11 - Analytic Narratives and Case Studies

Summary

Margaret Levi and Barry Weingast focus on a particular type of case in which the subject is an outcome that results from strategic interaction, when one person’s decision depends on what another does. “A weakness of case studies per se is that there typically exist multiple ways to interpret a given case,” they begin. “How are we to know which interpretation makes most sense? What gives us confidence in the particular interpretation offered?” An analytic narrative first elucidates the principal players, their preferences, key decision points and possible choices, and the rules of the game. It then builds a model of the sequence of interaction including predicted outcomes and evaluates the model through comparative statics and the testable implications the mode generates. Most analytic narratives model situations as extensive form games. However, although game theory is useful, there is no hard rule that requires us to formalize. In this kind of case study, the findings do not generalize to other contexts, but instead point to the characteristics of situations to which a similar strategic logic applies.

11.1 Introduction

Analytic narratives (Bates et al., 1998, 2000; Levi, 2002, 2004) involve selecting a problem or puzzle, then building a model to explicate the logic of an explanation for the puzzle or problem, often in the context of a unique case. The method involves several steps. First, the use of narrative to elucidate the principal players, their preferences, the key decision points and possible choices, and the rules of game, all in a textured and sequenced account. Second, building a model of the sequence of interaction, including predicted outcomes. And, third, the evaluation of the model through comparative statics and the testable implications the model generates. The analytic narrative approach is most useful to scholars who seek to evaluate the strength of parsimonious causal mechanisms in the context of a specific and often unique case. The requirement of explicit formal theorizing (or at least theory that could be formalized) compels scholars to make causal statements and to identify a small number of variables as central to understanding the case.

Case studies abound in the study of development. A weakness of case studies per se is that there typically exist multiple ways to interpret a given case. How are we to know which interpretation makes most sense? What gives us confidence in the particular interpretation offered? This problem is particularly difficult where the uniqueness of the situation precludes the collection of a data set that encompasses multiple cases. Many scholars augment their case study with a model. The model adds some discipline to the account. For example, observed choices must be consistent with the assumption about preferences in the model. A model is necessary, but not sufficient: that is, a model alone does not an analytic narrative make.

The analytic narrative approach provides a means to help get around these questions. The essence of many cases is unique, including the French Revolution, the American Civil War, or the surprising and quick development of Spain following the death of long-time dictator, Francisco Franco, in 1975. Analytic narratives deal with these cases by building a model that has multiple implications, and then testing an implication of the model that provides the possibility for both confidence in the claims and comparison across cases.

For example, Weingast’s (1998) case study of the American Civil War builds on a unique feature of American institutions to explain long-term political stability – namely, the “balance rule”: the idea that both Northern free states and Southern slave states would be admitted in pairs, giving each set of states a veto over national policy. This institution fell apart in the 1850s. Weingast tests his account of this failure by using game theory to reveal a path not taken, given the interest calculations of those making choices about what path to take. This enabled him to estimate a counterfactual involving what would have happened had a contingency in the case study not occurred. Thus, the main thesis of the case – the balance rule – is unique and cannot be tested directly, but other implications of the approach can be tested so as to give confidence in the overall account.

A second example addresses the effect on public goods provision of a 2004 decentralization reform in postconflict Sierra Leone (Clayton et al. 2015). The specificities of most laws are unique; even when the words are replicated, implementation varies across and within countries. In this instance, the narrative reveals the key stakeholders at the local level: elected councilors and paramount chiefs. The interests of the former should lead them to prefer successful implementation, ceteris paribus; but in some localities they conceded to the paramount chiefs, who preferred the status quo. Given other reforms that increased the power of the councilors and reduced that of the chiefs, the question becomes why the councilors deferred. What are the relevant comparative statics? This question produced a series of testable implications. The case, although unique, sheds light on the more general problem of the variation in the impact of decentralization on the delivery of health and education services.

This chapter proceeds as follows. In Section 11.2, we discuss criteria for case section. Section 11.3 discusses identifying processes and mechanisms, while Section 11.4 discusses the limitations of the approach. In Section 11.5, we briefly discuss the implications of the approach for the development context. Our take-aways follow.

11.2 Criteria for Case Selection

The analytic narrative approach combines a commitment to rational choice, a deep interest in a particular case, a method for devising a generalizable model of the case, and a means of providing empirical evidence, even in unique cases.

