The Chances of Explanation: Causal Explanation in the Social, Medical, and Physical Sciences

The Chances of Explanation: Causal Explanation in the Social, Medical, and Physical Sciences

by Paul Humphreys


View All Available Formats & Editions
Choose Expedited Shipping at checkout for guaranteed delivery by Thursday, March 28

Product Details

ISBN-13: 9780691605821
Publisher: Princeton University Press
Publication date: 07/14/2014
Series: Princeton Legacy Library , #1051
Edition description: Reprint
Pages: 182
Product dimensions: 9.00(w) x 6.00(h) x 0.50(d)

Read an Excerpt

The Chances of Explanation

Causal Explanation in the Social, Medical, and Physical Sciences

By Paul Humphreys


Copyright © 1989 Princeton University Press
All rights reserved.
ISBN: 978-0-691-07353-8


Traditional Causation

§1. Scope of the Work

Ontology, epistemology, and conceptual analysis constitute the great philosophical triad of causation. When pursuing ontological issues, we try to provide an account of the nature of causation: what differentiates causal relations from noncausal relations and what causal relations, as part of the world, are like. Epistemological interests, in contrast, focus on how causal relationships are discovered, how hypotheses about causal relations are tested and confirmed, when it is justifiable to assert a causal claim, and what kinds of causal inferences might be valid. Conceptual analyses, finally, are concerned with what a term such as 'causes' means, with constructing definitions of 'cause' and related terms, with their use in language, ordinary or technical, and with providing rules for the correct employment of such terms. A similar tripartite division can be made in the areas of probability and explanation, even if one of the divisions is considered to be empty, as, for example, those who have denied the existence of objective chance have held. Philosophical work in each of these areas is entirely legitimate, and the divisions cannot be entirely separated, yet confusion can easily arise if arguments that are based primarily on evidence from one of these areas are used as reasons to accept or reject a position in another area without this move being explicitly noted and justified. So it is as well to state right at the outset the orientation of this work. In each of these three areas — causation, probability, and explanation — my primary interest will be ontological. A moderate amount of attention will be paid to the epistemological, and the least attention will be devoted to conceptual analysis.

This ontic orientation may seem peculiar in the case of explanation, for explanation has traditionally been taken to be squarely in the realm of epistemology, with conceptual analysis contributing as a source of useful insights. The reasons for my own orientation will become clear as we proceed, but it derives from two axiomatic theses. The first is that an inviolable requirement of a satisfactory scientific explanation is that it be true. In the case of causal explanations this means that explanations must accurately capture the causal structure of systems and their interactions. False explanations, however psychologically beguiling and however widely accepted in a scientific community, are incapable of producing genuine understanding of the way the world works. The second axiom is that scientific methods, especially experimental methods, have been successful in discovering at least some of the causes that operate in the world and that they are more successful at this than are unsystematic, nonexperimental methods. We thus need to address and answer the philosophical question, "Why are experimental methods superior to nonexperimental methods in discovering and establishing the existence of causal relations?" The answer to this second question will move us away from a traditional kind of empiricist epistemology toward one that, although still empiricist, allows for a good deal of realism about unobserved entities, especially unobserved causes. This causal realism then allows us to address explanations as features of the world that one can discover, rather than construct, and the linguistic explanations that have been the focus of so much philosophical inquiry must be seen merely as flexible devices that convey the essential causal content from individual to individual, when required.

§2. Methodology

The avoidance of conceptual analysis mentioned above has a specific reason and a specific consequence. The consequence is that we should not be primarily concerned with what is said in ordinary language(s) about causation or forced to square our theory with every construction in English that appears to have causal content. The reason is this: there is no ground for supposing that languages which have evolved over the centuries in response to various needs, most of them noncausal and unscientific, should contain within them a coherent representation of causal truths, even at a nonsuperficial level of linguistic structure. One of the more remarkable features of the literature on causation and scientific explanation is the frequency of appeals to what we should ordinarily say about various cases. There is an important difference between our knowing that certain relations are elementary cases of causation and what we should ordinarily say about such cases, other than that they are causal.

The linguistic situation is even worse when we enter the realm of probabilistic causality, especially when, as we often shall be, we are concerned with causal relationships between quantitative variables. Most ordinary talk, and most classical philosophical terminology, is oriented toward deterministic causal relations between qualitative events. This presents anyone who is concerned with providing more than a purely mathematical or logical theory of causation with a potentially serious difficulty, akin to the difficulty that the early advocates of the Copenhagen interpretation of quantum mechanics noted. They claimed that because our epistemic access to quantum phenomena was ultimately made through the medium of experimental apparatus at the macroscopic level, and the conceptual apparatus at that macroscopic level of observability was thoroughly infused with classical mechanical modes of thought, it was profoundly difficult, and perhaps impossible, to construct a satisfactory mode of description for nonclassical quantum phenomena. Whether or not they were correct about that issue, we have to face a similar predicament in providing an understanding within the confines of ordinary English of a theory of probabilistic causality and explanation that is applicable to quantitative scientific contexts.

