20.08.02 · philosophy / phil-of-science

Scientific realism vs. anti-realism: structural realism, constructive empiricism (van Fraassen)

stub3 tiersLean: nonepending prereqs

Anchor (Master): Psillos, S. — Scientific Realism: How Science Tracks Truth (1999)

Intuition Beginner

When a scientific theory speaks of electrons, quarks, or black holes — things no one can directly see — should we believe those things exist? Scientific realists answer yes. They hold that our best-confirmed theories are approximately true, and that the unobservable entities the theories introduce genuinely populate the world.

The realist's strongest card is the "no miracles" argument, associated with Hilary Putnam. Modern science predicts the outcomes of experiments with striking precision. That predictive success would be a kind of miracle, the argument runs, if our theories were not at least roughly tracking how things really are. Accurate prediction points toward truth.

Anti-realists are not convinced. Bas van Fraassen argues that science aims at empirical adequacy — fitting what we can observe — rather than literal truth, so we should accept theories without believing their unobservable claims. Larry Laudan replies that many once-successful theories (phlogiston, caloric, the ether) later proved false. John Worrall splits the difference: trust the equations, not necessarily the entities.

Visual Beginner

Figure: A horizontal map of positions on the question "what may we believe?". From left to right: scientific realism (believe theories are true and their entities real), entity realism (believe entities, not theories), structural realism (believe the mathematical structure, not the entities), constructive empiricism (accept empirical adequacy only), and instrumentalism (theories are tools, nothing more). Two diagonal arrows cross the map: the no-miracles argument pulls leftward toward belief; the pessimistic meta-induction pushes rightward toward caution.

The positions differ on a single axis: how far acceptance of a theory commits us to the reality of what the theory describes. Every later section refines this map, adding nuance about which parts of a theory earn belief, and how much continuity survives when one theory replaces another.

Worked example Beginner

In 1818 Augustin-Jean Fresnel treated light as a wave travelling through the luminiferous ether. His equations made a sharp prediction: light reflecting off a transparent solid at one particular angle becomes completely polarised. For ordinary glass, with refractive index about 1.5, the predicted angle is roughly 56 degrees. The prediction was specific and testable.

Step 1. Experiment confirmed it. A beam reflected off glass at 56 degrees came out polarised exactly as Fresnel's equations foretold. The ether, the medium carrying the wave, seemed as real as the glass.

Step 2. Decades later, James Clerk Maxwell reconceived light as an electromagnetic wave. The ether was dropped — modern physics finds no such substance. The entity at the centre of Fresnel's theory does not exist.

Step 3. Yet Fresnel's equations survived. They reappear, almost unchanged, inside Maxwell's theory. The mathematics persisted; only the underlying substance was removed.

What this tells us: this single transition is John Worrall's showcase for structural realism — across a revolution that destroyed one ontology, the equations carried over intact.

Check your understanding Beginner

Exercise (easy, multiple choice).

What is the "no miracles" argument for scientific realism?

A. Genuine miracles occur inside scientific laboratories. B. The predictive success of science would be miraculous if our theories were not at least approximately true. C. Every successful theory turns out to be exactly true. D. Science cannot explain miracles, so theories must be false.

Hint

The argument treats accurate prediction as evidence. If a false theory predicted reliably, what would that imply about coincidence?

Answer

Option B. The no-miracles argument infers that the best explanation of science's predictive success is that successful theories are approximately true. Feedback-correct: B captures the abductive structure. Feedback-wrong: C is too strong (approximately, not exactly); A and D misread "miracle" literally.

Exercise (easy, multiple choice).

The "pessimistic meta-induction" (Laudan) is best described as the claim that:

A. Scientists are psychologically predisposed to pessimism. B. Because many past successful theories turned out false, current successful theories are probably false as well. C. Induction is an unreliable form of reasoning in every case. D. Realism makes scientists give up on research.

Hint

Laudan lists formerly successful theories — phlogiston, caloric, the ether — whose central entities we now reject.

Answer

Option B. The argument generalises from past false-but-successful theories to a warning about present ones. Feedback-correct: B states the inductive challenge to realism. Feedback-wrong: A and D misread "pessimistic" as a mood; C overstates the conclusion beyond science.

