Table 7.1 Rosenau’s Pre-Theory
When we speak of theoretical integration in FPA, we may consider two different types of integrative attempts. The first type of integration is fundamental to the purpose of FPA, and that is theoretical integration across levels of analysis to the end of producing an integrated explanation of foreign policy decisionmaking in particular cases. However, there is also a second form of integration that is also important to the subfield of FPA, that is, the desire for greater integration between FPA and IR. Despite the fact that FPA is seen as a subfield of IR, the relationship between the field of IR and its subfield FPA has been, somewhat counterintuitively, disengaged. In this chapter, we will examine each of these two integrative enterprises in turn. In chapter 8 we will discuss integration between FPA and fields beyond IR, as well.
Foreign Policy Analysis theory, as we have seen in this textbook, is rich, detailed, multilevel, multidisciplinary, and centered on foreign policy decisionmaking (FPDM) as it is performed by human beings. There is a catch. You may have noticed that this textbook’s chapters examined theory at each of several levels: personalities of decisionmakers, small group effects, large group effects, culture, domestic politics, national attributes, systemic influences. That is because it is relatively straightforward to examine each separate level of analysis. But the true promise of Foreign Policy Analysis (FPA) must be theoretical integration: the integration of theoretical insight across these several levels to develop a more complete perspective on foreign policy decisionmaking. Such an integration fosters several goals.
First, theoretical integration permits scholars to assess the interrelationship among factors at different levels of analysis. In a theoretical sense, this is quite important. Examining only variables at different levels of analysis is a bit like figuring out a chemical reaction taking place in a vacuum. In the real world, variables at different levels of abstraction are clearly interacting. For example, we can presume that any given high-level small group of foreign policymakers will contain a variety of individuals, each of whom: possesses unique personalities; is embedded in national and subnational cultures; is likely to either represent or favor particular organizations which play particular stakeholder roles in the bureaucracy; is aligned with larger political, ideological, or religious groups; and is living in a nation-state with specific national characteristics that help define its place in the international system. The permutations and contingencies are manifold. Adopting a self-consciously integrative lens reveals how relevant factors, regardless of the level of analysis from which they originate, will always interact with and be shaped by a range of other factors. New concepts, new propositions, and new generalizations may arise from attempting cross-level theoretical integration. Theory improves.
Second, explanation improves. The integration of theory demands that one assess the scope conditions under which variables at certain levels of analysis prove more important in an explanatory sense than variables at other levels of analysis. There may also be nonintuitive interaction effects, where variable 1 by itself might augur for X behavior, variable 2 by itself might also augur for X behavior, but variables 1 and 2 together might augur for Y as a behavioral predisposition. Analysts begin to gain a more nuanced and hopefully more accurate perspective on the use of FPA variables to explain a particular decisionmaking episode as they attempt integration.
Third, estimation improves. For foreign policy analysts working in a professional setting where FPA is used to gauge likely behavior of other nation-states over time, only theoretical integration will permit coherent estimation. If an analyst is tasked with figuring out what the North Korean regime will do in response to a new U.S. policy initiative, it is possible to combine in ad hoc fashion the variables at different levels of analysis in one’s mind and come up with a rough projection. But surely that integration process is inferior to one in which the interrelationships between these variables have been made explicit, have been worked out in some detail, and have been probed for validity and reliability by subjecting them to historical and counterfactual testing.
For all these reasons, theoretical integration is an imperative for FPA. Nevertheless, there are many obstacles to integration, and it remains a promise yet to be fulfilled in FPA. Most FPA integration is instantiated through qualitative case studies in which the particulars of the case under investigation suggest relatively ad hoc case-specific principles of cross-level integration. One excellent example of this approach is Steve Yetiv’s in-depth analysis of President George H. W. Bush’s decision to invade Iraq in 1990–1991 (2011). While suggesting that groupthink characterized Bush’s inner decision-making circle in this case, Yetiv also examines several other levels of analysis, including domestic politics, international politics, and cognitive factors. According to Yetiv, foregrounding process is key to the integrative enterprise as it allows for a theoretical form of “layered thinking” to take place (210). That is, the cognitive level provides insight into preference formation among key actors, for preferences form as “a function of the discursive environment created by the decisionmakers” (204). This discursive environment is shaped by the identity of the group’s members, which in turn may be molded by bureaucratic role. Discourse is broadened by the framing and counterframing inherent in domestic politics.
In the case of the decision to launch Operation Desert Storm, Yetiv argues that bureaucratic politics did not play an important processual role. Meanwhile, the Bush administration was able to dominate the framing at the level of domestic politics. With domestic level factors bracketed off from influencing decisionmaking process, factors operating closer to the individual level came to the fore. Yetiv finds the groupthink tendencies of Bush’s inner circle, alongside Bush’s own adoption of the “Munich frame”—a mindset which is skeptical of the prospects of appeasing leaders who seek territorial expansion and takes its name from the failure of the Munich Pact to stem Hitler’s aggressions in Europe—were the most relevant factors at play. Indeed, these two factors (groupthink and leader framing) reinforced each other.
Yetiv’s qualitative case study approach to integration allows him to explore the “unique confluence of historical, global, regional, bureaucratic, group and, above all, individual dynamics at play here that help explain how and why the United States went to war” (183). This approach to integration is arguably the norm in FPA; the FPA scholar presents arguments about which level of analysis were more or less important in a given case, with group decisionmaking processes being at the theoretical heart of the enterprise. There is nothing amiss in this approach, and there is good reason why it remains the norm.
Yet it is also true that some FPA scholars have sought a more rigorous theory-building enterprise; one which is less ad hoc, less case-based, and allows for a more systematized understanding of how the level of analysis intersects and interacts. While the prospective outcomes of such an enterprise are clearly desirable, realizing them is exceedingly challenging. Those few attempts at such systematization of cross-level integration have faced daunting obstacles, as we will see shortly when we examine the prominent systemized integration efforts in the history of FPA. Before we do so, however, it is worth exploring the nature of these obstacles as a way of unpacking what theoretical integration entails.
The first obstacle to theoretical integration is that FPA data is impressive in its quantity and diversity. A full FPA explanation along the lines described in this textbook usually synthesizes a vast quantity of information. And that information may be at various levels of measurement precision: categorical or nominal, ordinal or ranked, interval, or ratio level. Whatever means are devised to perform theoretical integration must be able to manipulate and integrate large amounts of information that may be difficult to weave together in a straightforward methodology.
