Case Studies and Theory Development in the Social Sciences - Alexander L. George [90]
In terms of the discussion in the comparative politics literature about whether researchers should select cases that are as similar as possible or as different as possible, the authors of DSI “recommend a different approach,” namely one that foregoes or minimizes reliance on the comparative method and focuses instead on identifying the potential observations in a single case “that maximizes leverage over the causal hypothesis.”348
At the same time, DSI concedes that comparative small-n studies that use careful matching techniques can produce useful results, even though matching can never be complete or reliable. They mention three studies that they regard as having been reasonably successful in matching cases.349 DSI concludes with a somewhat more receptive view of small-n studies that achieve adequate, if not perfect, controls than in their earlier comments: “With appropriate controls—in which the control variables are held constant, perhaps by matching—we may need to estimate the causal effect of only a single explanatory variable, hence increasing the leverage we have on a problem.”350
Let us examine more closely DSI’s preferred method of assessing theories via their observable implications. The method they espouse is well known, but it is spelled out in great detail and considerably extended in Designing Social Inquiry. The familiar concern with “too many variables, too few cases” takes the form of concern with “too few observations.” One merit of their discussion is its emphasis on the possibility of attributing a large number of observations to a theory even when working with a single instance or a small number of cases. In the final chapter of the book, “Increasing the Number of Observations,” two strategies for doing so are presented.351 First, the reader is presented with a “very simple formal model.” The simplification of the model includes use of a linear regression assumption and a focus “on the causal effect of one variable”; all other variables “are controlled” in the model “in order to avoid omitted variable bias or other problems.”
However, discussion of this model in DSI and the example chosen address not a single case or small-n research, but a large-N type of study. This would seem an inappropriate example for addressing the problem in qualitative research. The possibility of multicollinearity is recognized, but finessed by suggesting it could be dealt with by more observations; besides, assurance is given that “it is often possible to select observations so as to keep the correlation between the causal variable and the control variable low.”352 Later, the assumption of linearity is addressed and nonlinearity is briefly discussed. In the end, the authors of DSI acknowledge that they cannot provide a precise answer to the question of how many observations will be enough, which will always apply, and that “most qualitative research situations will not exactly fit the formal model,” although its “basic intuitions do apply much more generally.”353
The authors then turn to the second strategy for increasing the number of observations by “making many observations from few.” This is accomplished by “reconceptualizing” a qualitative research design “to extract many more observations from it.”354 Since DSI rests much of its argument on this strategy, we need to examine it closely.