Loading field map
Why does a scientific explanation satisfy us? Is it enough to subsume an event under a general law, or must we trace the causal mechanisms that produce it? These questions have driven a half-century of debate in philosophy of science. The subfield of explanation and causation examines what makes an explanation genuinely explanatory, how causal claims are grounded, and whether explanation and causation are even the same kind of relation. The story begins with a bold attempt to reduce explanation to logical deduction, then fractures into competing causal, unificationist, and pragmatic accounts, and finally settles into a pluralist landscape where different frameworks serve different scientific domains.
In 1948, Carl Hempel and Paul Oppenheim proposed the Deductive-Nomological (DN) Model, which held that a scientific explanation is an argument whose conclusion describes the phenomenon to be explained and whose premises include at least one universal law. The model was elegant: explanation was logical subsumption. If you could deduce the occurrence of an event from initial conditions plus a law, you had explained it. The DN Model dominated for two decades, but its very clarity exposed deep problems.
First, the model could not handle statistical regularities. Many scientific explanations invoke probabilistic laws—say, why a particular atom decayed—but a probabilistic law does not deductively entail its outcome. Second, the DN Model failed to capture causal asymmetry. A flagpole's height and the sun's position can be used to deduce the length of its shadow, but the shadow's length cannot explain the flagpole's height, even though the deduction works both ways. Third, the model allowed irrelevant information to count as explanatory. A barometer reading can be used to deduce a storm, but the reading does not explain the storm. These failures showed that logical form alone cannot distinguish genuine explanation from mere derivation.
Wesley Salmon directly confronted the DN Model's shortcomings. His first response was the Statistical Relevance (SR) Model (1960–1980), which replaced deductive subsumption with a statistical criterion: an explanation identifies factors that make a difference to the probability of the event. If a factor changes the probability of the outcome, it is statistically relevant and therefore explanatory. The SR Model handled probabilistic cases that the DN Model could not, but it still suffered from the asymmetry problem. Statistical relevance is symmetric: if smoking is relevant to lung cancer, lung cancer is equally relevant to smoking. Yet we do not explain smoking by citing lung cancer.
Salmon then moved to a deeper framework: the Causal-Mechanical Model (1970–2000). He argued that explanation requires tracing the causal processes and interactions that produce the phenomenon. A causal process transmits a mark or signal from one spacetime region to another; a causal interaction modifies such a process. Explanation, on this view, is not about argument or probability but about revealing the causal structure of the world. The Causal-Mechanical Model solved the asymmetry problem—causal relations are directional—and it gave a natural account of why irrelevant factors fail to explain: they are not part of the causal chain. However, the model struggled with explanations in domains where causation is difficult to track, such as fundamental physics or highly abstract theories.
Not everyone agreed that explanation must be causal. Philip Kitcher and Michael Friedman developed the Unificationist Model (1974–2000), which argued that explanation is a matter of reducing the number of independent phenomena we must accept. A theory explains by unifying diverse facts under a small set of patterns or argument schemas. The more phenomena a schema covers, the better the explanation. Unificationism captured something the causal models missed: many scientific explanations, especially in physics, work by showing that apparently different events are instances of the same law. The unificationist goal of economy directly contrasted with the causal-mechanical goal of tracing production. Where Salmon wanted to show how an effect is brought about, Kitcher wanted to show how it fits into a coherent system. The two frameworks coexisted in productive tension, each highlighting a different virtue of explanation.
Bas van Fraassen's Pragmatic Theory of Explanation (1980–2000) challenged the very idea that explanation is an objective relation between theory and world. Van Fraassen argued that explanation is a three-place relation among a theory, a phenomenon, and a context of inquiry. What counts as a good explanation depends on the question being asked and the background knowledge of the audience. The same event can be explained differently in different contexts—for example, a car crash might be explained by mechanical failure in one context and by driver distraction in another. The Pragmatic Theory did not replace causal or unificationist accounts; instead, it shifted attention to the role of the explainer and the audience, revealing that earlier frameworks had assumed a single correct explanation for any phenomenon. This pluralist insight would prove influential even as philosophers continued to seek objective constraints on explanation.
By the 1990s, philosophers of science had grown dissatisfied with purely logical or statistical accounts of causation. James Woodward's Interventionist Theory of Causation (1990–Present) offered a non-circular definition: X causes Y if and only if an ideal intervention on X changes Y, while other variables are held fixed. This account is especially useful for the special sciences—biology, psychology, economics—where controlled experiments are the gold standard. Interventionism does not require a deep metaphysics of causal powers; it works with the manipulability relations that scientists actually test. The framework provided a rigorous foundation for causal inference, distinguishing genuine causation from mere correlation, and it connected naturally to the statistical methods of causal modeling.
Around the same time, philosophers of biology and neuroscience developed the Mechanistic Explanation framework (1990–Present). A mechanism is a set of entities and activities organized so that they produce a regular change from start to finish. Mechanistic explanation decomposes a system into its parts and shows how their interactions generate the phenomenon of interest. This framework refined the earlier Causal-Mechanical Model by specifying the organizational features—spatial, temporal, and hierarchical—that make a mechanism explanatory. Where Salmon's model focused on causal processes in general, Mechanistic Explanation zeroes in on the internal workings of complex systems, from protein synthesis to neural firing. It has become the dominant framework in philosophy of biology and cognitive science, precisely because it matches how scientists in those fields actually reason.
Inference to the Best Explanation (IBE) (1965–Present) occupies a unique position: it is both a theory of explanation and a theory of confirmation. According to IBE, when we infer that a hypothesis is true, we do so because it provides the best explanation of the available evidence. This framework bridges the subfield of explanation and causation with the subfield of confirmation and evidence. IBE does not compete directly with causal or mechanistic accounts; rather, it assumes that explanation is a guide to truth. Critics argue that IBE is too vague—what counts as "best"?—but its defenders reply that scientists routinely use explanatory considerations to choose among hypotheses. IBE remains a live tradition, especially in debates about scientific realism.
Today, the subfield is marked by pluralism. The leading frameworks—Interventionist Theory of Causation, Mechanistic Explanation, and Inference to the Best Explanation—agree that explanation is not merely logical subsumption and that causation plays a central role in many explanatory contexts. They disagree, however, on the scope of causation. Interventionists hold that manipulability is the key to causation, while mechanists emphasize internal organization. IBE proponents argue that explanatory power is a mark of truth, a claim that interventionists and mechanists need not endorse. The earlier fault lines between causal and unificationist approaches have not disappeared; they have been absorbed into a broader recognition that different scientific domains require different explanatory tools. Physics may prize unification, biology may prize mechanism, and psychology may prize intervention. The subfield's central tension—what makes an explanation genuinely explanatory—remains unresolved, but the frameworks now coexist as specialized resources rather than as exclusive competitors.