NOTE-137: Hazelwood — Emerging dilemma for reciprocal causation

Contribution

The paper takes the EES slogan “reciprocal causation” and makes it precise as a claim about processes (RC). It then shows that RC is incompatible with the dominant causalist view of natural selection, and also with every alternative to it. What is new is that the reciprocal-causation debate is joined to the metaphysics of emergence (Wilson, Pence), the interlevel-causation literature (Kim, Craver & Bechtel) and Woodward’s interventionism. It also salvages the EES’s substantive aim: niche construction is a cause of evolutionary outcomes by partly composing selection.

Key insight

Selection is a population-level process composed of individual-level goings-on such as niche construction. The relation between them is therefore part–whole, not cause–effect. Counterfactual dependence of selection on niche construction is compositional dependence, and the two variables are not independently fixable, so interventionism does not license a causal arrow between the processes. Niche construction makes its evolutionary difference through selection, as part of it, not to selection as a cause of it.

Assumptions

• RC is read in the strong, process–process sense. Its textual basis is Laland et al. 2015: “process A is a cause of process B and, subsequently, process B is a cause of process A”.
• Emergence: Wilson’s two conditions, synchronic dependence plus ontological and causal autonomy (irreducibility, novel laws or multiple realizability), argued by cases. Realization is taken to be (or entail) composition.
• Selection’s relata are population-level: variation in trait fitness and change in trait frequencies (Millstein 2006; Sober 2013). A ceteris paribus clause (Sober 1984) applies.
• Niche construction is an individual-level, developmental process (Laland et al. 2013: “environment-altering activities of phenotypes during ontogeny”). §4.2 argues the conclusion does not depend on this.
• Causation criterion (§4.3): Woodward’s interventionism with the independent-fixability condition (IF). A part–whole relation precludes causation because it fails IF (Woodward 2021, 230).
• Fitness is a propensity (Mills & Beatty 1979), used in fn. 14.
• “Levels” talk is replaceable by “scales”. The argument needs only compositional relations (fn. 5).

Key results

• Horn 1 (affirm EC), §4.1: niche construction (individual level) and selection (population level) are related by composition. Their counterfactual dependence is “deceptive” (the analogy is mean kinetic energy and temperature), so RC is “conceptually incompatible with the emergent character of natural selection” (p. 603).
• §4.2: even population-level niche construction (coal mining) does not rescue RC, since variation in fitness is still “synchronously realized, not caused” by it.
• §4.3: interventionism allows interlevel causation, but only between independently fixable variables. Intervening on “is a niche constructor” “simultaneously intervene[s] on (or perhaps introduce[s]) the emergent variation in fitness”, so niche construction and selection “qua processes are … not sufficiently distinct to stand in a causal relationship” (pp. 607–8). The same intervention can cause an evolutionary outcome, so niche construction is “a bona fide cause of evolution — the product of natural selection”.
• Horn 2 (deny EC), §5: statisticalism denies selection is causal, so RC fails a fortiori (Walsh 2019 is quoted calling RC a “category error”). Individual-level causalism is incoherent because relative fitness is a property of at least a pair.
• Conclusion, §6: “No available metaphysical account of natural selection is compatible with the view that selection is both a cause and effect of developmental processes such as niche construction” (p. 610). Niche construction “is undoubtedly an evolutionary process”.

Claims

id	claim	strength	support
C1	The EES endorses reciprocal causation between the processes of selection and niche construction, not only between organism and environment	moderate	textual evidence: Laland et al. 2015 and Laland & O’Brien 2011 quoted (pp. 597–98)
C2	Natural selection satisfies Wilson’s conditions for metaphysical emergence	informal argument	argument by cases (evolvability, population-genetic laws, multiple realizability); relies on Haug 2007 and Pence 2021
C3	Realization entails or coincides with composition, at least for selection	informal argument	Aizawa & Gillett, Wimsatt, Wilson; reply to Polger that “the underlying matter matters”
C4	If selection is emergent, niche construction cannot cause it	informal argument	composition vs causation (Craver & Bechtel 2007); temperature analogy
C5	Interventionism does not rescue RC because niche construction and selection fail independent fixability	informal argument	Woodward 2015, 2021 IF condition; the cholesterol analogy; the process/product distinction
C6	Niche construction is a genuine cause of evolutionary outcomes	informal argument	interventionist reading plus Clark et al. 2020 cited as empirical illustration
C7	Statisticalism entails rejecting RC	moderate	follows directly from statisticalism’s denial that selection is causal; Walsh 2019 concurs
C8	Individual-level selection is conceptually incoherent	informal argument	relative fitness is relational (Millstein 2006, Sober 2013); departs from Pence 2021 and Bourrat 2018 without engaging their replies in detail
C9	The dilemma is exhaustive, so RC must be abandoned	informal argument	the disjunction “emergent causal or not” is exhaustive, but horn 2’s split (non-causal vs individual-level causal) assumes those are the only non-emergent options