The combination also entails an aim most area specialists lack: to go beyond detailing the case to elaborate more general conditions for the problem or puzzle. This exercise requires criteria for selection of cases other than their intellectual appeal as puzzles demanding solutions. Standard approaches to case selection emphasize the bases for choice among a sample of cases which are informative about the causal chain of interest, because of the absence, presence, or extreme values of key variables. One traditional method advocates pairs of cases that are either “most similar,” hopefully allowing the analyst to identify similar mechanisms in the two cases, or “most different,” hopefully allowing the analyst the ability to isolate a mechanism that accounts for the differences. These traditional methods fail when more than one causal variable is relevant.

Bearing similarities to the analytic narratives approach is process tracing (George and Bennett 2005; Collier 2011; Bennett, Chapter 4, this volume), which shares an emphasis on both sequencing and fine-grained description as means for making causal inferences. Process tracing also shares a concern with generating testable implications, but its emphasis is on key variables rather than the key actors, their interaction, and their strategies. This makes game-theoretic analysis largely irrelevant to process tracing.

Analytic narratives include features that make the cases amenable to modeling, which not all puzzles or problems are. Essential to the model building is the choice of cases in which the key actors interact strategically. That is, the choices of one actor depend on the choices of the other. In addition, analytic narratives consider situations that can be modeled as an extensive-form game, which generates a subgame perfect equilibrium.

Another necessary feature of an analytic narrative is the opportunity to get at an important process or mechanism not easily accessible through other means. For example, the extensive-form game allows the analysis to demonstrate the existence of a self-enforcing institution that often solves an important economic or political problem through creating a credible commitment. The advantage of the game is that it reveals the logic of why, in equilibrium, it is in the interest of the players to fulfill their threats or promises against those who leave the equilibrium path.

The formalization itself is not a requirement of a successful analytic narrative; indeed, in some cases, there are too many actors and no benefit from reducing the multiple players to the small number required for a game-theoretical model. Levi’s case on conscription (Bates et al. 1998) illustrates how one can still use the logic of extended form games to assess the strategies and actions – and paths not taken – without formalization. Another example is Ferrara’s (2003) analytic narrative of the Burmese uprising in 1988 as a means to understand both a particular historical event and the more general question of the relationship between coercion and protest.

The final expectation of an analytic narrative is that the causal mechanisms and the structures or relationships must be generalizable to other cases under specifiable conditions. We deal with this issue below.

11.3 Identifying Sequence and Mechanisms

Analytics, in this approach, refer to the building of models derived from rational choice, particularly the theory of extensive-form games.¹ The steps toward building the model include:

• First, extracting from the narratives the key actors, their goals, the sequence of options available to an actor at a given moment, and the effective rules that influence actors’ behaviors.

• Second, elaborating the strategic interactions that produce an equilibrium that constrains some actions and facilitates others. By making clear and explicit the assumptions about who the key actors and their preferences are, it is possible to challenge the assumptions to produce new insights and competitive interpretations of the data.

• Third, the equilibrium analysis leads to comparative static predictions that produce testable implications even if they’re not the main assertion of the case.

We emphasize this third criterion. An important advantage of relying on game theory is that this method often produces comparative statics – that is, predictions about how the equilibrium shifts in response to changes in the exogenous variables. This approach allows the analyst to identify the reasons for the shift from one equilibrium to another. It therefore produces expectations of behaviors in the form of testable implications if the key actors are staying on the equilibrium path and if they are not. A case study that includes a model may involve the first two criteria, but generally not the third. This is especially true for unique cases where it is hard to test the model directly. Both authors have written many case studies of this type. These cases may provide insights, but they are not analytical narratives (see, e.g., Levi 1988; Weingast 2004). Analytic narratives require testable implications derived from the comparative statics that the narrative helps reveal.

The narrative of analytic narratives establishes the principal players, their goals, and their preferences while also illuminating the effective rules of the game, constraints, and incentives. Narrative is the story being told but as a detailed and textured account of context and process, with concern for sequence, temporality, and key events. By meeting these criteria, the narrative offers a means to arbitrate among possible explanations for observational equivalences – that is, two distinct processes that lead to the same outcome.

