SYMPOSIUM combining game theory models and process tracing: potential and limits david kuehn Institute of Political Science, Heidelberg University, Bergheimer Str. 58, D-69115 Heidelberg, Germany doi:10.1057/eps.2012.9; published online 30 March 2012

Abstract This article argues that process tracing provides a useful, although underestimated, avenue for empirically testing deductive game theoretical arguments. In-depth case analysis allows for a systematic evaluation of the crucial assumptions underlying the models and for making internally valid measurements of the model’s core concepts. Furthermore, process tracing enables the researcher to overcome the weak conceptions of processes encountered in many game theoretic arguments. After outlining the usefulness of combining deductive game theory and process tracing, as well as discussing the limits of such an approach, the article illustrates the argument with an example from substantive research on civil–military relations in new democracies.

Keywords

case study; civil–military relations; game theory; process

tracing

M

ost of the recent interest in Multi-Method Research (MMR) has focussed on the combination of large-n analyses with in-depth case studies and the combination of deductive theoretical models and statistics. In comparison, there is little systematic debate on the promises and problems of combining deductive theorising with qualitative case studies. This dearth of methodological discussion is also reflected in applied research. One review of methodological trends in Political Science found

52

that only 5 to 16 per cent of published MMR articles have actually combined formal models and case studies (Bennett et al, 2003: 377).1 In fact, case studies have been considered useful merely as a ‘first step toward building a connection between a formal model and the empirical world’, but not as a substitute for more rigorous statistical analysis (Morton, 1999: 134). Hence, it is unsurprising that most of these applied studies employ the case study method for purely illustrative purposes (Bennett and Braumoeller, 2006).

european political science: 12 2013 (52 – 63) & 2013 European Consortium for Political Research. 1680-4333/13 www.palgrave-journals.com/eps/

This is also reflected in the methodological discussion on the topic. While for some time there has been a lively debate on the systematic combination of quantitative methods and rational choice modelling (Goldthorpe, 1996; Aldrich et al, 2008), there is little discussion on the methodological promises and problems of combining deductive theory and case study empirical analysis (but see Bates et al, 1998). This article argues that this lack of interest is unfortunate as the combination of case study analysis and deductive theorising can significantly contribute to the accumulation of knowledge in the Social Sciences. Focussing on game theory as one particularly important strand of deductive theorising, the article first identifies analytical rigour and theoretical completeness as inherent strengths of deductive models, but also points out that these desirable principles are in danger if the underlying assumptions are not empirically sound and if the status of the modelled processes is unclear. On the basis of this, the second section shows how process tracing can somewhat alleviate these challenges by empirically evaluating a model’s assumptions and generating theoretical arguments on social processes. The third section discusses and evaluates the limits of this combinatory approach. In the penultimate section, the arguments are illustrated with an example from the study of civil–military relations in new democracies. The article concludes with a short summary.

THE CHARACTERISTICS OF GAME THEORETIC MODELS Game theoretic models are stylised representations of the real world, which are employed for deducing theoretical propositions and hypotheses to explain the phenomenon of interest. They contain assumptions on those entities and their

‘y there is little discussion on the methodological promises and problems of combining deductive theory and case study empirical analysis y’ actions, which are considered relevant for bringing about the explanandum (Morton, 1999: chapter 2). Substantively, game theory focuses on those social phenomena that result from the strategic interactions of two or more goal-oriented actors. As such, it is useful for many political phenomena. In addition to their substantive content, game theoretic models have two further characteristics, which make them attractive for theory building in the Social Sciences: rigour and completeness.2 They impose analytical rigour as modelling forces the researcher to clearly specify all prior assumptions, relevant elements and ‘moving parts’ of the causal argument. This includes the relevant actors; their goals and actions; the contextual factors, which determine the payoffs; and the mechanism of choice, which provides the link between the individual elements of the model (Geddes, 2003: 191). As such, they contain complete theoretical arguments, which not only include statements on the expected co-variation of the independent and the dependent variable(s), but also of the ‘microfoundations’ (Little, 1998: chapter 10) through which the outcome is brought about. While theoretical clarity and completeness are necessary conditions for good theory (Gerring, 2001: chapter 5), the litmus test of any explanatory theoretical argument lies in its concordance to the empirical phenomena it is supposed to reflect. It is here, however, that game theoretic models might confront challenges david kuehn