A second obstacle is that foreign policy decisionmaking is dynamic and full of contingencies and creative agency. Social science methods, generally speaking, are not well equipped to handle dynamic systems, especially those defined at lower levels of measurement precision. Furthermore, data must be tracked almost continuously to identify contingency points. For example, at a critical meeting of the COMOR (Committee on Overhead Reconnaissance) group during the Cuban missile crisis, the director of central intelligence (DCI), John McCone, was absent because he was on his honeymoon. Had he been present, the outcome of that meeting, which was to have the U-2s fly only around Cuba’s periphery, might have been different because of his strong feelings on the matter. It was not until early October, when he had returned to COMOR meetings, that McCone finally won the right to make overland U-2 flights of Cuba, and those flights discovered the Soviet missile construction there. If he had not been on his honeymoon during the latter part of September, McCone might have secured those overland flights two weeks earlier, and the course of the Cuban missile crisis might have been quite different. In addition to these types of contingencies, the analyst in quest of theoretical integration must also take into account the possibility of human creativity. Human creative agency may produce forces that cannot be modeled; it may produce outcomes that have never been seen before. No social science model could ever possibly capture the entire horizon of foreign policy decisionmaking.
A third obstacle is that some data is likely to be missing. Whether because the regime is highly secretive or for some other reason difficult to obtain information about, or because the analytic task must be performed in real time where situations are changing rapidly, it is quite often the case that some data, perhaps pivotal data, will not be available to the analyst. This means that theoretical integration as a scholarly endeavor is likely to proceed on the basis of extremely well-documented historical cases where most data points are accessible—and that it may be difficult to apply such integrated theory in cases where key data points are unobtainable. It is no surprise, therefore, that the most ambitiously integrative efforts tend to re-interrogate well-examined cases, as was the case with Yetiv example we highlighted earlier, or other oft-studied episodes such as the Cuban missile crisis.
A fourth obstacle is the question of what the explanatory or applied output of theoretical integration in FPA would look like. Does it look like a probability distribution? If so, a distribution over what? “Types” of foreign policy behavior or choice? And if so, how are such “types” defined, and at what level of abstraction? Or perhaps the output looks like a contingency diagram, or a series of if-then statements? Perhaps the output is a set of statements concerning the generalized predispositions of a particular regime at a particular point in time? Perhaps the output is an actual point prediction? The methods chosen to implement the integration and the purposes for which the analyst desires integration will all affect the output of their endeavor. This reality is evident in each of the four types of systematized cross-level integration efforts that we now move to review: Rosenau’s Pre-Theories; the frameworks proposed by Brecher (qualitative) and Wilkenfeld (quantitative); Rule-based Production Systems; and CREON II efforts.
Rosenau’s 1964 “Pre-theories” article (published in Farrell, 1966), discussed in chapter 1, is noteworthy not only for its significance as one of the founding articles of FPA, but also for its precocious attempt at systematized theoretical integration. After all, the field did not really even exist in 1964, and Rosenau was already looking forward to the day when integration would be at the top of every foreign policy analyst’s agenda! In that early article, Rosenau points out:
To identify factors is not to trace their influence. To uncover processes that affect external behavior is not to explain how and why they are operative under certain circumstances and not under others. To recognize that foreign policy is shaped by internal as well as external factors is not to comprehend how the two intermix or to indicate the conditions under which one predominates over the other. (1966 [written 1964], 98)
Rosenau does not stop with a call for integration. He tries his hand at it in an exercise that he labels “pre-theorizing,” by which he means developing a meta-theoretical approach to integration. He takes as his departure the metaphor of Mendelian genetics, with its distinction between genotype and phenotype. One of Mendel’s remarkable achievements was to demonstrate that genotype determined phenotype, and that two similar phenotypes might have different underlying genetics. Rosenau posited that there was a genotype of nations, and that this underlying genotype would tell us about the relative importance of variables at different levels of analysis in FPA.
To demonstrate how this approach to integration would work, Rosenau needed to choose “genotypic” variables as well as names for the “clusters” of variables to be found at different levels of analysis. The genotypic variables Rosenau chose were size, wealth, and political system, all dichotomized (large/small; developed/underdeveloped; open/closed). The clusters of explanatory variables whose importance he ranked according to genotype are individual-level variables (e.g., personalities of leaders), role variables (e.g., national role conception), governmental variables (e.g., domestic politics), societal variables (e.g., national attributes and more cultural variables such as level of national unity), and systemic variables (e.g., bipolar, multipolar, etc.).
Rosenau presents a diagram to illustrate how integration might take place. A revised version of that diagram appears in table 7.1.
We can immediately notice a few generalizations that Rosenau is making. The individual-level variables have the least significance for developed/open states, and the most significance for underdeveloped states. Role variables are most important for developed states but are never lower than third rank for any genotype of nations. Systemic effects are much more important for small states than for large states. Governmental variables are never higher than third rank for any type of nation and are least important for underdeveloped/open nations. Societal-level variables are least important for closed nations. And so on.
Rosenau also felt that it was important to measure the degree to which the internal and external environments of nations were meshed and posited another variable of “penetration.” Furthermore, he felt that these rankings might differ according to the specific issue area involved.
It was an ambitious beginning, and it is easy to be critical in hindsight. However, the importance of Rosenau’s contribution is hard to overstate: this was the first meaningful attempt to suggest how one might accomplish a more systematized and less ad hoc cross-level integration of FPA explanations. And yet we must admit that Rosenau’s pre-theory still does not give us the necessary scope conditions or a substantive understanding of the integration of these variables, even though Rosenau himself held these up as the benchmarks for success.
Michael Brecher and Jonathan Wilkenfeld worked together on several projects involving crisis and crisis behavior. They also individually attempted to offer a vision of FPA theoretical integration. What is striking is that these two collaborators offered very different methodological approaches to integration. Brecher’s instantiation of theoretical integration was to be accomplished through qualitative historical case study. Wilkenfeld’s instantiation was through multiple regression methodology.
Brecher’s The Foreign Policy System of Israel (1972) remains a classic of FPA. Through exhaustive historical case studies and content analysis, Brecher integrates an examination of many different FPDM variables. Brecher develops a theoretical framework showing the interrelationship between several clusters of input and output variables by positing two important environments for decisionmaking: the operational environment and the psychological environment.
