What is a scientific explanation?
Apr 9, 2019
Criteria of Adequacy
Any satisfactory account of explanation must accord both with common intuitions about what suffices as an explanation and with how they are used in science. I suggest a few criteria of adequacy (perhaps not exhaustive) as follows:
- The aim of explanation is to promote understanding.
- Satisfactory explanations appeal to accepted background knowledge—accepted theory and observable facts.
- Explanation differs from description; an explanation is an answer to a why question—a context-specific request for information.
- The context of a demand for explanation (the contrast-class of interest to the questioner) decides what explanations satisfy the demand.
The first criterion is intuitive; something fails to explain if it does not render the phenomenon of interest more intelligible to a subject. If an explanatory account appeals to facts and theory that are themselves not well understood, then it can hardly be said that explanation has been achieved; further explanation is needed for these facts/theory to suffice as an answer. This criterion, in conjunction with the second, necessitates that the adequacy of an explanation is dependent on the understanding and background knowledge of a subject—meaning it cannot be objectively judged as satisfactory. The second criterion is closely related to the first. As Van Fraassen argues, to say that one has an explanation is to say that one has a theory that is acceptable, explains, and is consistent with what is observed. Each of these is relative to a subject. A theory’s acceptability hinges on its empirical adequacy relative to what the subject can observe, and its intelligibility to the subject is what determines its explanatory aspect.
How an explanation must address a why question is illustrated in the following example. While it may be deduced from the length of a shadow and the angle of a sun that a pole is 100 meters tall, most would not consider these facts an explanation. An explanation of the pole’s height might be better explained by the builder’s intentions when constructing it or by some city regulation that requires that poles not exceed 100 meters. The why question focuses on some notion of causality: the phenomenon as a consequence of a combination of theory and fact.
The context-dependence of explanation is aptly related using Van Fraassen’s notion of contrast-classes. A contrast-class is the set of alternative possibilities that the explanandum is being compared to. For instance, in asking why a tree is positioned in a certain park, one might be wondering why it is a tree there as opposed to some other landmark like a statue or a bench. One could also be asking why that tree is in that park as opposed to across the street from the park. Clearly, the appropriate explanation is context-dependent—the contrast-class of interest being the relevant context. With these criteria in mind, I present a few accounts of scientific explanation and evaluate each in turn.
The Deductive-Nomological Model
Hempel and Oppenheim suggest that explanation exists when the phenomena to be explained (explanandum) can be deduced from the explanatory laws and facts. The conditions of the model are as follows:
- The explanans (that which does the explaining) must deductively entail the explanandum.
- The deduction must make essential use of general laws.
- The explanans must have empirical content.
- The sentences in the explanans must be true.
The first condition requires that the explanandum must follow from the laws and facts appealed to in order for them to count as an explanation. The second means that facts alone do not suffice as explanation. Only laws paired with initial conditions can deductively entail the explanandum. A series of facts only serves as description without any causal aspect. The third condition requires that an explanation must be testable. The laws appealed to must make definite predictions about the world that render them subject to confirmation or disconfirmation. Finally, the fourth condition means that a sound explanation cannot make use of falsehoods. Hempel acknowledges the possibility that some explanans might be well-supported by evidence initially but later disconfirmed. In such a case, he says, the explanation should be regarded as initially correct but later incorrect upon being disconfirmed.
The deductive-nomological account does not satisfy all of the criteria of adequacy that I have outlined above. It is true that Hempel’s account of explanations aims at understanding and appeals to relevant background knowledge, but sound explanations according to this account may not always answer a why question. The aforementioned example of the 100 meter pole shows how theory and fact can be used to deduce the explanandum, but the length of the shadow and angle of the sun do not account for why the pole is 100 meters tall. Moreover, an appeal to the intentions of the builder does respond to a why question but does not deductively entail that the pole will be 100 meters tall; it merely renders the observation more probable or intelligible. As such, Hempel’s demand for deductive certainty rules out probabilistic explanations, which are ubiquitous in actual scientific practice. For instance, certain genetic predispositions do not guarantee the incidence of cancer (it cannot be deduced), but geneticists would certainly consider them explanatory. Both the failure to tie explanations to why questions and the inability to account for probabilistic explanations are serious defects of Hempel’s account, and, for these reasons, it is inadequate as a model of explanation.