european political science: 12 2013

53

that undermine the very benefits outlined above. First, because of their deductive logic, models are crucially dependent on the correct specification of the prior assumptions. Since models are abstracted representations, they include many simplifications that are false when compared with the complexities of the real world. As long as the model is a close enough representation of those elements thought to be relevant for capturing the core logic which produces the explanandum, these simplifications are justified (Morton, 1999: 38–41). However, assumptions are a problem if their substantive contents cannot be justified based on our knowledge of the real world: conclusions drawn from substantively false assumptions are logically true but empirically false. This undermines the usefulness of the model for capturing the logic of real world phenomena and for explaining the outcome of interest. For instance, if the researcher assigns wrong preferences to the actors or wrongly specifies their informational resources, the model will yield wrong predictions on these actors’ behaviour. The second problem concerns the modelling of processes. Often, models are not clear on how the modelled sequences relate to actual empirical processes. Of course, all models include assumptions on the sequence of decisions: extensive form and repeated games ‘take sequence into account and highlight its significance for outcomes’ (Bates et al, 1998: 14), and there is a wide variety of models that explicitly model changes over time (e.g., North, 2005). However, often researchers are not clear what exactly the modelled sequences refer to in the real world. In some models, the sequences are thought to be abstracted representations of actual empirical processes (e.g., Bates et al, 1998). In other models, they are considered as depictions of the various steps the participating actors

54

european political science: 12 2013

‘y often researchers are not clear what exactly the modelled sequences refer to in the real world’. go through mentally in order to make their decisions. As such, they are representations of what Rohlfing (this issue) calls ‘anticipated processes’ (e.g., Feaver, 2003). Finally, some sequential models function merely as conceptual shortcuts for linking model-exogenous explanatory factors to the outcome without reference to the actual processes we should expect to see in the real world.

EMPIRICALLY EVALUATING GAME THEORETIC MODELS: THE POTENTIAL OF PROCESS TRACING Process tracing (George and Bennett, 2004) or systematic process analysis (Hall, 2008) provide adequate methodological tools to address these challenges.3 The core of these case study methods is to investigate the historical sequence, which leads to the outcome, and to compare this empirical process with the observable implications of the theoretical argument that is supposed to explain the outcome. As such, process tracing is employed to uncover the underlying causal processes that bring about the explanandum. The primary inferential power of process tracing and other case study techniques, therefore, lies in allowing for a more in-depth test of the theoretical argument. Large-n empirical tests of game theoretic models typically focus on testing a single observable implication of the model: the co-variance of the dependent variable with a given factor thought to be causally relevant (or a set of such factors) (e.g., Geddes, 2003: chapter 3).4 Case

combining game theory models and process tracing

studies, on the other hand, allow for a much more varied empirical analysis on a larger range of observable implications and multiple levels of analysis. This makes them useful to test game theoretic arguments and address the two challenges outlined above. First, case study methods are well suited to empirically evaluate a model’s assumptions. Many of the theoretical concepts that underlie the assumptions are hard to measure. However, even if one can clearly define and operationalise the assumptions, it might be highly problematic to provide empirical evidence to justify or falsify their adequacy. As access to data on assumptions is usually limited and the reliability of such data, even if available, is hard to evaluate (see below), the construction of large enough data sets for statistically testing a model’s assumptions across a large number of cases will often be impossible outside of experimental research (cf. Morton, 1999: chapter 5). It is here that case studies are particularly useful. Not only are case studies well equipped to cope with complex and hardto-measure concepts (Collier and Mahon, 1993). They also allow the researcher to ‘dig deep’ into the cases and uncover inside information for evaluating the model’s assumptions. Of course, even for in-depth case studies, the aim of putting assumptions to a definite test is too high a goal. This is particularly true, for instance, for the rationality assumption that is clearly defined, but almost impossible to (dis)prove (Elster, 1986). However, using process tracing, case study researchers can try to uncover at least some pieces of empirical evidence on the assumptions, for instance showing that actors tried to inform themselves on their environment before making a decision. Most often, such basic evidence will be sufficient to accept the rationality assumption as a realistic ‘conceptual bridge in a causal claim’ (Parsons, 2007: 63), especially if there is robust evidence

