Thursday, October 15, 2009

Examples of Suggestive Evidence

I argued in my last post and in the lively discussion in the comments that evidence has a wider variety of epistemic functions that is commonly appreciated and that those epistemic functions are not reducible to the "support" function. In other words, I want to characterize the essential features of inquiry not as an attempt to gather information that confirms or contradicts some theory of hypothesis, but rather as a complex, multi-phase, multi-directional set of interactions between evidence and theory, as well as problem-statement and other elements.

To be clear, I'm defining evidence in a very broad way: observationally or experimentally generated information about empirical particulars, including not only "raw data" but also models of data, basic interpretations of the data, and even rather generic statements about the phenomena. What I don't mean is what Greg Frost-Arnold suggested in his comment: that we reserve "evidence" for that set of data or information which is settled at the end of an inquiry as figuring in the "support" relation. That is, Greg suggested a dichotomy between "data," which serves all the complex roles I'm interested in, and "evidence," which is the end-product serving the traditional "support" role. While we can go far on this amendment to my argument, I don't ultimately accept it. A basic part of the problem is that philosophers tend to cut off the product from the process and then focus exclusively on the former. This is one of the cases where I think the best way to avoid the strategy is to insist that the product cannot be treated independently. Thus, the broader sense of evidence.

One forceful objection to my view (pressed especially by Thomas and Gabriele Contessa in comments) is that the only way evidence could serve any other function is derivative on and reducible to its function of supporting. Let's focus on my claim that evidence not only supports hypotheses but also suggests them. I think this works in the following way: the inquirer, faced with a problematic situation, surveys the preliminary evidence, which allows them to generate first an attempt to state the problem to be investigated and subsequently a hypothesis for solving the problem. In scientific inquiry, the ability to suggest hypotheses worth investigating further depends heavily on training and "tacit knowledge," but it is, I think, just as essential in naive, commonsense inquiry.

Let's look at some examples:

  1. Consider a child attempting to grow flowers in a flower bed. She's rather upset by the fact that some of the flowers aren't growing very well, while those in a small part of the bed are growing rather well. Looking carefully at the garden during the day, she notices that the small part of the garden with healthy flowers gets full sunlight, whereas the other parts are shaded by the fence or a nearby tree. This suggests to her that maybe the difference is due to the amount of sunlight the plants receive. (Probably the initial observations suggest several other hypotheses as well, but she decides to choose this one.) Now, in order to really support her hypothesis, she'll have to do an experiment. . . .

  2. A case from John Dewey's How We Think (1910) of inquiry involving an experiment:

    In washing tumblers in hot soapsuds and placing them mouth downward on a plate, bubbles appeared on the outside of the mouth of the tumblers and then went inside. Why? The presence of bubbles suggests air, which I note must come from inside the tumbler. I see that the soapy water on the plate prevents escape of the air save as it may be caught in bubbles. But why should air leave the tumbler? There was no substance entering to force it out. It must have expanded. It expands by increase of heat or by decrease of pressure, or by both. Could the air have become heated after the tumbler was taken from the hot suds ? Clearly not the air that was already entangled in the water. If heated air was the cause, cold air must have entered in transferring the tumblers from the suds to the plate. I test to see if this supposition is true by taking several more tumblers out. Some I shake so as to make sure of entrapping cold air in them. Some I take out holding mouth downward in order to prevent cold air from entering. Bubbles appear on the outside of every one of the former and on none of the latter. I must be right in my inference. Air from the outside must have been expanded by the heat of the tumbler, which explains the appearance of the bubbles on the outside. . . . (pp. 70-71, emphasis added)

  3. A case from John Snow's work on cholera in the 19th Century: various kinds of evidence has shown that the effluvial hypothesis for the transmission of the disease was unworkable. From reports and his observation of cholera patients, Snow saw that the pathology of the disease began with intestinal symptoms, rather than symptoms of systematic infection such as fever. This suggested the hypothesis that some morbid material ejected from sick patients was subsequently ingested by those who became infected (and several of the histories of Snow describe it in just this way, that the pathology along with other facts suggested the hypothesis about transmission). Snow then went on to support hist hypothesis by certain epidemiological evidence. . . .
Now, I guess we can't deny that suggestion is a real process going on in inquiry. What is at issue is whether it is both distinctive and epistemic, which I'll attempt to defend in a subsequent post.


  1. How do you understand the support relation? Until I know what you mean by support, I don't think I can judge whether suggestion and support are distinct. I'm leaning toward the contrary view that suggestion is simply weak support that seems to demand *additional* evidence, but I'm worried that support is something different for you and me.

    Let me put it concretely in a different setting. Suppose I have a bunch of data points that I want to fit via regression. I decide that since I have a lot of them, I will leave out some to use in a cross-validation step. With the first chunk, I fit a regression line and then I look at how well that line fits the left-over points. Would you say that the first chunk suggests but does not support my choice of regression line? Does the cross-validating set support or fail to support (or even confute) the regression line? (Same questions for your examples.)

  2. This would be my take on all three cases. In all three cases, the "data" do in fact suggest the evidence. But in the way that support is usually understood, they also support the very hypotheses they suggest.

    In your earlier post, you suggested that regardless if one were a Baysian, or an Inductivist, or Falsificationist, one ought to expand one's conception of the role of evidence. This suggests that you had in mind that even on the conception of support held by each of these evidence does more than support. So, lets just think of it from the point of view of the Baysian.

