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)

Mathematical modeling is a convenient way for characterization of complex ecosystems. This approach was applied to study the dynamics of zooplankton in Lake Sevan (Armenia) at different stages of anthropogenic eutrophication with the use of a novel method called discrete modeling of dynamical systems with feedback (DMDS). Simulation demonstrated that the application of this method helps in characterization of inter- and intra-component relationships in a natural ecosystem. This method describes all possible pairwise inter-component relationships like “plus–plus,” “minus–minus,” “plus–minus,” “plus–zero,” “minus–zero,” (...) and “zero–zero” that occur in most ecosystems. Based on the results, a working hypothesis was formulated. It was found that the sensitivity to weak external influence in zooplanktons was the greatest during the mid period of eutrophication in Lake Sevan, whereas in the final stages of eutrophication, an outbreak in the biomass production of cyanobacteria was evident. To support this approach, a weak external disturbance in the form of magnetic storm was used to see its effect on species Daphnia longispina sevanica. A statistically significant correlation between the frequency of magnetic storms and the number of this species was revealed and an increase in the number of toxic cyanobacteria species as a consequence of eutrophication. This paper, for the first time, suggests a DMDS method, to diagnose impact of anthropogenic eutrophication on environment. (shrink)

This paper has two main parts. In the first part, we motivate a kind of indeterminate, suppositional credences by discussing the prospect for a subjective interpretation of a causal Bayesian network, an important tool for causal reasoning in artificial intelligence. A CBN consists of a causal graph and a collection of interventional probabilities. The subjective interpretation in question would take the causal graph in a CBN to represent the causal structure that is believed by an agent, and interventional probabilities in (...) a CBN to represent suppositional credences. We review a difficulty noted in the literature with such an interpretation, and suggest that a natural way to address the challenge is to go for a generalization of CBN that allows indeterminate credences. In the second part, we develop a decision-theoretic foundation for such indeterminate suppositional credences, by generalizing a theory of coherent choice functions to accommodate some form of act-state dependence. The upshot is a decision-theoretic framework that is not only rich enough to, so to speak, ground the probabilities in a subjectively interpreted causal network, but also interesting in its own right, in that it accommodates both act-state dependence and imprecise probabilities. (shrink)

A few purported counterexamples to the Adams thesis have cropped up in the literature in the last few decades. I propose a theory that accounts for them, in a way that makes the connections between indicative conditionals and counterfactuals clearer.

A main message from the causal modelling literature in the last several decades is that under some plausible assumptions, there can be statistically consistent procedures for inferring (features of) the causal structure of a set of random variables from observational data. But whether we can control the error probabilities with a finite sample size depends on the kind of consistency the procedures can achieve. It has been shown that in general, under the standard causal Markov and Faithfulness assumptions, the procedures (...) can only be pointwise but not uniformly consistent without substantial background knowledge. This implies the impossibility of choosing a finite sample size to control the worst case error probabilities. In this paper, I consider the simpler task of inferring causal directions when the skeleton of the causal structure is known, and establish a similarly negative result concerning the possibility of controlling error probabilities. Although the result is negative in form, it has an interesting positive implication for causal discovery methods. (shrink)

Much of the recent work on the epistemology of causation has centered on two assumptions, known as the Causal Markov Condition and the Causal Faithfulness Condition. Philosophical discussions of the latter condition have exhibited situations in which it is likely to fail. This paper studies the Causal Faithfulness Condition as a conjunction of weaker conditions. We show that some of the weaker conjuncts can be empirically tested, and hence do not have to be assumed a priori. Our results lead to (...) two methodologically significant observations: (1) some common types of counterexamples to the Faithfulness condition constitute objections only to the empirically testable part of the condition; and (2) some common defenses of the Faithfulness condition do not provide justification or evidence for the testable parts of the condition. It is thus worthwhile to study the possibility of reliable causal inference under weaker Faithfulness conditions. As it turns out, the modification needed to make standard procedures work under a weaker version of the Faithfulness condition also has the practical effect of making them more robust when the standard Faithfulness condition actually holds. This, we argue, is related to the possibility of controlling error probabilities with finite sample size (“uniform consistency”) in causal inference. (shrink)