on other steps in the causal argument. This is particularly true for the actors’ preferences and their beliefs concerning the results of their actions in relation to these outcomes. Through in-depth analysis of archival documents, diaries, interviews, newspaper reports and other sources, the researcher can put the model’s assumptions to the empirical test. Of course, this information will not ‘prove’ or ‘disprove’ the truth of the models’ assumptions, as actors might have an incentive to misstate their true goals or interpretation of an historical situation. However, finding empirical evidence for (or against) the assumed preferences and beliefs greatly enhances (or reduces) the overall persuasiveness of the model and might suggest directions for re-specifying it. Second, process tracing also helps addressing the status and persuasiveness of the modelled sequences. In general, in-depth within-case analysis allows for testing a large number of different hypotheses, which can be derived from the model (King et al, 1994: 224–228). Next to ‘relational predictions’ (Morton, 1999: 102) concerning the co-variation of the dependent and the independent variables, process tracing can be used to evaluate ‘dynamic or process predictions’, which compare ‘real-world dynamic processes with a model’s predicted dynamic path’ (ibid.). Depending on the concept of the modelled sequences, process tracing can in various ways contribute to theory development. This is straightforward if the model is assumed to represent actual empirical processes. Process tracing, then, is used to ‘trace the behavior of particular actors, clarify sequences, describe structures, and explore patterns of interaction’ (Bates et al, 1998: 10) and to compare the empirical processes with the modelled path. If the empirical sequence follows the general logic of the theorised process, this provides strong empirical support for the model (but see Rohlfing, this issue). david kuehn

european political science: 12 2013

55

A similar case can be made if the modelled sequence is supposed to reflect mental decision-making processes. Indepth analysis of decision-making procedures allows for reconstructing the actors’ reasoning in concrete situations. Through in-depth interviews or personal records, actual thought processes could, in principle, be retraced and compared with the modelled sequences. Through reconstructing the mental aspects of actual decision making, process tracing can also evaluate evidence on counterfactual, ‘offthe-path’ behaviour. Since much of a model’s explanatory power depends on those hypothetical actions, which do not lead to the identified equilibrium, researchers can bolster their causal argument by showing that actors actually considered the alternative actions (and their outcomes) proposed by the model and made their decisions in the light of these alternatives. Finally, even if the modelled sequences are not supposed to be abstracted simplifications of actual physical or mental processes, but are employed as ‘conceptual shortcuts’ to capture the general logic of real world interactions, process tracing can be useful. First, it can be employed to inductively derive hypotheses on causal processes. These theoretical propositions can, then, be tested on other cases and be used in combination with the deductive model to provide a more complete, though less parsimonious, explanation. Second, even if one is not interested in generating theoretical arguments on causal processes, process tracing can be used to test a multitude of within-case implications, which can be derived from the model. In this, it is particularly useful to test propositions on the temporal sequencing of individual decision situations. From many models, hypotheses can be derived on the temporal sequencing of decisions, even though these paths are not explicitly modelled. A simple bargaining model,

56

european political science: 12 2013

‘y process tracing can also evaluate evidence on counterfactual, “offthe-path” behaviour’. for instance, could suggest that different substantive issues are bargained for in a certain sequence, depending on the actors’ interests in the substance matter and the distribution of bargaining power (see below). Process tracing can corroborate or falsify such a hypothesis, and, thus, put the model to an additional test beyond the examination of its relational predictions.

