Case Studies and Theory Development in the Social Sciences - Alexander L. George [63]
When a deviant case leads to the specification of a new theory, the researcher may be able to generalize about how the newly identified mechanism may play out in different contexts, or he or she may only be able to suggest that it should be widely relevant. As an example of the former, Andrew Bennett, Joseph Lepgold, and Danny Unger undertook a study of burden sharing in the 1991 Gulf War partly because several countries’ sizeable contributions to the Desert Storm coalition contradicted the collective action theories that then dominated the literature on alliances and would have predicted more free-riding. The authors found that pressure from the United States, the coalition leader, explained the large contributions by allies dependent on the United States for their security, most notably Germany and Japan. While pressure from a powerful state is not a novel hypothesis in explaining international behavior, the finding suggested that the collective action hypothesis was generally less determinative in alliance behavior than had been argued. While the temptation of free riding grows as one state becomes more powerful relative to others, so does the ability of the powerful state to coerce dependent allies as well. As these forces offset one another, other factors—domestic politics and institutions, the nature of the public good of alliance security, and so on—help tilt the balance toward or away from a contribution. In short, the authors developed fairly detailed contingent generalizations on how the understudied factor of alliance dependence would play out in different contexts.227
Theory Testing
When theories are fairly well developed, researchers can use case studies for theory testing. The goal here is rarely to refute a theory decisively, but rather to identify whether and how the scope conditions of competing theories should be expanded or narrowed. This is a challenging process: when a theory fails to fit the evidence in a case, it is not obvious whether the theory fails to explain the particular case, fails to explain a whole class of cases, or does not explain any cases at all. Should we blame a theory’s failure on a flaw in the theory’s internal logic or on contextual conditions that rendered the theory inapplicable (which would require only a narrowing of the theory’s scope conditions to exclude the anomalous case), or on some combination of the two? We should not be too quick to reject general theories on the basis of one or a few anomalous cases, as these theories may still explain other cases very well. Conversely, there is a danger of too readily retaining a false theory by narrowing its scope conditions to exclude anomalous cases, or by adding additional variables to the theory to account for anomalies.
An additional difficulty in theory testing is that tests are partly dependent on the causal assumptions of theories themselves. For example, theories that posit simple causal relations, such as necessity, sufficiency, or linearity can be falsified by a single case (barring measurement error). Theories are harder to test if they posit more complex causal relations, such as equifinality and interactions effects. Still, such theories, which are often the kind that most interest case study researchers, may be subjected to strong tests if they assume high-probability (but not necessarily deterministic) relations between variables and posit a manageably small number of variables, interactions, and causal paths. Theories are hardest to subject to empirical tests if they involve