Abstraction Traps

In scientific research, we can see several prominent types of traps concerning what level of analysis to bound a relevant system at both arising from a form of infinite regress. Assume a system of interest S. S has N relevant causal levels needed to successfully explain its behavior. How can we derive these levels? Most disciplines do so by assuming an arbitrary subset of N, essentializing the haphazard process by which they originally derived it. For example, statements like “the social cannot be explained by the psychological” or “we can predict the behavior of an complex organization by assuming a representative agent” do not speak for themselves. They may be correct or incorrect but certainly not for all problems at all times.

Those seeking to escape ad hocery, however, can make two mistakes of their own. First, they can take S as the starting point and then move progressively lower and lower until they are explaining the destinies of nations in terms of a set of levels beginning with S and ending with the most granular levels of physics. Another is to take S as the lowest inherent level and then move up in terms of levels. A given social outcome is explained not due to the choices made by participants, but in terms of a Rube Goldberg-like contraption that begins with a high-level social force that works progressively down a causal chain until it acts on individuals. The former mistake is often seen in calls for a discipline that is “unrealistic” to embrace more of the “human” elements of decision. But adding lower levels without a valid compositional strategy is fruitless. Likewise, criticism that an approach is not systemic and lacks core elements that S is embedded within will not yield useful results unless the composition of the elements is rigorously formulated and justified.