The operational environment refers to the set of potentially important factors that arguably set the parameters for FPDM. In Brecher’s framework the operational environment consists of two parts: external and internal. In the external realm, relationships and issues are variously placed at the global level, issue- or geography-based subsystem level, and/or bilateral level. The internal sector of the operational environment examines attributes of the state and the polity, such as military capability, economic capability, political structure, interest groups, and competing elites. Information about the operational environment is conveyed to the decisionmaking elite through a variety of means, including firsthand knowledge, media reports, and contacts with members of society.
This information is then filtered through the psychological environment of the decisionmakers. This environment is also characterized by two aspects: elite images and what Brecher terms the “attitudinal prism.” Elite images refers to the interpretation elites have placed upon the information communicated to them about the operational environment. As Brecher notes, perception of reality might not correspond to reality, but even so the perceptions held by elites may be much more formative of foreign policy than objective measures of the operational environment. The attitudinal prism refers to the attitudes generally held in society concerning their identity and history, which also color elite beliefs and attitudes.
Brecher’s framework then addresses the formulation of decisions, typologized into several categories for ease in determining which parts of the operational and psychological environments are most pertinent, and also the implementation thereof. Feedback loops help update the system. Overall, Brecher’s framework for integration resembles the diagram in figure 7.1.
Brecher’s framework helps us order our integration effort, pointing out clusters of variables and suggesting how they might interrelate. Issues of scope conditions or rules of integration are left unspecified. Like Rosenau’s effort, though much more specified and accompanied by case studies of how such integration can be accomplished in the particular case of Israel, Brecher’s framework is more an explication of a philosophical approach to integration.
Jonathan Wilkenfeld, a sometime collaborator with Brecher, offered a vastly different approach to integration, with his coauthors Gerald Hopple, Paul Rossa, and Stephen Andriole, in the book Foreign Policy Behavior: The Interstate Behavior Analysis (IBA) Model (1980). Wilkenfeld and colleagues articulated a very detailed model, with independent, intervening, and dependent variables.
Independent variables included clusters such as psychological, political, societal, interstate, and global. Subclusters included variables such as the following:
The intervening variables are a classification of state types. The two main characteristics examined are state capabilities and governmental structure. Capabilities involve size (area, population, gross national product), military power (military manpower, defense expenditures, defense expenditures per capita), and resource base (percentage of energy consumed that is domestically produced). Governmental structure involves political development (number of parties, power distribution, local government autonomy), political structure (selection of executive, legislative effectiveness, selection of legislators), and political stability (coups, constitutional changes, major cabinet changes, executive changes). Wilkenfeld and his colleagues perform a Q-sort factor analysis to come up with a fivefold typology of states as Western, closed, large developing, unstable, or poor.
Their classification of the dependent variable is also quite involved. Though they desire to create a six-dimensional classification for foreign policy behavior (spatial, temporal, relational, situational, substantial, and behavioral), given data collection constraints they are forced to perform a factor analysis on World Event/Interaction Survey (WEIS)–coded data. This produces three broad factors: constructive diplomatic behavior, nonmilitary conflict behavior, and force.
Thus, the integrative framework looks as shown in figure 7.2.
Unlike Brecher, the approach to implementing this integrative framework is quantitative. Using partial least squares regression, Wilkenfeld and his colleagues give us the partial correlation coefficients between each rectangle above, and also the correlations for the independent variables given different values of the intervening variables. These coefficients are to tell us the “relative potency” of each variable and variable cluster as it relates to accounting for the variance in the three types of foreign policy behavior. Results are explained in discourse such as the following excerpt:
Several noteworthy findings emerge. First, the overall model explains 94 percent of the variance in foreign policy behavior. In terms of the three behavioral dimensions, it explains 74 percent of the variance in constructive diplomatic behavior, 61 percent of the variance in nonmilitary conflict, and 50 percent of the variance in force. Clearly the model does quite well in explaining foreign policy behavior, although the more routine actions, particularly of a diplomatic nature, are better explained than the force and conflict acts. (Wilkenfeld et al., 1980, 197)
It should be clear from the above quotation why, though certainly an integrative effort, the IBA Project was eventually abandoned. Relative potency testing is well and good but does not address the more important questions of FPA theoretical integration. Furthermore, the specification of the dependent variable is simply too crude to be useful. Wilkenfeld and his coauthors state, “No attempt has been made to develop more sophisticated causal models, building upon the results of the relative potency tests. Such models should now begin to stress the complex types of interrelationships among the clusters of determinants, as well as a variety of mediated relationships between the determinants of foreign policy and its various behavioral manifestations” (Wilkenfeld et al., 1980, 243). These more sophisticated models were never realized.
It is noteworthy that two of the coauthors on this project, Stephen Andriole and Gerald Hopple, also worked on the EWAMS (Early Warning and Monitoring System) project for the Defense Advanced Research Projects Agency (DARPA) in the late 1970s and early 1980s, as mentioned in chapter 1. EWAMS also used events data, in this case to monitor and predict the use of force.
During the late 1980s and early 1990s, advances in computer technology and artificial intelligence allowed foreign policy analysts to play with new means of implementing integration. One of the most oft-experimented-with approach of this time period was that of rule-based production systems.
A rule-based production system is simply a group of interrelated “if-then” statements. The very simple diagram in figure 7.3 could be used to describe how, say, actions upon waking up in the morning are produced:
This very simple diagram explicates the rules that a person is using to produce behavior. “If” the alarm clock rings, “then” turn off the alarm clock. At the second tier, note that we can make a probability distribution if we so desire: “If have turned off the alarm clock, then 10 percent of the time fall back asleep/90 percent of the time get out of bed.” Notice that we have also specified two different routes to brushing your teeth: you could fall back asleep, have your kids jump all over you, and then brush your teeth. Or you could turn off your alarm clock, get out of bed, and brush your teeth. If we wanted to, we could include a numeric variable: “If snowfall is greater than three inches, then go back to bed.” Notice we could also make this into a computer program:
In other words, a rule-based production system is one of the most flexible instruments for theoretical integration one could imagine. If you can conceive of a specific relationship between any two variables—regardless of their form or level of measurement precision—you can make a rule-based production system. And if you can make a rule-based production system, you can simulate, by programming a computer, the entity you are investigating. Some of the most intriguing early rule-based production systems in FPA were of individuals, such as the system JFK, or of states, such as the system CHINA_WATCHER, or more general decisionmaker systems, such as POLI or EVIN (Thorson and Sylvan, 1982; Tanaka, 1984; Taber, 1997).“Situational Predisposition” (SP) was the name of the rule-based production system that Hudson created many years ago (Hudson, 1987). The aim was to create a very simple system based on few variables that would produce predictions of foreign policy behavior. Assessing the accuracy of the simple system would allow one to say under what conditions one would not need all of the detail of a full-fledged FPA-style analysis and under what conditions one would. An overview of what Hudson had to do to create that system will explain both the promise and the downside of using rule-based production systems.
