms as something other than intervening variables filling in the black box between inputs and outputs. Rather, social mechanisms should be viewed as relatively abstract concepts that can travel from one specific instance or episode (Tilly 2001, 26) of causation to another. In this paper we seek to demonstrate that this understanding of mechanisms implies that it is not the mechanism itself that causes an outcome, but rather the interaction between a mechanism and a given context. We then ask what are the consequences for historical institutionalist analysis of this understanding of social mechanisms. We focus in particular on the problems that arise for historically-oriented scholars in periodization and selecting starting points, both of which are central to specifying the temporal context within which a causal process plays out. And we emphasize the particular difficulties that attend to these tasks when we understand context to be composed of multiple, unsynchronized layers of institutions, policies, and background conditions. We close the paper by offering some partial and rather tentative 1 Paul Diesing (1991) How Does Social Science Work? Reflections on Practice, 91, cited in Davis (2005, 168). 2 Bunce (2003, 184). 3 In fact Norkus (2005, 350) bemoans the proliferation of definitions and warms that ill-informed mechanisms talk may come to replace theoretically-informed discussion of the concept a charge of which, we hope, we remain innocent. 1

5 solutions to these problems, centering on the goal of building middle-range theories by making theory-guided choices about contextualization and periodization. CAUSAL MECHANISMS AND CONTEXT Despite a growing interest in causal mechanisms in the social sciences, expressed equally by scholars who subscribe to different epistemological and methodological traditions, there is very little consensus in the literature about what causal mechanisms are. Mahoney (2001, ) identifies twenty-four definitions of causal mechanisms proposed by sociologists, political scientists, and philosophers of science in the last thirty five years, and even more definitions can be added to that list, some of which we discuss below. In political science, the plurality of definitions of causal mechanisms has, however, disguised some underlying similarities. 4 First, causal mechanisms are most often conceptualized as links between inputs, or independent variables, and outcomes, or dependent variables. They serve to open the black box of law-like or probability statements that simply state the concurrence or correlation of certain phenomena or events. Statements of the type if I then O (I O) become if I, through M, then O (I M O). Second, most definitions of causal mechanisms sustain that they should apply to units of a lower level of aggregation than the level of the phenomena the researcher seeks to explain. In other words, a macro-input through the operation of a micro-mechanism leads to a macro-output (Macro-I Micro-M Macro-O). Finally, 4 In other disciplines such as history, sociology, and even public policy, debates about causal mechanisms seem to have followed slightly different contours, perhaps reflecting historical differences in the dominant ontological and epistemological positions held in these disciplines as compared to political science. 2

6 and partially as an extension of the second common trait, most definitions of causal mechanisms used in political science are embedded in the methodological individualist paradigm, such that explanations of macro-level phenomena ultimately rest on mechanisms that apply to individual agents, their psychologies, or their cognitive skills. In this section we discuss each of these common factors, explain our position toward them, and elaborate what we see as a more tenable definition of causal mechanism and its relation to context. From I M O to I M + C O There is little disagreement, even among scholars subscribing to different research traditions within political science, about the usefulness of causal mechanisms for opening the black box that connects inputs and outputs. For example, King, Keohane, and Verba argue that an emphasis on causal mechanisms makes intuitive sense: any coherent account of causality needs to specify how its effects are exerted (1994, 85-6), and Kitschelt sustains that [t]o accept something as a cause of a social phenomenon, we must identify the mechanism(s) that brought it about (1999, 8). For King, Keohane, and Verba, the identification of mechanisms is an operational procedure (1994, 87) consisting of connecting the original posited cause and the ultimate effect in a causal chain of intervening variables. Thus, greater minority disaffection under a presidential regime and lesser governmental decisiveness under a parliamentary regime are some of the hypothetical mechanisms that explain the effect of a political system type (presidential or parliamentary) on democratic stability (King et al. 1994, 86). Interested in explaining the increasing polarization of regime types among post-communist 3