We examine a formal semantics for counterfactual conditionals due to Judea Pearl, which formalizes the interventionist interpretation of counterfactuals central to the interventionist accounts of causation and explanation. We show that a characteristic principle validated by Pearl’s semantics, known as the principle of reversibility, states a kind of irreversibility: counterfactual dependence (in David Lewis’s sense) between two distinct events is irreversible. Moreover, we show that Pearl’s semantics rules out only mutual counterfactual dependence, not cyclic dependence in general. This, we argue, (...) suggests that Pearl’s logic is either too weak or too strong. (shrink)

In the artificial intelligence literature a promising approach to counterfactual reasoning is to interpret counterfactual conditionals based on causal models. Different logics of such causal counterfactuals have been developed with respect to different classes of causal models. In this paper I characterize the class of causal models that are Lewisian in the sense that they validate the principles in Lewis’s well-known logic of counterfactuals. I then develop a system sound and complete with respect to this class. The resulting logic is (...) the weakest logic of causal counterfactuals that respects Lewis’s principles, sits in between the logic developed by Galles and Pearl and the logic developed by Halpern, and stands to Galles and Pearl’s logic in the same fashion as Lewis’s stands to Stalnaker’s. (shrink)

Models developed using machine learning are increasingly prevalent in scientific research. At the same time, these models are notoriously opaque. Explainable AI aims to mitigate the impact of opacity by rendering opaque models transparent. More than being just the solution to a problem, however, Explainable AI can also play an invaluable role in scientific exploration. This paper describes how post-hoc analytic techniques from Explainable AI can be used to refine target phenomena in medical science, to identify starting points for future (...) investigations of causal relationships, and to generate possible explanations of target phenomena in cognitive science. In this way, this paper describes how Explainable AI—over and above machine learning itself—contributes to the efficiency and scope of data-driven scientific research. (shrink)

Comparisons of rival explanations or theories often involve vague appeals to explanatory power. In this paper, we dissect this metaphor by distinguishing between different dimensions of the goodness of an explanation: non-sensitivity, cognitive salience, precision, factual accuracy and degree of integration. These dimensions are partially independent and often come into conflict. Our main contribution is to go beyond simple stipulation or description by explicating why these factors are taken to be explanatory virtues. We accomplish this by using the contrastive-counterfactual approach (...) to explanation and the view of understanding as an inferential ability. By combining these perspectives, we show how the explanatory power of an explanation in a given dimension can be assessed by showing the range of answers it provides to what-if-things-had-been-different questions and the theoretical and pragmatic importance of these questions. Our account also explains intuitions linking explanation to unification or to exhibition of a mechanism. (shrink)

The attributable fraction (population) has attracted much attention from a theoretical perspective and has been used extensively to assess the impact of potential health interventions. However, despite its extensive use, there is much confusion about its concept and calculation methods. In this article, we discuss the concepts of and calculation methods for the attributable fraction and related measures in the counterfactual framework, both with and without stratification by covariates. Generally, the attributable fraction is useful when the exposure of interest has (...) a causal effect on the outcome. However, it is important to understand that this statement applies to the exposed group. Although the target population of the attributable fraction (population) is the total population, the causal effect should be present not in the total population but in the exposed group. As related measures, we discuss the preventable fraction and prevented fraction, which are generally useful when the exposure of interest has a preventive effect on the outcome, and we further propose a new measure called the attributed fraction. We also discuss the causal and preventive excess fractions, and provide notes on vaccine efficacy. Finally, we discuss the relations between the aforementioned six measures and six possible patterns using a conceptual schema. (shrink)

As many philosophers agree, the frame problem is concerned with how an agent may efficiently filter out irrelevant information in the process of problem-solving. Hence, how to solve this problem hinges on how to properly handle semantic relevance in cognitive modeling, which is an area of cognitive science that deals with simulating human's cognitive processes in a computerized model. By "semantic relevance", we mean certain inferential relations among acquired beliefs which may facilitate information retrieval and practical reasoning under certain epistemic (...) constraints, e. g., the insufficiency of knowledge, the limitation of time budget, etc. However, traditional approaches to relevance—as for example, relevance logic, the Bayesian approach, as well as Description Logic—have failed to do justice to the foregoing constraints, and in this sense, they are not proper tools for solving the frame problem/relevance problem. As we will argue in this paper, Non-Axiomatic Reasoning System (NARS) can handle the frame problem in a more proper manner, because the resulting solution seriously takes epistemic constraints on cognition as a fundamental theoretical principle. (shrink)