SP used the Comparative Research on the Events of Nations (CREON) events data set as its dependent variable. CREON used four main dimensions to describe events: affect (positive/negative), level of commitment (words/ deeds), instrumentality (diplomatic/economic/military), and target (identity of the “direct object” of the event). The goal would be to “postdict”: that is, using the independent variables of the SP framework, Hudson would create “postdictions” as to what one would expect would have happened in an actual historical event. These postdictions could then be compared with the CREON events to determine if the SP system had postdicted accurately or not.
But first Hudson had to create the rule-based production system and explicating the steps of this creation will be instructive as to the pros and cons of this approach. The independent variables were situational roles of the nations involved, type of situation (derived from situational roles), prior affect between the acting nation and the other role occupants, capabilities of the acting nation relative to the other role occupants, and salience of the other role occupants for the acting nation. There were three basic situational roles: actor, source, and subject—with the last two roles defined from the perspective of the acting nation.
In any foreign policy situation, there is a problem that is the occasion for decision. The acting nation (or actor) whose behavior we want to explore must decide which entity or entities have caused the problem—that is, who occupies the role of “source.” The actor must also decide what entity or entities are directly affected by the problem caused by the source—that is, who occupies the role of “subject.” Thus we can see that different types of situations are defined by the identification of the other role occupants by the actor. So, for example:
There turn out to be five main types of situations defined in terms of role occupants. However, in order to get from situation to behavior, more information is necessary. The actor must assess its relationship to the other role occupants. This relationship is defined in terms of some very simple questions: Do I like them or not? (prior affect). Are they stronger than I am or not? (relative capabilities). Are they in some way especially important to me or not? (salience). Consider the same situation with two very different permutations of the relationship variables:
In the first case, we can make the prediction that X will forcefully, perhaps even aggressively, attempt to stop Y from continuing to create a problem situation for X. However, in the second case, we would predict that X will attempt to entreat with Y, as a much stronger and important friend, to recognize the problem Y is causing for it and persuade it to stop. Same situation, two very different behavioral predictions.
After collecting all of the data, the most important task facing Hudson was to create the “if-then” statements that would lead from each permutation of independent variables to a prediction on the dependent variables. Though ostensibly a very simple model, Hudson ended up having to posit 191 “if-then” statements to create a complete system to cover all possible permutations of the explanatory variables. Actually, it was not the sheer number of statements that was the problem. According to Hudson, “This brain-racking exercise demanded that I understand how each of the variables interacted with one another and how those differences in interaction would lead to differences in behavior. Sometimes that task seemed almost impossible, but it did force me to create three levels of rules: isolation rules, meta-rules, and interaction rules.”
Isolation rules are rules about how one particular variable will influence behavior without regard to what the other variables’ values are. Without isolation rules, no other level of rule is possible. So, for example, one example of an isolation rule is: If the SOURCE possesses a significant CAPABILITY ADVANTAGE over the ACTOR, the actor will most likely respond by using DIPLOMATIC INSTRUMENTS and will NOT use HIGH COMMITMENT. What is being posited is that when relative capabilities do not favor the actor, regardless of what else is going on in the situation, that variable will have impact on two dimensions of behavior: instrumentality and level of commitment.
But isolation rules aren’t enough. You have to figure out how all of the explanatory variables will interact with one another. A first step in this direction was the positing of “meta-rules,” or rules about rules. In the SP system, meta-rules took one of several forms: ignore/precedence rules, additive/augment rules, cancel/dampen rules. An ignore/precedence rule might be, If SITUATION is ASSISTANCE CONSIDERATION, PRIOR AFFECT takes PRECEDENCE over all other relational variables. An additive/augment rule might be, If SITUATION is CONFRONTATION and SOURCE and actor have NEGATIVE prior affect, when SOURCE is WEAKER than ACTOR, this will AUGMENT the effect of prior affect. A cancel/dampen rule would be the reverse: If SITUATION is CONFRONTATION and SOURCE and ACTOR have NEGATIVE prior affect, when SOURCE is much STRONGER than ACTOR, the effect of prior affect will be DAMPENED.
Now all these isolation and meta-rules are but precursors, then, to the final rules that must specify a “production” for each possible permutation of variable values. Hudson called these final rules the “interaction rules.” So a final interaction rule might look like this: In a CONFRONTATION SITUATION, if PRIOR AFFECT between the ACTOR and the SOURCE has been NEGATIVE, and the SOURCE is SALIENT for the ACTOR, and the actor’s relative capabilities are GREATER THAN those of the source, based on the ISOLATION and META-RULES for this combination of variable values, the likely behavior attribute values for the actor will be HIGH NEGATIVE AFFECT, MODERATE COMMITMENT, DIPLOMATIC INSTRUMENTS, with the SOURCE as TARGET.
This exercise brings to light some of the upsides of rule-based production systems, but also some of the downsides. On the positive side of the ledger, the very methodology demanded specificity about how FPA variables were to be integrated. The form of the integration was also completely up to Hudson, who was not stuck using mathematically based relationships such as addition and multiplication, for example. Second, the system was complete. Every possibility had to be examined—or the computer program would not run. Third, all sorts of variables could be combined together: nominal- and interval-level variables could easily be combined in one rule. Last, it was possible to actually say something about accuracy of postdictions made (in case you are interested, Hudson discovered that in about 30 percent of the 6,605 cases examined, SP was sufficient to accurately predict the resulting foreign policy behavior).
However, there were some definite downsides. This was a small model, very un-FPA-like in the number of variables it incorporated, and Hudson still ended up not only with 191 final rules, but also multitudinous isolation rules and meta-rules to boot. The exercise forced Hudson to go beyond what she knew about the variables’ anticipated interactions simply to accommodate the exponential growth of permutations. In a way, the dependent variables were overdetermined by the complexity of the rules, and as the rule maker, Hudson often felt overwhelmed in the attempt to find differences in the dependent variables based upon all the permutations of the independent variables. Overall then, despite the benefits this approach offers, this method of integration remains almost too demanding in light of the current lack of specificity of the theories that are to be integrated.
