Contribution
It generalizes Baumgartner and Gebharter’s evidential result (NOTE-1493) into a claim about constitution as such: there is no experimentum crucis for it, “not even under ideal discovery circumstances” (abstract). It then defines constitution in a way that fits that fact. On the No De-Coupling theory (NDC), a part is a constituent because whole and parts cannot be pulled apart by any cause. On MM it was because each can be used to move the other.
Key insight
The kinds of underdetermination differ. Causal models are underdetermined by noisy, confounded data, and an ideal experiment removes the confounding, since causes and effects can be separated (pp. 10–11). In constitutional discovery “data confounding is introduced by the very experimental manipulations intended to uncover constitution” (p. 11). No idealization removes it. What remains is a pattern: the coupling of whole and parts is never broken, however many causes are tried. A constitutive model explains that pattern, and a common-cause model only tolerates it (pp. 14–15).
Assumptions
- Constitution is non-causal, after Craver and Bechtel 2007 (p. 3).
- Phenomena non-reductively supervene on their constituents “relative to a given a mechanistic organization” (p. 3).
- Parthood is “containment in a phenomenon’s spatiotemporal extension”, axiomatized by Ground Mereology. A “thin” notion, not analysed (footnote 1, p. 3).
- Specific variables for behaviours of specific entities (p. 4).
Key results
- MM is unsatisfiable (pp. 6–8). If I_Ψ intervenes on Ψ and changes Φ₁, it causes both. The single-path option I_Ψ → Ψ → Φ₁ is excluded because Ψ and Φ₁ overlap in space and time, so I_Ψ is a common cause and not an intervention. By supervenience “every cause of a mechanism’s macro level is necessarily a common cause of the phenomenon and at least one of its constituents” (pp. 7–8). Footnote 5 rejects the reply that variable sets may contain only causally distinct variables: whether I_Ψ is surgical depends on “whether what is represented by I_Ψ surgically causes Ψ in the world”.
- Relaxing interventions does not help (pp. 8–10). Under Woodward 2015, common causes of macro and micro count as interventions and MM becomes applicable. But “Model 3a, which features constitutive dependencies, and model 3b, which does not, imply the very same correlations under manipulations”; “For every model featuring constitution there exists a pure common-cause model that entails the very same correlations under manipulations” (p. 10). Harbecke’s method-of-difference proposal fails for the same reason (p. 10).
- No crucial experiment (p. 11). Even if “Φ is a constituent of Ψ” is tested in isolation with the whole background taken as certain, “no evidence can be produced demonstrating that this hypothesis is true and its negation false.”
- Coupling, with redundancy-freedom (pp. 12–14). Every cause of Ψ is a common cause of Ψ and some element of Φ, and no proper subset of Φ satisfies this. Redundancy-freedom excludes sterile effects (the car’s air conditioning, driven by the engine) and detached parts (the ashtray, warped by the same accident that stops the car).
- NDC (p. 15). Φ₁(X₁) is constitutively relevant to Ψ(S) iff some variable set V contains Ψ and a Φ = {Φ₁,…,Φₙ} with (1) Parthood: every complex instance of Φ is part of an instance of Ψ; (2) Coupling, (i) and (ii) as above; (3) No De-Coupling: “The Coupling of Φ(X) and Ψ(S) cannot be broken (invalidated) by expanding V.”
- Logical form (p. 16). NDC defines constitution by a negative existential, no surgical cause, so it is a universally quantified relation, “much closer to non-causation or causal irrelevance than to causation”, and “can only be inductively corroborated”. Its extension is a proper subset of causal irrelevance.
- Extrinsic but not relative (pp. 16–17). Constitutive relevance is “an extrinsic property of that pair, which depends on whether Φi is contained in a constituting set of Ψ”, yet “does not turn it into a relativized notion”: it depends on the existence of any such set, and holds across all expansions or not at all.
- Discovery (pp. 17–18). A finding of a surgical cause of Ψ conclusively falsifies one candidate set’s coupling but not the constitutive relevance of its members, which may belong to another set; “every constitutional inference—be it to constitution or to non-constitution— inevitably involves an inductive leap.” How far a set must be expanded has “no universal rule” (footnote 10).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Under non-causal constitution and supervenience, every cause of a phenomenon is a common cause of it and some constituent | strong | pp. 6–8, deduction |
| C2 | For every constitutive model there is a pure common-cause model with the same correlations under all manipulations | strong | pp. 9–10; follows from C1 for correlational data |
| C3 | This underdetermination, unlike the causal kind, survives ideal discovery conditions | moderate | pp. 10–11; argued, and depends on the claim that ideal causal experiments are conclusive (Hofmann & Baumgartner 2011, cited) |
| C4 | Constitution is best defined by unbreakable, redundancy-free common-cause coupling (NDC) | weak | a proposal; no operationalization (“postpone … to another occasion”, p. 17) and no case study (p. 19) |
| C5 | Constitutional inference is abductive and only inductively corroborated | moderate | follows from NDC’s universal form, p. 16 |
Concepts
- Common-cause coupling — every cause of Ψ in V is a common cause of Ψ and some element of Φ (p. 15).