LIMITS: EXTERNAL VALIDITY AND THE DIFFICULTY TO ASSESSING ASSUMPTIONS The combination of game theoretic modelling and process tracing, however, is hardly a panacea and there are notable limits to what case study researchers can hope to uncover through within-case analyses. First, there is the well-known trade-off between in-depth theory testing on one or a few cases and the broad but less intense analysis of a larger set of cases (cf. Ragin, 1987). While process tracing generates a wealth of ‘causal process observations’ (Collier et al, 2004) and thus allows for testing a multitude of observable implications on the within-case level, on the cross-case level there will be only one or a few observations. Furthermore, for many research questions testing within-case implications does not have the same analytical value as hypotheses on the cross-case level. Empirical knowledge on decision-making processes, actors and how their interactions produce the outcome of interest, improves the internal validity of causal claims, but does not enhance the

combining game theory models and process tracing

robustness of the inferences on the cross-case level. To a certain degree, these problems can be countered with selecting cases that have particular theoretical weight or that are representative for the population, that is, ‘least likely cases’ or ‘typical cases’ (cf. Gerring, 2007: chapter 5). These selection rules, however, essentially depend on the robustness of knowledge on the distribution of cases which, in turn, necessitates some, at least, informal large-n analysis of the population (see also King et al, 1994: 212–214). Second, there are inherent limits to empirically evaluating core assumptions. While some assumptions are relatively easy to evaluate, such as the temporal sequencing of decisions, others, such as the rationality principle, will remain all but impossible to assert even with the best within-case data. Furthermore, even specialists with expert knowledge on the case and access to good data, such as firsthand information on actors and decisionmaking processes, will not be able to ultimately prove preference assumptions, not least because political actors act and speak strategically and there are considerable biases in interviews, (auto-)biographic statements and news reports (Thies, 2002). Finally, given their weaknesses in terms of external validity, case studies are less suited for inductively generating hypotheses on preferences that are supposed to hold for a larger population. In fact, large-n surveys and experimental research doubtlessly are superior for generating general propositions on preferences and preference distributions across different groups of actors. Third, in evaluating models that are supposed to reflect mental processes, case studies of course cannot truly reconstruct the complete mental and psychological processes in concrete decision-making situations, as thinking does not leave observable evidence indepen-

dent from the words and deeds of the actors. Hence, these processes are not amenable to simple ex post observational analysis. None of these reservations, however, are necessarily fatal and devalue the case study method per se or its possible contributions to testing game theoretic arguments. First of all, case studies might not be as useful for inductively generating assumptions on preferences as survey research, but they can still provide insights because case study researchers are much more likely to gain access to, and evaluate the quality of, data on what actors in concrete decisionmaking situations want than what is possible for most large-n alternatives. Of course, preferences usually cannot be observed directly and even if a political actor states a concrete goal, there will remain a considerable degree of uncertainty. However, sometimes preferences can be inferred from circumstantial evidence, and while qualitative researchers do not have at their disposal the formalised instruments of robustness checks that have been developed for statistical methods, they can increase the persuasiveness of inferences from ‘causal process observations’ by triangulating and evaluating the quality of diverse data sources (see Leuffen et al, in this issue). While this is not sufficient to prove any assumption to be true, it is much preferable to no evidence at all. Second, there is no need to actually reconstruct psychological decision-making processes even if the model is meant to represent such a process. The main point here is that, again due to access to varied in-depth data, case study research can uncover empirical evidence, which can support or weaken the argument that actors consciously reflected on ‘off-the-equilibrium path’ courses of action. Qualitative data from interviews or minutes could, for instance, suggest that (a) actors thought about multiple david kuehn

european political science: 12 2013

57

alternative decisions and their outcomes, that (b) these alternatives are sufficiently similar to the modelled alternatives, and that (c) actors decided on the ultimate course of action because the expected effects of the chosen path has been more beneficial for their – assumed or stated – preferences. Each of these data points can be compared with the concepts built into the model and together they will increase our trust in the model. Of course, the problems of uncertainty as discussed above remain, but so does the point that even little and weak evidence for process assumptions is preferable to no information at all. The main conclusion from this discussion is that in spite of the potential of case studies for assessing within-case hypotheses, the researcher needs to be wary of its limits and needs to always consider (and report) the degree of uncertainty.