Fortunately, the structure of these more complex causal and explanatory systems lends itself to a method that to a large extent avoids these difficulties. One of the principal themes of this work is that our understanding of probabilistic causation comes from principles that also apply to nonprobabilistic causation. A second theme is that causal knowledge comes from something like an inductive definition, with simple base cases of causation being used to discover knowledge about more complex cases, and that this base knowledge is nonlinguistic, perhaps even prelinguistic. These two facts about causation mean that one can appeal to clear cases of traditional causation as examples and then argue by analogy, via explicit connecting principles, to the probabilistic case. We can also, it transpires, apply pieces of standard talk about causes to the new cases. There are limits to this, and when there is a conflict between the consequences of a systematic philosophical theory of causation and casual talk of causation, it is the latter that has to give. This strain of revisionary metaphysics is explicit throughout the book, and I consider it to be an important contribution that philosophy can make, in conjunction with science, to our understanding of causation.

Our use of examples is also affected by these concerns. There is a perfectly legitimate tendency in philosophy to use the simplest examples that will illustrate the point at hand, with resort often being made to artificial and artificially simple examples. Yet a characteristic feature of science is its attention to the importance of detail — as noted in §4, the world is a complex place, and part of science's success stems from not ignoring that complexity. One consequence of the use of simple examples in the philosophical literature has been the almost total lack of consideration of quantitative examples, or of how causation and explanation operate with quantitative variables. So most of our examples will be at least moderately sophisticated real examples that reflect the detailed character of science. On occasion, however, everyday and simplified examples will be used where this will not prove misleading.

§3. An Example

In 1981, physicians in Los Angeles and New York began to notice an unusual cluster of cases of formerly rare symptoms — Kaposi's sarcoma, Pneumocystis carinii pneumonia, and other opportunistic infections, primarily in young men. At that time no specific etiology for the symptoms was known, and the repeatedly raised question, "What is the explanation of this man's illness?" set in motion a standard scientific procedure. The search for an explanation brought to bear segments of scientific methodology that are designed specifically for the discovery of causes, in this case epidemiology to identify risk groups, theories in molecular biology to identify possible causal factors, and controlled experimentation to isolate the specific causal factors that were responsible in each case. What this systematic search for an explanation was seeking was not a linguistic entity (such as an argument or a speech act) but a real thing, a cause or group of causes of the disease. As we now know, an explanation was found that included, among other causative factors, a group of retroviruses that cause AIDS. Subsequently, and only subsequently to this discovery, were the investigators in a position to answer why-questions, what-questions, or how-questions, and gradually to fill in the causal story so that groups with different interests — homosexuals, intravenous drug users, public-health officials, biomedical researchers, and so on — could have described to them the parts of the explanation in which they were most interested. Most notably, an explanation could be given even though it was incomplete. It was never claimed that there were no factors involved other than the retroviruses, factors that increased or decreased the risk for an individual, only that part of the internal causal mechanism leading to the illness in each case had been found, and a mechanism of transmission found that was causing the cluster of cases.

Any adequate philosophical theory of scientific explanation should be able to provide an acceptable account of such a case as this. Yet existing theories are all defective in one or more ways. For some, the causal theory that lies behind their presentation of causal explanations proves inadequate when there are multiple probabilistic factors at work. Others are so idealized that claims to have an explanation when we know some, but not all, of the causes influencing the phenomenon violate their criteria of adequacy for explanations. As we shall see, once we move beyond simple, idealized examples, the features of multiplicity, insufficiency, and incompleteness of causes is normal, rather than unusual.

§4. The Multiplicity of Causes

The world is a complex and messy place. If it were not, if it consisted solely of medium-sized atoms that were causally independent of one another, say, we should not need science to discover its structure. But the world is not self-presenting in that way. As the early natural philosophers slowly solved the problem of the difference between appearance and reality, they realized that theories about the immediately accessible parts of our world were generally either banal or false. At least as early as Galileo, scientific investigation was found to proceed most efficiently when investigating artificial phenomena produced in the clean and austere conditions of the laboratory, when only a single causal influence was at work. But science is also frequently called upon to investigate naturally occurring phenomena such as epidemics, tree diseases, rainfall distributions, migratory patterns, rainbows, the nonexistence of higher forms of life on Mars, and planetary movements. It is nowadays also often required to explain the results of applied science, such as rocket explosions, holes in the ozone layer, the properties of artificial elements, the effects of plastics on the environment, and presidential campaigns.

