The stochastic grouping effect: analytical dynamics, theoretical classification, and scope of random group division as a support for cooperation

A single-author bioRxiv preprint (Steiner, 2021) proposes the 'stochastic grouping effect': dividing a population into random groups in which a benefit contributed by cooperators is shared equally among all group members gives cooperators an innate statistical advantage — sufficient, in the preprint's agent-based simulations, to eliminate defectors, to maintain stable cooperator-majority equilibria despite a cooperator mortality cost, and to sustain a persistent 'turbulent' fluctuating regime — without assortment, synergy, group competition, reciprocity, or kin structure. The source leaves the effect theoretically unfinished: it does not classify the mechanism within evolutionary theory (group selection without inter-group competition, kin selection with sampling-induced relatedness, or a drift-like phenomenon), derives no analytical expressions for equilibrium frequencies or stability, maps no systematic boundaries between its three simulated dynamical regimes (stable equilibrium, turbulent fluctuations, population collapse), and leaves unmodeled the coevolution of group size, claiming only in a single passing sentence without details that sexual reproduction and graduated trait values behave alike. This problem asks for an analytical theory of the stochastic grouping model — its classification, equilibrium and regime-boundary structure, and its scope. The preprint has no published journal version and the questions below are not settled for this model by the framework literature cited in Background, although framework-level classification tools from multilevel-selection theory pre-date the source and bear directly on the classification component, which the source itself poses as an open question.

Background

The evolution of cooperation asks how behaviors that benefit others at a cost to the actor persist when defectors, who share the benefits while paying no cost, are favored within every mixed group. Classical answers require an extra structuring ingredient: genetic relatedness between actor and beneficiary (kin selection), repeated interactions with memory or reputation (direct and indirect reciprocity), spatial population structure (network reciprocity), or selection between groups (group selection) (Nowak 2012; Wilson & Wilson 2008). The group-selection route has a long, contested history. In the 'trait group' model of Wilson (1975), a population is subdivided each generation into temporarily isolated groups whose members interact only with each other before their offspring disperse and remix; in the 'haystack' model of Maynard Smith (1964), cooperation can persist inside colonies founded by few individuals even though it is selected against on average. Maynard Smith (1976), and in a widely cited numerical example Maynard Smith and Szathmáry in The Major Transitions in Evolution, argued that when groups form at random and benefits are shared equally, the frequency of altruists decreases — a critique that strongly colored the reception of 'pure' group selection.

Against this background, the preprint of Steiner (2021) proposes a 'stochastic grouping effect'. In a population of cooperators (each contributing a benefit $b=1$ to its group) and defectors (contributing nothing), with the total group benefit divided equally among all group members, the total benefit accruing to cooperators under a random division into groups is claimed never to be smaller — and that accruing to defectors never to be larger — than in the same population left undivided, with the advantage growing as groups get smaller. In the preprint's agent-based implementation, a fixed number of groups (16 in the reported runs) is treated as territories with a fixed maximum capacity and is initially populated by randomly assigned cooperators and defectors. Each generation, every member of a group holding $c$ cooperators among $g$ members survives with probability $c/g$, survival being random with respect to trait (so an all-defector group is eliminated and an all-cooperator group survives intact); the survivors then face trait-specific intrinsic mortality that is independent of the group (for example 25% per generation for cooperators versus 20% for defectors); and each surviving individual produces $N$ asexual offspring ($N=1$ in the reported runs), each offspring switching trait with mutation probability $m$ ($m=0.02$ in the reported runs). Offspring first fill their natal group up to its capacity and then recolonize empty or partially filled groups; once all groups are full, the excess offspring are eliminated, so that total population size is endogenous up to the number of groups times the capacity. Some runs add random migration between groups at a given rate, and the maximum group size is varied across runs (5 in the base runs, 20 and 50 in the mutation-and-mortality runs, and swept in version 2). The reported simulations show defector extinction within about 10 generations without migration; full cooperator occupancy within about 20 generations with migration; a stable equilibrium at almost 90% cooperators in version 2 (around 80% in version 1) at maximum group size 20 despite a 5% mortality disadvantage; and, at maximum group size 50 with mortality 25% versus 15%, a persistent regime labelled '"turbulent" (chaotic?)' in the source, with strong fluctuations of trait frequencies, population size and number of empty groups over 600 generations without collapse (the fluctuation range is legible only in the source's figure), while under more extreme mortality differences the whole population quickly collapses and dies out. The preprint reads this as an innate, mechanism-free support for cooperation (no non-random assortment, synergistic fitness effects, group competition, reciprocity, kin selection, or spatial selection) and re-analyzes the Maynard Smith–Szathmáry numerical example (10 cooperators, 15 defectors; the group division used is shown only in the source's figure) as yielding 13 cooperators and 12 defectors.