Comparative statics are crucial for comparative research because they generate hypotheses of what could have taken place under different conditions. Comparative statics therefore clarify the relationship between the key endogenous and exogenous variables. Moreover, the consideration of “off-the-equilibrium-path” behavior typically reveals reasons and reasoning for why actors took one path and not another. Indeed, what actors believe will happen were they to make a different choice typically influences the choices they do make. As Niall Ferguson (1999) observed in his study of the causes of World War I, to understand why Britain entered what would otherwise have been a continental war, we need to know what the British believed would happen had they not entered the war. Another important aspect of the game-theoretic approach is that the off-the-path behavior of an equilibrium disciplines each player’s beliefs, for they must be consistent with all the other player’s strategies (see Weingast 1996).

For example, consider the illustrative “deterrence game” in the appendix of Analytic narratives (Bates et al. 1998). Two countries interact: the home country and an opponent. The home country maintains a large, expensive army; the opponent does not attack. Is the large army the reason for peace as it deters the opponent from attacking? Or is it a waste of resources because the opponent has no interest in attacking? These two hypotheses are observationally equivalent: both offer explanations for why the opponent does not attack, yet they differ dramatically as to the reason for the observation. Different people have different beliefs that can only be understood contextually:

[T]he observationally equivalent interpretations rest on markedly different theories of behavior. To settle upon an explanation, we must move outside the game and investigate empirical materials. We must determine how the opponent’s beliefs shape their behavior. This blend of strategic reasoning and empirical investigation helps to define the method of analytic narratives.

(Bates et al. 1998: 241)

This approach provides the researcher with some discipline. As the deterrence game illustrates, absent a game and an equilibrium structure, it is possible to posit a wide range of beliefs that motivate action. How do we choose among these different accounts? In the context of a game, beliefs about another player’s actions are part of the equilibrium. Not just any sets of beliefs will work. In the deterrence game, the opponent must have a belief about how the home country will react to an attack; and, in equilibrium, this belief must hold in practice.

This form of explicit theory provides criteria to enable the researcher to distill the narrative and ensure that the explanation need not rely too much on factors outside the model.

11.4 Overcoming the Limits of Analytic Narratives

The analytic narrative approach, at least in its original formulation, had several potential limitations, some recognized by the authors and others revealed by various critiques.

11.4.1 Generalizing

The Achilles’ heel of analytic narratives – as with any approach to case studies – is in the capacity to generalize, given that each narrative represents an effort to account for a particular puzzle in a particular place and time with a model and theory tailored to that situation. Even so, it is possible to use the cases to make some more general points.

Although the approach is not straightforwardly deductive, it nonetheless relies on rational choice, which is a general theory of how structures shape individual choices and, consequently, collective outcomes. Rational choice, particularly in its game-theoretic form, highlights certain properties of the structure and strategic choices that arise. Although the specific game may not be portable, it may yield explanations that can be tested in the form of collective action problems, principal–agent issues, credible commitments, veto points, and the like. Analytic narratives provide a way to suggest the characteristics of situations to which these apply and in what ways. For example, the models of federalism, as initially developed by William Riker (1964) and further developed by Weingast and his collaborators (Weingast 1995; Montinola, Qian, and Weingast 1995), are useful in explicating a large number of problems in a wide range of countries, including the case Weingast (1998) addresses in his Analytic narratives chapter.

Moreover, the analytic narrative approach also demands identification of causal mechanisms. A wide range of mechanisms, such as emotions, resentment, and other aspects of behavioral economics, can offer a fine-grained explanation of the link between actions and alternatives (Elster 1998, 1999). Others have fruitfully made these links in such situations as insurgency in El Salvador (Wood 2001) and violence in Eastern Europe and the Balkans (Petersen 2002, 2011) while meeting the requirement that they “generate new predictions at the aggregate or structural level” (Stinchcombe 1991: 385).

