Case Studies and Theory Development in the Social Sciences - Alexander L. George [93]
DSI’s misunderstanding of process-tracing leads to a failure to recognize that it can often provide an alternative method for testing theories. Thus, by utilizing process-tracing, a theory can be assessed by identifying a causal chain that plausibly links the independent variable of a theory with its dependent variable. Process-tracing does not regard each component of the intervening space between independent and dependent variables as simply an “observable implication” but rather as a step in a causal chain. Such a causal chain, if there is sufficient data for identifying it, can—and should—be supported by appropriate causal mechanisms.
There is another important issue that DSI deals with in an idiosyncratic way. The authors make no reference to assessing the predictive or explanatory power of a theory, a subject much emphasized by other writers. Instead, DSI focuses solely on assessing the validity of a theory via its observable implications; this is not equated to or related to a given theory’s predictive or explanatory power. Perhaps this question is ignored because DSI favors assessing a theory’s validity and usefulness not by its ability to explain or predict variance on a given dependent variable, but by the considerable number and variety of observable implications it can generate which they believe increases a theory’s “leverage”—i.e., its ability to explain more with less.371
Why do DSI’s authors not assert that the observable implications of a theory, once established, constitute its predictive capacity? The answer would seem to be that the subjects of observable implications as construed by DSI can vary so widely and, as already noted, that the initial theory itself changes during the process of searching for observable implications. Accordingly, it would make little sense to regard the totality of the various observed implications as reflecting the theory’s ability to predict or explain a given dependent variable. A large number of observable implications does not give assurance that the revised theory is capable of predicting variation in the values of the dependent variable that was postulated in the initial theory. If one wants to focus on the task of establishing the predictive and explanatory power of a given theory, then the congruence method discussed in Chapter 9 should be of interest.
In any case, as this chapter has noted, a variety of procedures, including that proposed in Designing Social Inquiry, for dealing with the “too many variables, too few cases” problem are available from which researchers can choose.
In DSI the phenomenon of equifinality—i.e., when the same type of outcome in different cases may have quite different causes—is recognized in the discussion of research on revolutions.372 The authors later provide a more detailed discussion of “multiple causality” (a phrase sometimes used by social scientists as a synonym for equifinality). However, the discussion (especially in its hypothetical research examples) confuses equifinality with something quite different—namely the fact that explanations of complex phenomena encompass a number of independent variables.373
Within-Case Methods of Causal Inference: The Congruence and Process-Tracing Approaches
There is an alternative that compensates for the limits of both statistical and comparative case analyses: within-case analysis.374 The methods we have discussed in this chapter are all what Charles Ragin has termed “variable-oriented” approaches; they attempt to establish the causal powers of a particular variable by comparing how it performs in different cases.