The interpretive machinery needed to situate this claim is standard. The Price equation (Price 1970, 1972) decomposes trait-frequency change into a covariance between group-mean trait value and group-mean fitness (the between-group, or group-selection, component) plus an expectation of within-group change; 'group selection' in this broad sense requires only that groups differ in productivity, not direct competition or extinction of groups. Damuth and Heisler (1988) distinguish multilevel selection of type 1 (MLS1), where groups are fitness-determining contexts for the individuals that are the real units of selection, from multilevel selection of type 2 (MLS2), where groups themselves reproduce; Wade (1985) connected soft selection — where population regulation acts within groups rather than through differential productivity of groups — to group and kin selection; and Queller (1992), Lehmann et al. (2007), and Marshall (2011) established that group-selection and kin-selection accounting are formally equivalent, random formation of finite groups inducing a positive relatedness between groupmates even without genealogical kinship, one that strengthens as groups shrink. The same phenomenon family — random compartmentalization maintaining 'cooperative' replicators — appears in prebiotic 'stochastic corrector' models, where random sampling of small packages of early replicators preserves favorable replicator mixtures (Szathmáry and Demeter 1987), and in transient-compartment models, where finite groups repeatedly form by bottleneck sampling, grow in isolation, and merge (Arenzon and Peliti 2023); stochasticity-driven pathways to cooperation in finite populations are analyzed by Fumagalli and Rice (2019); group-structured populations resolving the tragedy of the commons are treated by Killingback et al. (2006); and coevolution of cooperation and group size by Lerch and Abbott (2020). Whether the stochastic grouping effect is a genuinely distinct mechanism or an instance of these classical frameworks — and what its analytical dynamics actually are — is what the source leaves open.

Problem Statement

For the stochastic grouping model of Steiner (2021) — a fixed number of territory-like groups with a fixed maximum capacity, initially populated by randomly assigned cooperators and defectors; per-generation survival with probability equal to the within-group cooperator proportion $c/g$, followed by trait-specific intrinsic mortality; $N$ asexual offspring per survivor with trait-switching mutation probability $m$; fill-then-recolonize placement of offspring with elimination of the excess once all groups are full, so that population size is endogenous; and, in some runs, random migration between groups at a given rate, all as specified in Background — the source leaves a connected set of unresolved questions whose common objective is an analytical theory of the stochastic grouping effect, its classification, and its scope.

(1) Classification. Determine, within the modern multilevel-selection framework (Price covariance partition, MLS1/MLS2 distinction, group-selection/kin-selection equivalence), what the stochastic grouping effect is: group selection without inter-group competition, kin selection with sampling-induced relatedness, a drift-like phenomenon, or something genuinely distinct. The source states that the effect is not classical group selection because there is no competition between groups and that it is an emergent property of subdivision, and — in version 1 — asks whether it is 'also a kind of genetic drift, since grouping causes a random change in the gene frequencies'; it explicitly leaves this classification open ('Questions that I leave open here').

(2) Analytical dynamics. Derive analytical expressions for the equilibrium cooperator frequencies and their stability conditions, and systematically map the boundaries in parameter space (maximum group size, number of groups, cooperator-defector mortality differential, mutation rate, random-migration rate) between the three regimes that the source demonstrates only through example simulations and one-dimensional sweeps: stable cooperator-majority equilibrium, turbulent fluctuating coexistence, and population collapse. Here 'equilibrium cooperator frequency' is understood in the source's sense, as the settled frequency of the stochastic process (a stationary or quasi-stationary mean measured after an initial settling period), and 'population collapse' as extinction of the whole population within a bounded time horizon. This includes settling the source's own '(chaotic?)' question: whether the turbulent regime is a genuine chaotic regime — in which case the deterministic limit or map in which chaos is meant must be specified — or a stochastic quasi-stationary fluctuation regime.

(3) Scope. Establish how far the effect extends: the coevolutionary consequences of group size itself (the tendency acknowledged but explicitly not modeled in the source, in which cooperators benefit from smaller groups and defectors from larger ones while pressures from the advantages and disadvantages of group size ('group effects') counteract), and its generality beyond asexual binary traits (sexual reproduction and graduated trait values), for which the source claims comparable equilibria in a single passing sentence without presenting details.

A complete answer addresses all three parts. It establishes whether random group division alone is, by the standards of modern social-evolution theory, a distinct and analytically predictable support for cooperation, states the conditions under which each of the demonstrated dynamical regimes occurs, and establishes the scope of the effect in the sense of part (3). Because 'support for cooperation' carries two non-equivalent operational readings in the record above — the preprint's absolute-benefit reading (the total benefit accruing to cooperators never smaller, and that accruing to defectors never larger, than in the same population left undivided) and the frequency reading at issue in the Maynard Smith–Szathmáry critique (the change in the frequency of altruists), which can hold simultaneously — the verdict is to be stated relative to the reading or readings it establishes.