7 countries some moving toward becoming full democracies while others slide into increased authoritarianism Kitschelt argues that the first mechanism leading to different post-communist regime outcomes has to do with the presence or absence of ingredients of professional versus patronage bureaucracy in communist polities, and the second chain of conditions and mechanisms ha[s] to do with the organization of civil society before and under communism (1999, 24, 27). Although the two works draw from different methodologies (King, Keohane, and Verba s cite a large-n study and propose the use of experimental design in their example, whereas Kitschelt utilizes historical narrative combined with typological and path dependent analyses), they similarly treat causal mechanisms as intervening variables that can be directly observed and measured. However, as Mahoney convincingly argues, this notion of mechanisms as intervening variables ultimately falls back on correlational assumptions: [A] variable s status as a mechanism as opposed to an independent variable is arbitrary a correlation is explained simply by appealing to another correlation of observed variables (Mahoney 2001, 578). In our view, as in Mahoney s, mechanisms cannot simply be an attribute of an input, not even an intervening input e.g. something describing the size, situation, orientation, etc. of something else that is (even very proximate to) an outcome. Rather, mechanisms are relational and processual -- they link inputs and outputs by explaining how you get from one to the other. Both attributes of inputs in a causal chain (independent and intervening variables) and the causal mechanism(s) that link these inputs to the outcome (dependent variable) are often lumped together casually as causal factors. However, it should be clear from the preceding 4

8 discussion that while we see all causal mechanisms as causal factors, not all causal factors are causal mechanisms. While we agree with Mahoney s (2001) claim that causal mechanisms are not reducible to intervening variables, we take issue with some of the implications he derives from this axiom. In seeking to move away from the notion of mechanisms as variables, Mahoney requires that a causal mechanism be an unobserved entity that when activated generates an outcome of interest. This definition assumes that [ ] if the mechanism actually operates, it will always produce the outcome of interest (Mahoney 2001, 580). Two implications derive from this definition. First, as unobserved entities, [c]ausal mechanisms are posited relations or processes that the researcher imagines to exist; they do not refer to any particular set of empirical conditions (Mahoney 2001, 581). Second, mechanisms, according to Mahoney, are the bases of deterministic law-like statements. While we agree with Mahoney that causal mechanisms cannot be reduced to intervening variables without losing their explanatory leverage, we differ from him as to whether mechanisms are (a) necessarily unobservable and (b) deterministic in their operations. We understand causal mechanisms to be portable concepts that can travel from one sequence or trajectory of events or context to another. As such, mechanisms are of a higher level of abstraction than intervening variables and they are not specific to any singular type of process. They refer to the underlying logic that connects initial conditions and outcomes and accounts for how processes evolve. The same mechanism can thus operate in different contexts. But because mechanisms interact with the context in which they operate, the outcomes of the process cannot be determined a priori by 5

9 knowing the type of mechanism that is at work. Several implications derive from our definition. 1. Causal mechanisms are portable concepts. Hedström and Swedberg (1998) provide a good example of an individual-level mechanism that serves as the explanatory link connecting behavior and outcomes in three different sociological theories. First, Merton s (1968 [1948]) theory of self-fulfilling prophecy states that an initially false situation evokes behavior that eventually makes the false conception come true. Second, Coleman, Katz, and Menzel s (1957) study of the diffusion of a new drug found that physicians position in various professional networks influenced the diffusion process of the new drug being introduced. Finally, Granovetter s (1978) threshold theory of collective behavior argued that an individual s decision whether or not to participate in collective behavior often depends on how many other actors have already decided to participate. According to Hedström and Swedberg, the same individual-level causal mechanism, namely belief-formation, operates in the three theories as well as in the different contexts and the effects that they evoke. In their own words: the core characteristic of these theories is the general belief-formation mechanism which states that the number of individuals who perform a certain act signal to others the likely value or necessity of the act, and that this signal will influence other individuals choices of action. It is this belief-formation mechanism that is at the heart of the self-fulfilling prophecies of Merton, the network effects of Coleman, and the bandwagon effects of Granovetter. On the fundamental level of mechanisms, the run on the bank, the prescription of the drug, and the emergence of the collective movement, all are analogous (Hedström and Swedberg 1998, 21). 6

