Explanatory Power and Unification

Core Idea

Beyond logical confirmation, scientists value theories for their explanatory power—their ability to explain diverse phenomena and reveal underlying unity. Explanatory power is treated as a virtue in theory choice: other things equal, a more unified theory is preferred. This raises philosophical questions: Is explanatory power merely a psychological preference, or does it track truth? Can we distinguish genuine unification from mere appearance? How does explanatory power relate to empirical adequacy?

Explainer

You already know the unification model of explanation: on this view (developed by Friedman and Kitcher), a theory explains by reducing the number of independent phenomena we must accept as brute facts, showing how diverse observations follow from a single set of patterns or principles. You also know confirmation theory: evidence confirms hypotheses when it raises their probability (Bayesian) or survives attempts at falsification (Popperian). Explanatory power is distinct from both but interacts with both in theory choice.

Consider Newton's mechanics as the paradigm case. Before Newton, terrestrial mechanics (Galileo's falling bodies, Kepler's planetary laws, tidal patterns) were treated as separate phenomena requiring separate descriptions. Newton showed that a single inverse-square gravitational law, combined with his three laws of motion, derives all of these from one unified framework. The result is not merely that Newton's theory fits more data — it is that fewer independent assumptions are required to account for the same empirical range. Friedman's formal criterion captures this as a reduction in the number of "accepted phenomena" that serve as primitive inputs to explanations. Kitcher's rival account formalizes it as minimizing the number of argument patterns needed to systematize science — a theory is more unified if it explains more using fewer, more general schematic patterns.

The key philosophical question is whether explanatory virtue is truth-conducive or merely pragmatic. One view (explanatory realism) holds that when a theory genuinely unifies, this is evidence it has latched onto real structure in the world — the unity of Newton's framework is a sign that gravity really is one phenomenon, not many. A rival view (instrumentalism or van Fraassen's constructive empiricism) holds that unification is a feature we bring to science for pragmatic or cognitive reasons — it helps us organize knowledge and make predictions, but it does not provide additional evidence that the theory is true beyond what the data directly confirm. Van Fraassen explicitly argues that "loveliness" (explanatory virtue) and "likeliness" (probability of truth) are distinct, and that inference to the best explanation conflates them.

A subtler problem is distinguishing genuine unification from mere conjunction. Suppose theory T₁ explains phenomenon A and theory T₂ explains phenomenon B, and someone proposes T₃ = "T₁ and T₂" to explain both A and B. T₃ is not more unified — it is just a concatenation. Genuine unification requires that the same principles, applied in the same way, account for both A and B, revealing that A and B are not independent at all. Kitcher's argument-pattern approach tries to formalize this: T₃ uses two separate patterns for A and B, while a truly unified theory uses one pattern for both. The challenge is making this formal distinction rigorous without collapsing back into a purely syntactic criterion that misses the explanatory point.

In practice, explanatory power interacts with confirmation in theory choice. When two empirically equivalent theories (fitting all the same data equally well) are available, scientists regularly prefer the more unified one. This shows that explanatory virtues function as epistemic tiebreakers. Whether this practice is rationally justified depends on whether explanatory power carries evidential weight beyond empirical fit. Philosophers of science who defend inference to the best explanation (IBE) say yes: the best explanation of why unified theories succeed empirically is that they track real structure. Critics like van Fraassen say the success of unified theories only requires that they are empirically adequate — we have no further reason to believe them true. This debate about IBE is one of the central disputes in contemporary philosophy of science.