Abstract
Robustness analysis is widespread in science, but philosophers have struggled to justify its confirmatory power. We provide a positive account of robustness by analysing some explicit and implicit uses of within and across-model robustness in evolutionary theory. We argue that appeals to robustness are usually difficult to justify because they aim to increase the likeliness that a phenomenon obtains. However, we show that robust results are necessary (under certain conditions) for explanations of phenomena with specific properties. Across-model robustness is necessary for how-possibly explanations of multiply instantiated phenomena, while within-model robustness is necessary for explanations of phenomena with multiple evolutionary starting points. In such cases, the appeal of robustness is explanatory rather than confirmatory.
Similar content being viewed by others
Notes
McElreath and Boyd (2007) also mention one of the benefits of modelling approaches as providing ‘existence proofs’ (p. 6).
This idea has been recently defended by Ylikoski and Aydinonat (2014), according to whom ‘the main point of theoretical how-possibly explanations is to refine, systematize, and expand the menu of available explanations’ (31). They particularly develop and emphasise this latter point.
We are grateful to [redacted] for raising this issue.
Bokulich’s account has been attacked for being too liberal; for instance, Saatsi and Pexton (2013) doubt “that capturing a mere abstract structure ‘up to isomorphism’ of counterfactual relations can be explanatory” (223–224), if such counterfactual relations cannot be causally interpreted. This need not concern us, as the explanatory targets of evolutionary models are fundamentally causal—although not necessarily governed by factors on which one could intervene.
Brian Skyrms’ (2010) analysis of signalling systems contains other examples of this kind of reasoning.
As noted by an anonymous referee, this is not to say that basin comparisons cannot be unproblematic in some cases, for instance when we compare basin sizes in two models that accurately represent two different systems.
When the models differ E should not be considered as the same evidence for both, as it stands for “the behaviour of interest was obtained in some runs of the simulation”; strictly speaking we should write E1 and E2. However, this modification would only strengthen our point that the direct comparison between P(E|M1) and P(E|M2) is not necessarily insightful.
One may further object that even if we accept that we can attribute probabilities to models, we could never assess these probabilities precisely. This does not matter here, as our discussion does not hinge on the precise values of the P(Mi) but on their equality.
However, authors such as Sugden (2000) would disagree: he sees models as ‘credible fictions’, where credibility is not to be expressed in terms of a probability of being true – models are credible in the way fictions are.
Two possible objections to the use of Bayesian formalism are that P(E) cannot be determined, as it involves considering an infinite number of models. However, only finitely many models can ever be considered. Moreover, the precise value of P(E) does not impinge on our discussion, provided it is not 0.
We thank [redacted] for recommending the use of this Bayesian perspective.
Specifically, if an equilibrium is a global, asymptotically stable attractor then the population reaches that equilibrium eventually. Alexander (2007) relies on similar robustness claims for spatially structured populations.
This might for instance be true of models in which preferential interaction is not based on conditional strategies but on other sources of correlation such as spatially structured interactions or limited dispersal (not included in Skyrms’ aforementioned models).
Since the model uses simulations to estimate basin sizes we cannot, in this case be sure that such a requirement is met.
In a similar spirit, D’Arms et al.’s (1998) criticism of some earlier work of Skyrms can be read as mitigating conclusions drawn from considerating basins by emphasising the greater prior plausibility of alternative modelling assumptions.
Skyrms at least has claimed that within-model robustness has explanatory value, although for unstated reasons: “The fact that the signalling system equilibrium is a global attractor […] and remains so if the replicator dynamics is replaced by any qualitatively adaptive dynamics, strikes me as explanatorily powerful.” (2000, 110).
Note that in the case of physical robustness, basin comparison approaches could be defended; our criticisms of “The basin comparison problem” section only bear on such approaches in the context of within-model robustness.
Odenbaugh and Alexandrova (2011) also highlight a heuristic (rather than confirmatory) role of robustness analysis: it helps us identify acceptable candidate hypotheses. For Forber (2010), this role has indirect confirmatory consequences, as it affects the list of competing hypotheses later to be confirmed or disconfirmed.
Again, this is not to say that models always need to represent the robustness of properties, if only because phenomena of interest may be robust with respect to myriad of parameters not represented in the model.
Note that although Sugden (Ibid., 21–23) takes robustness as warranting model-to-model inferences, our defence argues for a model-to-world inference, in which a property of the target system should be similar to the robust model itself.