Judging an act’s causal efficacy plays a crucial role in causal decision theory. A recent development appeals to the causal modeling framework with an emphasis on the analysis of intervention based on the causal Bayes net for clarifying what causally depends on our acts. However, few writers have focused on exploring the usefulness of extending structural causal models to decision problems that are not ideal for intervention analysis. The thesis concludes that structural models provide a more general framework for rational (...) decision makers. (shrink)

One way social scientists explain phenomena is by building structural models. These models are explanatory insofar as they manage to perform a recursive decomposition on an initial multivariate probability distribution, which can be interpreted as a mechanism. Explanations in social sciences share important aspects that have been highlighted in the mechanisms literature. Notably, spelling out the functioning the mechanism gives it explanatory power. Thus social scientists should choose the variables to include in the model on the basis of their function (...) in the mechanism. This paper examines the notion of ‘function’ within structural modelling. We argue that ‘functions’ ought to be understood as the theoretical underpinnings of the causes, namely as the role that causes play in the functioning of the mechanism. (shrink)

Numerous papers have investigated the transitivity principle of ‘better-than.’ A recent argument appeals to the principle of mereological dominance for transitivity. However, writers have not treated mereological dominance in much detail. This paper sets out to evaluate the generality of mereological dominance and its effectiveness in supporting the transitivity principle. I found that the mereological dominance principle is vulnerable to a counterexample based on Simpson’s Paradox. The thesis concludes that the mereological dominance principle should be revised in certain ways.

In this paper we give a positive answer to a problem posed by Hofer-Szabó and Rédei (Int. J. Theor. Phys. 43:1819–1826, 2004) regarding the existence of infinite Reichenbachian common cause systems (RCCSs). An example of a countably infinite RCCS is presented. It is also determined that no RCCSs of greater cardinality exist.

It is surely obvious that medicine, like any other rational activity, must be based on evidence. The interest is in the details: how exactly are the general principles of the logic of evidence to be applied in medicine? Focussing on the development, and current claims of the ‘Evidence-Based Medicine’ movement, this article raises a number of difficulties with the rationales that have been supplied in particular for the ‘evidence hierarchy’ and for the very special role within that hierarchy of randomized (...) controlled trials (and meta-analyses of the results of randomized controlled trials). The point is not at all to question the application of a scientific approach to evidence in medicine, but, on the contrary, to indicate a number of areas where philosophers of science can contribute to a proper implementation of exactly that scientific-evidential approach. (shrink)

A number of writers, myself included, have recently argued that an “interventionist” treatment of causation of the sort defended in Woodward, 2003 can be used to cast light on so-called “causal exclusion” arguments. This interventionist treatment of causal exclusion has in turn been criticized by other philosophers. This paper responds to these criticisms. It describes an interventionist framework for thinking about causal relationships when supervenience relations are present. I contend that this framework helps us to see that standard arguments for (...) causal exclusion involve mistaken assumptions about what it is appropriate to control for or hold fixed in assessing causal claims. The framework also provides a natural way of capturing the idea that properties that supervene on but that are not identical with realizing properties can be causally efficacious. (shrink)

This article defends the use of interventionist counterfactuals to elucidate causal and explanatory claims against criticisms advanced by James Bogen and Peter Machamer. Against Bogen, I argue that counterfactual claims concerning what would happen under interventions are meaningful and have determinate truth values, even in a deterministic world. I also argue, against both Machamer and Bogen, that we need to appeal to counterfactuals to capture the notions like causal relevance and causal mechanism. Contrary to what both authors suppose, counterfactuals are (...) not "unscientific" - a substantial tradition within statistics and the causal modelling literature makes heavy use of them. (shrink)