11.4.2 Surprise, Contingency, and Conjunction

Daniel Carpenter’s (2000) critique of Analytic narratives raised several issues that the approach needed to confront to fulfill its promise. Carpenter worried that we narrowed the conceptualization of narrative in a way that was likely to neglect the surprises history offers, the contingencies that affect outcomes, and the conjunctures that make parsimony so difficult. The first and last are easiest to address since nothing about the method precludes either. The approach actually makes it possible to take surprises into account since they often take the form of events that would change comparative static outcomes. De Figueiredo, Rakove, and Weingast (2006) illustrate one means by which game-theoretic models can be generalized to encompass surprises. Those American colonists already suspicious of Britain were apt to believe the worst interpretation of any British act and to believe that large-scale rebellion was inevitable. The result was a self-confirming equilibrium to explain the surprise element in the eruption of the American Revolution. Nothing is foreordained by an analytic narrative, which, on the contrary, often reveals factors as significant that we might not otherwise have noted. For example, in Gretchen Helmke’s (2005) analysis of courts in autocratic regimes, her counterintuitive finding is that a nonindependent judiciary has the power, under certain circumstances, to rule against its government.

Carpenter also raises conjunctural analysis: the idea that multiple, interlacing factors occur at once – say, a war and a depression – so that the causal factors are difficult to disentangle. Carpenter’s concern (2000: 657–658) is that “[i]f one changes the values of two variables at once, or renders the values of one variable dependent on those of another – precisely as historians who rely on conjunctures tend to do – then the embedded independence assumption comes, well, with high costs.” Skocpol (2000) and Katznelson and Milner (2002) share Carpenter’s concern.

Conjunctures are a problem for every form of analysis, not just analytic narratives. Moreover, by relying on game theory, analytic narratives may be uniquely suited to addressing conjunctures. By providing a specific model of events, a game-theoretic model helps disentangle conjunctures by potentially making predictions about what would have happened had only one of the conjoining events occurred instead.

Carpenter claims that contingency disappears from the analytic narrative approach because, as he perceives it, there is less likelihood of multiple equilibria – that is, alternative stable states of the world. He goes on to say that “[i]t would have been theoretically appealing for the authors to give examples where history in some way ‘selects’ some equilibria and makes others impossible (kind of like a trembling hand, or stability, or coalition-proofness criterion)” Carpenter (2000, 657).

But this criticism reflects a misreading of the analytic narrative approach. The use of game theory means that in many instances multiple equilibria will arise. Hence, the existence of multiple equilibria is part of the analytic narrative approach even if the case studies in the original volume do not make that evident. Contingency in the form of multiple equilibria is therefore a feature of the approach.

Even when there are clear focal points and strategies, factors in the situation can change unexpectedly. Some contextual changes may have clear and significant consequences, others have butterfly effects, and still others have little or no effect. The narrative is crucial here for sorting out what matters for what. In Rosenthal’s Analytic narratives chapter, the potential birth of a Catholic heir to James II affects the calculations of both monarch and elites, but its importance lies in how it changes the strategies of the elites even unto the point of revolution (Rosenthal 1998: 92). Why elites resorted to revolution rather than peaceful institutional change becomes apparent through the narrative and the associated model.

Uncertainty and lack of information are prevalent features of the unraveling of events in history, and they are major bases of contingency. Ahlquist and Levi’s work on leadership illustrates the effect of uncertainty (Ahlquist and Levi 2011, 2013). They find that followers, members, and citizens are very concerned to have competent representation; followers, members, and citizens therefore do their best to figure out who will be a good leader based on the track record of potential candidates. Nonetheless, unknowns remain, often in the form of other variables that are uncertain. For example, no one can know for sure how opponents will react to a given leader, what the economy will do, or how leaders will respond under circumstances distinctive from those in which they were selected. This uncertainty has direct consequences for other facets of the organization, such as its governance arrangements and mechanisms of accountability. If members knew and understood all the implications of their original choice, they might make a different one – if they could. Yet, uncertainty instead leads them to coordinate around a specific leader and leadership style, and they may well continue to maintain that person in office for years.

Analytic narratives must include problems of randomness or contingency, but not if they are too extreme. The example of unions makes the point. Members address their leadership problem in the face of uncertainty about the occurrence of strikes and only partial information about the reaction of employers to their demands. Because the interactions between unions and managers involve unpredictable elements, and because leaders cannot always deliver what they promise, leadership turnover may result. However, as Ahlquist and Levi (2013) show, this turnover is not only relatively rare but also highly delimited by the organizational culture and governance arrangements that ensure new leaders will share many of the characteristics of their predecessors.

The analytic narrative approach rests on cases where there is some, but hardly complete, contingency in the path of history, cases that the model helps in understanding what was likely to happen. Nothing about the approach, however, limits it to cases of determinateness or low contingency. Extensive-form games have long proved useful in studying settings of high uncertainty and contingency.