No discussion of systematized integration would be complete without an examination of the most ambitious integration project in FPA history: the CREON Project, mentioned in chapter 1 as an events data project. Here we will examine what we term CREON II; this was the explicit effort later in the 1980s and 1990s to integrate across levels of analysis in a nonarithmetic fashion. Before we look at CREON II per se, it is worthwhile to examine a preliminary integration exercise undertaken by CREON investigators with regard to decisions taken in 1972 and 1973 by the Soviet Union concerning the sale of advanced weaponry to Egypt. CREON II would continue some of the same themes as this earlier integration exercise.
The preliminary integration exercise was undertaken by a team comprised of Philip Stewart, a country expert on the Soviet Union; Margaret Hermann, a political psychologist; and Charles Hermann, who studies group processes in decisionmaking. The team sought to understand why the Soviet Union refused to send weaponry to Egypt in 1972, but then reversed course and sent it in 1973 (P. Stewart, Hermann, and Hermann, 1989). This was an interesting case because at the time the article was written, the Soviet Union was a closed regime about which little information was publicly available. Could this team use unclassified information to answer this foreign policy question?
The strategy of theoretical attack was also noteworthy. First, the country expert, Stewart, was asked to determine what type of decision group the Soviet Politburo was at this particular time period. Stewart decided it had an oligarchic power structure, where some members mattered more than others. Stewart identified the strongest members of the Politburo as being Brezhnev, Kosygin, Podgorny, and Suslov.
Stewart then handed off the baton to Margaret Hermann, whose task was to inform the team about the background, expertise, preferences, and strength of preferences of the members of the Politburo. She examined the background of each man, noting whether they were generalists or careerists and from which parts of the bureaucracy they had arisen, in an attempt to determine the strength and nature of their organizational affiliation. She also content analyzed their speeches to decide how important Middle East issues were to each man, what their stance was and how strongly they held it, and whether they were generally sensitive or insensitive in a cognitive sense to the world around them. She discovered that there was only one member of the Politburo who had any strong feelings about the Egyptian situation, and that was Marshal Grechko.
Margaret Hermann then passed the baton to Charles Hermann, who on the basis of the information he had been provided classified each man according to whether he was an advocate for a particular position, a cue taker who would generally follow the direction in which the majority were moving, or a broker whose support was necessary before any advocate could succeed and whose opinions could be swayed by advocates. Hermann then created a decision tree to help him decide what position each man would take in group deliberations. Figure 7.4 shows the branch of the decision tree for Grechko.
Charles Hermann then traced the change in group dynamics over the period from 1972 to 1973, noting that in 1973 Kosygin, a sensitive cue taker, moved to support Grechko’s position due to changes in Anwar Sadat’s foreign policy orientation. Suslov, a broker, supported this move. Brezhnev, normally a broker, acquiesced. Thus Stewart, Hermann, and Hermann are not only able to show why the Politburo changed its position on this issue, but exactly how that change came about.
This integrative exercise contains some of the same elements the later, larger CREON II project would incorporate. First, foreign policy analysts worked hand in hand with country experts, using information from country experts as inputs to the model, and asking country experts to comment on the workings of the model itself. Second, the central element of the analysis became the decision unit—the actual individual or group who will make the decision. Third, there was a series of successive “cuts” at the analysis, with each specialist making a contribution upon which other team members in other specialties could build.
Given that the CREON II model was centered on decision units, the overall CREON model at its most abstract level appears in figure 7.5.
We’ve already met situational predisposition (SP); here it becomes an input to the central element of ultimate decision unit. (The Societal Structure and Status model subcomponent was never created.) Foreign policy behavior is operationalized along the lines discussed with situational predisposition, i.e., actor, affect, commitment, instrument, and target. The most important contribution of CREON II, however, is the set of rules that will integrate theories pertinent to the decisionmaking of the ultimate decision unit.
The ultimate decision unit can take one of three basic forms, with variants. The predominant leader decision unit is where a single leader has the power to decide, on his or her own if desired, what the foreign policy behavior of the nation will be. There are two variants of this type: the leader who is insensitive to contextual information, and the leader who is sensitive to contextual information. This will be determined by psychological analysis of the leader’s personality. The second type of decision group, the single group decision unit, has variants that revolve around the nature of the loyalties of group members as well as the nature of the decision rules. The analyst’s task is to identify groups where the loyalty is to the group itself; groups where loyalties are to entities outside of the group and where a majority is required for decision; and groups where loyalties are again outside of the group but where unanimity is required for decision. Multiple autonomous groups as the third type of ultimate decision unit are fairly infrequent, but they do occur, as in the case of military juntas. Here the variants depend upon whether the groups have established means of working with one another, especially in the case of conflict of opinion about the desired course of action. Variants include multiple groups where unanimity is required for action; multiple groups where there is an established process of working majority decisionmaking; and multiple groups where there is no real established process for decision.
Determination of the type of ultimate decision unit carries with it theoretical implications concerning which FPA theories would be most relevant to examine to understand what this type of group is likely to decide. The CREON researchers developed a decision unit typology summarized in table 7.2.
This was a real theoretical contribution: by putting type of decision unit at the heart of the analysis, one could highlight the insights of theories most pertinent to that particular type of decision unit. However, this step still wasn’t enough. The analyst still couldn’t get to foreign policy behavior specification from a chart like this. So the CREON II researchers came up with the decision trees needed to put this integration together in a way that could lead to behavioral projections. Figure 7.6 shows just one tree, for one variant of the single group decision unit.
For this particular type of ultimate decision unit, the chart shows that the authors have used the theories of coalition formation to try and decide what questions they should be asking about this group. Behavioral predictions are given, but they are admittedly still at a fairly abstract level, such as “lopsided compromise.” Such charts will not really mean much until they are put into action. So CREON II decided to do just that. They asked country experts to develop case studies of the different variants of decision units. Then the applicable decision tree would be used by the country expert, with the expert providing the answers to the questions in the tree, and the expert would look to see if the tree led to a behavioral projection that matched, or at least described well, the real outcome. Then the country expert was asked to provide feedback about the whole approach and also the particular tree used in their case study. It was an immensely ambitious undertaking. There is no other integrative effort in FPA that even comes close to what CREON II attempted (M. Hermann, 2001).