AN ILLUSTRATIVE EXAMPLE: CIVIL–MILITARY RELATIONS IN NEW DEMOCRACIES In order to illustrate the usefulness of process tracing for evaluating game theoretic claims, I will provide an example from research on the institutionalisation of civilian control over the military in new democracies.5 Almost all authoritarian regimes grant their militaries wide-ranging political prerogatives and institutional autonomies in order to co-opt them into the regime coalition and to safeguard their loyalty. One of the core challenges for civilian elites after the transition to democracy, therefore, is to reduce the military’s political prerogatives and to install institutions, which allow meaningful control over all politically relevant issue areas that have been previously under control of the armed forces (cf. Croissant et al, 2010). The general

58

european political science: 12 2013

argument I make is that the process of institutionalisation of civilian control in new democracies can be understood as a series of interactions between the elected civilians and the military leadership. I model these interactions as a series of one-shot extended crisis bargaining games between the civilian and military elites. Starting with the assumption that civilians have an interest in extending their political decision-making power over former military prerogatives and the military is interested in keeping them, the decision tree begins with civilians deciding on challenging or not-challenging the military’s remaining prerogatives. If civilians decide not to challenge, the game ends with the status quo. If civilians challenge, the military leadership has to decide if it accepts the institutional change and complies to the reduction of its prerogatives, or if it risks a power struggle with the civilians. If the military complies, a new institutional status quo is reached, and the military has lost its prerogative. If the military decides to challenge the civilians, the latter need to decide if they back down or if they continue to press for an institutional change and risk a power struggle, the concrete form of which can range from simple bargaining to open conflict, or even a coup d’e ´tat. The actual outcome of these interactions depends on exogenous factors, which define the payoffs, that is, the relative costs and benefits of the various possible outcomes, the weight the respective players pay to the prerogative at hand and the power resources available to the actors.6 In the end, it runs down to a cost–benefit calculation: civilians estimate the likelihood that the military will try to defend its prerogative and the possible costs of the power struggle in relation to the importance they attach to strip the military of that particular prerogative. The military, on the other hand, estimates the costs of the power struggle versus the

combining game theory models and process tracing

civilians in relation to the weight they attach to keeping the prerogative. For evaluating the model’s predictions, I employ process tracing of attempts to institutionalise civilian control over various political prerogatives during the democratisation of Taiwan and South Korea. As I do not consider the model to be a simplified reference of actual occurring interactions but merely as a conceptual approximation to the logic underlying such situations, the modelled micro process is not tested. However, the model yields predictions on the temporal sequencing of the various steps of institutionalising civilian control: as the outcome depends on the relation of power resources and the relative importance of the substance matter to the respective players, ceteris paribus civilians will first attempt to challenge those prerogatives which they attach a lot of weight to substantively and which are at the periphery of the military’s core interests. Since civilians are subject to electoral competition and depend on voter support that derives inter alia from delivering policies, this means that civilians will try to disband the military’s influence over elite recruitment, the processes of electoral competition and public policymaking in economic, social welfare and other non-security issue areas before they address issues that are closer to the military’s core issues, for instance, defence and security policy. This proposition can be evaluated by analysing the sequence of the complete macro-level process of change (or stagnation) in civil–military relations after the transition to democracy. And indeed, in both countries, civilians rapidly addressed these issues and succeeded in abolishing military influence in these core political arenas in rather short time while a reform of defence policymaking was only initialised long after the transition to democracy. Taiwan’s authoritarian one-party state, for instance, had been under martial law for almost four