Formal definition Intermediate+

Scientific realism is standardly analysed as a conjunction of three theses. The metaphysical thesis: there exists a mind-independent world, and the way it is does not depend on our beliefs about it. The semantic thesis: scientific theories should be read at face value — their claims, including those about unobservable entities, are literally true or false. The epistemic thesis: mature, predictively successful theories are approximately true, and the unobservable entities they introduce (electrons, fields, quasars) genuinely exist.

Definition (empirical adequacy, van Fraassen). A theory $T$ is empirically adequate iff what $T$ says about the observable phenomena is true — equivalently, at least one model of $T$ matches the actual world on every observable structure. To accept $T$ , in van Fraassen's sense, is to believe that $T$ is empirically adequate, not to believe that $T$ is true.

Definition (structural realism, Worrall). Theory $T_{2}$ preserves the structure of predecessor $T_{1}$ when the empirical-content equations of $T_{1}$ — the relations among measurable quantities — reappear within $T_{2}$ , even if $T_{2}$ 's ontology (the entities it posits) differs from $T_{1}$ 's. Epistemic structural realism (ESR) claims we can know the structure but not the natures of unobservables; ontic structural realism (OSR) claims that, at the fundamental level, there are no entities, only structure.

Distinguishing the positions Intermediate+

Realism vs entity realism. Realism commits to the approximate truth of whole theories; entity realism (Hacking, Cartwright) commits only to the reality of entities we can manipulate, whatever the truth of the laws.
Constructive empiricism vs instrumentalism. Instrumentalism treats theories as mere calculating devices; constructive empiricism treats theories as literally true-or-false but limits rational belief to their empirical adequacy.
Observable vs unobservable. Van Fraassen anchors observability to what unaided human senses could detect: the moons of Jupiter are observable (an astronaut could see them); electrons are not, no matter how powerful the instrument.

Key argument — the no-miracles argument and the pessimistic meta-induction Intermediate+

The case for realism rests on an inference to the best explanation, the no-miracles argument (Putnam, Boyd).

Premise N1. Mature scientific theories yield novel, precise, often surprising predictions that are subsequently confirmed.

Premise N2. The best explanation of this predictive success is that the theories are at least approximately true — that their theoretical terms latch onto real features of the world.

Conclusion N. We are warranted in believing that mature, successful theories are approximately true, and that their unobservables are real.

The alternative, realists urge, is to treat success as a cosmic coincidence — a "miracle".

Larry Laudan's pessimistic meta-induction turns the historical record against N2.

Premise P1. Many past theories were genuinely successful by any reasonable measure — phlogiston, caloric, the crystalline spheres, the luminiferous ether.

Premise P2. The central unobservable entities of these theories do not exist.

Conclusion P. By induction, the central unobservable entities of our current successful theories are probably non-existent as well; predictive success is no reliable guide to truth.

The two arguments collide head-on. N-miracles says success licences belief; P-meta-induction says it does not. Each feeds on the other's weakness. Structural realism and selective realism are the principal compromise strategies, each conceding enough history to Laudan while preserving enough truth-connection to satisfy Putnam.

Selective realism (Psillos, Kitcher) answers Laudan by separating working from idle parts of a theory. The idle posits — those not responsible for the theory's predictive success — may be discarded without loss; the working posits are the ones preserved across theory change. On this reading the ether was idle to the predictive work, while the equations carried the load.

Bridge. This argument reconstruction builds toward 20.08.03 pending, where inference to the best explanation reappears as the engine of theories of causation and explanation, and the no-miracles argument appears again in 20.08.03 pending as the paradigmatic IBE. The foundational reason the realism debate resists easy resolution is that IBE is doing double duty: the realist uses IBE to license belief in unobservables, while the anti-realist asks why that same mode of inference should be trusted when its past outputs (caloric, ether) so often proved false. This is exactly the tension that generalises into every "inference to unobservables" across the philosophy of science — from 20.03.02 pending (the reality of the quantum state) to 20.05.04 pending (genes as theoretical entities in the molecular revolution). The bridge is between epistemology and the special sciences: how we reason about what we cannot see.

Exercises Intermediate+

Exercise 1 (easy, argument reconstruction).

Reconstruct the no-miracles argument in explicit premise-conclusion form, and state precisely which premise the anti-realist rejects.

Hint

The argument has two premises: one about success, one about the best explanation of success. Which one does the constructive empiricist or the pessimistic inductivist deny?

