Case Studies and Theory Development in the Social Sciences - Alexander L. George [11]
We take up our disagreements with DSI (in this chapter and in Chapter 8); here, for reference, we merely list them, starting with our epistemological differences and proceeding to our methodological ones. One critique is that although DSI disavows being “a work in the philosophy of the social sciences,” it implicitly makes many important philosophical assumptions regarding highly contested issues in the philosophy of science.24 For example, DSI suggests that causal mechanisms are in some sense less fundamental to causal explanation than what DSI defines as “causal effects.”25 This runs counter to our view that causal mechanisms and causal effects are equally important to causal explanation. More generally, in our view DSI’s treatment of causal mechanisms is unsatisfactory, as we detail in Chapter 8. Robert Keohane has given a clearer exposition of the nature and importance of causal mechanisms for explanation in his later publications.26
We also critique DSI for emphasizing almost exclusively the epistemic goal of hypothesis testing (sometimes known as the “logic of confirmation”), neglecting other aspects of theory development, such as the formation of new hypotheses or the choice of new questions to study. DSI relegates these goals, the “logic of discovery,” to a quotation from Karl Popper that “there is no such thing as a logical method of having new ideas… . Discovery contains ‘an irrational element,’ or a ‘creative intuition. ’”27
We agree that there is no linear logic of discovery, but we emphasize theory development, focusing on hypothesis formation and the historical explanation of individual cases, as well as the testing of general hypotheses. We outline procedures that are conducive to the generation of new hypotheses, such as the study of deviant or outlier cases.
Another concern is that DSI pays little attention to problems of causal complexity, particularly equifinality and multiple interactions effects. It addresses these subjects very briefly, discussing only the simple case of two-variable interactions, and it tends to be optimistic on how easily statistical models can address complex interactions within a realistic sample size.28 We emphasize that various kinds of complex causal relations are central concerns of the social sciences, including not only equifinality and multiple interactions effects, but also disproportionate feedback loops, path dependencies, tipping points, selection effects, expectations effects, and sequential interactions between individual agents and social structures. Our approach to the problem of complexity is to recommend process-tracing as a means of examining complexity in detail and to suggest typological theorizing as a way to model complexity; DSI does not distinguish between typological theories, which model causal relations of equifinality, and mere taxonomical typologies.29
On the methodological level, we take issue with DSI’s arguments on case selection criteria, the value of single-case studies and “no variance” research designs, the costs and benefits of increasing the number of cases studied, and the role of process-tracing. On case selection criteria, DSI gives the standard statistical warnings about selection on the dependent variable and argues that single-case research designs are seldom valuable. 30 This advice overlooks the opportunities for studying deviant cases and the dangers of certain forms of selection bias in case studies that can be more severe than those in statistical studies.
DSI also argues for increasing the number of observable implications of a theory both within cases and across them. While we agree that increasing the number and diversity of observable implications of alternative theories is generally extremely useful, DSI tends to understate the dangers of “conceptual stretching” that can arise