10 Another example of a portable causal mechanism is boundary control. Rokkan (1983) proposed this concept to analyze the defensive strategy of cultural peripheries against encroachments from the center. In a recent article, Gibson (1995) adapted and further elaborated upon the same concept to refer to the strategies of subnational authoritarian leaders in maintaining their regional hegemonic power in the context of nationally democratic polities. Finally, although he does not use this concept, the same mechanism of boundary control could be applied to explain the strategies of the barons of the Italian academic system in keeping it fairly insulated from the rest of the world, as described by Gambetta (1998, 108). In all these cases, the same concept is used to refer to the strategies of either individual or collective actors who play in different contexts. Despite radical differences among the three contexts, they all constitute subunits of larger entities, in which those who exert local domination seek to protect themselves from external influences. Although the specific contexts are different, for the purposes of the causal explanations advanced by the authors, the contexts are analytically equivalent. Hence, whereas boundary control is not a universal mechanism, it is portable; and what makes it portable is precisely the analytical equivalence of the contexts to which the mechanism can be applied. As can be appreciated in these two examples of causal mechanisms, they are neither context-specific nor reducible to intervening variables. The Appendix provides some further examples of causal mechanisms found in recent works of political methodology and comparative politics. 2. Causal mechanisms do not have to be unobservable. Scholars have different opinions on the issue of observability of mechanisms. If causal mechanisms are defined 7

11 as intervening variables, they will tend to be operationalizable and measurable. They can be observed. If, instead, mechanisms are defined as the final movers of outcomes in the world (Mahoney 2003, 5) or as ultimately unobservable physical, social, or psychological processes through which agents with causal capacities operate (George and Bennett 2005, 137), then unobservability is inevitably related to a higher ontological status assigned to mechanisms when compared to variables. According to the latter conception, if improvement of measurement techniques makes it feasible to observe mechanisms, they stop being mechanisms and become variables. At that point, the border between the observable world and the unobservable ontological level where causal mechanisms reside has moved, and explanations via causal mechanisms should be sought at a lower (or at least different and unobservable) level (George and Bennett 2005, 143). While we disagree with any view that reduces causal mechanisms to intervening variables, we argue that unobservability is not a definitional element of causal mechanisms. Whereas many causal mechanisms are indeed unobservable (e.g. adaptive expectations, belief-formation, or power reproduction, just to name a few), others can be measured empirically. Take for example the positive feedbacks mechanisms considered by Pierson (2000). Mechanisms such as coordination effects or large set-up costs (see the Appendix for definitions) could easily be operationalized and measured. The same could be said of social system mechanisms such as technical interrelatedness or system scale economies proposed by David (1985). Yet, none of these four concepts is reducible to a context-specific indicator. They can be measured and observed, but they also maintain their portability property and travel to different contexts and situations. 8

12 3. The same causal mechanism can lead to different outcomes. We depart from the view that mechanisms lead to deterministic outcomes (Bunge 1997; Mahoney 2001). While we are interested in mechanisms as portable concepts and what is constant in them, we agree with Elster that they are triggered under generally unknown conditions or with indeterminate consequences (1998, 45). To us, the interaction between mechanism and context is what determines the outcome. Given an initial set of conditions, the same mechanism operating in different contexts may lead to different outcomes, as represented in schematic form in Figure 1. Figure 1. The I M O model in Different Contexts Context A Context B I M O a I M O b In other words, the indeterminacy is not in the mechanism but in the context. Pawson (2001) takes a similar approach when he states that [w]hether [a] mechanism is triggered depends on context (5), and warns policy-makers about the risk of mechanically transferring successful policy programs to contexts in which the underlying mechanism may not lead to the same outcome. This is the reason why it is so important to disentangle causal mechanisms from context, and at the same time provide very clear definitions of the mechanism(s) at work and the context in which they are operating. 9

13 Degree of conceptual abstraction Having defined mechanisms as portable concepts, we recognize that a hierarchy of mechanisms by their level of conceptual abstraction can be established. 5 There are umbrella mechanisms of higher levels of conceptual abstraction that may encompass other portable mechanisms. For example, increasing returns is a more general mechanism than positive feedback, which in turn is more general than path-dependent technological innovation or large set-up costs. Or, to return to the example of boundary control, this mechanism is more general than those of parochialization of power, the nationalization of influence, or monopolization of national-subnational linkages identified by Gibson (1995, 11) in the relationships between centers and peripheries. In our view, the most appealing causal explanations are those that open two boxes. First, they open the black box that connects inputs (I) and outcomes (O). Second, they also open the box of higher level of abstraction mechanisms and identify the mechanisms of lower levels of abstraction (or sub-mechanisms) that offer a more accurate depiction of how we get from I through sub-m to O. But this submechanism, while closer to the events of interest than the umbrella-mechanism, remains a portable concept and cannot be simply reduced to an intervening variable. Clearly, mechanisms, like other social science concepts, suffer from the intensionextension tradeoff identified by Sartori (1970). Our preference for identifying mechanisms at a lower level of abstraction is linked as well to our conviction that because the outcomes of mechanisms are linked to the context within which the 5 Note that this is a different issue than whether or not mechanisms should always apply to units of lower level of analysis than the phenomena we are interested in. 10

