Akaike’s framework for thinking about model selection in terms of the goal of predictive accuracy and his criterion for model selection have important philosophical implications. Scientists often test models whose truth values they already know, and they often decline to reject models that they know full well are false. Instrumentalism helps explain this pervasive feature of scientific practice, and Akaike’s framework helps provide instrumentalism with the epistemology it needs. Akaike’s criterion for model selection also throws light on the role of (...) parsimony considerations in hypothesis evaluation. I explain the basic ideas behind Akaike’s framework and criterion; several biological examples, including the use of maximum likelihood methods in phylogenetic inference, are considered. (shrink)

Proponents of Bayesian confirmation theory believe that they have the solution to a significant, recalcitrant problem in philosophy of science. It is the identification of the logic that governs evidence and its inductive bearing in science. That is the logic that lets us say that our catalog of planetary observations strongly confirms Copernicus’ heliocentric hypothesis; or that the fossil record is good evidence for the theory of evolution; or that the 3oK cosmic background radiation supports big bang cosmology. The definitive (...) solution to this problem would be a significant achievement. The problem is of central importance to philosophy of science, for, in the end, what distinguishes science from myth making is that we have good evidence for the content of science, or at least of mature sciences, whereas myths are evidentially ungrounded fictions. The core ideas shared by all versions of Bayesian confirmation theory are, at a good first approximation, that a scientist’s beliefs are or should conform to a probability measure; and that the incorporation of new evidence is through conditionalization using Bayes’ theorem. While the burden of this chapter will be to inventory why critics believe this theory may not be the solution after all, it is worthwhile first to summarize here the most appealing virtues of this simple account. There are three. First, the theory reduces the often nebulous notion of a logic of.. (shrink)

Despite the widespread use of key concepts of the Neyman–Pearson (N–P) statistical paradigm—type I and II errors, significance levels, power, confidence levels—they have been the subject of philosophical controversy and debate for over 60 years. Both current and long-standing problems of N–P tests stem from unclarity and confusion, even among N–P adherents, as to how a test's (pre-data) error probabilities are to be used for (post-data) inductive inference as opposed to inductive behavior. We argue that the relevance of error probabilities (...) is to ensure that only statistical hypotheses that have passed severe or probative tests are inferred from the data. The severity criterion supplies a meta-statistical principle for evaluating proposed statistical inferences, avoiding classic fallacies from tests that are overly sensitive, as well as those not sensitive enough to particular errors and discrepancies. Introduction and overview 1.1 Behavioristic and inferential rationales for Neyman–Pearson (N–P) tests 1.2 Severity rationale: induction as severe testing 1.3 Severity as a meta-statistical concept: three required restrictions on the N–P paradigm Error statistical tests from the severity perspective 2.1 N–P test T(): type I, II error probabilities and power 2.2 Specifying test T() using p-values Neyman's post-data use of power 3.1 Neyman: does failure to reject H warrant confirming H? Severe testing as a basic concept for an adequate post-data inference 4.1 The severity interpretation of acceptance (SIA) for test T() 4.2 The fallacy of acceptance (i.e., an insignificant difference): Ms Rosy 4.3 Severity and power Fallacy of rejection: statistical vs. substantive significance 5.1 Taking a rejection of H0 as evidence for a substantive claim or theory 5.2 A statistically significant difference from H0 may fail to indicate a substantively important magnitude 5.3 Principle for the severity interpretation of a rejection (SIR) 5.4 Comparing significant results with different sample sizes in T(): large n problem 5.5 General testing rules for T(), using the severe testing concept The severe testing concept and confidence intervals 6.1 Dualities between one and two-sided intervals and tests 6.2 Avoiding shortcomings of confidence intervals Beyond the N–P paradigm: pure significance, and misspecification tests Concluding comments: have we shown severity to be a basic concept in a N–P philosophy of induction? (shrink)

Russell (1948), Reichenbach (1956), and Salmon (1975, 1979) have argued that a fundamental principle of science and common sense is that "matching" events should not be chalked up to coincidence, but should be explained by postulating a common cause. Reichenbach and Salmon provided this intuitive idea with a probabilistic formulation, which Salmon used to argue for a version of scientific realism. Van Fraassen (1980, 1982) showed that the principle, so construed, runs afoul of certain results in quantum mechanics. In this (...) paper a new formulation of the principle is offered that emerges from its use in evolutionary theory. This characterization identifies fairly general conditions in which postulating common causes will be more explanatory than postulating separate causes. (shrink)