Nevertheless, this integration effort also had some significant shortcomings, which undermined its ability to be seen as a final solution to the problem of FPA theoretical integration. The country experts who were asked to use the decision trees came back with some important feedback. First, the idea of “occasion for decision,” that is, the foreign policy problem that allows one to start moving down the branches of the trees, is a bit messier than that assumed by the tree framework. For example, most important foreign policy decisions are not made in one sitting; they may be drawn-out affairs in which a mix of decision units may be involved. Indeed, one suggestion made was that decision units may need to be understood more as a dependent variable than as a starting input variable. Furthermore, the actual occasion for decision might have layers of predecession, where policymakers have dealt with this same situation or same entities before. Memory of these antecedent occasions for decision are an important input into any particular occasion for decision, but there is no current way of making these memories part of the decision trees. The trees appear to treat the occasion for decision as a tabula rasa, rather than as sequences of linked decisions.
The CREON II integration also does not consider issues of organizational and bureaucratic implementation, which we have seen in previous chapters, may have considerable impact on the foreign policy behavior actually produced. The boxed outcomes at the end of the decision trees are also fairly broadly defined, leaving one to wonder what degree of falsification an output such as “lopsided compromise” or “paper over differences” may afford the researcher.
The country experts also found that they themselves—the experts!—were at a loss to answer every question in the decision tree. The data requirements to use these trees were so high that some of the experts resorted to informed guessing. Furthermore, the questions in the tree had to be answered in fairly definitive fashion, whereas in the real situation being explained it is possible that the members of the decision unit itself might have suffered from a sense of uncertainty. In some cases, the fact that the decision unit–based integration of the CREON II effort did not include variables such as culture was to be lamented. One expert, commenting on the use of the trees to examine a foreign policy decision in Sweden, noted that cultural norms of consensus and consensus building made the decision less one of political bargaining and more one of joint problem solving, but such a distinction could not be made in the existing decision trees.
Even as the field of Foreign Policy Analysis was first being formed, the goal of theoretical integration was put forward as an essential task by its founders, such as James Rosenau. On a case-by-qualitative-case basis, such integration takes places normally on a relatively ad hoc basis. And yet principles for a more systematic theoretical integration cannot yet be said to have been reached, despite several ambitious efforts to do so. At the same time, such theoretical integration cannot be impossible. After all, foreign policy decisionmakers act every day. Somehow they are, in a sense, integrating variables at many different levels of analysis in order to make a decision. If decisionmakers are able to do this implicitly, surely it could be modeled explicitly by foreign policy analysts. And as noted, most detailed foreign policy analyses do perform an ad hoc final integration across explanatory levels of analysis in qualitative case studies (Yetiv, 2011).
Might theoretical integration move closer as we understand how decisionmakers actually themselves integrate across levels while making decisions? Some researchers have experimented with “think aloud” protocols, asking decisionmakers to verbalize what they are thinking of as they make a decision, in order to “see” how integration is actually taking place in real time. But what we are finding is that we are only barely beginning to understand the capabilities and complexity of human reasoning. The new wave of neuroscientific studies that visually map what the brain is doing during thought and emotion are only scratching the surface of what we will uncover in the next several decades. Is it possible that the task of theoretical integration in FPA must await the findings from this new exploration of the human mind? Are we missing necessary elements of theory, methodology, and perhaps even technology? Theoretical integration in FPA must be possible, but it remains a promise yet to be fulfilled for the time being.
As FPA scholars, we find it remarkable that IR and FPA are not more theoretically integrated. As Hudson has argued in chapter 1, the choices made by human decisionmakers in regard to foreign policy constitute the theoretical ground of IR. Nevertheless, there is a noticeable lack of engagement between FPA and IR, even though FPA is considered a subfield of IR. Kaarbo (2015) comments
Current International Relations (IR) theory is marked by a paradox concerning the role of domestic politics and decision making: Domestic politics and decision making are simultaneously everywhere and nowhere. On the one hand, recent developments in realism, liberalism, and constructivism have incorporated domestic level and psychological factors. Compared to 20 years ago, domestic political and decision-making concepts are very much part of contemporary IR theory and theory-informed empirical investigations. On the other hand, much of IR theory ignores or violates decades of research in foreign policy analysis (FPA) on how domestic political and decision-making factors affect actors’ choices and policies. (189; see also Houghton, 2007)
This disconnect is most likely due to the difference in style of theorizing, with mainstream IR favoring actor-general theory as instantiated in schools such as neorealism, and FPA favoring actor-specific theory. But surely this disengagement is not the optimal state of affairs for either side; surely the work of each set of scholars could inform that of the other. And, in a sense, FPA does utilize theories of the international system, as we have seen in the last chapter, and other actor-general theories such as the “rational actor model” in constructing its own theoretical approach. One could say the disengagement appears to be more one-sided, upon closer inspection. Indeed, Kaarbo advances the argument that FPA can offer a bridge across IR paradigms:
Foreign policy analysis is not a conglomeration of realism, liberalism, and constructivism—it would challenge critical, ontological, and theoretical aspects of each. But as IR theories have turned toward domestic and decision-making variables, the FPA perspective can bring them together. This integration is important as each theory is developing along different trajectories with regard to these factors. Neoclassical realists tend to focus on elites, liberals on institutions and societal constraints, and constructivists on ideas and discourse. Consequently, FPA has a separate response to each of these developments, has something to offer each of these avenues of thought, and covers all of them, thus offering a bridge for this significant domestic and decision-making turn in IR theory. That they each have turned is a major point of this article; that they have turned in different directions is an opportunity for FPA to integrate this transtheoretical development. (2015, 207)
In the remainder of this chapter, we examine two bodies of work that seem to bridge most explicitly between IR and FPA: neoclassical realism and behavioral IR. Interestingly, while the latter is a self-conscious attempt to bridge the two fields, the former seems to disavow any such desire. Let us examine each in turn.