14 mechanism operates, middle-range theorizing is the appropriate goal for mechanismic explanations. From Methodological Individualism to Scopes of Application Many of the scholars who seek to unearth the mechanisms that link causes and effects work within a paradigm of methodological individualism (e.g., Boudon 1998; Elster 1998, 47; Kitschelt 1999, 8). Boudon writes: [a]nalyzing social mechanisms requires at least ideally making them the outcome of individual beliefs, actions, attitudes (1998, 199). In political science, the concept of rationality as an individuallevel mechanism permeates a large body of literature. More recent studies in public opinion and electoral behavior are also moving toward the identification of cognitive and psychological individual-level mechanisms (such as learning) that explain macro-political effects. However, those working outside of a strictly methodological individualist paradigm are more apt to recognize that not all causal mechanisms must operate at the level of the individual. McAdam, Tarrow and Tilly s (2001, 25-6) relational, environmental, and cognitive mechanisms may occur at a variety of levels of analysis. Similarly, Ekiert and Hanson (2003, 15-48) identify different causal mechanisms according to the level of analysis (structural, institutional, or interactional) and type of context (temporal or spatial) to which they refer. And George and Bennett (2005, 142) argue that the appropriate level of analysis of causal mechanisms will vary depending on the particular research question and the objectives under investigation. As Mahoney (2003, 5) writes, mechanisms need not be specified at any particular level of 11

15 analysis, and micro mechanisms are not necessarily superior or more basic than macro mechanisms. Stinchcombe proposes methodological collectivism and states this position even more forcefully when he writes: Where there is rich information on variations at the collective or structural level, while individual-level reasoning (a) has no substantial independent empirical support and (b) adds no new predictions at the structural level that can be independently verified, theorizing a the level of [individual] mechanisms is a waste of time (Stinchcombe 1991, 380). In the Appendix, we identify a number of causal mechanisms and order them according to the level of analysis to which they refer. Some are strictly individually based (such as adaptive expectations or beliefformation), while others apply to collective actors (power reproduction, policy-ratchet effects, layering) or to social systems (large set-up costs, functional consequences). There are also mechanisms that are equally applicable to two contiguous levels of analysis, such as those listed under the individual or collective actors and collective actors or social systems subheadings in the Appendix. Causal Mechanisms in context So far we have defined what we mean as a causal mechanism and have discussed how our definition relates to previous ones. We have established that a causal mechanism may apply to individual or collective actors; it may be unobservable or not; it may be of a higher or lower level of abstraction; but above all, it is portable. That is, it is sufficiently distinct from the context in which it operates to merit the ontological status of a mechanism, rather than simply another set of intervening variables that link input to outcome. 12

16 It should be noted that defining a mechanism as a portable concept does not mean that it will operate in every context. Some mechanisms (e.g. rationality, in the sense of individuals acting to maximize their perceived utility) seem quite general, and are even presumed by some to operate universally. But many other mechanisms are not nearly so ubiquitous. In some cases, contextual factors may allow for some mechanisms to apply, but not others. For example, Lynch (2006) argues that a positive feedback mechanism links particularistic political competition to elderly-oriented welfare state spending in a context in which welfare state programs are linked to labor market status; but this mechanism does not function fully in a context defined by universal social welfare programs. Other mechanisms may apply only to a subset of all possible contexts. Boundary control, as discussed above, is one such example; another is the circular flow of power in Leninist regimes. 6 Additionally, as we have seen in Figure 1 above, the same mechanism operating in different contexts may produce different results. Hence, the scope of application of a mechanism and the mechanisms essential indeterminacy make context as important in generating the outcome of interest as the mechanism itself. If we view mechanisms as portable concepts that produce indeterminate outcomes, as we argue they should be viewed, then only context and mechanism taken together constitute a full causal explanation. In the next section we define what we mean by context, and problematize the use of critical junctures as a marker of context in historical institutionalist analysis. CONTEXT AND TEMPORALITY What is context? 6 Thanks to Steven Hanson, personal communication, for the latter example. 13

