Loading field map
The philosophy of explanation and causation asks two questions that are simple to state and notoriously difficult to answer: What does it mean to explain something? And what does it mean for one thing to cause another? These questions sit at the heart of science, where explanation is the stated goal of most inquiry and causation is the concept that connects explanation to prediction, intervention, and control. The subfield is not one unified theory but a landscape of competing approaches, each addressing a different aspect of the problem. Some focus on the logic of explanation, others on the metaphysics of causal connection, and still others on the practical role of explanation in human inquiry.
The central puzzle of explanation is that it is more than description. To explain why a window broke, one does not merely describe the window breaking; one cites something—a thrown rock, a sudden temperature change, a structural flaw—that makes the breaking intelligible. The same distinction applies to science: Newton's laws do not just summarize planetary positions; they explain them. But what exactly is added? What makes an explanation explanatory?
The earliest systematic answers came from Aristotle, who distinguished four kinds of "because": the material cause (what something is made of), the formal cause (its essence or form), the efficient cause (the agent that produced it), and the final cause (its purpose or end). For Aristotle, a full explanation of a natural phenomenon could involve all four. This framework dominated Western thought for nearly two millennia, but with the rise of modern science in the seventeenth century, the focus narrowed. Galileo, Descartes, and Newton increasingly treated explanation as a matter of efficient causation—finding the mechanical or dynamical processes that produce phenomena—and demoted final causes to the realm of theology or biology. This shift was not merely doctrinal; it changed the very standard of what counted as a satisfying scientific explanation. A phenomenon was understood when its underlying mechanisms were identified, not when its purpose was discerned.
The modern subfield, however, did not emerge until the mid-twentieth century, when philosophers began treating explanation and causation not as background assumptions but as explicit objects of analysis. The catalyst was the rise of logical empiricism, a school of thought that sought to reconstruct scientific knowledge in logical terms. For the logical empiricists, the key question was how to distinguish scientific explanation from mere description or prediction. Their answer, proposed by Carl Hempel, became the first modern theory of explanation.
Hempel's deductive-nomological (D-N) model, sometimes called the "covering law" model, held that an explanation is an argument. To explain an event, one must show that the event's description follows logically from two premises: a set of initial conditions and at least one general law. The law must be genuinely universal—not merely accidental—and the argument must be valid. Explanation, in this view, is deduction from laws. To explain why a pendulum has a certain period, one deduces the period from the law of pendulum motion and the length of the string.
The model was attractive because it made explanation rigorous and objective. It also unified explanation with prediction: if an explanation is a valid deductive argument, then the same argument, constructed before the event, would have been a prediction. The difference between explanation and prediction was merely temporal and epistemic.
But the D-N model faced severe objections. The most famous came from the philosopher Wesley Salmon, who pointed to cases that fit the model's form yet are clearly not explanations. A classic example involves a flagpole: one can deduce the length of a flagpole's shadow from the height of the pole, the angle of the sun, and the law of light propagation. But this deduction, while valid, does not explain the length of the shadow—at least not in the way one can explain the height of the pole from the length of the shadow and the angle of the sun. The asymmetry shows that explanation is not merely logical deduction; it has a direction that the D-N model cannot capture. Other objections concerned relevance: one can deduce the fact that a man does not get pregnant from his birth-control pills, but the pills do not explain his failure to become pregnant. The D-N model included irrelevant information as long as the deduction was valid.
Hempel also proposed a weaker variant, the inductive-statistical model, for explanations that rely on probabilities rather than universal laws. But this model inherited the same problems and added new ones: what makes a statistical relation explanatory rather than merely statistical? Hempel's answer—that the explanation must make the event highly probable—failed in the face of cases where rare events are perfectly explainable (a genetic mutation, a lottery win) and frequent events are not (a man's not becoming pregnant).
The D-N model is now largely rejected as a complete theory, but its influence persists in two ways. First, it set the agenda: subsequent theories defined themselves partly by how they addressed its failures. Second, it correctly identified that laws or law-like regularities play some role in many explanations, even if the precise role is more subtle than Hempel thought.
One major line of response to the D-N model's failures was to shift attention from the logical structure of explanation to its context. The philosopher Bas van Fraassen proposed a pragmatic theory: an explanation is not a special kind of argument but an answer to a "why-question." Whether something counts as an explanation depends on the contrast class (why this event rather than that alternative?) and on the interests and background knowledge of the questioner. For van Fraassen, there is no such thing as "the" explanation of an event, only explanations relative to a context. The D-N model's asymmetry problems dissolve: the flagpole's shadow explains the pole's height if the questioner is in that context, and vice versa, because explanation is fundamentally a communicative act.