The Statistical Relevance Account
An alternative account, put forth by Wesley Salmon, focuses on the various factors that are statistically relevant to the explanandum of interest. If the explanans sufficiently bears on the probability of the observation to be explained, it satisfies Salmon’s characterization. Nancy Cartwright’s criticisms of Salmon’s account are quite compelling. Offering two examples, she argues that statistical relevance is neither necessary nor sufficient for explanation to be achieved. Statistical relevance is not necessary because some event can seemingly cause the explanandum without affecting its preexisting probability. For instance, assume that the probability of paresis given either epilepsy or syphilis is 0.1, and every member of Jones’s family has one of these two conditions. If Jones develops paresis, the answer to the why question is either “because Jones had syphilis” or “because Jones had epilepsy”. However, because everyone in his family is afflicted by one of these conditions, Jones having either one has no bearing on the probability of his developing paresis—it was 0.1 from the outset. However, our intuitions tell us that Jones having syphilis would still count as an explanation. Neither is statistical relevance sufficient as a basis for explanation. Cartwright offers the example of poison ivy that is sprayed with defoliant. Even if such plants tend to perish 90% of the time, on the occasion that one does not, the statistical relevance of the defoliant does not qualify it as an explanation for the poison ivy’s survival. As an alternative to Salmon’s, I suggest the following statistical criterion: only events that are thought to increase or leave unaffected the probability of the explanandum are admissible as possible explanans. Those that decrease the probability of the phenomenon to be explained do not render its observation more intelligible.
Van Fraassen’s Account
Van Fraassen proposes that explanation serves a pragmatic end: by highlighting salient parts of a causal net, one is able to make sense of observed phenomena in a satisfactory and useful way. For instance, a police officer who can identify the facts and theories that explain a car accident can do the job better. Van Fraassen’s account of explanation is one that is relative to the user; an explanation succeeds insofar as it aids the understanding of the person seeking it. Because an explanation is an answer to a why question, its adequacy is evaluated in light of the question being asked (what contrast-class is being considered).
The scientific aspect of an explanation arises when an explanation makes use of an accepted scientific theory. To think that a certain combination of theory and fact suffices as an explanation does not require belief in its truth. As Van Fraassen argues, the asking of a why question need not involve any greater epistemic commitment than that involved in theory acceptance. It is not necessary to believe in the causal significance of an event to invoke it for explanatory purposes.
Two additional virtues of Van Fraassen’s account are as follows. First, it accounts for the possibility that explanations can be false and still render phenomena intelligible—the pragmatic end of explanation. Imagine that a young boy wants to know why he is not allowed to watch television in the mornings. His mother responds that watching television in the mornings will make him sick. While the explanation given is not true, it succeeds in helping the child understand the rule in question. In support of this claim, Van Fraassen offers numerous examples of scientific theories that are now considered false but remain explanatory: Newton’s theory as an explanation of tides, Lavoisier’s oxygen hypothesis as an explanation of combustion, etc. A second virtue is that Van Fraassen’s account permits explanations that are not causal in a statistically discernible way. For instance, evolutionary accounts of human biology may be sufficiently complex that they cannot be concretely observed or proven. Yet, the selectionist pressures that are asserted nonetheless serve the explanatory task faced by an evolutionary biologist. Furthermore, by a statistical relevance account, how can one accommodate final explanations of the teleological kind? For instance, a woodpecker’s beak is shaped to carve out holes in trees. Van Fraassen’s user-centric account allows for this kind of explanation, which renders it much more viable as a reflection of actual scientific practice. For these reasons along with the fulfillment of the criteria of adequacy that I have outlined, I consider Van Fraassen’s account the most defensible of the ones I have presented.