17 Pawson s (2000) middle-range realism posits context as causal mechanism s partner concept (296). Outcomes of causal mechanisms are not fixed, but rather dependent on the context within which they occur. Pawson illustrates with a favourite physical science example of contextual contingency: gunpowder. Gunpowder has a chemical makeup that gives it the potential to explode in the presence of a spark, but it only does so when certain contextual conditions (eg. the amount of humidity in the air or gunpowder) are conducive (296). But what are the elements of context that are likely to affect social mechanisms? If theorizing about social life requires attention to context, to what, precisely, are we supposed to pay attention? Bunge s (1997) notion of systemness provides some clues. According to Bunge mechanisms operate within systems, which are defined by their composition (the set of parts that make up the system), their environment (which Bunge does not define), and their structure (how the constituent parts are connected to each other and to things in the environment that influence or are influenced by the constituent parts) (416). Aspects of Bunge s notion of systemness particularly environment and structure -- contribute to our definition of context. We find it quite useful to note that elements that are not directly on the I-M-O path, but rather reside in some other aspects of the system, may nevertheless affect the functioning of a mechanism and hence the nature of O. Drawing on Pawson and Bunge, we define context broadly as the relevant aspects of a setting (analytical, temporal, spatial, or institutional) in which a set of initial conditions leads to an outcome of a defined scope and meaning via a specified causal mechanism or set of causal mechanisms. From this definition it follows that a causal explanation requires the analyst both to specify the operative causal mechanism, and to 14

18 delineate the relevant aspects of the surroundings -- i.e. those that allow the mechanism to produce the outcome. Herein lies the rub. How can we know what aspects of the context are relevant to the outcome until we have an explanation for the outcome? Our view of causation depends on a definition of context that is tied to the outcome. Does this not simple give researchers license to explain something by selecting in an ad hoc way the contextual factors that contribute to its occurrence? If it were not for the fact that researchers make these kinds of decisions routinely (albeit often in a less than fully conscious way), we might worry. But the case-based research paradigm has at its disposal a variety of tools for determining before the fact what aspects of a context are likely to be relevant to the outcome under study. We turn our attention now to one specific aspect of context, the temporal context within which causal processes play out. We argue that paying more attention to how we define and select the causally relevant aspects of a temporal context will aid us in constructing valid causal explanations. Causation in time Historically-oriented political science research is notable for its theoreticallybased expectation that the temporal aspects of the context within which a causal process plays out matter for explaining outcomes. Sequencing -- when things happen --, either in world-historical time (Wallerstein 1974), in relation to signal events in political development like the development of working-class parties, or to more contingent events or processes closer at hand (e.g. the availability of certain policy models), may affect how and whether a specific mechanism works. For example, Falleti (2005) shows that if a 15

19 process of decentralization begins with a political decentralization type of reform, the process will likely produce a policy-ratchet effect (mechanism) namely, the creation of a group of subnational supporters that will affect the second round of decentralization reforms. This mechanism, however, will most likely be absent if political decentralization occurs after administrative decentralization has taken place. Tempo and duration -- how long things take -- may suggest a likely set of plausible mechanisms. Outcomes that come about slowly, gradually, or after a long lag are likely to be produced by different kinds of mechanisms (policy drift, friction, increasing returns) than those that occur swiftly or suddenly (tipping points, rational choice) (see Pierson 2004, Chapter 3). Indeed, those political scientists who focus on longue durée processes have tended to emphasize structuralist, systems-oriented, and/or macro-social causal mechanisms while scholars interested in the consequences of shorterterm processes often are more attuned to mechanisms posited at the level of the individual or collectivities of individuals (see the Appendix for examples). Our focus here, however, is concerned with a third aspect of temporality: when things start. Starting points have had particular relevance for historical institutionalists because the notion of path dependence, which is at the center of many historical institutional analyses, relies on a well-specified starting point. Historical institutionalist scholars typically use starting points and critical junctures to delineate one context, "before," in which a mechanism (often path-dependent) does not function, from a second context, "after," in which it does function. We argue, however, that the act of periodizing as a way of marking shifts in context is often insufficiently theorized in historical institutionalism, and runs into particular difficulties when confronted with causal 16

