Shapiro and Sober claim that Walsh, Ariew, Lewens, and Matthen give a mistaken, a priori defense of natural selection and drift as epiphenomenal. Contrary to Shapiro and Sober’s claims, we first argue that WALM’s explanatory doctrine does not require a defense of epiphenomenalism. We then defend WALM’s explanatory doctrine by arguing that the explanations provided by the modern genetical theory of natural selection are ‘autonomous-statistical explanations’ analogous to Galton’s explanation of reversion to mediocrity and an explanation of (...) the diffusion ofgases. We then argue that whereas Sober’s theory of forces is an adequate description of Darwin’s theory, WALM’s explanatory doctrine is required to understand how themodern genetical theory of natural selection explains large-scale statistical regularities. 1 Introduction2 Shapiro and Sober’s ‘Epiphenomenalism Do’s and Don’ts’3 WALM’s Explanatory Doctrine4 Galton’s Autonomous-Statistical Explanation5 A Second Example: The Statistical Explanation of the Diffusion of Gases6 Distinguishing Two Theories of Evolution by Natural Selection7 A Possible Objection: Are Statistical Laws Sufficient for Explanation?8 Conclusion. (shrink)

Through his S–R model of statistical relevance, Wesley Salmon offers a solution to the scientific explanation of objectively improbable events.

Really statistical explanation is a hitherto neglected form of noncausal scientific explanation. Explanations in population biology that appeal to drift are RS explanations. An RS explanation supplies a kind of understanding that a causal explanation of the same result cannot supply. Roughly speaking, an RS explanation shows the result to be mere statistical fallout.

Some philosphers of science of an empiricist and pragmatist bent have proposed models of statistical explanation, but have then become sceptical of the adequacy of these models. It is argued that general considerations concerning the purpose of function of explanation in science which are usually appealed to by such philosophers show that their scepticism is not well taken; for such considerations provide much the same rationale for the search for statistical explanations, as these philosophers have characterized (...) them, as they do for lawlike explanations. But, it is further argued, a significant piece of what is frequently offered as an explanation of well known phenomena in statistical mechanics, fails to meet this general "pragmatic rationale" for statistical, or indeed any kind of, explanation. The question then arises whether the physicists have misconstrued the value of this piece of physical theorizing, ergodic theory, taking it to be explanatory when it is actually not; or whether, instead, the philosopher's account of just what is genuinely explanatory is too narrow. (shrink)

Probabilistic phenomena are often perceived as being problematic targets for contrastive explanation. It is usually thought that the possibility of contrastive explanation hinges on whether or not the probabilistic behaviour is irreducibly indeterministic, and that the possible remaining contrastive explananda are token event probabilities or complete probability distributions over such token outcomes. This paper uses the invariance-under-interventions account of contrastive explanation to argue against both ideas. First, the problem of contrastive explanation also arises in cases in (...) which the probabilistic behaviour of the explanandum is due to unobserved causal heterogeneity. Second, it turns out that, in contrast to the case of pure indeterminism, the plausible contrastive explananda under causal heterogeneity are not token event probabilities, but population-level statistical facts. (shrink)

The purpose of this paper is to provide a systematic appraisal of the covering law and statistical relevance theories of statistical explanation advanced by Carl G. Hempel and by Wesley C. Salmon, respectively. The analysis is intended to show that the difference between these accounts is inprinciple analogous to the distinction between truth and confirmation, where Hempel's analysis applies to what is taken to be the case and Salmon's analysis applies to what is the case. Specifically, it (...) is argued (a) that statistical explanations exhibit the nomic expectability of their explanandum events, which in some cases may be strong but in other cases will not be; (b) that the statistical relevance criterion is more fundamental than the requirement of maximal specificity and should therefore displace it; and, (c) that if statistical explanations are to be envisioned as inductive arguments at all, then only in a qualified sense since, in particular, the requirement of high inductive probability between explanans and explanandum must be abandoned. (shrink)

The statistical aspects of quantum explanation are intrinsic to quantum physics; individual quantum events are created in the interactions associated with observation and are not describable by predictive theory. The superposition principle shows the essential difference between quantum and non-quantum physics, and the principle is exemplified in the classic single-photon two-slit interference experiment. Recently Mandel and Pfleegor have done an experiment somewhat similar to the optical single-photon experiment but with two independently operated lasers; interference is obtained even with (...) beam intensity so small that only one photon is in the apparatus at a time. The result can be understood in terms of the superposition of states; or, in terms of the Uncertainty Principle, which is found to forbid the determination of which of the two lasers is the source of a given photon (if conditions for interference are to obtain). The Mandel-Pfleegor experiment gives a physical argument against the continuous localization of a photon that is assumed in the hidden variable theories and therefore gives further support for the generally accepted view that an observed entity (observed state) is created in the observation event. This aspect of quantum physics implies a subjectivism on the level of individual quantum-level occurrences, since there is in quantum theory no basis for asserting the existence of the event independently of observation of it. Extension of this subjectivism to large scale, non-quantum phenomena falls within the principles of quantum theory; counter considerations that argue against such an extension are noted. (shrink)

