Tuesday, October 6, 2009

The Varieties of Evidence

One of the things that seems to me to distort a lot of the discussions of evidence in philosophy of science (and related areas of epistemology) is an over-simplification of the role of evidence in science and inquiry. In particular, many accounts tend to treat evidence as mono-functional, where the only important function evidence plays is support. The nature of that support relation might vary from account to account (Bayesianism, falsification, inductivism, etc.), but many accounts agree that this is how evidence should be understood.

In contrast, I think we can point to a number of equally essential roles that evidence plays in inquiry. When first setting out to investigate the problem, early observational evidence can help locate or specify the problem. When you have an outbreak of disease, or an unexpected astronomical event, you first have to gather as much evidence about the nature of the problem as possible, before you can pose hypotheses or explanations for testing / support. Gathering evidence can also actually suggest hypotheses. A first look at the evidence suggests that this problem might be best analyzed by Fourier analysis, or a simple retrospective study design, or spectral analysis, etc., or by hypothesizing that the disease is malaria, that the new object in our telescope is a type of quasar, etc. Gathering evidence can also help with the elaboration of a hypothesis, specifying, clarifying, or improving it. And not only can it provide support for a hypothesis by testing or confirming its predictions, but experimental testing can be understood as a type of testing by application. If we understand experiment as a kind of intervention on the basis of a theory or hypothesis, then it is really a type of application of the hypothesis to some situation (often a highly controlled one).

So, to sum up, a (probably partial) list of the various functions of evidence: locating the problem, suggesting a hypothesis, elaborating the hypothesis, supporting the hypothesis, and testing it by application. Probably I should say something about helping specify initial conditions, too, though I'm not sure everyone would be willing to call that "evidence."

One way that idealizing evidence as mono-functional, focusing exclusively on the support relation, might go wrong, is the tendency to worry overmuch about the independent status of evidence; i.e., if evidence is to be judged solely by its suitability for providing firm support, then all of the problems of the "empirical basis" start to rear their ugly heads. I suspect that when we have a more complex picture of the functions of evidence, we can use it to develop a multi-scale analysis of the functional fitness of that evidence, which gives as a way of assessing the adequacy of it to stand as evidence.


  1. This may be nitpicking, but I would argue that evidence can only "suggest" a hypothesis by supporting it. I.e., we observe something that would count as evidence for the truth of a proposition (it supports that proposition) and once we realize this we can say the evidence "suggested" the hypothesis. Likewise, evidence only "locates" a problem by supporting a problem statement that focuses on it (it is now evidence for the truth of a description of situation as problematic in a particular way). As to the last point: I don't really understand how evidence can enter into testing except as support for the truth of a proposition (or its negation).

    I don't think "support for a conclusion" is an especially narrow definition. I think it captures a broad range of evidence (there many different kinds of "support"). Now, if what you mean is that propositions (truths) offered as evidence may also be used to some other end, then, yes, of course. But they then cease to function as evidence.

    That, anyway, is my initial reaction to this idea.

  2. A report on my idiolect: what you describe as the better, wider sense of 'evidence', I would call 'data' -- reserving the word 'evidence' for the narrow view you are arguing against.

    As motivation for considering my idiolect not entirely idiosyncratic: think about everything investigators learn/ collect at a crime scene, before they have any clue about who did it (or why). They certainly collect a lot of information that helps suggest hypotheses and frame problems, but it would be weird (I think) to call _all_ of that information gathered 'evidence.'

  3. @Thomas: Your suggestion seems to me like verbal gymastics or "rational" reconstruction wholly disconnected from the actual cognitive processes at work in scientific inquiry. However the process of "suggestion" works, it seems clear to me that it is not by a process of enumerating a number of candidate hypotheses and determining which is most supported by the evidence.

    @Greg: Another thing to like about your idiolect is that you're implicitly marking a distinction between something initial, the data, some of which may be thrown out, revised, re-organized and re-presented, etc., vs. something final, the evidence, which stands at the end of the particular investigation as an organized collection supporting the conclusion.