20 mechanisms that operate at the aggregate- or structural-level rather than the individual level. Context and periodization If causal mechanisms are themselves portable, but operate in a context-dependent fashion, then in order to develop causal theories we must be able to identify analytically equivalent contexts as well as specify where one context ends and another begins. For historical researchers, the passage of time is often the most obvious indication that a context has changed. So it is no surprise that in historically-informed analyses, periodization plays an important role in the development of causal theories. But even self-consciously methodological works examining periodization in causal analysis often fail to adequately specify how to place the markers that designate contextually homogeneous periods in time, or to specify which aspects of a context must be constant within a given period and which may be allowed to vary. Büthe (2002), for example, issues a plea for more careful attention to the placement of starting- and ending-points in research that utilizes historical narratives as case material. Büthe sees a tension between formal models, which provide an explicit, deductively sound statement of the theoretical argument, separate from a particular empirical context (482), and the analysis of complex causal processes over time, which often involve feedback loops or other forms of endogeneity. Decontextualized formal models may lead to invalid causal claims if they fail to consider the sequencing (when things happen) with which a specific causal mechanism plays out. Büthe sees the analysis of historical narrative as a solution to this problem, but recognizes the difficulty 17

21 of knowing where to start and end a narrative. In particular, he asks how do we delineate a sequence of events so as to justify the imposition of a narrative beginning and end onto a continuous empirical record? How does the imposition of a narrative closure affect the generality of our conclusions? (482). Ultimately, however, he is only able to offer the advice that the specification of the explanandum provides the criteria for choosing the beginning and end of the narrative (488). Where the process to be explained does not have a clear starting point (e.g. an exogenous shock) and/or has not run its course, Büthe himself notes that even this advice will prove inadequate (487). 7 Büthe advocates delineating the beginning of a new context with reference to the onset of the causal mechanism that produces the outcome. Analyses that use critical junctures to delineate the beginning of a period are one example of this strategy. Critical junctures are often defined ex post as the starting point of a path dependent causal mechanism that leads to the outcome of interest. Many analyses situate the critical juncture at the point of some exogenous shock (war, depression, shift in commodity prices, etc.); nevertheless, the most widely read classic examples of critical junctures analysis (e.g. Collier and Collier 1991; Lipset and Rokkan 1967; Moore 1966) embed critical junctures in a richly detailed context, and make it clear that the outcome of the causal process that begins with the critical juncture may also be influenced by a variety of other features of the environment. Collier and Collier (1991, Chap 1), for example, note that the duration of a critical juncture need not be short, and that longer critical junctures 7 Büthe also does not address the issue of how one might periodize within the grand historical narrative in order to achieve more causal leverage. 18

22 often incorporate background conditions and cleavages into the production of the outcomes in question. 8 Some more recent analyses employing critical junctures also recognize that although a juncture may be the starting point of a new "path," it is also a product of what came before (see for example Hacker 2002). However, a new strain of theorizing about critical junctures tends to emphasize the contingency of such moments, which highlights their status distinctive break-points with the previous context. Mahoney (2000), demands that the start of a path dependent process be "contingent", by which he means that the probability of any one particular path/outcome being chosen at the critical juncture is equal to the probability that any other (plausible) path/outcome will be chosen (514). Mahoney eschews the notion that critical junctures represent truly random choices, instead noting that "In the actual practice of research, social analysts will consider an event to be contingent when its explanation appears to fall outside of existing scientific theory," e.g. an unpredictable exogenous shock or a particular decision made by a political actor with a proper name. (514). Regardless of whether we think of critical junctures as truly random or as simply not predictable, however, this formulation departs significantly from Collier and Collier's (1991) notion that critical junctures may have strong links to the past. Similarly, Capoccia and Keleman (2005) emphasize the delinking from context that occurs at a critical juncture: "Critical junctures are characterized by a situation in which for a relatively short phase the 'structural' (i.e. economic, cultural, ideological, organizational) influences on political action are substantially relaxed" (3). Capoccia and 8 In some analyses (e.g. Capoccia and Keleman 2005; Collier and Collier 1991), critical junctures are labeled as causes of the outcomes of interest, but we believe this designation to be inconsistent with notions of causality employed by these same authors. 19