Arguably no concept is more fundamental to science than that of causality, for investigations into cases of existence, persistence, and change in the natural world are largely investigations into the causes of these phenomena. Yet the metaphysics and epistemology of causality remain unclear. For example, the ontological categories of the causal relata have been taken to be objects (Hume 1739), events (Davidson 1967), properties (Armstrong 1978), processes (Salmon 1984), variables (Hitchcock 1993), and facts (Mellor 1995). (For convenience, causes and effects (...) will usually be understood as events in what follows.) Complicating matters, causal relations may be singular (Socrates’s drinking hemlock caused Socrates’s death) or general (Drinking hemlock causes death); hence the relata might be tokens (e.g., instances of properties) or types (e.g., types of events) of the category in question. Other questions up for grabs are: Are singular causes metaphysically and/or epistemologically prior to general causes or vice versa (or neither)? What grounds the intuitive asymmetry of the causal relation? Are macro-causal relations reducible to micro-causal relations? And perhaps most importantly: Are causal facts (e.g., the holding of causal relations) reducible to non-causal facts (e.g., the holding of certain spatiotemporal relations)? (shrink)

Hacking’s Law of Likelihood says—paraphrasing—that data support hypothesis H1 over hypothesis H2 whenever the likelihood ratio for H1 over H2 exceeds 1. But Hacking later noted a seemingly fatal flaw in the LR itself: it cannot be interpreted as the degree of “evidential significance” across applications. I agree with Hacking about the problem, but I do not believe the condition is incurable. I argue here that the LR can be properly calibrated with respect to the underlying evidence, and I sketch (...) the rudiments of a methodology for so doing. (shrink)

Bayesianism and likelihoodism are two of the most important frameworks philosophers of science use to analyse scientific methodology. However, both frameworks face a serious objection: much scientific inquiry takes place in highly idealized frameworks where all the hypotheses are known to be false. Yet, both Bayesianism and likelihoodism seem to be based on the assumption that the goal of scientific inquiry is always truth rather than closeness to the truth. Here, I argue in favor of a verisimilitude framework for inductive (...) inference. In the verisimilitude framework, scientific inquiry is conceived of, in part, as a process where inference methods ought to be calibrated to appropriate measures of closeness to the truth. To illustrate the verisimilitude framework, I offer a reconstruction of parsimony evaluations of scientific theories, and I give a reconstruction and extended analysis of the use of parsimony inference in phylogenetics. By recasting phylogenetic inference in the verisimilitude framework, it becomes possible to both raise and address objections to phylogenetic methods that rely on parsimony. (shrink)

Bayesianism and likelihoodism are two of the most important frameworks philosophers of science use to analyse scientific methodology. However, both frameworks face a serious objection: much scientific inquiry takes place in highly idealized frameworks where all the hypotheses are known to be false. Yet, both Bayesianism and likelihoodism seem to be based on the assumption that the goal of scientific inquiry is always truth rather than closeness to the truth. Here, I argue in favour of a verisimilitude framework for inductive (...) inference. In the verisimilitude framework, scientific inquiry is conceived of, in part, as a process where inference methods ought to be calibrated to appropriate measures of closeness to the truth. To illustrate the verisimilitude framework, I offer a reconstruction of parsimony evaluations of scientific theories, and I give a reconstruction and extended analysis of the use of parsimony inference in phylogenetics. By recasting phylogenetic inference in the verisimilitude framework, it becomes possible to both raise and address objections to phylogenetic methods that rely on parsimony. 1Introduction2Problems with the Law of Likelihood3Introducing Verisimilitude-Based Inference4Examples of Verisimilitude-Based Inference Procedures4.1Parsimony inference over theories4.2Parsimony inference in phylogenetics5Conclusion. (shrink)