decades, the military leadership had always been included into the Kuomintang’s (Chinese Nationalist Party) top decisionmaking bodies, and Hau Pei-tsun, the Chief of the General Staff, was the nation’s second most powerful political figure. However, immediately after coming into office in 1988, civilian President Lee Teng-hui and his coalition of liberal civilians manoeuvred to neutralise Hau’s political clout and started to push the military out of the party leadership. Finally, with Hau’s departure from politics in 1993, the military lost all its former influence in the central political decisionmaking arena only 5 years after President Lee had come into office (Kuehn, 2008). A similar process can be shown in South Korea’s transition. The country had been under military rule since the 1960s and made the transition to democracy in 1987. Kim Young-sam, the first civilian President, came into office in 1993 and immediately acted against the military’s remaining political influence, purging around 1,000 officers from the army and putting the leaders of the former authoritarian regime on trial (Croissant, 2004). Compare these rapid and decisive actions with the protracted processes of enhancing control over defence and security policy: in Korea, meaningful defence reform was only initialised in the late 2000s while in Taiwan, the military command was restructured and civilian control over defence policy institutionalised only with the passing of the ‘Two Defence Laws’ in 2000. A more in-depth, micro-level analysis of Taiwan’s defence reform is illustrative for how process tracing can be used to evaluate hypotheses derived from the model, as well as testing some of the model’s assumptions. One conclusion drawn from the model is that civilians will challenge prerogatives which the military values highly, such as their internal autonomy or decision-making power over defence policy, only if the balance of david kuehn

european political science: 12 2013

59

bargaining power is strongly in their favour. The process of defence reform in Taiwan corroborates this hypothesis. In the early 1990s, individual civilians, particularly from the political opposition, had demanded the reorganisation of Taiwan’s military command structure. The existing command structure was bifurcated, with the operational command responsible for actual military operations, running directly from the president to the general staff, and the administrative command running from the president via the defence ministry (that was accountable to the legislature) to the general staff. As such, the military’s internal affairs and its de facto control over defence policy was shielded from any effective civilian influence and control: Since the president did not have a bureaucratic staff to actually control the day-to-day activities of the military, this bifurcation ensured that civilian control over the military’s operational activities were sketchy at best and highly dependent on the president’s interest and expertise in defence issues. Demands to change the existing rules were strongly opposed by the military and, thus, not taken up by the civilian government. This only changed after the Taiwan Missile Crisis in 1995/1996 when the Chinese military fired a barrage of ballistic missiles into Taiwan’s coastal waters. Frustrated with the cooperation between their militaries, the United States in the aftermath of the crisis put enormous pressure on Taiwan’s defence community to integrate their military decision-making channels and to increase the role of civilians in defence policy (Huang, 2010). Since Taiwan’s security ultimately depends on US political support and military hardware, this pressure not only strengthened civilian interest in pushing for defence reform, but at the same time significantly weakened the military’s bargaining position for keeping their prerogative and obstructing the process that led to the enactment of the reform.

60

european political science: 12 2013

Finally, in-depth analysis of the process also allowed me to check the plausibility of some of the model’s core assumptions, most importantly the interest of civilians in changing the status quo and the preference of the officer corps to let go of its prerogatives. Circumstantial evidence for Taiwan, for instance, suggests that civilian decision makers actually were interested in minimising military influence over politics, even though their immediate actions suggested otherwise. Confronted with the powerful former general and his conservative supporters in the party leadership, President Lee, in 1992, appointed Hau Pei-tsun as Prime Minister. This was followed by a huge public outrage in the press, constant attacks from the opposition and wide spread street protests against the alleged ‘militarisation’ of politics and Lee’s seeming buckling before the military (Chen and Ko, 1993; Lo, 2001). Together with survey data, which clearly show the rejection of any political activity by the military (Chu et al, 2001: 130), these various observations suggest that there was a broad social consensus on keeping the military out of politics. Being dependent on democratic support, this was doubtlessly of great concern to the president, even if continuing military clout meant that he could not simply push Hau out without political manoeuvring. Furthermore, (auto-)biographic writings and interviews with contemporaries from both sides of the civil– military spectrum clearly give evidence that Lee wanted to decrease military political influence as much as possible. Similarly, the case study of Taiwan supports the assumption that the military is interested in keeping civilians out of defence and military policy as much as possible. Direct evidence for this can be found in interviews with civilian defence bureaucrats who lament that the military is doing everything to shut them out, as well as in memoirs of, and interviews with,

combining game theory models and process tracing

military officers who criticise civilians attempting to influence defence policy without having the necessary ‘professional expertise’ (e.g., Hau, 1999; Tsai, 2010).