Concepts

• RC (strong reciprocal causation) — the process of natural selection is both cause and effect of processes such as niche construction.
• Weak reciprocal causation — organisms affect environments and vice versa. Called “unobjectionable”.
• EC — natural selection is an emergent causal process.
• Metaphysical emergence (minimal, fn. 6) — a higher-level process realized (not caused) by lower-level processes, synchronically dependent on them yet ontologically and causally autonomous (Wilson 2016).
• Composition / realization — the part–whole relation between individual-level processes and population-level selection. Treated as coextensive here.
• Independent fixability (IF) — Woodward’s condition that each variable can be set by intervention to any of its values while the others are independently set. Part–whole relata fail it.
• Process vs product of selection — the process is variation in fitness producing change; the product is differential reproductive success or changed trait frequencies (Millstein 2002; Brandon 2005).
• Statisticalism — selection is a statistical summary of individual births and deaths, not a cause (Walsh, Matthen, Ariew).

Connections

This is the other side of the reciprocal-causation debate the record holds through Baedke, Fábregas-Tejeda & Prieto (LIT-110). As noted in corrections, their organism–environment model falls under what Hazelwood calls the weak, unobjectionable sense; the direct target is Laland and colleagues. Baedke’s book (LIT-167) develops the organism-centred perspective that Hazelwood grants is compatible with his conclusion (organismal agency, citing Baedke & Fábregas-Tejeda 2023). Trappes on evolutionary versus ecological niches (LIT-169) bears on what “niche construction” is supposed to change. The composition-not-causation argument applies directly to the physicists’ emergence accounts in Rizi (LIT-150) and McKenzie (LIT-141). Both endorse “downward causation” as higher-level constraint (Rizi p. 7; McKenzie §2.10, reporting Noble), and Hazelwood’s independent-fixability test is the question neither asks. The paper is also a small worked case for the causal-emergence measures (LIT-025, LIT-027): those measures quantify macro-level “causal power” without saying whether macro and micro variables are independently fixable.

Metaphysical thesis. The paper defends emergentist population-level causalism about selection, together with a compositional account of realization. It defends the claim that part–whole relations cannot be causal (via failure of independent fixability), and that interlevel causation is possible only between distinct, independently fixable relata. These are theses about causation, emergence and composition, so the metaphysics tag is justified, and it should lead.

Bearing on the record

• No THEORY and nothing for ML practice. The paper is about evolutionary theory. Its composition-versus-causation criterion is, however, the right check on any claim in either record that a macro-level variable “causes” its own micro-constituents’ behaviour. Examples are “emergent” capabilities said to cause behaviour, or causal-emergence measures read as causal power.
• LIT-110’s standing note and NOTE-137’s “read the two together” should be corrected: Hazelwood’s target is Laland et al., and LIT-110 is cited only in passing.
• Pairs naturally with LIT-150 and LIT-141 as the philosopher’s constraint on the physicists’ “downward causation without extra causes”.

Note of 2026-10-16: THEORY-612 now cites LIT-156 as a source.

Limitations

• The process/product distinction does much of the work in §4.3. It is defended by citation (Millstein, Stephens, Brandon), not argued afresh. An EES proponent who identifies selection with differential reproduction would not accept that §4.3 has closed the gap.
• The composition reading of realization is defended briefly. The paper itself says a full defence of Pence’s compositional account is beyond its scope (p. 601).
• The temporal structure of reciprocal causation is not engaged in depth: A at t₁ affects selection at t₂, which changes A’s frequency at t₃. Hazelwood’s reply is that the diachronic arrow reaches only the product; a sequential, unrolled model such as LIT-110’s could press this.
• Horn 2 considers only statisticalism and individual-level causalism. Other non-emergent causal accounts (e.g. dispositional ones, Triviño & Nuño de la Rosa, cited but not discussed) are grouped with individualism.
• No formal model. Clark et al. 2020 is the only empirical work cited, and only as an illustration.

Open questions

• Does a diachronic, unrolled causal model (LIT-110) give a reciprocal structure between niche construction and later selection that satisfies independent fixability, because the relata are at different times?
• Is there a principled account of selection as both a population-level process and a cause of later individual-level developmental processes? That is the other half of RC (“selection causes niche construction”), which the paper treats only through the process/product split.
• Can the independent-fixability test be made quantitative for causal-emergence measures, as a criterion for when macro “causal power” is causal rather than compositional?

