Case selection in case-study research should have two main objectives: (1) obtain a representative sample with (2) useful variation on the dimensions of theoretical interest.1 In order to satisfy these objectives, I employ what John Gerring refers to as the "diverse case method.”  This technique aims to select a set of cases that represent the full range of values that characterize a specific relationship between an explanatory variable (in this case, the interaction between two measures of US financial interests) and an outcome of interest (ESF credits). Figure 6.1 presents a modified version of Figure 5.3 from the previous chapter. Each of the cases is placed into one of the four quadrants, based on whether their values on the two key explanatory variables—SIFI bank exposure and the systemic risk index—are above or below their respective sample means.
Case Distribution across Theoretical Dimensions of Interest
Thus, those in quadrant B represent cases where both bank exposure and systemic risk are above average for the sample. Those in quadrant D represent cases where SIFIs were exposed at above-average levels but systemic risk was below average. The case in quadrant A exhibits below-average bank exposure but above-average systemic risk. Thus, these cases present a range ofvariation across the two explanatory variables. Additionally, Figure 6.1 also indicates the value of the outcome of interest: whether or not an ESF credit was provided in each case. A plus symbol (+) indicates that a bailout was provided, while a minus symbol (-) indicates the absence of a credit. Thus, these cases also present both possible outcomes of the dependent variable. This case-selection strategy has an additional benefit: It makes it easy to identify "typical” cases that fit the argument’s expectations and "deviant” cases that do not.
The Polish (1989) and Argentine (1983) cases are deviant cases in that they disconfirm a deterministic relationship between the explanatory variables and the outcome of interest. My argument appears to poorly explain Treasury’s decision to bail out Poland in 1989 given that SIFIs had little exposure to that economy. Similarly, given that big banks were highly exposed to Argentina in 1983 and systemic risk facing the US financial system was high, my argument would predict a bailout—yet one did not occur in this case. Exploring these cases allows us to move beyond a simplistic, monocausal story by identifying alternative explanations of US ILLR actions. Mexico (1982), Brazil (1982, 1983), and Argentina (1984) are examples of typical cases—those that appear to be well explained by the model. Close within-case analysis of these rescues presents the opportunity to probe whether or not the historical evidence actually validates the argument. All three observations in quadrant D— Mexico (1995), Korea, Indonesia, and Thailand (1997)—fall somewhere in between deviant and typical cases. Here, SIFI exposure was high yet systemic risk was below the sample average. Thus, conducting within-case analysis ofthese particular events represents an important opportunity to assess the extent to which the private interests of the banks may have independently influenced US bailouts. If so, this would suggest that at least some US bailouts are not designed to protect the public interest.
Finally, these cases also present significant variation over time. Five cases are from the 1980s; four are from the 1990s. This variation across time is important given the argument I made in chapter 4 about changes in the structure of the global financial system and how this affected the IMF’s adequacy as an ILLR. In the 1980s, IMF unresponsiveness should be linked to the need for US intervention. In the 1990s, IMF resource insufficiency should be a key concern of policymakers. Thus, due to the variation over time, the case studies will also help identify whether the supposed weaknesses of the Fund were in fact linked to US bailout decisions.
-  Seawright and Gerring 2008, p. 296.
-  Gerring 2007.
-  For more on this, see Gerring 2008 (p. 300) and Seawright and Gerring 2007(pp. 97-101).
-  Admittedly, using the sample mean as the cutoff point is somewhat arbitrary. Seawrightand Gerring (2008) note that when dealing with continuous variables, as is the case here,it is difficult to identify where these lines should be drawn. The primary goal, however, is toensure that there is meaningful variation across these variables of interest. The approachI use here accomplishes this.
-  For more on typical and deviant cases, see Gerring 2007.