Answer

Premise N1. Mature theories yield novel, confirmed predictions. Premise N2. Approximate truth is the best explanation of that success. Conclusion. Mature successful theories are approximately true. The anti-realist rejects N2: van Fraassen offers empirical adequacy as a competing explanation of the same success that does not require truth; Laudan denies that success tracks truth at all, citing false-but-successful ancestors. Either way the inference from success to approximate truth is blocked.

Exercise 3 (medium, case analysis).

The caloric theory of heat treated heat as a fluid (caloric) that flowed from hot bodies to cold ones. It successfully predicted many thermal phenomena, including the results Sadi Carnot used to analyse heat engines, before being replaced by thermodynamics. Reconstruct how Laudan uses this kind of case against the realist, and state the selective realist's reply.

Hint

Caloric was predictively fruitful yet its central entity does not exist. Which parts of the theory were preserved, and which were discarded?

Answer

Caloric was predictively fruitful — it underwrote Carnot's analysis of heat engines, which survived into thermodynamics — yet its central entity, the caloric fluid, does not exist. Laudan's pessimistic meta-induction uses exactly such cases: success did not track truth, so realism's key inference (success implies approximate truth) is unreliable.

The selective realist (Psillos) replies by separating the working from the idle parts of caloric theory. The quantitative relations — the cycle, later entropy — did the predictive work and were preserved, while the fluid ontology was idle and was shed. On this reading the case shows not that realism fails, but that we should believe only the working posits, which is the weaker, selective claim.

Exercise 4 (medium, symbolic).

On van Fraassen's semantic view, a theory $T$ is identified with a family of models $M (T)$ . Write the empirical-adequacy condition in the form: there exists $M \in M (T)$ such that the observable substructure of $M$ is isomorphic to the observable substructure of the actual world $W$ . Then explain why, on this formulation, two empirically equivalent theories can disagree on unobservable structure.

Hint

Empirical adequacy constrains only the observable part of the models. What is left unconstrained?

Answer

Formally, $T$ is empirically adequate iff $\exists M \in M (T) : Obs (M) ≅ Obs (W)$ , where $Obs (\cdot)$ extracts the observable substructure. Two theories $T$ and $T^{'}$ can share the same observable consequences, $Obs (M (T)) = Obs (M (T^{'}))$ , while differing on their full models — that is, on unobservable structure. Because empirical adequacy constrains only the observable part, both can be empirically adequate and yet disagree on what the world is like beyond observation. This gap is van Fraassen's lever for agnosticism about unobservables.

Exercise 5 (hard, argument reconstruction).

Critically assess whether the Fresnel-to-Maxwell transition genuinely supports structural realism rather than selective realism. Does the survival of Fresnel's equations show that structure is what is preserved, or merely that the equations were the working parts? State which reading you find stronger and defend it.

Hint

The two readings agree on the history but diverge on the moral. Ask: in this case, does the preserved structure coincide with the preserved working posits, or come apart?

Answer

The structural realist reads the survival of Fresnel's equations as evidence that mathematical structure is the continuous thread across theory change. The selective realist reads the same case differently: the equations survived because they were the working parts doing the predictive labour, while the ether — an idle posit — was discarded. The two readings agree on the history but diverge on the moral.

The structural reading is stronger where one can show that structure persists even when no entity is preserved, as some argue for quantum-theoretic cases (see 20.03.02 pending). The selective reading is stronger where continuity is better explained by the retention of specific working posits. The Fresnel-Maxwell case alone does not decide between them, because here the preserved structure coincides exactly with the preserved working parts: the same equations did the work and constituted the structure. The case is consistent with both, so it cannot, by itself, adjudicate.

Exercise 6 (hard, critique).

Van Fraassen anchors the observable/unobservable distinction to unaided human perception. Evaluate this criterion. Consider: a distant exoplanet seen only through a telescope; a microorganism seen through a microscope; a molecule detected only by inference. Where does the line fall, and is the line epistemically principled?

Hint

Ask whether observability is an all-or-nothing, human-physiology-anchored notion, or whether it grades off at the edges.

Answer

On van Fraassen's criterion the exoplanet (in principle observable by an appropriately placed observer) and the microorganism (visible under magnification, on his reading of microscopes as extending vision) count as observable, while the molecule — detectable only via theory-laden inference — counts as unobservable. The line is principled only if observability is an all-or-nothing notion anchored to human physiology. But it grades off at the edges (electron microscopy, cloud-chamber tracks), and tying epistemic significance to the contingent limits of the unaided human body looks arbitrary: a being with different senses would draw the line elsewhere.