In Newcomb’s paradox you can choose to receive either the contents of a particular closed box, or the contents of both that closed box and another one. Before you choose though, an antagonist uses a prediction algorithm to accurately deduce your choice, and uses that deduction to fill the two boxes. The way they do this guarantees that you made the wrong choice. Newcomb’s paradox is that game theory’s expected utility and dominance principles appear to provide conflicting recommendations for what (...) you should choose. Here we show that the conflicting recommendations assume different probabilistic structures relating your choice and the algorithm’s prediction. This resolves the paradox: the reason there appears to be two conflicting recommendations is that the probabilistic structure relating the problem’s random variables is open to two, conflicting interpretations. We then show that the accuracy of the prediction algorithm in Newcomb’s paradox, the focus of much previous work, is irrelevant. We end by showing that Newcomb’s paradox is time-reversal invariant; both the paradox and its resolution are unchanged if the algorithm makes its ‘prediction’ after you make your choice rather than before. (shrink)

We use our knowledge of causal relationships to imagine possible events. We also use these relationships to look deep into the past and infer events that were not witnessed or to infer what can not be directly seen in the present. Knowledge of causal relationships allows us to go beyond the here and now. This chapter introduces a new theoretical framework for how this very basic concept might be mentally represented. It proposes an epistemological theory of causation — that is, (...) a theory that specifies the nature of people's knowledge of causation, the notion of causation used in everyday language and reasoning. (shrink)

Fodor's thinking on modularity has been influential throughout a range of the areas studying cognition, chiefly as a prod for positive work on modularity and domain-specificity. In _The Mind Doesn't Work That Way_, Fodor has developed the dark message of _The Modularity of Mind_ regarding the limits to modularity and computational analyses. This paper offers a critical assessment of Fodor's scepticism with an eye to highlighting some broader issues in play, including the nature of computation and the role of recent (...) empirical developments in the cognitive sciences in assessing Fodor's position. (shrink)

That one person's modus ponens is another's modus tollens is the bane of philosophy because it strips many philosophical arguments of their persuasive force. The problem is that philosophical arguments become mere pantomemes: arguments that are reasonable to resist simply by denying the conclusion. Appeals to proof, intuition, evidence, and truth fail to alleviate the problem. Two broad strategies, however, do help in certain circumstances: an appeal to normal informal standards of what is reasonable (nisowir) and argument by interpretation. The (...) method of explication features prominently in both strategies, and the paper extends this method to apply to nisowir, introducing the concept of canonical explication. The paper illustrates the two strategies with examples of arguments from formal epistemology and suggests that an appeal to nisowir might help to defend against philosophical scepticism by shifting the burden of proof to the sceptic. (shrink)

Systems medicine is a promising new paradigm for discovering associations, causal relationships and mechanisms in medicine. But it faces some tough challenges that arise from the use of big data: in particular, the problem of how to integrate evidence and the problem of how to structure the development of models. I argue that objective Bayesian models offer one way of tackling the evidence integration problem. I also offer a general methodology for structuring the development of models, within which the objective (...) Bayesian approach fits rather naturally. (shrink)

I put forward several desiderata that a philosophical theory of causality should satisfy: it should account for the objectivity of causality, it should underpin formalisms for causal reasoning, it should admit a viable epistemology, it should be able to cope with the great variety of causal claims that are made, and it should be ontologically parsimonious. I argue that Nancy Cartwright’s dispositional account of causality goes part way towards meeting these criteria but is lacking in important respects. I go on (...) to argue that my epistemic account, which ties causal relationships to an agent’s knowledge and ignorance, performs well in the light of the desiderata. Such an account, I claim, is all we require from a theory of causality. (shrink)

In recent years, realism?particularly critical realism?has become an important philosophical and methodological foundation for social science. A key feature is that of natural necessity, but this coexists alongside an acceptance of contingency in the social world. I argue in this paper that there cannot be any natural necessity in the social world, but rather the real nature of the social world is that it is contingent. This need not lead to an abandonment of realism, and indeed I argue that a (...) contingent form of realism allows us to bring ontological probability back into realism. (shrink)

Social Epistemology, Volume 25, Issue 2, Page 167-172, April 2011.