    But I would resist this usage insofar as it would lead us to focus on the latter by itself. What suits something to stand as evidence in your sense is, in part, its playing the right roles as "data," and surviving the entire process. In other words, what gets lost by treating evidence as defined solely by support is the larger process which is essential to its status. To correct this, I'd rather threat evidence as the process.

  4. But, Matthew, would you resist the usage Greg suggests in the case of the criminal investigation?

    If a murder happens on a downtown sidewalk, investigators sometimes go door-to-door and take copious notes as they talk to neighbors, etc. Lots of information. Lots of "data". But only some of that will count as evidence. And it will come to count as evidence precisely as it begins to support one or another hypothesis about whodunnit, right?

    My point was only that the data becomes evidence at the precise moment that it supports a conclusion. So it's evidence in the standard sense. That is, we can describe the complex reality of inquiry (both cognitive and social) using the word "evidence" in the narrow sense, but only, of course, if we don't restrict ourselves to using only that word.

  5. @Thomas - Why does it matter? At this point you're just haggling over "semantics" in the pejorative sense. I'm interested in understanding how inquiry comes to warranted conclusions, and in particular in how the processes of observation and experimentation that produce, gather, and refine "facts," "data," "evidence," etc. I'm trying to point out that there's a problem reducing the epistemic process to a single, abstract "support" relation between evidence on one side and hypotheses, theories, predictions, etc. on the other.

    Greg is right to point out the distinction between the "facts of the case" in an inquiry ongoing and the "facts of the case" presented as the results of the investigation. As long as we're recognizing the essential epistemic roles of all of these processes, the words we use don't really matter. But then, if we want to mark that distinction using "data" and "evidence," then what counts as "the evidence" at the conclusion of a forensic investigation becomes only "data" again for the legal trial, until it is validated by the determiner of facts in the case (the jury or judge depending on the trial). That seems like slightly a strange way to talk to me.

  6. Wait a minute... wasn't the point of the original post that there is something wrong with our current semantics of the word "evidence"? I.e., that our usage is too narrow? I'm merely defending the old sense against an attempt at your, if you'll pardon it, "haggling".

  7. @Thomas: My original post is about epistemology, not word usage. Sorry if that intent was unclear.

  8. Intriguing. It seems clear to me (and this is not just a matter of idiolect) that the sort of data collection that you do prior to hypothesizing is not rightly called "gathering evidence". What happens is that the idea that it's a quasar comes up; it will, as you rightly say, be suggested by various superficial features of the object; and then the astronomers set about to find evidence to support that hypothesis.

    Interestingly, there may be Gettier-style reasons that the original superficial indication of quasar-hood turns out not to be evidence, even though the object does turn out to be a quasar. I think that may be a substantive epistemological reason to stick with the mono-functional approach to evidence. Evidence is as evidence does; it supports. But there's much more going on science than gathering evidence (in that narrow sense). That's something we can agree on.

  9. So maybe thinking of evidence as a type of language, which may have various meaning and connotations, would account for this wider depiction of the role of evidence.

    There is, after all, no inherent reason why the world cannot speak in a way humans understand - or, well, understand at least as well as they understand each other.

  10. Maybe this is just more language-haggling, but I would say that "evidence" is always relative to a question or proposition. When the police go door-to-door or scour a crime scene, they are looking for evidence *of* a crime or evidence *for* (or against) a particular proposition (suspect A did it).

    The point that we should pay more attention to data mining (note -- not "evidence" mining) and various other things that one might do with data is well-taken. But that seems to suggest that what is being over-simplified and distorted is the acquiring, organizing, and evaluating of data. Surely, that is compatible with taking "evidence" in a restricted sense?