23 Keleman distinguish their conception of a critical juncture, quite rightly, from Lipset and Rokkan's (1967) and Collier and Collier's (1991), both of whom "embed" their critical junctures in antecendent conditions (Capoccia and Keleman 2005, 5) and in so doing downplay the agency of individual actors. The periodization strategies advanced in both Büthe s (2002) framework and in critical junctures analysis following Mahoney (2000) and Capoccia and Keleman (2005) implicitly or explicitly define the initiation of a (path dependent) causal mechanism as the starting point of the context within which the causal process unfolds. In other words, the context surrounding the I M O pathway is bounded precisely by the beginning and end of the causal mechanism (M) of interest. This conceptualization of a starting point is a useful tool for identifying the beginning of a path dependent process and may also highlight mechanisms that take place at the level of individuals or groups of individuals. It is not a good guide, however, to continuity and change in other important aspects of the context in which the causal mechanism plays out and that may have an important effect on the outcome of interest. Lieberman (2001), in an article devoted to periodization strategies in historical institutionalist analysis, helpfully goes beyond critical junctures in his search for possible starting points. Lieberman's typology identifies four types of starting points: the origination of a new institution (which for him equals the independent variable) of interest, or an important change in such an institution; an exogenous shock that changes the conditions in which the institution operates; or a change in some "rival independent variable" present in the background (1019, Table 1). This typology usefully points to the variety of potential markers of the beginning of a new context, which need not all 20

24 coincide with the onset of the mechanism presumed to be responsible for the outcome of interest. It also emphasizes that periodization may be based on activity in numerous layers (proximate institutions, background, truly exogenous events) of the context within which a causal process plays out, a point to which we will return in a moment. But while Lieberman introduces the idea that changes in a variety of different aspects of the context surrounding a causal mechanism may be consequential for the outcome of interest, his typology does not leave room for causal processes that might be generated by interaction or friction between the different aspects of the context. Yet, as Orren and Skowroneck (1994, 321) note, the multiple layers or "orders" of institutions that constitute the polity or context at any given time are not "synchronized in their operations." Rather, these orders "abrade against each other and, in the process, drive further change." Lieberman's strategies for periodization focus the attention on "important events, changes or turning points that can be conceptualized as markers of variation in a potentially important explanatory variable" (1017), which have "potentially important impact on the outcomes under investigation." Lieberman notes that such events are relatively rare: "Within a mass of historical observations, only a few events define a period, whereas most other events and processes are explained as taking place during a period" (1017). This relatively narrow definition is nicely operational, but it also seems to us to simultaneously raise and fail to grapple with the central fact about context, i.e. its multi-layered character. Causation in multi-layered contexts 21

25 Social processes take place in a context characterized by multiple overlapping layers of institutions and structures that govern the relationships between inputs and outcomes. Pawson (2000) cites as an example of this layered social reality the process of writing a check. Checks are "routinely accepted for payment only because we take for granted [their] place within the wider (institutional) rules of the banking system. The capacities that bring custom and order to the transaction do not reside solely with particular objects (checks) or agents (cashiers) but also belong to the institutional regulations (credit), legal constitutions (charters) and organizational structures (chancelleries)" (294). In historical institutionalist analysis, we are likely to be concerned with a variety of contextual layers: those that are quite proximate to the input (I) -- for example the electoral system in a study of the emergence of radical right wing parties --; exogenous shocks quite distant from (I) that might nevertheless effect the functioning of the mechanism and hence the outcome (eg. a rise in the price of oil that slows the economy and makes voters more sensitive to higher taxes); and also the middle-range context that is neither completely exogenous nor tightly coupled to (I) and that may include other relevant institutions and structures (the tax system, social solidarity, demography) as well as more atmospheric conditions such as the rate of economic growth, flows of immigrants, trends in partisan identification, and the like. Lieberman (2001) conceives of this background context as the locus of "rival causes." However, we believe that recent research (e.g. Hacker 2002; Lynch 2006; Streeck and Thelen 2005) bears out Orren and Skowroneck s (1994) contention that the interaction of different layers of context may itself generate important outcomes. 22