The pragmatic theory captured something real: explanations are offered to particular audiences for particular purposes. But critics charged that it made explanation too subjective, reducing it to whatever a questioner finds satisfying. It risks collapsing into psychology, losing the sense that some explanations are objectively better than others.
A second response, developed by Philip Kitcher and Michael Friedman, was the unificationist theory. This view held that explanation is not about deduction from laws but about reducing the number of independent phenomena we must accept. To explain something is to fit it into a pattern with other phenomena, so that fewer brute facts are needed. Kitcher's version was precise: an explanation is a derivation that uses a minimal set of argument patterns to cover a maximal range of cases. The best explanation is the one that systematizes the most with the fewest resources. This approach handles the asymmetry problem gracefully: the flagpole's height explains its shadow because deriving the shadow from the height fits a broader pattern (geometrical optics) that also covers many other phenomena, whereas deriving the height from the shadow requires an ad hoc pattern.
The unificationist theory is attractive because it ties explanation to a genuine scientific virtue—unification, like that achieved by Newton's mechanics or Darwin's natural selection. But it faces its own problems. It is not clear that all explanations unify; a geologist explaining a specific rock formation may cite contingent historical processes that unify nothing beyond that case. And the theory struggles to distinguish unification from mere economy of description—not every compact theory is explanatory.
A third response to the D-N model's failures was to embrace what Hempel had tried to avoid: explanation is fundamentally about causation. This "causal turn" has become the dominant approach in the late twentieth and early twenty-first centuries, though it takes many forms.
The most influential version was developed by Wesley Salmon, who, after criticizing the D-N model, proposed the causal-mechanical theory. Salmon argued that to explain an event is to trace the causal processes and interactions that produce it. A causal process is a physical entity that transmits marks or signals over time—a moving ball, a light ray, a sound wave. A causal interaction is a point where processes modify each other's trajectories. Explanation, on this view, is the exhibition of the causal network that links the explanandum to its causes. Unlike the D-N model, this makes explanation essentially non-linguistic: it is about the world's causal structure, not about arguments in a language.
The causal-mechanical approach handled the asymmetry problem directly: causation has a direction (from cause to effect), so explanation inherits that direction. It also handled the relevance problem: irrelevant deductions are excluded because they do not trace real causal connections. But it faced a different difficulty: it seemed to require that every explanation specify a detailed physical process, which many scientific explanations do not do. Explaining why a gas has a certain pressure by citing the ideal gas law does not trace the motions of individual molecules. Salmon's later work acknowledged that many explanations are "statistical-relevant" rather than fully mechanical, and he broadened his theory to include statistical relevance relations.
A different version of the causal turn came from philosophers who analyzed causation itself. The most prominent among them is David Lewis, who proposed a counterfactual theory: an event C causes an event E if and only if, had C not occurred, E would not have occurred. This made causation a matter of counterfactual dependence across possible worlds. The theory was elegant and connected causation to our ordinary reasoning about what would have happened if things had been different. But it faced the problem of preemption: if a rock hits a window and breaks it, we say the rock caused the break, even though if the rock had not hit the window, a second rock thrown immediately after would have broken it anyway. The counterfactual test fails because the break would still have occurred. Lewis's own response involved a complex machinery of "influence" rather than simple dependence, but the problem persists in various forms.
James Woodward later synthesized the causal turn with the tools of interventionism. On Woodward's account, an explanation is an answer to a "what-if-things-had-been-different" question: to explain a phenomenon is to show how it would change under interventions on its causes. This made explanation tightly connected to manipulability and control. A variable X causes a variable Y if an intervention on X changes Y, holding other variables fixed. This account is especially well suited to scientific practice, where experiments manipulate variables to determine causal structure. Woodward's interventionism is not a rejection of Salmon's mechanism but a more general framework that accommodates both mechanical and non-mechanical explanations, as long as they support counterfactual claims about interventions.
The causal turn has been enormously influential in the philosophy of the special sciences—biology, psychology, economics—where causal claims are central but universal laws are rare. It also links the subfield to statistics and computer science through the work of Judea Pearl and others on causal inference from observational data. Pearl's structural causal models, with their directed acyclic graphs and do-calculus, provide a formal language for the very interventionist ideas Woodward articulated philosophically. This is a rare case where a philosophical theory found direct practical application in data science.
A distinct strand of the causal turn arose from attention to the actual practice of the life sciences. Philosophers such as William Bechtel, Carl Craver, and Peter Machamer argued that explanation in biology and neuroscience is neither deductive-nomological nor purely interventionist, but mechanistic: to explain a phenomenon is to show the mechanism that produces it. A mechanism, in the standard formulation, is a set of entities and activities organized such that they produce a regular change from start to finish. Explaining neural transmission involves specifying the ion channels, neurotransmitters, and their organized activities—not invoking a law of neural transmission.