    Incidentally, do you see the problem arising specifically in the literature about evidence? Do you think that philosophers of science generally ignore other functions of data or is it just a problem in the literature on evidence? If the latter, then the terminological point seems more pressing. After all, if I think "evidence" just means "data that supports a hypothesis" then it would be odd for me, in a paper on evidence, to talk about how evidence can be used to formulate a new hypothesis.

  11. I tend to agree with Thomas's first comment and Greg's and I don't see how what Thomas was doing was '"haggling" over semantics' to use your words (and in particular I don't see why you took him to be doing so but not Greg)

    The point Thomas was making, I think, was that evidence can perform all those other functions only insofar as it performs its primary function, which is to support some hypotheses but not others. So, the point (like the point of your original post) was a point about the function(s) of evidence and more specifically it was a point about how the support function relates to other functions of evidence by being more fundamental.

    Compare this with someone who claims 'People think of the function of hammers too narrowly. The function of hammers is not just that of hitting things with a certain force, but hammers can do much more they can put nails in the wall, smash windows and crack nuts'. While it seems to be true that hammers can be used to perform all those functions it also seems to be true that they can do so only insofar as they can perform their primary function of hitting things with a certain force. So, we can debate whether it is true that data function as evidence only insofar as it supports some hypothesis or other (and I think it's interesting to do so) without '"haggling" over semantics'.

  12. Yes, that was my point. Or at least part of it. I think some of the functions Matthew suggests depend on the primary support function, like the hammers primary "hitting" function. But I think there may also be cases when an observation that might at one point have served as evidence (i.e., as support) is used to some other end. This would be akin to using a hammer as a doorstop or a paperweight. It leverages (to mix my metaphors) some of the properties of the hammer (weight) that also serve it well in its primary function. It's still, arguably, a "hammer". But its possible use as a paperweight doesn't (certainly shouldn't) complicate our understanding of what a hammer is.

  13. @Gabriele - I would agree with you that in Thomas's first comment, he was not necessarily haggling over words. But then, I think his (and your) attempt to explain all of the various essential functions in terms of the support-function, taken to be the only fundamental function of evidence, is an extreme distortion of the phenomena in question. I pointed out why I don't think this works for suggestion already. I think it is also too far of a stretch to analyze the role of evidence in locating the problem or posing the question for the inquiry in this way. How is a question or a problem something which can receive support? Thomas provided the beastly suggestion that "it is now evidence for the truth of a description of situation as problematic in a particular way." This attempts to save a pet notion of evidence at the cost of any sort of cognitively realistic description of the situation. The initial attempt to gather evidence / data in order to take stock of the situation and determine the nature of the problem at hand, to sort the ordinary and familiar aspects of the situation from the puzzling and problematic ones, doesn't actually look anything like assigning likelihoods, credences, or what have you to competing propositions of the form "Problem-statement P1 accurately describes what is problematic about this situation." (One important reason being that an adequate problem-statement has to be much more than true; accurate but unfruitful problem-statements are no more helpful than inaccurate ones.)

    So, your analogy of the hammer is not apt. There is no one essential feature from which all of these functions can be derived. Likewise, in contrast to Thomas's suggestion, I don't think that any of these functions can be sloughed off as themselves inessential. Arriving at the best problem-formulation is just as important a function of evidence as determining which solution is best supported. Likewise, hypotheses don't appear out of nowhere, nor is there development entirely adventitious. Both processes are guided by evidence.

    Now, I think Greg raises an important distinction, as I said, between the role of evidence during an inquiry and the presentation of the evidence at the conclusion of the inquiry, and I think he's right that in the latter case, the support function is primary. However, Greg seems to admit that all such uses are epistemically salient, and so does Thomas, in his second comment. Once we admit that all these things are epistemically salient and not all reducible to the same thing, then it seems to me whether we mark the distinction with the terms "data" and "evidence" in some other way is just haggling over words. Given that philosophers have tended to ignore essential features of the process of inquiry and focus exclusively at the end-product, it seems to me to be a tactical error to use the words in the way suggested by Greg and Thomas; however, if one is sufficiently clear and careful to include all of the epistemically-relevant functions of those things which begin as "data" and then are transformed into "evidence," I've got no complaint. I think I'd just prefer to have one name to refer to the multi-functional observationally-or-experimentally produced particulars with epistemic functions throughout the inquiry.