    Surely the probability of the Hypothesis that the flowers grow faster with more sunlight is higher when it is condinalized on the fact that the flowers that got more sun grew faster. surely the hypothesis that heating air makes it expand is higher when conditionalized on the fact that the bubbles appeared. And surely the hypothesis that colera was transmitted via excretions was higher when conditionalized on the fact that the disease starts in the gut.

    So, it does not seem to me that you have given examples of where evidence suggested a hypothesis without supporting it.

    Now, I know one thing that you might say in response, because you said it in a comment in the last thread:

    Suggestion is the process that produces hypotheses. Ipso facto, there is nothing for the evidence to support.

    In otherwords, the fact that the flowers grew faster in teh sun couldnt support the hypothesis that sun causes flowers to grow because the latter didnt yet exist in the child's mind.

    But this seems to misunderstand the kind of relation that support is. As I pointed out in my last commment, suggestion is a causal relation, but support is an epistemic relation. Hence, with respect to suggestion, the temporal co-incidence of the two relata matters. First you have the evidence, then you have to have the appearance of the hypothesis in someone's brain. But with repect to support, the temporal relation between the two relata is irrelevant. Evidence supports, or doesnt support, a hypothesis, period--regardless of whether that hypothesis exists in anyone's brain. The fact that Q follows from P and P->Q is a timeless fact. Its is true regardless of whether anyone is contemplating Q. Similarly, the fact that E supports H (relative to some fixed backround knowledge) is a timeless fact, and it can hold long before any human being entertains H in their head.

    That last fact seems to me to be true on ANY account of support.

  3. @Jonathan - The point is supposed to be a bit agnostic on what account of evidence you start with, as Eric gets at. Rather, I'm trying to point out a shared mistake amongst otherwise quite different accounts of evidence. The larger point is that in an inquiry, evidence plays many roles, indexed to the different phases of inquiry, starting with trying to figure out how to state the problem / post the question for research, up to finally accepting a hypothesis and presenting the support for its acceptance. Nothing in principle prevents some bit of evidence that was used in one way at one stage as being used at another, and I'd hazard to say that any "final" evidence has been used in various other ways during the inquiry.

    Of course, taking such things into consideration will likely change how we think about support...

    @Eric - You manage to bring up another shared mistake of many accounts of evidence: namely, that "support" is a timeless, abstract relation that holds between abstract objects like propositions. This is a mistake of hypostatizing the formal framework we use to analyze the relation. Support is a contextual process that takes place in actual inquiries between measurement records (and other types of evidence) and a hypothesis (theory, etc.).

    Now, I think there are plenty of ways in which we want to say, apart from some inquirer taking evidence to support some hypothesis, that the evidence does (or does not) actually support it. (This is crucial in criticism, for instance.) But we can just as easily say the same for suggestion ("this data should have suggested a different approach"). And we don't need the vantage of Plato's heaven to do it.

  4. It would be worth clarifying wether you mean, as you say in this post, that "evidence not only supports hypotheses but also suggests them," or, as I had assumed in the previous discussion, that evidence can suggest hypotheses without supporting them. I may have been wrong in that assumption. But it seemed to me that you were arguing that evidence has functions that are independent of its function as support for claims (like those contained in hypotheses). But if your point is only that evidence sometimes suggests the hypotheses it (always) supports then I don't have much of an objection.

    I would emphasize, as Eric seems to, that the suggestive capacity of a piece of evidence depends on its supportive capacity. It's a bit like a four-by-four lying on the ground on a building site. It would serve nicely as a beam, say, in a structure. But it obviously does not actually provide support just lying there on the ground. It has to be put into a context. I think all Eric was saying is that the "beam" has particular objective properties, which might suggest its use as a beam in a particular part of the structure being built, but that it has those properties before actually being so used. Its structural properties (its capacity to offer support) are in that sense "timeless".

  5. @Thomas - I would say that evidence has functions that are independent of its support function, that its function to suggest does not depend on its supportive function, but that any particular bit of evidence can function in more that one way in an inquiry, and that it often does, including both supporting and suggesting (though, coming as they do at opposite ends of the inquiry, we shouldn't be surprised if they come apart from time to time).

    I'm happy with your analogy to the beam, insofar as I of course think there are objective qualities and relational properties of the board that give it the potential to be used in various ways in different contexts of inquiry. I just don't think that the potential to support in certain contexts is more fundamental than some of the other functions, nor do all the epistmeic potentials of evidence depend on the potential to support.

  6. Sure, but one point of the analogy is that a use of a given piece of wood for something other than support (employing its shape, or its colour, or its weight only) would hardly make it a beam.

    Suppose there's a pile of sand on the building site and the wind begins to pick up. "Cover it with a tarp," say the foreman. "But it'll just blow away," says the apprentice. "Well, put some heavy things on it around the edges; those beams over there will do fine." Now, we could easily construct an example of scientific inquiry where "evidence" plays the role of those "beams". But surely they aren't really beams in that function?

    Likewise, an observation becomes evidence when used as support for a proposition in the larger context of an inquiry in which all kinds of things are going on, and towards an overall structure (a theory) in which not all elements are supporting other elements, and perhaps not all of which are epistemic.

  7. Science is mostly about patterns (and matching patterns). We spend a lot of time generating and visualizing patterns, and whether we see the patterns we visualize/describe or generate as suggestive or supportive of some understanding (explanation) of them is a bit arbitrary (both the suggestive and the supportive patterns are vital to the explanation).