CONCLUSION This article has argued that process tracing can be profitably combined with game theoretic models. Particularly interesting is the ability to empirically evaluate some of the assumptions of the model and the chance to compare the modelled processes and temporal sequences with empirical evidence. While process tracing is certainly no panacea and cannot substitute the rigorous analysis of a model’s relational predictions and comparative statics over a larger number of cases, it is a useful addition to the social scientist’s methodological tool-box, which allows for a deep and fine-grained empirical evaluation of the multitude of observable im-

‘y allows for a deep and fine-grained empirical evaluation of the multitude of observable implications which operate on different levels of game theoretic models’. plications that operate on different levels of game theoretic models.

Acknowledgements I wish to thank the participants of the Oldenburg workshop and one anonymous reviewer for their valuable comments on this article. I am particularly grateful to Bernhard Kittel for organising the workshop.

Notes 1 The overall share of MMR of all reviewed articles was between 15 and 19 per cent (Bennett et al, 2003: 377). 2 These aspects are true for both, verbal and formal, applications of game theoretic models. Formal mathematical theorising has the additional advantage of guaranteeing the internal consistency of the theoretical arguments. 3 George and Bennett’s (2004) conception of ‘process tracing’ and Hall’s (2008) ‘systematic process analysis’ differ in their focus on theoretical predictions. Hall puts greater emphasis on the role of alternative theory testing and explicitly calls for deriving and evaluating a variety of within-case implications. George and Bennett are mostly concerned with the concept of causal mechanisms and hypotheses on the causal process itself. They share the common core of a priori theorising, deriving within-case implications and comparing them with data derived from the retracing of empirical sequences. As the discussion at hand centres on these commonalities, this article will not differentiate between the methods. 4 The recent EITM (Empirical Implications of Theoretic Models) movement is aimed at better integrating game theoretic modelling and empirical testing, inter alia by increasing the numbers (and levels) of observable implications to be tested (Aldrich et al, 2008). 5 This analysis is based on research for the author’s Ph.D. dissertation, which will be submitted to the Faculty of Economy and the Social Sciences, Heidelberg University in 2012. 6 In a second model specification, which includes incomplete information, the outcome additionally depends on the degree of uncertainty over the respective other player’s power resources and the weight they attach to the issue.

References Aldrich, J., Alt, J. and Lupia, A. (2008) ‘The EITM Approach: Origins and Interpretations’, in H.E. Brady and J.M. Box-Steffensmeier (eds.) The Oxford Handbook on Political Methodology, Oxford: Oxford University Press, pp. 828–843. david kuehn

