Michael Strevens

The dissertation concerns the use of physical probability in higher level scientific theories such as statistical mechanics and evolutionary biology. My focus is complex systems--systems containing large numbers of parts that move independently yet interact strongly, such as gases and ecosystems. Although the underlying dynamics of such systems are prohibitively complex, their macrolevel behavior can often be predicted given information about physical probabilities. ;The technique has the following form: probabilities are assigned to the movements of the elements of the system, and these probabilities are then aggregated to produce a simple macrolevel dynamics. For example, knowing that members of a given class of rabbits all have a probability of one tenth of dying in a given month, it is easy to infer that about one tenth of such rabbits will die in the course of a month. Such reasoning is paramount in both sciences mentioned above. ;My aim is to say what, if the technique is to yield results, these physical probabilities must be. Two problems in particular must be solved: Why are the probabilities stable? That is, why can statistics from one ecosystem be used to infer probabilities for another ecosystem? This question seems especially puzzling given the complexity of the underlying dynamics. Why are the probabilities independent? The independence is perplexing because of the strong causal interactions between various parts of most complex systems. ;I show that the best current view about higher level physical probabilities--the frequency view--is unable to solve these problems. I provide an alternative view, and solve both problems for substantially large classes of systems. The answers to the questions come not so much from considering the composition of my probabilities but rather from an examination of their interrelations. I mean both relations between probabilities at the same level of description and relations between probabilities at different levels

What do the words "ceteris paribus" add to a causal hypothesis, that is, to a generalization that is intended to articulate the consequences of a causal mechanism? One answer, which looks almost too good to be true, is that a ceteris paribus hedge restricts the scope of the hypothesis to those cases where nothing undermines, interferes with, or undoes the effect of the mechanism in question, even if the hypothesis's own formulator is otherwise unable to specify fully what might constitute such undermining or interference. I will propose a semantics for causal generalizations according to which ceteris paribus hedges deliver on this promise, because the truth conditions for a causal generalization depend in part on the—perhaps unknown—nature of the mechanism whose consequences it is intended to describe. It follows that the truth conditions for causal hypotheses are typically opaque to the very scientists who formulate and test them.

I aim to reconcile two apparently conflicting theses: (a) Everything that can be explained, can be explained in purely physical terms, that is, using the machinery of fundamental physics, and (b) some properties that play an explanatory role in the higher level sciences are irreducible in the strong sense that they are physically undefinable: their nature cannot be described using the vocabulary of physics. I investigate the contribution that physically undefinable properties typically make to explanations in the high-level sciences, and I show that when they are explanatorily relevant, it is in virtue of their extension (or something close) alone. They are irreducible because physics cannot capture their nature; this is no obstacle, however, to physics' more or less capturing their extension, which is all that it need do to duplicate their explanatory power. In the course of the argument, I sketch the outlines of an account of the explanation of physically contingent regularities, such as the regularities found in most branches of biological inquiry, at the center of which is an account of the nature of contingent, empirical bridge principles.

Evolutionary biology distinguishes differences in survival and reproduction rates due to selection from those due to drift. The distinction is usually thought to be founded in probabilistic facts: a difference in (say) two variants' average lifespans over some period of time that is due to selection is explained by differences in the probabilities relevant to survival; in the purest cases of drift, by contrast, the survival probabilities are equal and the difference in lifespans is a matter of chance. When there is a difference in actual average lifespans there is always a difference in causal histories, but in drift, the causal differences make no contribution to the relevant probabilities. Some writers have claimed that the framework that determines which causes contribute to the probabilities, and so the framework that determines when the relevant probabilities differ and when they are equal, is imposed by modelers rather than being an objective fact of nature, and so that the distinction between selection and drift, however practically useful, is unreal. This paper uses my work on biological probabilities (Bigger than Chaos) and probabilistic explanation (Depth) to argue for the objectivity of, and the importance of the objectivity of, the distinction between selection and drift.

The three cardinal aims of science are prediction, control, and explanation; but the greatest of these is explanation. Also the most inscrutable: prediction aims at truth, and control at happiness, and insofar as we have some independent grasp of these notions, we can evaluate science’s strategies of prediction and control from the outside. Explanation, by contrast, aims at scientific understanding, a good intrinsic to science and therefore something that it seems we can only look to science itself to explicate.

The generalizations found in biology, psychology, sociology, and other high-level sciences are typically physically contingent. You might conclude that they play only a limited role in scientific investigation, on the grounds that physically contingent generalizations offer no or only feeble counterfactual support. But the link between contingency and counterfactual support is more complex than is commonly supposed. A certain class of physically contingent generalizations, comprising many, perhaps the vast majority, of those in the high-level sciences, provides strong counterfactual support of just the sort that appears to be scientifically important. This paper explains why.