References

1. Steiner, Klaus F. (2021). The Good, the Bad and the Stochastic: How Living in Groups Innately Supports Cooperation. bioRxiv preprint (v1 2021-02-23, v2 2021-05-19; no published journal version). DOI: 10.1101/2021.02.21.431661. https://doi.org/10.1101/2021.02.21.431661
2. Wilson, D. S. (1975). A theory of group selection. Proceedings of the National Academy of Sciences 72(1):143-146. DOI: 10.1073/pnas.72.1.143. https://doi.org/10.1073/pnas.72.1.143
3. Maynard Smith, J. (1964). Group selection and kin selection. Nature 201(4924):1145-1147. DOI: 10.1038/2011145a0. https://doi.org/10.1038/2011145a0
4. Maynard Smith, J. (1976). Group selection [review essay of M. E. Gilpin, Group Selection in Predator-Prey Communities]. The Quarterly Review of Biology 51(2):277-283. DOI: 10.1086/409311. https://doi.org/10.1086/409311
5. Maynard Smith, John; Szathmáry, Eörs (1997). The Major Transitions in Evolution. Oxford University Press. DOI: 10.1093/oso/9780198502944.001.0001. https://doi.org/10.1093/oso/9780198502944.001.0001
6. Price, George R. (1970). Selection and covariance. Nature 227(5257):520-521. DOI: 10.1038/227520a0. https://doi.org/10.1038/227520a0
7. Price, George R. (1972). Extension of covariance selection mathematics. Annals of Human Genetics 35(4):485-490. DOI: 10.1111/j.1469-1809.1957.tb01874.x. https://doi.org/10.1111/j.1469-1809.1957.tb01874.x
8. Wade, Michael J. (1985). Soft selection, hard selection, kin selection, and group selection. The American Naturalist 125(1):61-73. DOI: 10.1086/284328. https://doi.org/10.1086/284328
9. Queller, David C. (1992). Quantitative genetics, inclusive fitness, and group selection. The American Naturalist 139(3):540-558. DOI: 10.1086/285343. https://doi.org/10.1086/285343
10. Damuth, John; Heisler, I. Lorraine (1988). Alternative formulations of multilevel selection. Biology & Philosophy 3(4):407-430. DOI: 10.1007/bf00647962. https://doi.org/10.1007/bf00647962
11. Lehmann, Laurent; Keller, Laurent; West, Stuart; Roze, Denis (2007). Group selection and kin selection: two concepts but one process. Proceedings of the National Academy of Sciences 104(16):6736-6739. DOI: 10.1073/pnas.0700662104. https://doi.org/10.1073/pnas.0700662104
12. Marshall, James A. R. (2011). Group selection and kin selection: formally equivalent approaches. Trends in Ecology & Evolution 26(7):325-332. DOI: 10.1016/j.tree.2011.04.008. https://doi.org/10.1016/j.tree.2011.04.008
13. Traulsen, Arne; Nowak, Martin A. (2006). Evolution of cooperation by multilevel selection. Proceedings of the National Academy of Sciences 103:10952-10955. DOI: 10.1073/pnas.0602530103. https://doi.org/10.1073/pnas.0602530103
14. Nowak, Martin A. (2012). Evolving cooperation. Journal of Theoretical Biology 299:1-8. DOI: 10.1016/j.jtbi.2012.01.014. https://doi.org/10.1016/j.jtbi.2012.01.014
15. Wilson, David Sloan; Wilson, Edward O. (2008). Evolution 'for the good of the group'. American Scientist 96:380. DOI: 10.1511/2008.74.1. https://doi.org/10.1511/2008.74.1
16. Szathmáry, Eörs; Demeter, László (1987). Group selection of early replicators and the origin of life. Journal of Theoretical Biology 128(4):463-486. DOI: 10.1016/s0022-5193(87)80191-1. https://doi.org/10.1016/s0022-5193(87)80191-1
17. Fumagalli, Sarah E.; Rice, Sean H. (2019). Stochasticity and non-additivity expose hidden evolutionary pathways to cooperation. PLOS ONE 14:e0225517. DOI: 10.1371/journal.pone.0225517. https://doi.org/10.1371/journal.pone.0225517
18. Arenzon, Jeferson J.; Peliti, Luca (2023). Emergent cooperative behavior in transient compartments. Physical Review E 108(3):034409. DOI: 10.1103/physreve.108.034409. https://doi.org/10.1103/physreve.108.034409
19. Killingback, Timothy; Bieri, Jonas; Flatt, Thomas (2006). Evolution in group-structured populations can resolve the tragedy of the commons. Proceedings of the Royal Society B: Biological Sciences 273(1593):1477-1481. DOI: 10.1098/rspb.2006.3476. https://doi.org/10.1098/rspb.2006.3476
20. Lerch, Brian A.; Abbott, Karen C. (2020). Allee effects drive the coevolution of cooperation and group size in high reproductive skew groups. Behavioral Ecology 31(3):661-671. DOI: 10.1093/beheco/araa009. https://doi.org/10.1093/beheco/araa009