The mechanistic approach differs from Salmon's causal-mechanical theory in an important way: it emphasizes the organization of components into a working system, and it allows for mechanisms that are not continuous physical processes in Salmon's sense. It also differs from Woodward's abstract interventionism: mechanistic explanations typically provide concrete detail about how the system actually works, not just what would happen under interventions. The two are compatible, though: a mechanistic explanation usually supports interventionist claims, and interventionist claims are often made precise by mechanistic details. Many contemporary philosophers treat mechanisms and interventions as complementary aspects of a single, richer picture.
The mechanistic approach was a response to a real gap in earlier theories: Hempel's laws are rarely found in biology, and Salmon's processes are hard to identify at the level of, say, gene expression or neural computation. Mechanisms provided a middle ground—regular, intelligible, but not universal. The approach has its own limits, however. It is less clear how it handles explanations that are purely statistical or that involve very abstract, non-mechanistic models, such as those in population genetics or macroeconomics. It also faces the question of when a mechanism is "complete," since mechanisms are often decomposed into sub-mechanisms, and it is not clear where the decomposition should stop.
Given this landscape, a natural question arises: is there one correct theory of explanation, or are there several? Many contemporary philosophers of science have become pluralists, arguing that different kinds of explanation suit different kinds of scientific problem. The physicist explaining a gas's pressure may use a statistical law; the neuroscientist explaining a behavior may describe a mechanism; the economist explaining a recession may cite an interventionist model. These are not rival accounts of the same thing but different tools for different jobs.
This pluralism is not mere tolerance; it reflects a substantive view about the nature of scientific inquiry. Science is not one homogeneous enterprise but a collection of practices with different goals, and explanation follows function. The philosopher of science Helen Longino has argued that the very standards of explanatory adequacy are shaped by contextual values and local practices. Others, like Hasok Chang, have defended a "complementary" view: rival explanations can coexist because they capture different aspects of the same phenomenon, and science makes progress by integrating them over time.
Pluralism has its critics. Some argue that it gives up on the philosophical project of saying what explanation really is, leaving only a descriptive catalog of what scientists happen to do. Others hold that a deeper unity underlies the apparent plurality: perhaps all good explanations support counterfactuals (Woodward), or all involve some version of unification (Kitcher), or all reflect the world's causal structure (Salmon). The debate between monists and pluralists remains open, and it is one of the liveliest areas of the field.
The current subfield is characterized by productive traffic between philosophy, statistics, computer science, and the special sciences. The interventionist approach has become a lingua franca across these disciplines, especially in epidemiology, econometrics, and artificial intelligence, where causal inference from data is a pressing practical problem. Philosophers contribute by clarifying the assumptions behind these methods, such as when observational data can support causal conclusions and what the notion of an "intervention" actually requires.
At the same time, philosophers continue to debate the metaphysics of causation: whether causation is a real feature of the world or a projection of our inferential practices, whether it reduces to physical processes or is irreducible, whether it is local or global. These questions are not merely academic; they inform how scientists interpret their models, how statisticians define causal effects, and how policymakers justify interventions.
A notable recent development is the attention to explanatory asymmetry and contrast in the context of actual scientific practice. Philosophers now study not just ideal theories but the ways scientists actually explain: how they choose contrasts, how they handle imperfect knowledge, how they integrate different levels of explanation. This "philosophy of science in practice" movement has blurred the boundary between philosophy and the history and sociology of science, and it has made the subfield more empirically informed.
Another emerging area is the relationship between explanation and understanding. Some philosophers, following van Fraassen, argue that explanation is ultimately about producing understanding in humans, which is a psychological achievement. Others, following Salmon, insist that explanation is about fitting the world's causal structure, independent of human psychology. This debate connects the subfield to epistemology and cognitive science, raising questions about what kind of cognitive achievement understanding is and whether it can be modeled computationally.
The field of explanation and causation is thus not a settled doctrine but a set of live debates structured by a few recurring questions: whether explanation reduces to causation or is something more, whether there is one explanatory ideal or many, whether the standards of explanation are objective or context-dependent, and how philosophical theories connect to the actual practices of scientists. Each major approach—Hempel's logical empiricism, van Fraassen's pragmatism, Kitcher's unificationism, Salmon's causal mechanism, Woodward's interventionism, and the mechanistic school—answers these questions differently, and each has left a permanent mark on how philosophers (and increasingly, scientists themselves) think about what it means to explain the world.