A characteristic feature of these natural and unnatural phenomena is that they are usually the result of multiple causal influences. For example, the rate of enzyme-catalyzed reactions is affected by the enzyme concentration, the substrate concentration, the temperature, the pH of the substrate, oxidation of the sulfhydryl groups of an enzyme, and high-energy radiation; the first two increasing the rate of reaction, the last two decreasing it, while the actions of the third and fourth have maximal points of inflection at optimal temperature and pH, respectively. The occurrence of familiar, everyday phenomena is also characterized by this multiplicity. Successively adding (1) a smoking level of twenty cigarettes a day, (2) medium-high blood pressure (140/88), and (3) medium-high serum cholesterol levels (250 mg/dl) increases the probability of having a heart attack within the next twelve years for a forty-six-year-old man from .03 to (1) .05, (2) .075, and (3) .15. The flight of any commercial airliner is the result of many causal factors: the thrust from the jets, uplift from the airflow over the wings, tail and head winds, downdrafts, thermals, and so on. Some of these are more important than others, and some aid the flight whereas others hinder it, but for a full understanding of the flight as it actually occurs, they must be identified and listed. Such multiplicity of causes is commonplace in the social sciences, where for practical, ethical, or theoretical reasons, many phenomena occur in nonexperimental contexts, and the frequent use in those areas of multifactorial models is a direct consequence of the need to represent the numerous and distinct causal influences affecting such phenomena.

A number of different things may be meant by 'multiple causation', among which are:

1. That specific effect E was produced by specific causes A and B, both of which were present, and each of which separately contributed to E.

2. That specific effect E was produced by the interaction of specific factors A and B, both of which were present.

3. That effects of kind E can be produced, as a kind, by either kind-A or by kind-B causes.

4. That a specific event E was overdetermined by specific factors A and B, both of which were present.

Case 1 is the one that our theory covers, and it will apply to probabilistic as well as to deterministic cases, to quantitative as well as to qualitative factors. Cases 2 and 3 concern, respectively, causal invariance and causal generalizations, which are both discussed in §25. Case 4 precludes us from calling either factor a cause, and will be discussed in §7.

§5. The Incompleteness of Causal Knowledge

A common epistemic consequence of the multiplicity of causes is the incompleteness of our knowledge about them. Rarely are we in a position to provide a complete list of all the influences that affected a given outcome. Hence if those causal influences constitute the explanation of that outcome or are an integral part of its explanation, we have two options open to us. We can either allow that an incomplete specification of explanatory causes does provide explanatory information and that cumulative additions to our knowledge of the causes is a common feature of scientific explanatory procedures, or we can require that a complete specification of the causes is necessary for an adequate explanation. The second of these options is, I think, to be avoided if at all possible. In both the AIDS example and the enzyme-catalyst reaction example, there is no pretense that a complete explanation of the respective phenomena has been discovered. Yet we have good reason to suppose that what has been offered as an explanation is true and explanatorily informative. Successive supplementation of such partial explanations does not undermine the accuracy of the previous explanations. Yet, as we shall see (§38), many contemporary models of explanation have the consequence that they cannot separate true from complete explanations, and I take this to be a serious defect in those accounts.

We begin, then, with the fundamental notion that many scientific explanations consist in a specification of the causes of the phenomenon to be explained, that those causes will be numerous, and that our knowledge of them will ordinarily be incomplete. We now need to examine various types of causes in order to assess their suitability for inclusion in causal explanations of this kind.

§6. The Principle of Causal Contribution

Underlying various accounts of causation in terms of sufficiency, of necessity in the circumstances, of INUS conditions, of probabilistic relevance, and so on seems to be this general principle:

Principle of Causal Contribution. A factor X is a cause of another factor Y if and only if X's existence contributes to Y's existence.

As a schema, rather than a fully interpreted generalization, this principle is, obviously, too broad. For, casually construed, it would allow the middle third of a tub of lard (X) to be a cause of the whole barrel (Y). By variously, but specifically, interpreting 'factor' and 'contributes to', however, one can see identifiable accounts of causation emerging from this very general principle.


Excerpted from The Chances of Explanation by Paul Humphreys. Copyright © 1989 Princeton University Press. Excerpted by permission of PRINCETON UNIVERSITY PRESS.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.

Table of Contents

  • FrontMatter, pg. i
  • CONTENTS, pg. vii
  • PREFACE, pg. ix
  • CHAPTER ONE. Traditional Causation, pg. 3
  • CHAPTER TWO. Probabilistic Causation, pg. 22
  • CHAPTER THREE. Cause and Chance, pg. 61
  • CHAPTER FOUR. Scientific Explanations, pg. 98
  • APPENDIX ONE. Covariance Measures, pg. 143
  • APPENDIX TWO. Extension of the Basic Quantitative Theory, pg. 145
  • APPENDIX THREE. Transitivity and Negative Links, pg. 153
  • REFERENCES, pg. 158
  • INDEX, pg. 167

Customer Reviews

Most Helpful Customer Reviews

See All Customer Reviews