Jun Otsuka’s excellent book, Thinking about Statistics - the Philosophical Foundations (Otsuka 2023) is mostly organized around the idea that different statistical approaches can be illuminated by linking them to different ideas in general epistemology. Otsuka connects Bayesianism to internalism and foundationalism, frequentism to reliabilism, and the Akaike Information Criterion in model selection theory to instrumentalism. This useful mapping doesn’t cover all the interesting ideas he presents. His discussions of causal inference and machine learning are philosophically insightful, as is his (...) idea that statisticians embrace an assumption that is similar to Hume’s Principle of the Uniformity of Nature. I discuss these topics in what follows, sometimes disagreeing with details while at other times adding ideas that complement those presented in the book. (shrink)

Maximum Parsimony (MP) and Maximum Likelihood (ML) are two methods for evaluating which phlogenetic tree is best supported by data on the characteristics of leaf objects (which may be species, populations, or individual organisms). MP has been criticized for assuming that evolution proceeds parsimoniously -- that if a lineage begins in state i and ends in state j, the way it got from i to j is by the smallest number of changes. MP has been criticized for needing to assume (...) some model or other of the evolutionary process even though biologists often do not know what that process really is. This paper critically evaluates both criticisms. (shrink)

This paper is a sympathetic critique of the argument that Reichenbach develops in Chap. 2 of Experience and Prediction for the thesis that sense experience justifies belief in the existence of an external world. After discussing his attack on the positivist theory of meaning, I describe the probability ideas that Reichenbach presents. I argue that Reichenbach begins with an argument grounded in the Law of Likelihood but that he then endorses a different argument that involves prior probabilities. I try to (...) show how this second step in Reichenbach's approach can be strengthened by using ideas that have been developed recently for understanding causation in terms of the idea of intervention. (shrink)

A common view among statisticians is that convergence (which statisticians call consistency) is a necessary property of an inference rule or estimator. In this paper, this view is challenged by appeal to an example in which a rule of inference has a likelihood rationale but is not convergent. The example helps clarify the significance of the likelihood concept in statistical inference.

‘There are known knowns; there are things we know we know. We also know there are known unknowns; that is to say we know there are some things we do not know. But there are also unknown unknowns—the ones we don’t know we don’t know.’ —Donald Rumsfeld, 2003, President George W. Bush’s Secretary of Defense, on the subject of the U.S. government’s failure to discover weapons of mass destruction in Iraq.

Akaike's framework for thinking about model selection in terms of the goal of predictive accuracy and his criterion for model selection have important philosophical implications. Scientists often test models whose truth values they already know, and they often decline to reject models that they know full well are false. Instrumentalism helps explain this pervasive feature of scientific practice, and Akaike's framework helps provide instrumentalism with the epistemology it needs. Akaike's criterion for model selection also throws light on the role of (...) parsimony considerations in hypothesis evaluation. I explain the basic ideas behind Akaike's framework and criterion; several biological examples, including the use of maximum likelihood methods in phylogenetic inference, are considered. (shrink)

I discuss two versions of the doomsday argument. According to ``Gott's Line'',the fact that the human race has existed for 200,000 years licences the predictionthat it will last between 5100 and 7.8 million more years. According to ``Leslie'sWedge'', the fact that I currently exist is evidence that increases the plausibilityof the hypothesis that the human race will come to an end sooner rather than later.Both arguments rest on substantive assumptions about the sampling process thatunderlies our observations. These sampling assumptions have (...) testable consequences,and so the sampling assumptions themselves must be regarded as empirical claims.The result of testing some of these consequences is that both doomsday argumentsare empirically disconfirmed. (shrink)

“Absence of evidence isn’t evidence of absence” is a slogan that is popular among scientists and nonscientists alike. This article assesses its truth by using a probabilistic tool, the Law of Likelihood. Qualitative questions (“Is E evidence about H ?”) and quantitative questions (“How much evidence does E provide about H ?”) are both considered. The article discusses the example of fossil intermediates. If finding a fossil that is phenotypically intermediate between two extant species provides evidence that those species have (...) a common ancestor, does failing to find such a fossil constitute evidence that there was no common ancestor? Or should the failure merely be chalked up to the imperfection of the fossil record? The transitivity of the evidence relation in simple causal chains provides a broader context, which leads to discussion of the fine-tuning argument, the anthropic principle, and observation selection effects. (shrink)