Cases of overdetermination or preemption continue to play an important role in the debate about the proper interpretation of causal claims of the form "C was a cause of E". I argue that the best treatment of preemption cases is given by Mackie's venerable INUS account of causal claims. The Mackie account suffers, however, from problems of its own. Inspired by its ability to handle preemption, I propose a dramatic revision to the Mackie account – one that Mackie himself would certainly have rejected – to solve these difficulties. The result is, I contend, a very attractive account of singular causal claims.

Robert Merton observed that better-known scientists tend to get more credit than less well-known scientists for the same achievements; he called this the Matthew effect. Scientists themselves, even those eminent researchers who enjoy its benefits, regard the effect as a pathology: it results, they believe, in a misallocation of credit. If so, why do scientists continue to bestow credit in the manner described by the effect? This paper advocates an explanation of the effect on which it turns out to allocate credit fairly after all, while at the same time making sense of scientists' opinions to the contrary.

Does the Bayesian theory of confirmation put real constraints on our inductive behavior? Or is it just a framework for systematizing whatever kind of inductive behavior we prefer? Colin Howson (Hume's Problem) has recently championed the second view. I argue that he is wrong, in that the Bayesian apparatus as it is usually deployed does constrain our judgments of inductive import, but also that he is right, in that the source of Bayesianism's inductive prescriptions is not the Bayesian machinery itself, but rather what David Lewis calls the ``Principal Principle''.

This paper examines the standard Bayesian solution to the Quine–Duhem problem, the problem of distributing blame between a theory and its auxiliary hypotheses in the aftermath of a failed prediction. The standard solution, I argue, begs the question against those who claim that the problem has no solution. I then provide an alternative Bayesian solution that is not question-begging and that turns out to have some interesting and desirable properties not possessed by the standard solution. This solution opens the way to a satisfying treatment of a problem concerning ad hoc auxiliary hypotheses.

How can a model that stops short of representing the whole truth about the causal production of a phenomenon help us to understand the phenomenon? I answer this question from the perspective of what I call the simple view of understanding, on which to understand a phenomenon is to grasp a correct explanation of the phenomenon. Idealizations, I have argued in previous work, flag factors that are casually relevant but explanatorily irrelevant to the phenomena to be explained. Though useful to the would-be understander, such flagging is only a first step. Are there any further and more advanced ways that idealized models aid understanding? Yes, I propose: the manipulation of idealized models can provide considerable insight into the reasons that some causal factors are difference-makers and others are not, which helps the understander to grasp the nature of explanatory connections and so to better grasp the explanation itself

Elliott Sober argues that the statistical slogan “Absence of evidence is not evidence of absence” cannot be taken literally: it must be interpreted charitably as claiming that the absence of evidence is (typically) not very much evidence of absence. I offer an alternative interpretation, on which the slogan claims that absence of evidence is (typically) not objective evidence of absence. I sketch a definition of objective evidence, founded in the notion of an epistemically objective likelihood, and I show that in Sober’s paradigm case, the slogan can, on this understanding, be sustained.

The system for awarding credit in science—the priority rule—functions, I have proposed elsewhere, to bring about something close to a socially optimal distribution of scientists among scientific research programs. If all goes well, then, potentially fruitful new ideas will be explored, unpromising ideas will be ignored, and fashionable but oversubscribed ideas will be deprived of further resources. Against this optimistic background, the present paper investigates the ways in which things might not go so well, that is, ways in which the priority rule might fail to realize its full potential as an incentive for scientists to work on the right things. Several possible causes of herding—an outcome in which a single research program ends up with a number of researchers well in excess of the optimum—are considered.

Hall has recently argued that there are two concepts of causality, picking out two different kinds of causal relation. McGrath, and Hitchcock and Knobe, have recently argued that the facts about causality depend on what counts as a “default” or “normal” state, or even on the moral facts. In the light of these claims you might be tempted to agree with Skyrms that causal relations constitute, metaphysically speaking, an “amiable jumble”, or with Cartwright that ‘causation’, though a single word, encompasses many different kinds of things. This paper argues, drawing on the author’s recent work on explanation, that the evidence adduced in support of causal pluralism can be accommodated easily by a unified theory of causality—a theory according to which all singular causal claims concern the same fundamental causal network