  14. Beastly? I grant that I left out an indefinite article, but the point seems very tame. Certainly as tame as "...one name to refer to the multi-functional observationally-or-experimentally produced particulars with epistemic functions throughout the inquiry."

    Am I right to think that you are suggesting that, given three distinct but related phenomena, we should, in general, avoid giving them separate names. In this case we have perfectly good senses of "data", "information", and "evidence" that would help us to track, say, the development of efforts to identify an object in the night sky. We make a number of observations of its location each night. We may also register its spectral properties. All this will count as data. Properly organized (suggesting a trajectory and correlating with spectral properties of elements) this data counts as information (about the object--where it might be going, what it might be made of). Finally, this information will serve as evidence for the hypothesis that the object is a planet, or comet, or extra-terrestrial spacecraft. Etc.

    You may be right that philosophers of science have not looked too closely at the processes that generate data (and they should look at those processes, I agree). But I'm not sure you're right about the "tactical" question. If I wanted to get someone to see something they think is distinct from some other thing, I wouldn't use their word for that other thing to point it out to them. Especially if they already have a word for the thing I'm trying to draw attention to.

    In any case, doesn't this "tactical error" arise somewhere between "merely semantic" issues and substantive epistemological ones? I guess my original reading of your post wasn't too far off. As a matter of tactics (philosophical rhetoric) surely it does matter what we call it?

  15. This thread may be dead by now, but I wanted to ask a question of Matt --

    Suppose someone demanded that 'evidence' be used in the narrower way, using 'data' or something like it for the broader notion. Would you be happy with expressing your substantive point as follows?

    Data have a much wider range of epistemological importance than is usually recognized, including X, Y, Z...

  16. Matt,

    I'm still unconvinced. Consider a simple case. Suppose that the fact that Green's prints are on the murder weapon is evidence that Green killed Mustard. Now, suppose that before finding out that Green's prints are on the murder weapon the police did not even consider him a suspect. Does the fact that Green's print are on the murder weapon suggest them the hypothesis that Green killed Mustard? I think so. Does it do so in virtue of supporting that hypothesis? I also think so. How could the fact that Green's prints are on the murder weapon suggest the hypothesis the he is the killer if not by supporting it? And how could it suggest the hypothesis without supporting it? On the other hand, it is clear how the fact that Green's prints are on the murder weapon could support the hypothesis that Green killed Mustard without suggesting that hypothesis (because the police already considered him the main suspect before finding out about that).
    Could you explain what's wrong with the above and why the hammer analogy is not apt?

  17. @Greg - I'm torn. My main target is the view that only support matters, who say what you call the narrow sense of "evidence" is all that's going on, or that the other functions are reducible to support. If you give me that, in different terminology, I've won the main battle. But I guess I would resist your formulation as well. Using different terms implies that you can treat "evidence in the narrow sense" as a separate topic from "data" and its multiple functions. I think these are all necessary steps of the process.

    @Gabriele - There's plenty I could say, here, but since you haven't given me any explanation for why you're unconvinced, I'm not sure where to go. Greg's original suggestion is one reason to reject your reductionism: support is something that evidence does in later stages of or at the end of an investigation, where a hypothesis (or hypotheses) has been forwarded and the evidence is relatively refined. Suggestion is often done by evidence of a rather less refined and more inchoate form, which isn't suitable for supporting anything because it isn't settled enough. Further, suggestion is the process that produces hypotheses. Ipso facto, there is nothing for the evidence to support. In some very nice cases, I suppose that the evidence could suggest a hypothesis that it also supported; though the amount of support would be minimal, I think, given that it was precisely the evidence on the basis of which the hypothesis was formulated. (This captures in pellucid terms, I think, why novel prediction is so important.)

