Most Different Systems Design

Conceptual Overview and Discussion

Mill's method of agreement powers most different systems design. This logic argues that for two cases identically composed except for one characteristic, when one case experiences an effect that the other does not, then the prior circumstance on which they differ must be the cause (or, at least, an integral part of it). As this method can only identify correlation and cannot confirm causal inference, the phenomenon equally may be either cause or effect. In logical terminology, if characteristics A B C D occur together with events w x y z, and if characteristics B C D occur together with events w x y, then A is the cause or the effect or part of the cause of z. This principle combines with Mill's method of difference in application as the joint method—which is the foundation for conditions of necessity and sufficiency.

MDSD can be either exploratory (as a case selection tool) or confirmatory (when evaluating hypotheses). The method assumes that cases are drawn from the same population, and questions whether the observations and relationships will hold across a large number of varied samples. Whereas a comparative design (such as MSSD) highlights differences wherein all other variation is either controlled or accounted for, a statistical design (such as MDSD) analyzes the few similarities in order to identify what is sufficiently common to produce the event. Generally, the cases feature very different values for multiple independent variables; then, if they take the same value for the dependent variable, the factors that covary must be productive.

Furthermore, MDSD presumes that the level of observation is lower than a system—specifically, the level that most reduces within-group variance—and systemic factors therefore are not special. Unlike MSSD, MDSD does not assume that systems' group characteristics can be eliminated individually; rather, group factors are removed together.

The situation that provides the best context in which to apply MDSD is maximum heterogeneity. First, the researcher identifies a set of variables projected to be related to the phenomenon of interest, as well as variables typically characteristic of members of the set or population. Factors should be included only if they are relevant (in order to avoid overdetermination), [Page 571]and must be included if they are related to either the proposed explanation or the outcome. Then, the researcher identifies cases that differ on all factors except one independent variable and the outcome; this procedure amounts to selecting on the dependent variable. Whereas each case scores some value (including extremes) on the continuous scale of each descriptor, MDSD does not employ exact matching in an attempt to introduce control variables; instead, it aims to eliminate irrelevant systemic factors in favor of sufficient explanatory individual factors. In this way, any extraneous sources of variance that persist across the range of circumstances must be generic. Second, the investigation begins with a hypothesis of no difference: through examining heterogeneous samples, the remaining similarity must be significant. Third, the researcher uses iterative comparison procedures (e.g., process tracing) in order to eliminate necessary causes—leaving only sufficient conditions. Fourth, analysis ends when it is no longer possible to draw general conclusions valid for all the subpopulations; this is analogous to the saturation point that terminates interview data collection. Finally, the researcher reports that among cases that differ on (the specified characteristics), similarities regarding (the event) can be attributed to the following (factor of interest).

...