Wesley Salmon has advanced a new model of explanations of particular facts which requires that the explanans contain laws. The laws used in explanations (according to this model) are of the form P(A· C1,B)=p1... P(A· Cn,B)=pn. A condition imposed by Salmon on these laws is that the reference classes, i.e. A· C1... A· Cn, be homogenous with reference to the property B. A reference class A is homogenous with reference to a property B if every property which determines a place (...) selection with reference to B within A is statistically irrelevant to B in A. It is argued here that the concept of homogeneity cannot in general be satisfied in scientific explanations and that even a weaker requirement, "epistemic homogeneity," may be too strong. (shrink)

An explanation of the Dunning–Kruger effect is provided which does not require any psychological explanation, because it is derived as a statistical artifact. This is achieved by specifying a simple statistical model which explicitly takes the boundary constraints into account. The model fits the data almost perfectly.JEL ClassificationA22; C24; C91; D84; D91; I21.

The intuitive notion of a statistical explanation has been explicated in different ways; recently it has even been claimed that there are no statistical explanations at all. In an attempt to clarify the disputed issue, the approaches adopted by Hempel, by Jeffrey, Salmon and Greeno, and by Stegmuller are analyzed critically, as far as they are concerned with the explanation of particular events. A solution of the controversy is proposed on the basis of a concept of (...) explanation which refers essentially to a causal analysis of the explanandum. The possibility of statistical explanations, then, becomes contingent upon the existence of indeterministic causation. In conclusion, therefore, a conception of causality is sketched which shows that indeterminism and causal connection are compatible, at least from an epistemological point of view, so that statistical explanation can be seen to represent a specific and possibly irreducible scientific activity. (shrink)

Really statistical explanation is a hitherto neglected form of noncausal scientific explanation. Explanations in population biology that appeal to drift are RS explanations. An RS explanation supplies a kind of understanding that a causal explanation of the same result cannot supply. Roughly speaking, an RS explanation shows the result to be mere statistical fallout.

Generally speaking, scientific explanation has been a topic of lively discussion in twentieth‐century philosophy of science; philosophers of science have endeavored to characterize rigorously a number of different types of explanation to be found in the various fields of scientific research. Given the indispensability of statistical concepts and techniques in virtually every branch of modern science, it is natural to ask whether some scientific explanations are essentially statistical or probabilistic in character. The answer would seem to (...) be yes. For example, we can explain why atoms of carbon 14 have a 1/4 probability of surviving for 11,460 years because the half‐life of that species is 5,730 years. As we shall see, explanations of this type are not especially problematic. As another example, we might explain why a particular weed withered by citing the fact that it received a dose of a herbicide, even though we know that the herbicide is not invariably effective. This means that the withering is related probabilistically to the herbicide treatment but is not necessitated by it. Explanations of this kind, by contrast, lead to severe difficulties. (shrink)

According to Lange,?Really Statistical explanations? constitute an important kind of non-causalscientific explanation. However, Roski has argued that all alleged RS explanations are either causalexplanations or not explanations at all. In so arguing, Roski has invoked Kahneman?s interpretation of onealleged RS explanation. I employ Roski?s arguments as an opportunity to elaborate and defend RS explanations. Iargue that?RS explanations? genuinely explain rather than deny the presuppositions of why-questions. I argue thatthe RS model is not excessively permissive in allowing some (...) explanations to work purely statistically rather than bydescribing causal relations. I argue that Roski?s view that some?RS explanations? operate by describing causalrelations fails to capture the kind of explanatory insight that RS explanations provide. I elaborate the notion of a?characteristically statistical phenomenon? that figures in the RS model and thereby explain how an RS explanationreveals that its explanandum is independent of any causal facts or specific chances, but rather depends only on ageneric sort of arrangement of chances. Finally, I argue that Roski misinterprets Kahneman, who actually holds thatthe explanation he discusses is a non-causal explanation that nicely fits the RS model. RS explanations constitutean important kind of non-causal scientific explanation. (shrink)

Standard models of statistical explanation face two intractable difficulties. In his 1984 Salmon argues that because statistical explanations are essentially probabilistic we can make sense of statistical explanation only by rejecting the intuition that scientific explanations are contrastive. Further, frequently the point of a statistical explanation is to identify the etiology of its explanandum, but on standard models probabilistic explanations often fail to do so. This paper offers an alternative conception of statistical (...) explanations on which explanations of the frequency of a property consist in the derivation of that frequency from a statistical specification of the mechanism by which instances of the relevant property are produced. Such explanations are contrastive precisely because they identify the determinate causal etiologies of their explananda. (shrink)