11.5 Analytic Narratives for Use in Development Policy and Practice

In this section we turn, briefly, to suggest the implications of the analytic narrative approach to problems of development.

Economists have long proposed an economic role for political institutions, such as the market infrastructure embodied in the provision of secure property rights, enforcement of contracts, and, generally, the provision of justice and the rule of law (Weingast 1995). Governments that use violence against minorities and opponents, confiscate citizens’ wealth, and create economic privileges (such as dispensing monopoly rights) fail to provide adequate market infrastructure. As Adam Smith recognized more than two centuries ago, the risk of violence and of plunder leads men to avoid hard work, initiative, and investment. In discussing settings in which “the occupiers of land in the country were exposed to every sort of violence,” Smith argued that “men in this defenceless state naturally content themselves with their necessary subsistence; because to acquire more might only tempt the injustice of their oppressors” (Smith 1776: III.iii.12:405). Further, a “person who can acquire no property, can have no other interest but to eat as much, and to labour as little as possible” (Smith 1776: III.ii.9:387–88).

But, if secure property rights, enforcement of contracts, and the provision of justice are necessary for economic development, how are such institutions built and, especially, sustained?

North and Weingast (1989) developed the hypothesis of credible commitments to answer this question. Governments seeking to implement the economists’ prescriptions for political institutions had to commit to honoring rights of citizens and to use agreed upon political procedures to make political decisions. They developed their hypothesis in the context of a unique case: the English Glorious Revolution of 1688–1689. Although this revolution and its institutional consequences were unique to that case, North and Weingast provided some important evidence favoring their larger, general argument about credible commitments. Focusing on public finance, they showed that the ability of the English government to borrow money changed dramatically. Government debt had never been much above 5 percent of estimated GDP in the seventeenth century. But in the eight years following the Glorious Revolution, it rose by nearly an order of magnitude, to 40 percent of estimated GDP. Because debt repayment depends critically on credible commitments, the massive increase in debt in a short time suggests that a new mechanism for making credible commitments had emerged.

Sure enough, subsequent studies have identified some of the devices used to create credible commitments and have leant support to the hypothesis.² First, Cox (2012) has shown that a number of other variables also increased dramatically, consistent with the credible commitment hypothesis. Specifically, per the North and Weingast narrative, Cox demonstrates that parliament gained control over taxation and the issuance of public debt. Similarly, the ministerial responsibility system emerged: while parliament faced difficulties in holding the king accountable for public decisions, they could hold the king’s ministers accountable, forcing them to honor parliament’s interests. Second, scholars have undertaken a range of studies of public debt at similar events. For example, Summerhill (2015) has shown that nineteenth-century imperial Brazil provided the institutions for credible commitment to public debt, yet it failed to provide the institutional foundations for private financial markets and hence this fundamental basis for economic development. Mo and Weingast (2013: ch 4) reveal the means by which the South Korean regime under President Park Chung provided credibility to its promises to honor property rights and a range of other programs, such as education, underpinning that country’s economic development.

As a second illustration, consider political stability, another element widely agreed as important for economic development. Coups, civil wars, ethnic conflict, and other forms of disorder cripple a country’s ability to develop. Cox, North, and Weingast (2019) show that disorder in the form of violent takeover of regimes occurs surprisingly often in the developing world: the median regime of the poorest half of countries lasts only seven years. Just how do a minority of countries provide for political stability?

Mittal and Weingast (2012) provide three conditions for political stability, one of which they call the “limit condition”: the idea that all successful constitutions reduce the stakes of power, for example, by providing incentives for political officials to honor a range of citizen rights. Limited government does not imply small government (as modern political debate suggests), but a government that can honor restrictions on its behavior, such as abiding by election results, refraining from the use of violence to repress enemies, and, generally, honoring citizen rights.

The logic of the limit condition is that high stakes make it much more likely that people who feel threatened by the government will support coups. For example, landowners in Chile under the presidency of Salvadore Allende supported the military coup in 1973 to protect themselves. Similar events led to disorder in Spain (1936–1939) and Kenya (2007–2008). One way that constitutions reduce the stakes of politics is through various forms of countermajoritarian institutions.