We present a new framework for combining logic with probability, and demonstrate the application of this framework to breast cancer prognosis. Background knowledge concerning breast cancer prognosis is represented using logical arguments. This background knowledge and a database are used to build a Bayesian net that captures the probabilistic relationships amongst the variables. Causal hypotheses gleaned from the Bayesian net in turn generate new arguments. The Bayesian net can be queried to help decide when one argument attacks another. The Bayesian (...) net is used to perform the prognosis, while the argumentation framework is used to provide a qualitative explanation of the prognosis. (shrink)

It is argued that causal understanding originates in experiences of acting on objects. Such experiences have consistent features that can be used as clues to causal identification and judgment. These are singular clues, meaning that they can be detected in single instances. A catalog of 14 singular clues is proposed. The clues function as heuristics for generating causal judgments under uncertainty and are a pervasive source of bias in causal judgment. More sophisticated clues such as mechanism clues and repeated interventions (...) are derived from the 14. Research on the use of empirical information and conditional probabilities to identify causes has used scenarios in which several of the clues are present, and the use of empirical association information for causal judgment depends on the presence of singular clues. It is the singular clues and their origin that are basic to causal understanding, not multiple instance clues such as empirical association, contingency, and conditional probabilities. (shrink)

Some have argued that our intuitive reactions to a number of cases of moral responsibility can only be preserved at the expense of a unified account of moral responsibility for acts and omissions. I argue against this conclusion, proposing that a plausible condition on responsibility, the Causal Condition can, when properly elaborated, justify the relevant intuitive data.

Making a Difference: Essays on the Philosophy of Causation Edited by BeebeeHelen, HitchcockChristopher and PriceHuwOxford University Press, 2017. xii + 336 pp.

Is the common cause principle merely one of a set of useful heuristics for discovering causal relations, or is it rather a piece of heavy duty metaphysics, capable of grounding the direction of causation itself? Since the principle was introduced in Reichenbach’s groundbreaking work The Direction of Time (1956), there have been a series of attempts to pursue the latter program—to take the probabilistic relationships constitutive of the principle of the common cause and use them to ground the direction of (...) causation. These attempts have not all explicitly appealed to the principle as originally formulated; it has also appeared in the guise of independence conditions, counterfactual overdetermination, and, in the causal modelling literature, as the causal markov condition. In this paper, I identify a set of difficulties for grounding the asymmetry of causation on the principle and its descendents. The first difficulty, concerning what I call the vertical placement of causation, consists of a tension between considerations that drive towards the macroscopic scale, and considerations that drive towards the microscopic scale—the worry is that these considerations cannot both be comfortably accommodated. The second difficulty consists of a novel potential counterexample to the principle based on the familiar Einstein Podolsky Rosen (EPR) correlations in quantum mechanics. (shrink)

Proponents of individual-based modeling in ecology claim that their models explain the emergence of population-level behavior. This article argues that individual-based models have not, as yet, provided such explanations. Instead, individual-based models can and do demonstrate and explain the emergence of population-level behaviors from individual behaviors and interactions.

Young children spend a large portion of their time pretending about non-real situations. Why? We answer this question by using the framework of Bayesian causal models to argue that pretending and counterfactual reasoning engage the same component cognitive abilities: disengaging with current reality, making inferences about an alternative representation of reality, and keeping this representation separate from reality. In turn, according to causal models accounts, counterfactual reasoning is a crucial tool that children need to plan for the future and learn (...) about the world. Both planning with causal models and learning about them require the ability to create false premises and generate conclusions from these premises. We argue that pretending allows children to practice these important cognitive skills. We also consider the prevalence of unrealistic scenarios in children's play and explain how they can be useful in learning, despite appearances to the contrary. (shrink)

A common strategy for simplifying complex systems involves partitioning them into subsystems whose behaviors are roughly independent of one another at shorter timescales. Dynamic causal models clarify how doing so reveals a system’s nonequilibrium causal relationships. Here I use these models to elucidate the idealizations and abstractions involved in representing a system at a timescale. The models reveal that key features of causal representations—such as which variables are exogenous—may vary with the timescale at which a system is considered. This has (...) implications for debates regarding which systems can be represented causally. (shrink)