The most compelling extant accounts of explanation casts all explanations as causal. Yet there are sciences, theoretical population biology in particular, that explain their phenomena by appeal to statistical, non-causal properties of ensembles. I develop a generalised account of explanation. An explanation serves two functions: metaphysical and cognitive. The metaphysical function is discharged by identifying a counterfactually robust invariance relation between explanans event and explanandum. The cognitive function is discharged by providing an appropriate description of this (...) relation. I offer examples of explanations from portfolio theory and population genetics that meet this characterisation. In each case the invariance relation holds between a statistical property of an ensemble and a change in structure of the ensemble. In neither case, however, does the statistical property cause the outcome it explains. There are genuine statistical, non-causal scientific explanations. (shrink)

In this note, I discuss the simplicity of rival statistical explanations of a correlation, couched in terms of Reichenbachian Common Cause Systems. Simplicity is analyzed in two components, the so-called intrinsic and contextual simplicity. I show that if one disentangles simplicity from explanatory power then the size of the system provides an adequate for simplicity in both of its dimensions.

There is a widespread philosophical interpretation of natural selection in evolutionary theory: natural selection, like mutation, migration, and drift are seen as forces that propel the evolution of populations. Natural selection is thus a population level causal process. This account has been challenged by the Statistics, claiming that natural selection is not a population level cause but rather a statistical feature of a population. This paper examines the nature of the aforementioned ontological debate and the nature of statistical (...) explanations given by population genetics. I claim that the Modern Synthesis provides good explanations of the changes in trait structure of populations without appealing to detailed causal information about the individual trajectories of the members of a population. (shrink)

A consideration of some basic problems that arise in the attempt to provide an adequate characterization of statistical explanation is taken to show that an understanding of the nature of scientific explanation requires us to deal with the philosophical problems connected with the nature of prior probabilities.

Causal modeling methods such as path analysis, used in the social and natural sciences, are also highly relevant to philosophical problems of probabilistic causation and statistical explanation. We show how these methods can be effectively used (1) to improve and extend Salmon's S-R basis for statistical explanation, and (2) to repair Cartwright's resolution of Simpson's paradox, clarifying the relationship between statistical and causal claims.

An argument has been recently proposed by Watkins, whose objective is to show the impossibility of a statistical explanation of single events. This present paper is an attempt to show that Watkins's argument is unsuccessful, and goes on to argue for an account of statistical explanation which has much in common with Hempel's classic treatment.

Carl G. Hempel has often stated that inductive-statistical explanations, as he conceives them, are inductive arguments. This discussion note raises the question of whether such arguments are to be understood as (1) arguments of the traditional sort, containing premises and conclusions, governed by some sort of inductive "acceptance rules," or (2) something more closely akin to Carnap's degree of confirmation statements which occur in an inductive logic which entirely eschews inductive "acceptance rules." Hempel's writings do not seem unequivocal on (...) this issue. It is suggested that adoption of construal (2) would remove the need for Hempel's high probability requirement on I-S explanations. (shrink)

The book explores several open questions in the philosophy of statistical mechanics. Each chapter is written by a leading expert in the field. Here is a list of some questions that are addressed in the book: 1) Boltzmann showed how the phenomenological gas laws of thermodynamics can be derived from statistical mechanics. Since classical mechanics is a deterministic theory there are no probabilities in it. Since statistical mechanics is based on classical mechanics, all the probabilities statistical (...) mechanics talks about cannot be fundamental. However, if probabilities are epistemic, how can they play a role, as they seem to do, in laws, explanation, and prediction? 2) Many physicists use the notion of typicality instead of the one of probability when discussing statistical mechanics. What is the connection between the two notions? 3) How can one extend Boltzmann’s analysis to the quantum domain, where some theories are indeterministic? 4) Boltzmann’s explanation fundamentally involves cosmology: for the explanation to go through the Big Bang needs to have had extremely low entropy. Does the fact that the Big Bang was a low entropy state imply that it was, in some sense, “highly improbable” and requires an explanation? 5) What exactly is the connection between statistical and classical mechanics? Is the one of theory reduction or there is no such thing? 6) Statistical mechanics has two main formulation: one due to Botzmann and the other due to Gibbs. What is the connection between the two formulations . (shrink)

(1988). Explanation, subjunctives and statistical theories. International Studies in the Philosophy of Science: Vol. 3, No. 1, pp. 80-96. doi: 10.1080/02698598808573326.