The criterion is stable in the clear cases and contentious exactly where the realism debate bites, which is both its strength (it handles the paradigms cleanly) and its weakness (the hard cases are the ones that matter). Whether this is a fatal objection or a tolerable boundary blur remains the live point of contention between van Fraassen and his realist critics.

Realism, anti-realism, and the structure of theory change Master

Constructive empiricism and the aim of science Master

Van Fraassen's constructive empiricism, set out in The Scientific Image (1980) ^{[van Fraassen 1980]}, relocates the realism debate from truth to aim. Science aims at empirical adequacy, not truth: a theory is acceptable when what it says about the observable phenomena is correct, and acceptance is belief in that adequacy — not belief in the theory's further claims about unobservables. Acceptance also carries a pragmatic commitment: to use the theory as a guide in one's research, to entrust oneself to its picture of the phenomena. The distinction between observable and unobservable carries the epistemic weight. Moons are observable — a human observer placed nearby could see them; electrons are not, since no human could perceive them unaided, however powerful the instrument.

The sharpest objection targets that distinction. Grover Maxwell argued that there is no principled line: every observation depends on some causal chain, and the difference between looking through a telescope and looking through a microscope is one of degree, not of kind. Van Fraassen replies that observability is a contingent, empirical matter about human physiology, not a philosophical essence — but this makes the boundary hostage to scientific findings, complicating its role as a fixed epistemic criterion. The debate connects outward to sensation and perception (see [29.03.*]).

Structural realism: epistemic and ontic Master

Worrall's "Structural Realism: The Best of Both Worlds?" (1989) ^{[Worrall 1989]} proposes that the pessimistic meta-induction defeats entity realism but not structural realism. Across the Fresnel-to-Maxwell transition the ether vanished while the equations persisted: what is retained through scientific revolutions is mathematical structure, not ontology. This promises the realist a reply to Laudan (continuity is real) without conceding too much (the continuity is structural, not entity-level).

Two readings diverge. Epistemic structural realism (ESR) holds that we can know the relational structure of the unobservable world while remaining ignorant of its intrinsic natures — there are entities, but structure is all we grasp. Ontic structural realism (OSR), developed by French and Ladyman, goes further: at the fundamental level there are no objects, only structure; particles are better understood as nodes in a relational pattern. OSR finds its strongest motivation in quantum mechanics, where individuality and identity behave in ways that strain any object-first metaphysics (see 20.03.02 pending). Critics charge OSR with incoherence — structure without relata is hard to make intelligible — while ESR is accused of collapsing back into either realism or empiricism depending on how "structure" is parsed.

Entity realism and the interventionist criterion Master

Hacking's Representing and Intervening (1983) ^{[Hacking 1983]} reframes belief: we are entitled to believe in an unobservable entity when we can use it to intervene in something else. If experimenters spray electrons onto a target to produce a predicted effect, they are committed to the electron's reality regardless of whether the surrounding theory is true. The criterion is manipulative, not representational. Nancy Cartwright's How the Laws of Physics Lie (1983) and The Dappled World (1999) press a related line: the entities are real, but the laws are idealised and frequently false.

The objection is that interventionism is both too weak and too narrow. Too weak, because past entities (phlogiston) were experimentally manipulated before being eliminated. Too narrow, because many central unobservables — quarks, confined and never isolated — cannot be manipulated in Hacking's sense, yet physics treats them as central. The criterion covers some entities well and leaves others outside its reach, yielding a patchwork realism rather than a general epistemology.

Selective realism and the working/idle distinction Master

Psillos's Scientific Realism: How Science Tracks Truth (1999) ^{[Psillos 1999]} and Kitcher's earlier work refine the realist defence into selective realism: believe only the parts of theories responsible for their predictive success. On this view the pessimistic meta-induction misfires because it treats theories as all-or-nothing packages. Once the working posits (Maxwell's equations, the gene) are separated from the idle baggage (the ether, pre-Mendelian blending assumptions), the working posits survive theory change and the idle ones do not. Realism is preserved for the working posits, and Laudan's counterexamples are explained away as failures of the idle parts.