In his influential 1998 article defining the school of neoclassical realism, Gideon Rose suggests that classical realism appreciated the linkages between external and internal politics of the state, and laments that neorealism appears so indifferent to them:
A neoclassical realist would return to their roots by explicitly incorporat[ing] both external and internal variables, updating and systematizing certain insights drawn from classical realist thought. Its adherents argue that the scope and ambition of a country’s foreign policy is driven first and foremost by its place in the international system and specifically by its relative material power capabilities. This is why they are realist. They argue further, however, that the impact of such power capabilities on foreign policy is indirect and complex, because systemic pressures must be translated through intervening variables at unit level. This is why they are neoclassical. (Rose, 1998, 146)
Relative power distributions are insufficient to explain the diversity of behavior exhibited in the international system, according to neoclassical realists. Why did Bush 41 not invade Iraq after victory in Desert Storm, but Bush 43 did? An examination of system-level variables alone will be insufficient to explain these differences. The examination of what neoclassical realists call “unit-level variables,” or what is happening within the nation-states, will be required:
Neoclassical realism posits an imperfect “transmission belt” between systemic incentives and constraints, on the one hand, and the actual diplomatic, military, and foreign economic policies states select, on the other. Over the long term, international political outcomes generally mirror the actual distribution of power among states. In the shorter term, however, the policies states pursue are rarely objectively efficient or predictable based upon a purely systemic analysis. (Taliaferro et al., 2009, 4)
Ironically, the system attribute of “anarchy” gives states license to define their own security interests and develop idiosyncratic means of assessing threats. States will vary according to their individual abilities to extract and mobilize resources, in their degree of elite cohesion, and also according to their domestic institutions (such as autocracy versus democracy, but also more fine-grained differences such as presidential versus parliamentary systems). These unit-level variables will affect such important foreign policy processes as threat assessment, strategic adjustment, domestic mobilization, and policy implementation.
Brawley (2009), for example, states that during the 1920s, Germany was clearly the major threat to the rest of the European powers, particularly France, England, and the Soviet Union. According to neorealist theory, some significant bandwagoning between these three powers should have occurred—but it did not. Brawley explains that threat assessment differed sharply among the three nations, with only France seeing an imminent threat from Germany at this time. Accordingly, France pushed for high reparations, while Britain, due to its need to resuscitate its own trading empire, did not agree, seeing in German consumption of British goods a means to this end. However, the British supported disarmament to stave off a future German threat. The Soviet Union, on the other hand, shut out from reparations and agreeing with the French that the Germans would move west first, cooperated with the Germans in order to buy themselves time to build up their military strength. Thus, faced with the very same threat, and faced with what would seem to be a system-imposed imperative to bandwagon, each of these three countries reacted differently, so much so that they could not band together against the German threat.
Taliaferro, Lobell, and Ripsman suggest that these types of unit-level variations lead neoclassical realists to ask questions such as:
How do states, or more specifically the decision-makers and institutions that act on their behalf, assess international threats and opportunities? What happens when there is disagreement about the nature of foreign threats? Who ultimately decides the range of acceptable and unacceptable foreign policy alternatives? To what extent, and under what conditions, can domestic actors bargain with state leaders and influence foreign or security policies? (2009, 1)
It probably comes as no surprise that we conclude that neoclassical realists are doing foreign policy analysis. The questions they are asking, and the levels of analysis at which they seek answers, are almost completely consistent with the tenets of FPA. And yet one would look in vain for any serious engagement with FPA scholarship in the bibliographies of work published by self-identified neoclassical realists, despite the fact that, as Kaarbo (2015) puts it, “FPA research can provide NCR [neoclassical realism] with considerable theoretical and empirical leverage” (205). Neoclassical realists do not regard themselves as bridge builders between IR and FPA; they see themselves as adumbrating a higher quality realist path, which does not require them to engage the subfield of FPA. FPA scholars are not their audience; FPA scholarship is not in their canon. What motivates this choice to reinvent the wheel of FPA? Is it a strategic decision to remain recognized as belonging to mainstream IR, which greater identification with FPA might preclude? The matter is unclear; if neoclassical realists do decide to one day turn and build that bridge to FPA, we imagine they would be stunningly successful.
If neoclassical realism reaches out from IR toward FPA (even without acknowledging it), then the behavioral IR approach comes at the divide from the opposite direction, from FPA toward IR, and very purposefully so. The term “behavioral IR” is meant to call to mind the field of “behavioral economics,” which revolutionized that field by introducing cognitive, psychological, and sociological considerations into the heretofore actor-general theories of microeconomics. In like fashion, behavioral IR would integrate cognitive, psychological, and sociological considerations into the heretofore actor-general theories of mainstream IR.
The two most organized efforts that reach from FPA toward IR are those of Stephen Walker and his students (hereafter, the Walker School), and those of Alex Mintz and his students (who utilize poliheuristic theory). Mintz and his colleagues have produced a significant corpus of work, much of it experimental, demonstrating that foreign policymakers do not engage in straightforward rational cost-benefit calculations. Instead, cues based on culture or domestic politics or advisors are likely to also be part of the calculus (see, for example, Geva and Mintz, 1997; Mintz et al., 1997; Geva et al., 2000; Mintz, 2002, 2004, 2005; Redd, 2002, 2005; Christensen and Redd, 2004). Through these efforts, Mintz and his colleagues begin to develop a “behavioral IR” blending cognitivism, rational choice, and realism. This is an explicit bridging effort from FPA toward more mainstream IR.
The Walker School’s attempt to bridge IR and FPA through behavioral IR is deserving of special mention for its theoretical ambition. The edited volume Rethinking Foreign Policy Analysis: States, Leaders, and the Microfoundations of Behavioral International Relations, which combines new material as well as previously published journal articles, is the definitive statement on the aims of this research program (Walker et al., 2011). It is well worth reading, even if one does not intend to join the Walker School, for it raises the bar for setting the objectives and organizing the activity of scholarship in IR and FPA.
The Walker School terms its efforts part of the neo-behavioral movement in IR. The “neo” derives from the fact that the Walker School builds upon older manifestations of behaviorism: behavioral IR and behavioral FPA. They “employ both the concepts of rationality and power and the concepts of beliefs, emotions, and motivations” (7). Noting that behavioral IR and behavioral FPA have been either cast as rivals or assumed to inhabit separate intellectual spheres entirely, the Walker School is determined to move beyond this stalemate. Their work can be characterized simultaneously as realist, rationalist, and cognitivist. Power politics, rational choice, and political psychology must be allied, argues Walker and his colleagues. As physicists have found, things look very differently from a microscopic versus a macroscopic point of view: what has been necessary is the development of mesoscopic theory that allows us to see the unity between what we see at the microscopic level and what we see at the macroscopic level in international affairs.