The phenomenon of regression toward the mean is notoriously liable to be overlooked or misunderstood; regression fallacies are easy to commit. But even when regression phenomena are duly recognized, it remains perplexing how they can feature in explanations. This article develops a philosophical account of regression explanations as “statistically autonomous” explanations that cannot be deepened by adducing details about causal histories, even if the explananda as such are embedded in the causal structure of the world. That regression explanations have (...) class='Hi'>statistical autonomy was first suggested by Ian Hacking and has recently been defended and elaborated by André Ariew, Yasha Rohwer, and Collin Rice. However, I will argue that these analyses fail to capture what regression’s statistical autonomy consists in and how it sets regression explanations apart from other kinds of explanation. The alternative account I develop also shows what is amiss with a recent denial of regression’s statistical autonomy. Marc Lange has argued that facts that can be explained as regression phenomena can in principle also be explained by citing a conjunction of causal histories. The account of regression explanation developed here shows that his argument is based on a misunderstanding of the nature of statistical autonomy. (shrink)

This paper considers the problem of causal explanation in classical and statistical thermodynamics. It is argued that the irreversibility of macroscopic processes is explained in both formulations of thermodynamics in a teleological way that appeals to entropic or probabilistic consequences rather than to efficient-causal, antecedental conditions. This explanatory structure of thermodynamics is not taken to imply a teleological orientation to macroscopic processes themselves, but to reflect simply the epistemological limitations of this science, wherein consequences of heat-work asymmetries are (...) either macroscopically measurable (entropy) or calculable (probabilities), while efficient-causal relationships are obscure or indeterminable. (shrink)

A common way of characterizing Boltzmann’s explanation of thermodynamics in term of statistical mechanics is with reference to three ingredients: the dynamics, the past hypothesis, and the statistical postulate. In this paper I focus on the statistical postulate, and I have three aims. First, I wish to argue that regarding the statistical postulate as a probability postulate may be too strong: a postulate about typicality would be enough. Second, I wish to show that there is (...) no need to postulate anything, for the typicality postulate can be suitably derived from the dynamics. Finally, I discuss how the attempts to give preference to certain stochastic quantum theories (such as the spontaneous collapse theory) over deterministic alternatives on the basis that they do not need the statistical postulate fail. (shrink)

For the case of statistical theories, the criteria of explanation, prediction, and testability can all be viewed as particular instances of a more general evaluation scheme. Using the ideas of a gain matrix and expected gain from statistical decision theory, these three criteria can be compared in terms of the elements in their associated gain matrices. This analysis leads to (1) further understanding of the interrelationship between the current criteria, (2) the proposal of an ordering for the (...) criteria, and (3) the suggestion of a new criterion. (shrink)

A subarea of the debate over the nature of evolutionary theory addresses what the nature of the explanations yielded by evolutionary theory are. The statisticalist line is that the general principles of evolutionary theory are not only amenable to a mathematical interpretation but that they need not invoke causes to furnish explanations. Causalists object that construction of these general principles involves crucial causal assumptions. A recent view claims that some biological explanations are statistically autonomous explanations (SAEs) whereby phenomena are accounted (...) for statistically and which prescind from micro-causal details. I raise three major problems for this account and then advance a view which unifies SAEs as mathematical explanations: the MSAE view. The MSAE view not only resolves the issues bedeviling the original SAE account but serves to importantly broaden the class of non-causal explanations in population biology. (shrink)

A hitherto neglected form of explanation is explored, especially its role in population genetics. “Statistically abstractive explanation” (SA explanation) mandates the suppression of factors probabilistically relevant to an explanandum when these factors are extraneous to the theoretical project being pursued. When these factors are suppressed, the explanandum is rendered uncertain. But this uncertainty traces to the theoretically constrained character of SA explanation, not to any real indeterminacy. Random genetic drift is an artifact of such uncertainty, and (...) it is therefore wrong to reify it as a cause of evolution or as a process in its own right. *Received July 2009. †To contact the author, please write to: Department of Philosophy, University of Toronto, 170 St. George St., Toronto, ON M5R 2M8, Canada; e‐mail: [email protected]. (shrink)

The concept of classical indistinguishability is analyzed and defended against a number of well-known criticisms, with particular attention to the Gibbs’paradox. Granted that it is as much at home in classical as in quantum statistical mechanics, the question arises as to why indistinguishability, in quantum mechanics but not in classical mechanics, forces a change in statistics. The answer, illustrated with simple examples, is that the equilibrium measure on classical phase space is continuous, whilst on Hilbert space it is discrete. (...) The relevance of names, or equivalently, properties stable in time that can be used as names, is also discussed. (shrink)

This paper argues that a notion of statistical explanation, based on Salmon’s statistical relevance model, can help us better understand deep neural networks. It is proved that homogeneous partitions, the core notion of Salmon’s model, are equivalent to minimal sufficient statistics, an important notion from statistical inference. This establishes a link to deep neural networks via the so-called Information Bottleneck method, an information-theoretic framework, according to which deep neural networks implicitly solve an optimization problem that generalizes (...) minimal sufficient statistics. The resulting notion of statistical explanation is general, mathematical, and subcausal. (shrink)