Corrections to the seeded skim

• Dossier: “Directly engages Baedke et al. (p035) on reciprocal causation; read the two together”. This is misleading. Hazelwood’s target throughout is Laland/Lala and colleagues (2011, 2013, 2015, 2016) and Laland & O’Brien 2011, whom fn. 1 says he cites “predominantly … because they have codified the tenets of the EES”. Baedke, Fábregas-Tejeda & Prieto 2021 (LIT-110) is cited only twice, in passing: once as critics of the “standard” synthesis’s black-boxing of the organism (p. 596), and in fn. 2 as a paper that also uses niche construction as its example. It is not argued against. LIT-110’s own “deeper reading” note, calling it “the main pro-reciprocal-causation paper that Hazelwood argues against”, repeats the error.
• The distinction Hazelwood draws bears on LIT-110 in a way the dossier’s check (“whether a process ontology of the kind Baedke et al. favour escapes it”) does not capture. He separates a “weak” sense of reciprocal causation, between organism and environment as entities, from the “strong” sense RC, between the processes of selection and niche construction. He calls the weak sense “unobjectionable” (p. 597; Lewontin’s coupled equations; “I can build a fire … and the fire can warm and sustain me”). As summarised in the record, LIT-110’s unrolled model is organism–environment causation, so on Hazelwood’s terms it would fall under the unobjectionable weak sense and is not what he targets. Whether LIT-110 also asserts process-level RC needs its full text; this reading did not check that. Process ontology (Nicholson & Dupré 2018) is mentioned once, as a reading under which organism and environment would themselves be processes, “and there is no evidence to suggest this” for Laland et al. (p. 597). It is not examined.
• Dossier: “§4.1: a Kimian exclusion problem — the counterfactual dependence of selection on niche construction reflects composition, not causation”. Right in substance. The operative premise is Craver & Bechtel 2007 (interlevel relata “are not distinct events … there is no possibility of their coming to spatiotemporally intersect”), with Kim invoked by analogy. Hazelwood does not endorse Kim’s exclusion argument. In §4.3 he says “Woodward is correct in claiming that interlevel causation is completely defensible under the right circumstances” (p. 606). His objection is specifically that niche construction and selection are not “independently fixable” (Woodward 2015, 2021; fn. 13 gives Woodward’s IF definition), as with Woodward’s cholesterol example.
• Dossier: “§4.3: considers and rejects a Woodwardian interventionist defence of interlevel causation”. Incomplete. Hazelwood rejects it only for the process of selection. He grants that intervening on “is a niche constructor” can cause “differences in reproductive success” at a later time (citing Clark et al. 2020), i.e. an evolutionary outcome. He diagnoses the interventionist objection’s appeal as conflating the process of selection with its product (Millstein 2002; Stephens 2004; Brandon 2005). He concludes that niche construction “can be a bona fide cause of evolution — the product of natural selection” (p. 608). This EES-friendly result is half the paper’s conclusion.
• The dossier omits §4.2. It considers population-level niche construction (Brandon’s coal-mining example) and replies that location on the hierarchy is “immaterial”: variation in trait fitness is “synchronously realized, not caused” by niche-constructing interactions (p. 605). Fn. 14 answers a referee with a propensity reading of fitness (Mills & Beatty), using the Batesian-mimic “Heimlich the Caterpillar” case.
• The dossier omits the notion of emergence used. It is deliberately minimal (fn. 6: a higher-level process “realized, not caused” by lower-level ones). It is taken from Wilson 2016 (synchronic dependence plus ontological and causal autonomy), and autonomy is shown in each of three disjuncts: irreducibility (evolvability; individuals “do not have the capacity to evolve”), novel laws (Hardy–Weinberg, Fisher’s theorem) and multiple realizability (1:2 → 1:3 beetle ratios). Realization is argued to entail composition (Aizawa & Gillett, Wimsatt, Wilson). Against Polger it is argued that “the underlying matter matters” for selection (p. 601).
• Dossier (§5) is correct. Denying EC gives either statisticalism (Walsh, Matthen, Ariew), which rejects RC a fortiori because it denies ultimate causes, or individual-level causalism, which is incoherent because relative fitness is relational (Millstein 2006; Sober 2013). Fn. 15 marks this as a departure from Pence 2021, the source of the paper’s compositional account.
• No missing topic. The paper’s subject is when a relation is causal (counterfactual dependence, interventionism, independent fixability, exclusion). There is no causation topic, but the metaphysics blurb names “causation, emergence and levels” explicitly, so the current tags cover it.
• metaphysics tag: justified. See Connections.
• primary topic: metaphysics. The argument is a dilemma built from the metaphysics of emergence, realization and composition and from theories of causation (Kim, Craver & Bechtel, Woodward). Hazelwood frames his conclusion as protecting “the contributions philosophy has made to the metaphysics of biology” (p. 611). philosophy-of-science, the current lead, is a close second because the subject is evolutionary theory, and should stay as a tag. natural-sciences is justified as a subject tag.