Some recent accounts of constitutive relevance have identified mechanism components with entities that are causal intermediaries between the input and output of a mechanism. I argue that on such accounts there is no distinctive inter-level form of mechanistic explanation and that this highlights an absence in the literature of a compelling argument that there are such explanations. Nevertheless, the entities that these accounts call ‘components’ do play an explanatory role. Studying causal intermediaries linking variables Xand Y provides knowledge of the (...) counterfactual conditions under which X will continue to bring about Y. This explanatory role does not depend on whether intermediate variables count as components. The question of whether there are distinctively mechanistic explanations remains open. (shrink)

Some recent accounts of constitutive relevance have identified mechanism components with entities that are causal intermediaries between the input and output of a mechanism. I argue that on such accounts there is no distinctive inter-level form of mechanistic explanation and that this highlights an absence in the literature of a compelling argument that there are such explanations. Nevertheless, the entities that these accounts call ‘components’ do play an explanatory role. Studying causal intermediaries linking variables Xand Y provides knowledge of the (...) counterfactual conditions under which X will continue to bring about Y. This explanatory role does not depend on whether intermediate variables count as components. The question of whether there are distinctively mechanistic explanations remains open. (shrink)

Within the causal modeling literature, debates about the Causal Faithfulness Condition have concerned whether it is probable that the parameters in causal models will have values such that distinct causal paths will cancel. As the parameters in a model are fixed by the probability distribution over its variables, it is initially puzzling what it means to assign probabilities to these parameters. I propose that to assign a probability to a parameter in a model is to treat that parameter as a (...) function of a variable in an augmented model. By combining this proposal with widely adopted principles regarding which variables must be included in a model, I argue that the various proposed counterexamples to CFC involving coordinated parameters are not genuine counterexamples. I then consider the cases in which CFC fails due not to coordination, but by coincidence, and propose explanatory and predictive bases for ruling out such coincidences without presuming that they are improbable. The aim of the proposed defenses is not to show that CFC never fails, but rather to argue that its use in a particular context may be defended using general modeling assumptions rather than by relying on claims about how often it fails. (shrink)

In a recent article in this journal, Federica Russo and Jon Williamson argue that an analysis of causality in terms of probabilistic relationships does not do justice to the use of mechanistic evidence to support causal claims. I will present Ronald Giere's theory of probabilistic causation, and show that it can account for the use of mechanistic evidence (both in the health sciences—on which Russo and Williamson focus—and elsewhere). I also review some other probabilistic theories of causation (of Suppes, Eells, (...) and Humphreys) and show that they cannot account for the use of mechanistic evidence. I argue that these theories are also inferior to Giere's theory in other respects. (shrink)

I argue that the best interpretation of the general theory of relativity has need of a causal entity, and causal structure that is not reducible to light cone structure. I suggest that this causal interpretation of GTR helps defeat a key premise in one of the most popular arguments for causal reductionism, viz., the argument from physics.

We propose a formal framework for interpretable machine learning. Combining elements from statistical learning, causal interventionism, and decision theory, we design an idealised explanation game in which players collaborate to find the best explanation for a given algorithmic prediction. Through an iterative procedure of questions and answers, the players establish a three-dimensional Pareto frontier that describes the optimal trade-offs between explanatory accuracy, simplicity, and relevance. Multiple rounds are played at different levels of abstraction, allowing the players to explore overlapping causal (...) patterns of variable granularity and scope. We characterise the conditions under which such a game is almost surely guaranteed to converge on a optimal explanation surface in polynomial time, and highlight obstacles that will tend to prevent the players from advancing beyond certain explanatory thresholds. The game serves a descriptive and a normative function, establishing a conceptual space in which to analyse and compare existing proposals, as well as design new and improved solutions. (shrink)

Necessity and sufficiency are the building blocks of all successful explanations. Yet despite their importance, these notions have been conceptually underdeveloped and inconsistently applied in explainable artificial intelligence, a fast-growing research area that is so far lacking in firm theoretical foundations. In this article, an expanded version of a paper originally presented at the 37th Conference on Uncertainty in Artificial Intelligence, we attempt to fill this gap. Building on work in logic, probability, and causality, we establish the central role of (...) necessity and sufficiency in XAI, unifying seemingly disparate methods in a single formal framework. We propose a novel formulation of these concepts, and demonstrate its advantages over leading alternatives. We present a sound and complete algorithm for computing explanatory factors with respect to a given context and set of agentive preferences, allowing users to identify necessary and sufficient conditions for desired outcomes at minimal cost. Experiments on real and simulated data confirm our method’s competitive performance against state of the art XAI tools on a diverse array of tasks. (shrink)