The anti-realist rejoinder is that the working/idle distinction is drawn retrospectively, with the benefit of knowing which posits later theories preserved. Applied prospectively, we cannot tell which parts of our current theories are doing the predictive work and which are idle decoration, so selective realism offers no advance guidance about what to believe now. The debate turns on whether the distinction can be made independently of the historical outcome it is meant to explain.

Bayesian, explanationist, and pluralist currents Master

The realism debate ramifies. Howson and Urbach's Scientific Reasoning: The Bayesian Approach argues that the no-miracles inference is not valid as standardly stated: its force depends on prior probabilities, and the "old evidence" problem complicates using past success to confirm present theories (see 20.01.02 pending). Lipton's inference-to-the-best-explanation programme supplies the realist's master move but also its vulnerability, since IBE is precisely what the anti-realist mistrusts (see 20.08.03 pending).

Cartwright's The Dappled World argues that the laws of physics form a patchwork rather than a unified pyramid; Dupré's The Disorder of Things urges pluralism about what counts as scientific knowledge. Galison's Image and Logic documents distinct cultures of scientific practice, undermining the picture of a single unified method. Longino's Science as Social Knowledge introduces contextual empiricism and the role of values; Harding and the feminist standpoint tradition argue that the value-free ideal produces less objective science. Decolonial work (Harding, Santos) presses further on whose knowledge counts. In climate science, Lloyd, Winsberg, and Parker analyse how models, uncertainty, and values interact with realism — climate skepticism is, partly, a realism dispute conducted in public (see 27.07.02 pending). Feyerabend's Against Method (1975) supplies the anarchist flank: if every proposed method has been violated by successful science, "anything goes" as methodology (see 20.08.01).

Synthesis. The foundational reason the realism debate resists closure is that every argument — no-miracles, pessimistic induction, structural preservation, interventionist manipulation — reconstructs the same historical record from a different epistemic vantage. The central insight is that "success" is not a unitary datum: it decomposes into predictive, explanatory, and manipulative strands, each licensing a different degree of belief. Putting these together with the Bayesian and pluralist currents, this is exactly why structural realism and selective realism converge — both try to locate the minimal, survival-proof content of a theory — and the pattern generalises from 20.03.02 pending (what the wave function describes) to 20.05.04 pending (whether species and genes are real) and 20.09.01 (mathematical Platonism as a parallel realism dispute). The bridge is between philosophy of science and the special-science realisms it underwrites; values, models, and historical contingency enter not as contaminants but as part of the very content over which the realism dispute is fought.

Connections Master

Demarcation, falsification, and paradigms 20.08.01. This unit's predecessor frames the realism debate: Kuhn's paradigms and the Duhem-Quine thesis supply the historical and logical pressure that makes the pessimistic meta-induction possible. Realism debates presuppose an account of what theories are and how they change, which is the demarcation unit's deliverable.
Causation and explanation 20.08.03 pending. Builds toward the theory of inference to the best explanation that the no-miracles argument instantiates. The realism dispute is, at its core, a dispute about whether IBE can carry belief all the way to unobservables — the very move that theories of causation and explanation make explicit.
Epistemology: knowledge, justification, and truth 20.01.01. Provides the upstream framework: what counts as justification, how truth relates to evidence, and how inference licences belief. Constructive empiricism is an epistemic stance first and a philosophy-of-science doctrine second.
Philosophy of quantum mechanics 20.03.02 pending. The hardest test case for structural realism: the wave function's ontic status, Bell's theorem trading realism against locality, and OSR's motivation from quantum individuality all replay the realism debate at the frontier of physics.
Reductionism in biology 20.05.04 pending and philosophy of mathematics 20.09.01. Genes and mathematical objects are parallel "unobservables": each domain asks whether successful theorising commits us to the reality of its theoretical posits, making the special sciences independent laboratories for the realism debate.

Historical and philosophical context Master

Scientific realism is ancient in impulse — we naturally read our best theories as describing a real world — but its modern form is a twentieth-century construction. Peirce's convergence thesis (the view that inquiry, indefinitely prolonged, would settle toward the real) is a proto-realist strand. The logical positivists, despite their verificationism, largely held a quiet realism about observables; it was the post-positivist generation, after the collapse of the observation/theory distinction, that made realism a thesis requiring explicit defence.