Collins (The Blackwell companion to natural theology, 2009) presents an argument he calls the ‘core fine-tuning argument’. In this paper, I show that Collins’ argument is flawed in at least two ways. First, the structure, depending on likelihoods, fails to establish anything about the posterior probability of God’s existence given fine-tuning. As an argument for God’s existence, this is a serious failing. Second, his analysis of what is appropriately restricted background knowledge, combined with the credences of a specially chosen ‘alien’, (...) do not allow him to establish the premise \( \Pr (LPU \mid NSU~ \& ~k') \ll 1\). (shrink)

An inductive logic is a system of inference that describes the relation between propositions on data, and propositions that extend beyond the data, such as predictions over future data, and general conclusions on all possible data. Statistics, on the other hand, is a mathematical discipline that describes procedures for deriving results about a population from sample data. These results include predictions on future samples, decisions on rejecting or accepting a hypothesis about the population, the determination of probability assignments over such (...) hypotheses, the selection of a statistical model for studying the population, and so on. Both inductive logic and statistics are calculi for getting from the given data to propositions or results that transcend the data. (shrink)

According to the probabilistic relevance account of confirmation, E confirms H relative to background knowledge K just in case P(H/K&E) > P(H/K). This requires an inequality between the rational degree of belief in H determined relative to two bodies of total knowledge which are such that one (K&E) includes the other (K) as a proper part. In this paper, I argue that it is quite plausible that there are no two possible bodies of total knowledge for ideally rational agents meeting (...) this requirement. Hence, the positive relevance account may have to be rejected. (shrink)

This paper describes a cubic water tank equipped with a movable partition receiving various amounts of liquid used to represent joint probability distributions. This device is applied to the investigation of deductive inferences under uncertainty. The analogy is exploited to determine by qualitative reasoning the limits in probability of the conclusion of twenty basic deductive arguments (such as Modus Ponens, And-introduction, Contraposition, etc.) often used as benchmark problems by the various theoretical approaches to reasoning under uncertainty. The probability bounds imposed (...) by the premises on the conclusion are derived on the basis of a few trivial principles such as "a part of the tank cannot contain more liquid than its capacity allows", or "if a part is empty, the other part contains all the liquid". This stems from the equivalence between the physical constraints imposed by the capacity of the tank and its subdivisions on the volumes of liquid, and the axioms and rules of probability. The device materializes de Finetti's coherence approach to probability. It also suggests a physical counterpart of Dutch book arguments to assess individuals' rationality in probability judgments in the sense that individuals whose degrees of belief in a conclusion are out of the bounds of coherence intervals would commit themselves to executing physically impossible tasks. (shrink)

Philosophers of science have often favoured reductive approaches to how-possibly explanation. This article identifies three alternative conceptions making how-possibly explanation an interesting phenomenon in its own right. The first variety approaches “how possibly X?” by showing that X is not epistemically impossible. This can sometimes be achieved by removing misunderstandings concerning the implications of one’s current belief system but involves characteristically a modification of this belief system so that acceptance of X does not result in contradiction. The second variety offers (...) a potential how-explanation of X. It is usually followed by a range of further potential how-explanations of the same phenomenon. In recent literature the factual claims implied by the second variety have been downplayed whereas the heuristic role of mapping the space of conceptual possibilities has been emphasized. I will focus especially on this truth-bracketing sense of potentiality when looking closer at the second variety in the paper. The third variety has attracted less interest. It presents a partial how-explanation of X. Typically it aims to establish the existence of a mechanism by which X could be and was generated. The third conception stands out as the natural alternative for the advocate of ontic how-possibly explanations. This article transfers Salmon’s view that explanation-concepts can be broadly divided into epistemic, modal, and ontic to the context of how-possibly explanations. Moreover, it is argued that each of the three above-mentioned varieties of how-possibly explanation occurs in science. To recognize this may be especially relevant for philosophers. We are often misled by the promises of various why-explanation accounts, and seem to have forgotten nearly everything about the diversity of how-possibly explanations. (shrink)