Mittal and Weingast develop their hypothesis in the context of the American case, where the institutional features of the US Constitution are unique. Subsequent work has revealed similar features in a range of cases of stable constitutions. Countermajoritarian provisions serve two valuable roles in preserving political stability. First, they often aid in the instantiation of democracy. When groups see themselves as potentially worse off under democracy, they are likely to resist democratization. The reality is that powerful – and sometimes inimical – groups often have the power to hold up democratization, such as slaveholders in the early American Republic, Whites in South Africa in the 1990s, the supporters of the authoritarian regime of Francisco Franco in Spain following the death of the dictator in 1975, the military dictatorship in Chile in the late 1980s and early 1990s, and the communist regime in Poland in 1989. In each of these cases, countermajoritarian provisions to protect powerful groups aided democratization and, eventually, the lightening of the relevant constraints. Similarly, electoral laws often bias elections in favor of constituencies that favor the previous regime (Chile and Spain). These provisions also become part of the limit condition once democracy has been initiated.

Others have also used analytic narratives to discuss the difficulties of achieving transitions to democracy. Ferrara (2003) suggests conditions under which widespread protest and uprising has little effect, given the strategic use of coercion. His case is Burma, but the implications are more general. On the other hand, Nalepa (2010) considers the conditions under which pacts and negotiated settlements among elites facilitate the transition from autocracy to more open access regimes. She finds that the transition will prove unstable (if it even takes place) unless a specific type of limit condition holds: namely, that key players receive credible commitments that the “skeletons in their closets” will not be revealed or that they will receive amnesty for politically problematic behavior during the old regime. Her analytic narrative focuses on Eastern Europe but is applicable to a wide range of cases where transitional justice is at issue.

Analytic narrative approaches are also useful in understanding why some reforms succeed while others do not in countries experiencing development. Methodologically, this demands explicit recognition of the comparative statics, on the one hand, and the off-the-path-behavior, on the other. These features distinguish analytic narratives from other case studies, enabling them to reveal processes and causal mechanisms that might otherwise go undetected and to provide the bases for generalizations that might otherwise not be possible. Some authors already self-consciously attempt analytic narratives (e.g., Hosman 2009 on Nigeria’s failed oil policies), but many do not. Even so, we could get additional leverage on reform by transforming existing case studies into analytic narratives.

Among the many accounts of reform, we have selected two categories of cases where analytic narratives clearly increase explanatory power for the particular instance as well as making the findings transportable to other situations.

The first set is where the same laws have different effects in different places. The study of the Sierra Leone decentralization reforms (Clayton, Noveck, and Levi 2015) display not only varying impact but also reasons for that variation. Sierra Leone has a long history of tension among elites at different levels of government and a more recent history of tensions among key local elites. This case explores the consequences of the latter for effective public service delivery once decentralization is introduced.

A rich literature (cited in the case study) reveals contradictory expectations of the effects of local interelite dynamics. The narrative reveals considerable county-level variations in power sharing between the traditional power-holders, the paramount chiefs, and the newer power-holders, the elected councilors. This variation provides an opportunity to derive expectations specific to the case and then assess their plausibility. Indeed, from the narrative the authors hypothesized that competitive relationships among the two improve services while collusion reduces their quality.

The first challenge was to offer a measure of elite dynamics to be used in a statistical investigation of the implications of differences. To test the impact of this relationship required, first, a measure of the nature of their interaction. The probability of collusion was operationalized by using data that reports on the following direct relationships: the median number of times councilors report having contacted a chief in the previous month, the percentage of councilors that report having had a dispute with a chief during the past month, and the percentage of councilors that report that they are related to a Paramount Chief either through blood or marriage.

But what accounts for the distinctiveness of power-sharing arrangements? The strategic interaction underlying the implementation of the reforms reveals two possible equilibria of collusion and competition. Digging deeper uncovers factors that possibly change the strategic interactions among the key actors and, thus, the comparative statics. The authors considered the gender, age, and party of the councilors as well as the degree of electoral competition. But the most telling explanatory factor was one that could only be known by knowing the case in depth: the proportion of councilors who were in the Civilian Defense Force (CDF) during the war. Former CDF councilors were likely to have forged deep ties with the paramount chiefs who were the primary sources of funding.