Although the mechanisms of social learning and of natural selection differ, they share similar principles: a trait’s fixation respectively depends on its effect on fitness or its consequences for rewards (see Benaim et al. 2004). Moreover, models and their basins have similar importance for both, as Skyrms’ analysis renders manifest. Accordingly, we consider our focus on evolutionary models to include social learning.
However, they do refer to general mechanisms and conditions not limited to signalling behaviour, such as imitative learning or freedom of interaction. These may constitute necessary conditions for a signalling model, but not its common core, because following Weisberg’s condition (2), the presence of such a common core would have to entail the evolution of signalling (to lead to a robust theorem), rather than being able to produce it. Imitative learning and freedom interaction are too general and high level to constitute a common core.
Still, this is not impossible, as pointed out by [redacted].
Note that if the phenomenon really is multiply instantiated, then we should only be satisfied with how-possibly explanations that are different enough, that is, that do not only aim to represent different causal scenarios but different causal mechanisms—a distinction found in Ylikoski and Aydinonat (2014). Where two causal scenarios are just alternative causal histories, which may disagree regarding only one fact, causal mechanisms are more general and involve types of causal links (and so may constitute ‘skeletons of causal processes’ (27) upon which detailed causal scenarios can be built).
Also note that it does not draw on Bokulich’s account of explanation.
Our positive take applies to robustness within one model and with respect to very different models, but probably not to robustness with respect to tractability or modelling assumptions, which have no clear real-world counterparts.
If robustness analysis à la Weisberg aims to identify the common causal core responsible for a target phenomenon that occurs in several (possibly similar) contexts, then the absence of homogenous across-model robustness would actually provide reasons to look for heterogeneous across-model robustness.
References
Alexander JM (2007) The structural evolution of morality. Cambridge University Press, Cambridge
Barrett J (2007) The evolution of coding in signaling games. Theor Decis 67:223–237
Barrett JA (2014) The evolution, appropriation, and composition of rules. Synthese. doi: 10.1007/s11229-014-0421-6
Benaim M, Schreiber SJ, Tarres P (2004) Generalized Urn models of evolutionary processes. Ann Appl Probab 14:1455–1478
Binmore K (2006) Why do people cooperate? Politics Philos Econ 5(1):81–96
Bokulich A (2011) How scientific models can explain. Synthese 180:33–45
Bourke A (2011) Principles of social evolution. Oxford University Press, Oxford
Brandon R (1990) Adaptation and environment. Princeton University Press, Princeton
Calcott B (2008) The other cooperation problem: generating benefit. Biol Philos 23:179–203
Calcott B (2011) Wimsatt and the robustness family: review of Wimsatt’s Re-engineering philosophy for limited beings. Biol Philos 26(2):281–293
Craver C (2006) When mechanistic models explain. Synthese 153:355–376
D’Arms J, Batterman R, Górny K (1998) Game theoretic explanations and the evolution of justice. Philos Sci 65:76–102
Diggle SP, Griffin AS, Campbell GS, West SA (2007) Cooperation and conflict in quorum-sensing bacterial populations. Nature 450:411–414
Dray W (1957) Law and explanation in history. Oxford University Press, Oxford
Dunbar R (1993) Coevolution of neocortical size, group size and language in humans. Behav Brain Sci 16:681–735
Elgin M, Sober E (2002) Cartwright on explanation and idealization. Erkenntnis 57:441–450
Forber P (2010) Confirmation and explaining how possible. Stud Hist Philos Biol Biomed Sci 41:32–40
Forber P, Smead R (2014) The evolution of fairness through spite. Proc R Soc Lond B 281:20132439
Frean M, Abraham ER (2001) ‘Rock–scissors–paper and the survival of the weakest. Proc R Soc Lond B 268(1474):1323–1327
Gibbard A, Varian H (1978) Economic models. J Philos 75:664–677
Gintis H (2011) Gene–culture coevolution and the nature of human sociality. Philos Trans R Soc B 366:878–888
Gore J, Youk H, van Oudenaarden A (2009) Snowdrift game dynamics and facultative cheating in yeast. Nature 459:253–256