Plausibly, when we adopt a probabilistic standpoint any measure Cb of the degree to which evidence e confirms hypothesis h relative to background knowledge b should meet these five desiderata: Cb > 0 when P > P < 0 when P < P; Cb = 0 when P = P. Cb is some function of the values P and P assume on the at most sixteen truth-functional combinations of e and h. If P < P and P = P then (...) Cb ≤ Cb; if P = P and P < P then Cb ≥ Cb. Cb – Cb is fully determined by Cb and Cbe – Cbe; if Cb = 0 then Cb + Cbe = 0. If P = P then Cb = Cb. (shrink)

Evolutionary gradualism, the randomness of mutations, and the hypothesis that natural selection exerts a pervasive and substantial influence on evolutionary outcomes are pair-wise logically independent. Can the claims about selection and mutation be used to formulate an argument for gradualism? In his Genetical Theory of Natural Selection, R.A. Fisher made an important start at this project in his famous “geometric argument” by showing that a random mutation that has a smaller effect on two or more phenotypes will have a higher (...) average fitness than a random mutation that has a larger phenotypic effect. Motoo Kimura’s demonstration that a gene’s probability of fixation depends on both the selection coefficient and the effective population size shows that Fisher’s argument for gradualism was mistaken. Here we analyze Fisher’s argument and explain how Kimura’s theory leads to a conclusion that Fisher did not anticipate. We identify a fallacy that reasoning about fitness differences and their evolutionary consequences should avoid. We then distinguish forward-directed from backward-directed versions of gradualism. The backward-directed thesis about a single mutation may be correct, but the forward-directed thesis is not. After that we consider a sequence of random mutations that all affect the same cluster of phenotypes and Allen Orr’s idea that there is an optimal sequence of mutation sizes, moving from larger to smaller as the population approaches a fixed optimum. This provides a likelihood justification for a forward-directed version of gradualism when there is a fixed optimum, but it also provides an argument against forward-directed gradualism if the optimum shifts substantially as the population evolves. Finally, we consider whether genome-wide association studies (GWASs) can furnish empirical evidence that bears on the truth of gradualism. The version of gradualism that GWASs directly bear on concerns the phenotypic effect size of a gene after it arises by mutation and has reached an appreciable frequency, not the phenotypic size of the mutation event itself. (shrink)

In this paper, I show how one might resist two influential arguments for the Likelihood Principle by appealing to the ontological significance of creative intentions. The first argument for the Likelihood Principle that I consider is the argument from intentions. After clarifying the argument, I show how the key premiss in the argument may be resisted by maintaining that creative intentions sometimes independently matter to what experiments exist. The second argument that I consider is Gandenberger’s :475–503, 2015) rehabilitation of Birnbaum’s (...) :269–306, 1962) proof of the Likelihood Principle from the more intuitively obvious principles of conditionality and sufficiency. As with the argument from intentions, I show how Gandenberger’s argument for his Experimental Conditionality Principle may be resisted by maintaining that creative intentions sometimes independently matter to what experiments exist. (shrink)

It is probable that, increasingly, genome investigations are going to be based on statistical formalization. This review summarizes the state of art and potentiality of using statistics in microbial genome analysis. First, I focus on recent advances in functional genomics, such as finding genes and operons, identifying gene conversion events, detecting DNA replication origins and analysing regulatory sites. Then I describe how to use phylogenetic methods in genome analysis and methods for genome‐wide scanning for positively selected amino acids. I conclude (...) with speculations on the future course of genome statistical modeling. BioEssays 25:266–273, 2003. © 2003 Wiley Periodicals, Inc. (shrink)

Deduction leads to causal explanation in phylogenetic inference when the evidence, the systematic character, is conceptualized as a transformation series. Also, the deductive entailment of modus tollens is satisfied when those kinds of events are operationalized as patristic difference. Arguments to the contrary are based largely on the premise that character-states are defined intensionally as objects, in terms of similarity relations. However, such relations leave biologists without epistemological access to the causal explanation and explanatory power of historical statements. Moreover, the (...) prediction-making to which those kinds of relations are limited in practice can lead to a category error—the mental conversion of an abstraction into a thing . The latter practices and problems characterize pattern cladistics, taxa being interpreted as homeostatic property cluster natural kinds, and other instrumentalist research programs. (shrink)