The next step was to determine the extent to which these different kinds of power-sharing arrangements – and the probable causes of them – actually influenced service delivery. Drawing out testable implications that could in fact be explored with the available survey material enabled the authors to provide additional confidence in the hypotheses they had derived. Further statistical tests indeed suggested that collusion, particularly that produced by the proportion of councilors who had been in the CDF, leads to far poorer service delivery than does competition.

Bangladesh, Honduras, China, and the United States all have similar labor laws on their books but very distinctive actual protection and enforcement of labor protections in the supply chains of global brands. Berliner et al. (2015a, 2015b) investigate the clusters of stakeholders and what transforms the relations of power among them. To do this, they consider the strategic interactions among key players and what transforms the current equilibrium or status quo. Using the logic of game theory but not formalizing it, they are aware that it is off the equilibrium path for workers to organize and make demands unless they are assured that they will not be punished for their actions by losing their jobs or being sent to jail.³ That only happens when brands find it in their interest to improve worker rights and benefits, and this only occurs if government is upholding its laws or the reputation of brands among consumers is being threatened.

Unfortunately, both of these circumstances are most likely to occur when there is an unexpected (if predictable) catastrophe such as a major fire or building collapse where workers’ lives are tragically lost. Reflecting comparative statics, such a shift leads to reform, but whether the commitments are credible depends on the creation of legal institutions that are hard to change and that incorporate sufficient administrative capacity to implement the rules. The testable implications may differ among the cases, but they are the organizing principles of the cases. The findings are not promising for labor rights. Honduras and Bangladesh lack the government capacity to maintain a positive labor rights regime over a long period of time. In China, the government has the capacity but not the will to establish meaningful labor rights, although it does ensure some protections. The United States, which once had both the will and capacity, now lacks the first and possibly the second and so has undergone a reversal. The result in all these instances is that the pressure on brands to discipline their supply chains is episodic or nonexistent.

The second set of cases document instances where societal interests come to trump private interests, making it possible to actually implement policies that will serve the population as a whole. As we saw with the labor illustration, it is difficult enough to ensure the protection of the interests of a neglected group within the society. It is arguably harder to protect general interests, as the case of corruption in Indonesia (Kuris 2012a, 2012b) documents. And it is arguably harder still to implement policies where the interests of the world at large are at issue, as the case of deforestation in Brazil (Jackson 2014, 2015) details.

Of course, in each of these cases a range of stakeholders are the beneficiaries or losers from policy change. To transform past practice required some combination of leadership, interests, expert knowledge disseminated widely, trust relationships, monitoring, new forms of direct enforcement, credible commitments, and mobilizations that changed the incentives of both government officials and recalcitrant stakeholders. All of these features are documented in these cases, and documented well. Lacking is a structure to the accounts that makes it possible to observe the causal mechanisms and derive testable implications. The comparative statics are not sufficiently explicit. While the Sierra Leone decentralization and labor standards cases do not provide an actual formalization of the game, the presentation of the material makes it possible not only to derive but also to test implications that enhance confidence in the claims of the authors and make them generalizable to other cases.

11.6 Creating Take-Aways

Multiple interpretations are inherent in the traditional case study method. Moving beyond traditional approaches, analytic narratives provide two methods for establishing the generalizability of findings from case studies. First, the model in an analytic narrative often affords a range of explanations and predictions. Although the main account of a unique case may not be testable, the model may yield other predictions that can be tested, either in this case or in other cases. Second, as with other methods, out-of-sample tests constitute an important route to generalization. The presumption today in social science research is that the authors will provide those tests themselves. However, seldom does the level of knowledge for the out of sample case rival the detailed understanding of the original case that puzzled the author. The demonstration of generalizability must rest on a larger community of scholars who take the findings applicable to one place and time to illuminate a very different place and time. Each case then becomes a case among many that are grist for the mill of scholars, experts on particular countries and sectors, and policy-makers who must work collaboratively to sort out the lessons learned.

In this chapter, we have outlined the analytic narrative approach and, in Section 11.5, suggested the potential value of the approach for problems of development. Reflecting the interest of the authors employing the approach, the applications tend to focus on political issues, such as political stability and violence. The approach also applies to case studies of particular economic reforms, and we believe it will produce valuable results in this area.

In summary, the goal of analytic narratives is to provide several forms of discipline on the structure of case studies, such as a game, with emphasis on comparative statics and on off-the-path-behavior, and on predictions that can be tested on aspects of the case even if the main assertion about the case cannot.