Walker uses the analogy of driving to illustrate what he means by mesoscopic theory. At the microscopic level, we may look at the specific movements of wheels and gears of the car; at the macroscopic level we may recognize a type of behavior called “driving the car.” What allies the two views of reality is “driving to Grandma’s house,” which will help us understand why the wheels and gears are moving as they do while also retaining a conception of the activity as part of a broader type of event. This would be a quantum theory, if one will, of international affairs, and reduce the dissonance that we call the agent-structure problem: “If we are successful in explaining the exercise of power in world politics with a robust behavioral model based on richer and more rigorous conceptualizations and measurements of rationality and power, then we can claim to make scientific progress in the study of International Relations and Foreign Policy Analysis without adding more elements” (17).
That’s the encouraging and understandable overview of the Walker School’s efforts. The nuts and bolts of how they build this mesoscopic theory is much more complex and may daunt all but the most determined. (For example, in just one table, Walker et al. outline 144 sequential games, but note there are over 576 possible.) Indeed, one must learn an entire set of acronyms, the most important of which are TIP, TOM, CUE, and VICS, but which also include P1 (not to be confused with P-1), BACE, and many others.
The Walker School’s theoretical framework links “the world in their [the leaders’] minds” with “the world of events,” and to do so, they must first be able to analyze the belief systems of leaders. The Walker School does this through binary role theory, instantiated through assessment of a leader’s operational code, operationalized through the Verbs in Context (VICS) automated content analysis coding system. More precisely, once they have content analyzed using the VICS scheme a leader’s speech texts for four particular elements of the operational code, specifically, I-1, P-4a, P-1, and P-4b (which refer to various philosophical and instrumental belief continua about Self and Other), Walker et al. are able to suggest what preference order each type of leader will have when they face situations of dyadic international conflict. (The Walker et al. framework is applicable primarily to dyadic relations, though they outline how the same framework could be applied to triads. Beyond triads, the complexity explodes exponentially.)
That is, Walker et al. have a Theory of Inferences about Preferences (TIP). The six rules of TIP will determine for each leader type, whether he or she prefers as an outcome to the conflict—Settlement (DD), Deadlock (EE), Domination (ED), or Submission (DE) —and in what order these four outcomes would be preferred. In a conflictual dyad, Walker et al. will determine the preference orderings of each side in the conflict on the basis of each leader’s operational code.
At this point, we begin to move into “the world of events.” A dyad wherein each side possesses a known preference ordering on outcomes in essence creates a 2×2 game. Walker et al. then turn to TOM, Steven Brams’s Theory of Moves (1993), to suggest how each rational choice game, played in four moves, will turn out (by looking at the Nash or the nonmyopic equilibria). The two players may or may not be playing the same game, but the TOM allows for that possibility and is still able to suggest what the subsequent moves in the game will be. (Indeed, using CUE, the Theory of Cues that Walker et al. have developed, the players’ learning during the game can also be gauged.) There are seventy-eight structurally different 2×2 games, and this number increases when one looks at the intersections created when the two players are playing different games.
These predictions can be checked against a record of what actually happened in the conflict. That is, events data sets can be probed for these dyadic sequential games. How? Walker et al. have developed a software system that “partitions an event series into a series of moves by each state toward the other with each actor’s moves bounded by the intervening words and deeds of the other … we recode each move as either escalatory (E) or de-escalatory (D)” (225) (all conflict events are coded as escalatory, and all cooperative events are coded as de-escalatory). Thus Walker School scholars are able to content analyze leaders’ speeches, posit how dyads involving those leaders will play their sequential games, and then check to see whether those projections match up with what actually occurred in the real world as captured by events data sets.
In sum, Rethinking Foreign Policy Analysis demonstrates that it is possible in IR to see thin accounts of rationality being replaced by thicker accounts informed by psychology. Nevertheless, as a bridge between IR and FPA, behavioral IR becomes, in a sense, an ironic enterprise. Walker and his coauthors know why the bridge must be built and understand that the theoretical and methodological stakes are high. What an impressive groundwork to have laid! And yet, the bridge leads directly to … game theory—dyadic game theory, with all its skeletal depiction of foreign policy (E or D). This comes across a bit as dropping one’s large, diverse, painstakingly assembled collection of tools (FPA) in exchange for one worn hammer. The bridge built with the one hammer instead of the tool kit is unlikely to attract travelers seeking a better route.
However, from the Walker School has come another promising branch that takes a less game-theoretic approach to the use of role theory to bridge between IR and FPA and may actually turn out to be a more fruitful research program (Walker, 1987). While role theory is not new to the FPA toolkit, as we have seen in previous chapters, the idea that role theory could serve as a bridging mechanism between FPA and IR is. Thies and Breuning (2012, 1) argue that “role theory offers the possibility of integrating Foreign Policy Analysis and International Relations theory.” More specifically, they point out that constructivist scholars’ identification of identity as a very important element in understanding foreign policy choice is clearly linked to national role conception. Thies and Breuning comment
FPA generally, as well as cognitive approaches specifically, and IR theory generally, as well as constructivism specifically, stand to benefit from the results of dialogue between the former’s largely agent-based role theory and the latter’s largely system-based agent–structure debate … there is so much common ground that bridging the divide between these two research traditions not only brings them closer together but also advances knowledge in both FPA and IR theory. (2)
Most of the works coming forth in this research program are qualitative case studies designed to illustrate the dynamics of role contestation, role learning, role conflict, and similar phenomena (Harnisch, Frank, and Maull, 2011; Thies, 2013; Walker, 2013; Harnisch, Bersick, and Gottwald, 2016. See also Thies and Breuning, 2012 and the special issue of Foreign Policy Analysis this article introduces). In addition, the study of role theory and FPDM may even offer traction on the issue of cross-level theoretical integration. To take but one example, Cantir and Kaarbo (2012) argue that, “An examination of the domestic political conflicts over roles would provide role theory the underlying mechanisms to account for the emergence of shared roles, the imposition of a dominant role, and the changes in roles and foreign policy when domestic political conditions change.” Role theory as a behavioral IR project with the potential to bridge FPA and IR theory bears continued attention.
In sum, then, the quest for cross-disciplinary theoretical integration between IR and FPA has made discernible progress, but currently remains as elusive as the question for cross-level integration within FPA proper. These quests, however, will certainly shape the future of FPA, a topic to which we now turn.