26 Steinmo and Thelen (1992, 16-7) illustrate how changes in background conditions may affect the outcome of an institutionally-structured process, even if the direct institutional inputs do not themselves change. Changes in the social, economic, or political context may make previously marginal institutions more central to political life; bring new actors into play who use the same institutions to achieve different ends; or cause the same actors to pursue new goals through existing institutions. In all of these cases, the institutional inputs do not change, the mechanism may or may not change, but the institutional outputs do change because the context has changed. Expanding her view beyond the operation of formal institutions, Lynch (2006) provides an extended illustration of how the multiple layers of context within which a path dependent causal mechanism operates play an essential role in generating the outcome of interest in this case, the extent to which social policies in different countries privilege the elderly over working-aged adults and children. Lynch begins by observing a correlation at the level of the nation-state between the age-orientation of social policies and the way that politicians use social programs to compete for votes. But this correlation is not linked to an obvious causal mechanism, suggesting that a timedependent process may be at work. With a nod to Lipset and Rokkan (1967) and Collier and Collier (1991), Lynch presents her argument as a classic branching tree. Two critical junctures mark choice points in the development of welfare state institutions, with the mode of political competition in a polity pushing towards particular institutional choices and reinforcing those choices over time through positive feedback mechanisms until the next critical juncture arrives. But the outcomes in question, the age-orientation of social policies in 23

27 different countries, cannot be satisfactorily explained within a framework that specifies critical junctures as moments of radical discontinuity. Rather, Lynch argues that processes occurring in three separate layers of context over the relatively longue durée of the twentieth century interact to produce the age-orientations observed in ca The layer of context closest to the causal mechanism in Lynch's argument is the political arena, where the policy preferences of parties and unions take shape, informed in part by the mechanism of voter recruitment (particularistic versus programmatic competition) in a polity. Following a critical juncture around the time of World War II, the political arena is linked via a path-dependent process of increasing returns to a second layer, the institutional arena of social policy programs. Changes in institutional structure over time shape the boundaries of labor market "insider" and "outsider" groups, in a process that shapes the relative weight of spending on different age groups. But the balance of pensioners, workers and children that fall into these groups is ultimately determined by what happens in a third layer of context, composed of slow-moving background processes: population aging, the gradual closure of many Continental European labor markets to younger job-seekers, and the development of public and private markets for old-age insurance. Lynch's explanation for the age-orientation of social spending shows how a single outcome may be determined by the interaction of multiple layers of context, including very slow-moving processes quite far removed from the institutional dynamics that originate in the critical juncture This analysis suggests that in addition to the causal mechanism driving the path-dependent processes of institutional choice following critical junctures, there is another important causal mechanism at work: "policy drift" (Hacker 24

28 2004; 2005), a mechanism that can in fact only operate in a system characterized by multiple layers of relevant context. If political contexts tends to be layered, with processes occurring at different speeds in different layers; and if some mechanisms are characteristic of the interaction of separate layers; then periodization in historical institutionalist analysis should be attuned to the start- and end-points as well as the tempo and duration of multiple processes in multiple layers. Consider a causal process that begins at time t I (for input) with a change in the main institution of interest, which is found in contextual layer L 1, as in Figure 2. A critical junctures analysis would start the clock at time t I, tracing the outcome occurring at time t O (for output) back to the change in the institution in L 1. In this case, the change in this institution follows closely (but not instantaneously) upon an exogenous shock E, which itself spans considerably less time than most other elements in this diagram, but does have some measurable duration. Preceding the exogenous shock and lasting well past the critical juncture at time t I, background condition B exerts a continuous influence on the unfolding of our causal process. A second causal process linked to a change in contextual layer L 2 also predates and persists through the critical juncture, although it should be noted that its start and end points do not coincide neatly with B, either. Another process of potential relevance to O occurs in L 3. Strictly speaking, portable causal mechanisms require that, under the same initial conditions, identical contexts produce identical outcomes. But which context is the relevant one in this diagram? Only the temporal context marked C 5 captures all of the major contextual layers, but it excludes the exogenous shock and resulting critical juncture. Context C 3 includes the 25

29 critical juncture, but misses the beginning and end points of some processes that might prove important. [Figure 2 on facing page] It is clear from this schematic representation of unfolding causal processes in a layered context that a perfect periodization scheme may prove elusive, and that care must be taken when making decisions about periodization to specify which layers of context are relevant and how. The use of critical junctures as starting points may pose particular problems in multi-layered contexts when important processes are not synchronized with start point of I M O pathway to which the critical juncture pertains. Figure 2: Periodization in Multi-Layered Contexts 26