Assumptions of stochastic independence are crucial to statistical models in science. Under what circumstances is it reasonable to suppose that two events are independent? When they are not causally or logically connected, so the standard story goes. But scientific models frequently treat causally dependent events as stochastically independent, raising the question whether there are kinds of causal connection that do not undermine stochastic independence. This paper provides one piece of an answer to this question, treating the simple case of two tossed coins with and without a midair collision

Complexity theory attempts to explain, at the most general possible level, the interesting behaviors of complex systems. Two such behaviors are the emergence of simple or stable high-level behavior from relatively complex low-level behavior, and the emergence of sophisticated high-level behavior from relatively simple low-level behavior; they are often found nested in the same system. Concerning the emergence of simplicity, this essay examines Herbert Simon's explanation from near-decomposability and a stochastic explanation that generalizes the approach of statistical physics. A more general notion of an abstract difference-making structure is introduced with examples, and a discussion of evolvability follows. Concerning the emergence of sophistication, this essay focuses on, first, the energetics approach associated with dissipative structures and the fourth law of thermodynamics, and second, the notion of a complex adaptive system.

Bayesian confirmation theory—abbreviated to in these notes—is the predominant approach to confirmation in late twentieth century philosophy of science. It has many critics, but no rival theory can claim anything like the same following. The popularity of the Bayesian approach is due to its flexibility, its apparently effortless handling of various technical problems, the existence of various a priori arguments for its validity, and its injection of subjective and contextual elements into the process of confirmation in just the places where critics of earlier approaches had come to think that subjectivity and sensitivity to context were necessary.

When new theoretical terms are introduced into scientific discourse, prevailing accounts imply, analytic or semantic truths come along with them, by way of either definitions or reference-fixing descriptions. But there appear to be few or no analytic truths in scientific theory, which suggests that the prevailing accounts are mistaken. This paper looks to research on the psychology of natural kind concepts to suggest a new account of the introduction of theoretical terms that avoids both definition and reference-fixing description. At the core of the account is a novel psychological process that I call introjection.

The two major modern accounts of explanation are the causal and unification accounts. My aim in this paper is to provide a kind of unification of the causal and the unification accounts, by using the central technical apparatus of the unification account to solve a central problem faced by the causal account, namely, the problem of determining which parts of a causal network are explanatorily relevant to the occurrence of an explanandum. The end product of my investigation is a causal account of explanation that has many of the advantages of the unification account.

Sciences of complex systems thrive on compositional theories – toolkits that allow the construction of models of a wide range of systems, each consisting of various parts put together in different ways. To be tractable, a compositional theory must make shrewd choices about the parts and properties that constitute its basic ontology. One such choice is to decompose a system into spatiotemporally discrete parts. Compositional theories in the high-level sciences follow this rule of thumb to a certain extent, but they also make essential use of what I call distributed ontologies: divisions of the system into entities or states of affairs each depending on sets of fundamental-level facts that to a large extent overlap. The point is developed using the example of statistical theories in which the probabilities are ontologically distributed.

The concept of causation plays a central role in many philosophical theories, and yet no account of causation has gained widespread acceptance among those who have investigated its foundations. Theories based on laws, counterfactuals, physical processes, and probabilistic dependence and independence relations (the list is by no means exhaustive) have all received detailed treatment in recent years---{}and, while no account has been entirely successful, it is generally agreed that the concept has been greatly clari{}ed by the attempts. In this magni{}cent book, Woodward aims to give a uni{}ed account of causation and causal explanation in terms of the notion of a manipulation (or intervention, terms which can be read interchangeably). Not only does he produce in my view the most illuminating and comprehensive account of causation on o{}er, his theory also opens a great many avenues for future work in the area, and has rami{}cations for many other areas of philosophy. Making Things Happen ought to be of interest not only to philosophers of causation and philosophers of science, but to any philosopher whose concerns involve assumptions about the nature of causation, laws, or explanation

According to principles of probability coordination, such as Miller's Principle or Lewis's Principal Principle, you ought to set your subjective probability for an event equal to what you take to be the objective probability of the event. For example, you should expect events with a very high probability to occur and those with a very low probability not to occur. This paper examines the grounds of such principles. It is argued that any attempt to justify a principle of probability coordination encounters the same difficulties as attempts to justify induction. As a result, no justification can be found.

Why are causal generalizations in the higher-level sciences “inexact”? That is, why do they have apparent exceptions? This paper offers one explanation: many causal generalizations cite as their antecedent—the \(F\) in \(Fs\,\, {\textit{are}}\,\, G\) —a property that is not causally relevant to the consequent, but which is rather “entangled” with a causally relevant property. Entanglement is a relation that may exist for many reasons, and that allows of exceptions. Causal generalizations that specify entangled but causally irrelevant antecedents therefore tolerate exceptions