As machine learning has gradually entered into ever more sectors of public and private life, there has been a growing demand for algorithmic explainability. How can we make the predictions of complex statistical models more intelligible to end users? A subdiscipline of computer science known as interpretable machine learning (IML) has emerged to address this urgent question. Numerous influential methods have been proposed, from local linear approximations to rule lists and counterfactuals. In this article, I highlight three conceptual challenges that (...) are largely overlooked by authors in this area. I argue that the vast majority of IML algorithms are plagued by (1) ambiguity with respect to their true target; (2) a disregard for error rates and severe testing; and (3) an emphasis on product over process. Each point is developed at length, drawing on relevant debates in epistemology and philosophy of science. Examples and counterexamples from IML are considered, demonstrating how failure to acknowledge these problems can result in counterintuitive and potentially misleading explanations. Without greater care for the conceptual foundations of IML, future work in this area is doomed to repeat the same mistakes. (shrink)

The goal of research in social epidemiology is not simply conceptual clarification or theoretical understanding, but more importantly it is to contribute to, and enhance the health of populations (and so, too, the people who constitute those populations). Undoubtedly, understanding how various individual risk factors such as smoking and obesity affect the health of people does contribute to this goal. However, what is distinctive of much on-going work in social epidemiology is the view that analyses making use of individual-level variables (...) is not enough. In the spirit of Durkheim and Weber, S. Leonard Syme makes this point by writing that just “as bad water and food may be harmful to our health, unhealthful forces in our society may be detrimental to our capacity to make choices and to form opinions” conducive to health and well-being. Advocates of upstream (distal) causes of adverse health outcomes propose to identify the most important of these “unhealthful forces” as the fundamental causes of adverse health outcomes. However, without a clear, theoretically precise and well-grounded understanding of the characteristics of fundamental causes, there is little hope in applying the statistical tools of the health sciences to hypotheses about fundamental causes, their outcomes, and policies intended to enhance the health of populations. This paper begins the process of characterizing the social epidemiological concept of fundamental cause in a theoretically respectable and robust way. (shrink)

The failure to recognize a correlation as spurious can lead people to adopt strategies to bring about a specific outcome that manipulate something other than a cause of the outcome. However, in a 2008 paper appearing in the journal Analysis, Bert Leuridan, Erik Weber and Maarten Van Dyck suggest that knowledge of spurious correlations can, at least sometimes, justify adopting a strategy aiming at bringing about some change. This claim is surprising and, if true, throws into question the claim of (...) Nancy Cartwright and others that knowledge of laws of association is insufficient for distinguishing effective and ineffective strategies. This paper examines the nature of spurious correlations and their value in crafting strategies for change. The conclusion of the paper is that while knowledge of a spurious correlation may have practical value, the value depends on either having knowledge of the causal structure underlying the correlation or it depends on the use of ‘causal criteria’. (shrink)

In their recent book, Is Inequality Bad for Our Health?, Daniels, Kennedy, and Kawachi claim that to “act justly in health policy, we must have knowledge about the causal pathways through which socioeconomic (and other) inequalities work to produce differential health outcomes.” One of the central problems with this approach is its dependency on “knowledge about the causal pathways.” A widely held belief is that the randomized clinical trial (RCT) is, and ought to be the “gold standard” of evaluating the (...) causal efficacy of interventions. However, often the only data available are non-experimental, observational data. For such data, the necessary randomization is missing. Because the randomization is missing, it seems to follow that it is not possible to make epistemically warranted claims about the causal pathways. Although we are not sanguine about the difficulty in using observational data to make warranted causal claims, we are not as pessimistic as those who believe that the only warranted causal claims are claims based on data from (idealized) RCTs. We argue that careful, thoughtful study design, informed by expert knowledge, that incorporates propensity score matching methods in conjunction with instrumental variable analyses, provides the possibility of warranted causal claims using observational data. (shrink)