Houkes W, Vaesen K (2012) Robust! Handle with care. Philos Sci 79:1–20
Huttegger S (2007) Robustness in signaling games. Philos Sci 74:839–847
Huttegger S, Zollman K (2010) Dynamic stability and basins of attraction in the Sir Philip Sidney game. Proc R Soc Lond B 277:1925–1932
Huttegger S, Zollman K (2013) Methodology in biological game theory. Br J Philos Sci 64:637–658
Huttegger S, Skyrms B, Tarrès Pierre, Wagner E (2014) Some dynamics of signalling games. Proc Natl Acad Sci 111(Supplement 3):10873–10880
Johnston MD, Edwards CM, Bodmer WF, Maini PK, Chapman SJ (2012) Examples of mathematical modelling: tales from the crypt. Cell Cycle 6(17):2106–2112
Jonsen ID, Flemming JM, Myers RA (2005) Robust state-space modeling of animal movement data. Ecology 86(11):2874–2880
Justus J (2010) Complexity, diversity and stability. In: Sarkar S, Plutynski A (eds) A companion to the philosophy of biology, vol 90. Wiley, New York, pp 321–350
Justus J (2012) The elusive basis of inferential robustness. Philos Sci 79(5):795–807
Kuorikoski J, Lehtinen A, Marchionni C (2009) Economic modelling as robustness analysis. Br J Philos Sci 61(3):541–567
Levins R (1966) The strategy of model building in population biology. Am Sci 54(4):421–431
Lloyd E (2010) Confirmation and robustness of climate models. Philos Sci 77:971–984
Mäki U (1994) Isolation, idealization and truth in economics. Poznan Stud Philos Sci Humanit 38:147–168
Maynard Smith J (1982) Evolution and the theory of games. Cambridge University Press, Cambridge
McCloskey D (1983) The rhetoric of economics. J Econ Lit 21:481–517
McElreath R, Boyd R (2007) Mathematical models for social evolution: a guide for the perplexed. University of Chicago Press, Chicago
McMullin E (1985) Galilean idealization. Stud Hist Philos Sci 16:247–273
Odenbaugh J, Alexandrova A (2011) Buyer beware: robustness analyses in economics and biology. Biol Philos 26:757–771
Orzack SH, Sober E (1993) A critical assessment of Levins’s the strategy of model building in population biology (1966). Q Rev Biol 68(4):533–546
Popat R, Crusz SA, Messina M, Williams P, West SA, Diggle P (2012) Quorum-sensing and cheating in bacterial biolfiilms. Proc R Soc Lond B 279:4765–4771
Raerinne J (2013) Robustness and sensitivity of biological models. Philos Stud 166:285–303
Saatsi J, Pexton M (2013) Reassessing Woodward’s account of explanation: regularities, counterfactuals, and noncausal explanations. Philos Sci 80(5):613–624
Sanchez A, Gore J (2013) Feedback between population and evolutionary dynamics determines the fate of social microbial populations. PLoS Biol 11(4):e1001547
Sinervo B, Lively CM (1996) The rock-paper-scissors game and the evolution of alternate male strategies. Nature 380:240–243
Skyrms B (1996) Evolution of the social contract. Cambridge University Press, Cambridge
Skyrms B (2010) Signals. Cambridge University Press, Cambridge
Smead R (2010) Indirect reciprocity and the evolution of moral signals. Biol Philos 25:33–51
Sugden R (2000) Credible worlds: the status of theoretical models in economics. J Econ Methodol 7(1):1–31
Wagner EO (2012) Deterministic chaos and the evolution of meaning. Br J Philos Sci 63(3):547–575
Weisberg M (2006) Robustness analysis. Philos Sci 73:730–742
Weisberg M, Reisman K (2008) The robust Volterra principle. Philos Sci 75:106–131
West SA, Griffin AS, Gardner A (2007) Social semantics: altruism, cooperation, mutualism, strong reciprocity and group selection. J Evol Biol 20:415–432
Woodcock S, Heath J (2002) The robustness of altruism as an evolutionary strategy. Biol Philos 17:567–590
Woodward J (2003) Making things happen: a theory of causal explanation. Oxford University Press, Oxford
Woodward J (2006) Some varieties of robustness. J Econ Methodol 13:219–240
Acknowledgments
We thank Conor Mayo-Wilson, Seamus Bradley, Catherine Herfeld and Aidan Lyon for comments on previous versions of this paper, as well as the audiences of the Philosophy of Biology in the UK (2012) and of the BSPS 2013 conferences. The remarks of several anonymous referees also proved extremely valuable. Cédric Paternotte was supported by the Alexander von Humboldt Foundation.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Paternotte, C., Grose, J. Robustness in evolutionary explanations: a positive account. Biol Philos 32, 73–96 (2017). https://doi.org/10.1007/s10539-016-9539-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10539-016-9539-x