- Redundancy-free — no proper subset of Φ is common-cause coupled with Ψ (p. 13).
- No De-Coupling — the coupling persists under every expansion of V (p. 15).
- Sterile effect — Craver’s term for a constitutively irrelevant downstream effect of a constituent (p. 13).
Connections
- Builds on Baumgartner & Gebharter (LIT-1934), dropping its claim that a temporal clause restores MM’s consistency, and on Scheines 2005 for fat-handed interventions.
- Cites Romero 2015 (LIT-1933) and Eronen & Brooks 2014 for the same fat-handedness result (footnote 6), and answers Harinen (LIT-1938) in footnote 8.
- Footnote 9 notes a parallel with Causey 1977: a biconditional bridge law is read as identity when every causal explanation of it has failed.
- Baumgartner, Casini & Krickel (LIT-1924) take the opposite route, a restricted intervention that gives a sufficient condition.
Bearing on the record’s own argument
- CLAIM-522: the same test turned round, in form only. CLAIM-522 separates an internal producer from an outside controller by an intervention that breaks one of them: disable the pathway inside, or interrupt the controller. NDC calls a set of parts constitutive when no cause can break its coupling with the whole. Both, then, read the structure off what can and cannot be decoupled. The outside controller plays the part of the cause NDC looks for: something that moves the capacities while running through nothing inside the declared boundary. That is a likeness of form, not shown to be the same relation. CLAIM-522’s producer–capacity relation is diachronic production, which can be broken surgically, not part–whole realization, which by C1 cannot. Nothing here makes the producer part of the system. Which pathways are “inside” is NDC’s Parthood condition, taken as given (footnote 1), and CLAIM-522’s declared boundary.
- QUESTION-284 and CLAIM-150. QUESTION-284 anticipated that if no evidence distinguishes “ontological constitution from ordinary causal modification”, that is itself an answer, CLAIM-150’s underdetermination. This paper proves the first-order version for part–whole constitution (C2, C3) and adds a second-order route: constitution is preferred when every attempt to decouple fails, and preferred for its explanatory power, “not forced upon the modeler by the evidence” (p. 3). So, for this kind of constitution, the answer is underdetermination by any finite set of manipulations, and abduction from the failure to decouple. The question’s “constitute” (relations fixing what a relatum is) is a different relation. Whether the same second-order route is open for it is a sharper form of the question (THEORY-1294’s Connections).
- CLAIM-477 and CLAIM-405. NDC makes constitutive relevance extrinsic to the part–whole pair, a matter of the part’s membership in a redundancy-free set coupled with the whole (p. 16). That is CLAIM-477’s “what an entity is, is partly fixed by the … configurations … it can enter”, for one relation: being a constituent is fixed by the set the part belongs to. And it is set-relative without being relativized (pp. 16–17), which is CLAIM-405’s “dependence does not make it arbitrary”: the verdict depends on a set, but once the phenomenon is fixed it can be false. Same structure, not shown same relation in either case.
- CLAIM-515’s “realized”. On this paper, if “realized” means constituted, “shown by interventions that discriminate implementations” is not available even in principle (C3). It is inferred abductively from the failure to decouple. CASE-163’s catalog identification separates candidate causal mechanisms, which is a different inference (CLAIM-528).
- TERM-088. “Constitutive” individuation in the term and constitutive relevance here are different relations. The paper does give a reason a constitutive sense needs its own evidence: it cannot be read off interventions as the dynamical sense can.
Limitations
- NDC is unoperationalized (p. 17) and untested against practice (p. 19). How to enumerate “every cause of Ψ in V”, and when V has been expanded enough (footnote 10), are left open.
- Parthood is assumed thin and given (footnote 1). Leuridan’s endosymbiosis counterexample (LIT-1929, from its abstract) is not discussed.
- The contrast with causation is overdrawn in one respect. The conclusiveness of ideal causal experiments is cited, not shown, and it holds only for mereologically independent variables, which is the assumption at issue.
- A unique constituting set is not required, so Ψ may have several, and whether the theory then picks out one mechanism is not discussed.
Open questions
- Can NDC be stated for dynamical models, where a coupling might be broken at one time scale and not at another?
- Is “no surgical cause of the whole exists” testable in a computational system where every cause can be enumerated? A model with a declared variable set and a complete list of exogenous inputs would let NDC be decided rather than corroborated.