Theories built from conjunctions and disjunctions resist simple regression-style testing.

Evidence

Our empirical analysis of Skocpol’s theory suggests that it is not clear how one might go about testing her explanation with statistical methods (e.g., Geddes 2003). It is a complex, multilevel model constructed using necessary and sufficient conditions, along with equifinality at the secondary level. As Pierson says in the context of the welfare state literature: “Different welfare state configurations are the products of complex conjunctural causation, with multiple factors working together over extended periods of time to generate dramatically different outcomes. There is no theoretical justification for arguing that a 10 percent shift in the value on one variable or another will have a simple direct effect on outcomes” (2000, 809–10). Braumoeller (2003) is a rare example of an attempt to model theories formulated in terms of AND and OR in a way that is faithful to the theory and estimatible using statistical techniques. As our various examples have illustrated, qualitative, comparative theories often are complex and multilevel. Much more needs to be done to understand what are the appropriate empirical (statistical, fuzzy set, or whatever) methods for evaluating such theories. — Gary Goertz, Social Science Concepts, 1755

Categorization

• QUESTION: What would a statistically estimable model of equifinal, multilevel causation actually look like? Braumoeller 2003 is the only cited example and the door is explicitly left open.
• XREF: Connects to Milgrom’s point that full-history information is unnecessary for efficient trade — both argue the standard modeling assumption (linear effects, complete information) is the wrong frame, not merely incomplete.