european political science: 12 2013

61

Bates, R.H., Greif, L., Levi, M., Rosenthal, J. and Weingast, B. (1998) Analytic Narratives, Princeton: Princeton University Press. Bennett, A., Barth, A. and Rutherford, K.R. (2003) ‘Do we preach what we practice? A survey of methods in political science journals and curricula’, PS: Political Science and Politics 36(3): 373–378. Bennett, A. and Braumoeller, B.F. (2006) ‘Where the model frequently meets the road: Combining statistical, formal, and case study methods’, Paper presented at the APSA Annual Meeting; August 2006, Philadelphia. Chen, S. and Ko, C. (1993) Guofang Heihezi yu Beipishu (Whitebook on the Black Box of National Defense), Taipei: Formosa Foundation. Chu, Y., Diamond, L. and Shin, D. (2001) ‘Halting progress in Korea and Taiwan’, Journal of Democracy 12(1): 122–135. Collier, D., Brady, H.E. and Seawright, J. (2004) ‘Sources of Leverage in Causal Inference: Toward an Alternative View of Methodology’, in H.E. Brady and D. Collier (eds.) Rethinking Social Inquiry: Diverse Tools, Shared Standards, Lanham: Rowman & Littlefield Publishers, pp. 229–266. Collier, D. and Mahon, J.E. (1993) ‘Conceptual “stretching” revisited: Adapting categories in comparative analysis’, American Political Science Review 87(4): 845–855. Croissant, A. (2004) ‘Riding the tiger: Civilian control and the military in democratizing Korea’, Armed Forces & Society 30(3): 357–381. Croissant, A., Kuehn, D., Chambers, P. and Wolf, S. (2010) ‘Beyond the fallacy of coup-ism: Conceptualizing civilian control of the military in emerging democracies’, Democratization 17(5): 950–975. Elster, J. (ed.) (1986) ‘Introduction’, in Rational Choice, Oxford: Oxford University Press, pp. 1–33. Feaver, P.D. (2003) Armed Servants: Agency, Oversight, and Civil–Military Relations, Cambridge: Harvard University Press. Geddes, B. (2003) Paradigms and Sand Castles: Theory Building and Research Design in Comparative Politics, Ann Arbor: University of Michigan Press. George, A.L. and Bennett, A. (2004) Case Studies and Theory Development in the Social Sciences, Cambridge: MIT Press. Gerring, J. (2001) Social Science Methodology: A Criterial Framework, Cambridge: Cambridge University Press. Gerring, J. (2007) Case Study Research: Principles and Practices, New York: Cambridge University Press. Goldthorpe, J. (1996) ‘The quantitative analysis of large-scale data-sets and rational action theory: For a sociological alliance’, European Sociological Review 12(2): 109–126. Hall, P.A. (2008) ‘Systematic process analysis: When and how to use it’, European Political Science 7(3): 304–317. Hau, P. (1999) Ba Nian Canmou Zongzhang Riji (Diary of Eight Years Chief of General Staff), Taipei: Commonwealth Publishing. Huang, A.C. (2010) ‘Personal interview with Alexander Chieh-cheng Huang’, Professor at the Graduate Institute of International Affairs and Strategic Studies, Tamkang University, Tamsui, Taiwan. King, G., Keohane, R.O. and Verba, S. (1994) Designing Social Inquiry: Scientific Inference in Qualitative Research, Princeton: Princeton University Press. Kuehn, D. (2008) ‘Democratization and civilian control of the military in Taiwan’, Democratization 15(5): 870–890. Little, D. (1998) Microfoundations, Method, and Causation: On the Philosophy of the Social Sciences, New Brunswick: Transaction. Lo, C. (2001) ‘Taiwan: The Remaining Challenges’, in M. Alagappa (ed.) Coercion and Governance: The Declining Political Role of the Military in Asia, Stanford: Stanford University Press, pp. 143–164. Morton, R.B. (1999) Methods and Models: A Guide to the Empirical Analysis of Formal Models in Political Science, Cambridge: Cambridge University Press. North, D.C. (2005) Understanding the Process of Economic Change, Princeton: Princeton University Press. Parsons, C. (2007) How to Map Arguments in Political Science, Oxford: Oxford University Press. Ragin, C. (1987) The Comparative Method: Moving Beyond Qualitative and Quantitative Strategies, Berkeley: University of California Press.

62

european political science: 12 2013

combining game theory models and process tracing

Thies, C.G. (2002) ‘A pragmatic guide to qualitative historical analysis in the study of international relations’, International Studies Perspectives 3(4): 351–372. Tsai, M.M. (2010) ‘Personal interview with Michael M. Tsai’, Former Minister of Defence, ROC.

About the Author David Kuehn is a Research Fellow at the Institute of Political Science, Heidelberg University. His research interests are civil–military relations, democratisation and social science methodology. He is co-author of a textbook on the military and politics; his articles have been published in Democratization, the Journal of East Asian Studies and Comparative Governance and Politics.

david kuehn

european political science: 12 2013

63