    Here's a more basic worry. The attempt to treat evidence and support in vacuo from the rest of the process of inquiry has been a dead end. If support is the only process going on, then I don't see any hope for the attempt to understand justification in science. My proposal is an attempt to get around that: perhaps if we flush out a more complex functional profile for evidence, and more generally, a more complex picture of the varied interactions and relations between facts, data, evidence, etc. on one side and theories, models, hypotheses, etc. on the other, we might find a more illuminating picture. Of course, the proof of the pudding is in the eating, and all I've proposed is a new recipe. Whether the suggestion proves fruitful depends on work not yet given (and which the comments of this entry are too small to contain ;).

    So, a problem with the hammer analogy: it is as if we knew that hammers can "put nails in the wall, smash windows and crack nuts," but we had no clue of what essential properties underlie all those functions, and your argument that it is "support" that underlies all the other functions of evidence is like insisting that all functions of a hammer are reducible to putting nails in the wall. "Really, it is in virtue of the ability to put nails in the wall that the hammer can crack nuts. How could the hammer crack nuts in any other way?" (Of course, your point is much more subtle than that.)

  18. I think a fleshed-out example of a non-supporting-but-suggestive evidence would help here. Like Gabriele, there's something here that's just not sinking in.

    Also, and this may just be because I'm a bit out of the epistemology loop: who treats evidence in vacuo?

  19. Its not hard to cook up an example of the kind Thomas wants: to play around with Gabriele's example, suppose that Green is the murder victim, and suppose that the police find a mustard stain on Green's shirt. This might suggest the hypothesis that Colonel Mustard is the killer, just by reminding the detective of his name. But it might be well know that Mr. Plum is an avid eater of hot dogs with mustard on them. So, the observation of the mustard on Green's shirt might suggest the hypothesis that Mustard did it, while in fact lowering the probability of that Mustard did because it is substantially raising the probability that the killer was Plum.

    Its a stupid example, but it illustrates a relevant point: suggestion is a causal relation, support is not. Whether E suggested H is a matter of historical fact indexed to a particular time. whether A supports B is a very different kind of fact; one which might vary over time, etc.

    So, the two seem relatively independent. The question then remains: when E causes some person to contemplate H, but E does not raise the probability of H, should it be counted as a kind of evidence? I'm inclined to say that the only reason we might be tempted to say yes is that support and suggestion are so often co-incident.

    A dream or a blow to the head might very well suggest a hypothesis, but they are not evidence. the kinds of examples where we are inclined to call the thing that suggests a hypothesis "evidence" are the ones where the thing also supports the hypothesis.

  20. Is that really the sort of example Matthew is thinking of? I completely agree that about the causal/epistemic distinction. Evidence, it seems to me, is an epistemic notion (supports the truth of a prop). Maybe that's what Matthew is arguing? Evidence is not just an epistemic notion? Jason Stanley treats "epistemic" in a similar way. More later.

  21. Hi Eric,

    Thanks for your example, that's exactly what I was looking for. However, while it's true that a hammer can be used as a door-stopper (as Thomas suggested above), that does not mean that one of the function of hammers is to keep doors open. What I was trying to get at is that pieces of evidence qua pieces of evidence only suggest hypothesis insofar as they support them. Qua facts (or whatever you take pieces of evidence to be) they can do a lot of other things. The stain of mustard on Green is also tasty and can be eaten but that does not make evidence tasty or edible, I guess.

  22. Hi Gabriele,

    I completely agree. That was more or less my point in the last paragraph. Many different sorts of (facts/data/dreams/blowstothehead/whatever) can suggest a hypothesis. But they only do so qua being evidence if they also support it.

  23. Hi Eric,

    Sorry, I misundertood the point of that remark--I thought it was meant to support Matt's suggestion that they are entirely separate functions.