J. J. C. Smart's Philosophy and Scientific Realism (1963) ^{[Smart 1963]} opened the modern defence, arguing that physics genuinely describes a micro-world of theoretical entities. Hilary Putnam and Richard Boyd formulated the no-miracles argument in the 1970s, making realism turn on the explanatory credentials of science's success. Larry Laudan's "A Confutation of Convergent Realism" (1981) ^{[Laudan 1981]} supplied the historical counterattack with its catalogue of successful-but-false theories. Van Fraassen's The Scientific Image (1980) ^{[van Fraassen 1980]}, delivered as the 1978 Hempel Lectures, reframed the dispute around empirical adequacy and the aim of science.

The 1980s and 1990s saw the position multiply. Hacking (1983) and Cartwright (1983) split off entity realism; Worrall (1989) introduced structural realism with the Fresnel-Maxwell case; Psillos (1999) consolidated selective realism. French and Ladyman developed ontic structural realism through the 2000s, drawing on the quantum-mechanical challenges to individuality. Concurrently, feminist philosophy of science (Longino 1990, Harding 1991) and decolonial science studies broadened the debate from "is realism true?" to "whose realism, serving which interests?", relocating part of the dispute in social epistemology.

The contemporary landscape treats the realism question as irreducibly plural: no single argument settles it, and the most productive work stages the disagreement between specific arguments (Bayesian no-miracles, structural preservation, interventionism) rather than between monolithic "realism" and "anti-realism". Climate science and the replication crisis have pushed the dispute back into the public square, where realism about models, about mechanisms, and about statistical claims is negotiated under pressure from policy and values.

Bibliography Master

Smart, J. J. C. — Philosophy and Scientific Realism (Routledge & Kegan Paul, 1963).
Putnam, H. — Mathematics, Matter and Method: Philosophical Papers, Volume 1 (Cambridge University Press, 1975).
Boyd, R. — "Scientific realism and naturalistic epistemology", PSA: Proceedings of the Biennial Meeting of the Philosophy of Science Association 1980 (2), 613–662 (1980).
Laudan, L. — "A confutation of convergent realism", Philosophy of Science 48 (1), 19–49 (1981).
van Fraassen, B. C. — The Scientific Image (Clarendon Press, Oxford, 1980).
Hacking, I. — Representing and Intervening: Introductory Topics in the Philosophy of Natural Science (Cambridge University Press, 1983).
Cartwright, N. — How the Laws of Physics Lie (Clarendon Press, Oxford, 1983).
Cartwright, N. — The Dappled World: A Study of the Boundaries of Science (Cambridge University Press, 1999).
Worrall, J. — "Structural realism: the best of both worlds?", Dialectica 43 (1–2), 99–124 (1989).
Kitcher, P. — The Advancement of Science: Science without Legend, Objectivity without Illusions (Oxford University Press, 1993).
Psillos, S. — Scientific Realism: How Science Tracks Truth (Routledge, 1999).
Ladyman, J. — Understanding Philosophy of Science (Routledge, 2002).
French, S. & Ladyman, J. — "Remodelling structural realism: quantum physics and the metaphysics of structure", Synthese 136 (1), 31–56 (2003).
Ladyman, J. & Ross, D. (with Spurrett, D. & Collier, J.) — Every Thing Must Go: Metaphysics Naturalised (Oxford University Press, 2007).
Howson, C. & Urbach, P. — Scientific Reasoning: The Bayesian Approach, 3rd ed. (Open Court, 2006).
Lipton, P. — Inference to the Best Explanation, 2nd ed. (Routledge, 2004).
Kuhn, T. S. — The Structure of Scientific Revolutions, 2nd ed. with postscript (University of Chicago Press, 1970).
Longino, H. E. — Science as Social Knowledge: Values and Objectivity in Scientific Inquiry (Princeton University Press, 1990).
Harding, S. — Is Science Multicultural? Postcolonialisms, Feminisms, and Epistemologies (Indiana University Press, 1998).
Feyerabend, P. K. — Against Method: Outline of an Anarchistic Theory of Knowledge (New Left Books, 1975).
Maxwell, G. — "The ontological status of theoretical entities", in Feigl & Maxwell (eds.), Scientific Explanation, Space, and Time, Minnesota Studies in the Philosophy of Science III (University of Minnesota Press, 1962), pp. 3–27.