N-reading: Explanations: Refining Articulations & Relevance
(Wright, 206-249)
Simple Evaluation
We began looking at diagnostic reasoning to develop a way to articulate our evaluation of explanations; we wanted to be able to say what it is about support that makes one rival better than another. We are now in a position to take the next step in that direction. Explanation is the link between evidence and rivals that determines how good a diagnostic argument is. The ranking of a rival depends on how plausibly it explains the trace data, given what else we know about the case (NTD and our general understanding). An argument is plausible if its conclusion explains the traces more plausibly than its rivals; and it is strong if it is a much better explanation: if its rivals are clearly far worse. But our judgment of argument strength depends on the way the individual bits of support relate to the serious rivals and to each other. So to provide detail in our picture of evaluation, we must turn to diagnostic relevance relations. That is, we must articulate the impact of diagnostic support, of trace data and non-trace data, on the rival explanations.
Diagnostic Relevance
To show relevance we do a little thought experiment: estimate what the serious rivals would look like first with and then without the bit of information in question. … If the list of serious rivals changes in either of two basic ways the information is relevant. Either what’s on the list changes (a rival joins or falls off) or something changes in the ranking. A change in ranking could mean that some rival leap frogs another or simply that one gains or loses strength in relation to another, that is, the gap may widen between the top rival and the next best rival. So what we'd like is the diagnostic articulation of what goes on when this happens in diagnostic arguments. Consider first the relevance of trace data.
CENTRAL TRACE DATA
If we try the experiment with some TD, it quickly becomes obvious that trace data can be relevant in two distinct ways: very generally or quite particularly. Recall that what raises a diagnostic IQ is usually some TD. Something happens and we ask, "What was that?" or "What's going on there?" Or we may ask more specifically, "What caused the crash?" or "Why is the trail littered with bones?" Because the IQ asks specifically about these traces, all of the serious rivals - any serious answer - will have to explain them. We call these "central" TD and their impact on the rivals is very general: they determine which rivals are serious; they shape the list. Whether a rival is serious or not is simply determined by how well it explains the crash, the bones, or my wilting plants. A rival is not serious if it cannot coherently explain these central pieces of TD.
Moreover, the central items explicitly mentioned in the IQ will have other features and properties that are just as central. The IQ asks about bones, but the bones are domestic animal bones. So the serious rivals will all have to explain this too. The IQ asks about a truck crash, but a key feature of this truck crash is that the tire marks went straight off the road; so the serious rivals will all have to explain this too. The point is that given our understanding of a case, the IQ may not have mentioned these other features, but it might just as well have. Instead of "Why is the trail littered with bones?", Daisy Hiker could as well have asked, "Why is the trail littered with domestic animal bones?" This would not be a different understanding of the case, just a more explicit representation of the understanding we had. So these features too have the same, very general impact: all of the rivals are trying to explain them and the serious list is simply the list of rivals that explains these central pieces of TD plausibly.
PERIPHERAL TRACE DATA
Of course not all TD will be central and relevant in this way to every rival. Some will be explained, better or worse, by one or two rivals, but will not be directly relevant to any others. We will call such TD "peripheral," and it can have a large or small impact, depending on the case. Let us say that, upon hearing a loud crash in the kitchen, we go to the kitchen and find shards on the floor that are obviously the shape and color of a particular drinking glass. The look of these pieces have a big impact on our argument: by themselves they put the "that glass broke" rival on the serious list. Without those shards, that rival is far too specific to be on the serious list. But with them, it explains both the shards and the noise (the central TD) so well that it clearly tops the ranking on what we know. On the other hand, the fragments would not be something all the rivals are trying to explain. The radio rival (noise came over the radio) is just trying to explain the noise, not the glass fragments. So the fragments would not be directly relevant to it.
Peripheral TD can have a subtler impact too. In the Flight 800 case, for instance, the exploding central fuel tank rival is already on the serious list. But suppose investigators figure out that it (the exploding tank rival) can explain why the explosion occurred exactly 13 minutes into the flight. Thirteen minutes after takeoff is exactly how long it would take for a defective fuel probe to set off the fumes in the overheated tank. Getting explained in this way would make the “13 minutes” trace data. But the other rivals might have nothing to say about the exact time of the explosion. They would not explain it or be hurt by not explaining it. It would still be trace data, because it is explained by one of the rivals; but it would not be central in the way the explosion itself is. Furthermore, it would not have a big impact, because the explaining rival is already serious, already on the list. Its impact is simply to boost that rival a bit in the ranking. It is a small but significant relevance.
Finally, it will turn out to be very useful to note that peripheral TD may be negatively relevant in this modest way. Sometimes one bit of TD will be especially tough for one rival to explain. As we noted earlier, the lack of blast damage on the engines of Flight 800 is a case in point. At least three rivals (bomb, missile, malfunction) explain the gross details of the crash equally well. All account easily for the fact that the plane exploded, how it came apart, and the general shape and contents of the debris field. But the missile rival cannot easily account for the absence of blast damage on any of the engines, the hot things on which a heat-seeking missile would home in. This hurts its plausibility, lowering it in the ranking. So the lack of engine damage belongs in the support and "hard to explain" articulates why. The hard-to-explain item is such a useful notion in diagnostic reasoning that we will coin a term just for it: “Explanatory Hurdle”. A bit of trace data is an explanatory hurdle for a certain rival if it is particularly difficult for that rival to explain. The absence of blast damage on the engines is an explanatory hurdle for the missile rival. …
NON-TRACE DATA & RELEVANCE
Since NTD are by definition not the object of explanation, not something the rivals even try to explain, their relevance cannot be articulated in anything like the way TD are. Nevertheless, the notion of explanation is crucial to describing the relevance of NTD too, just differently. For although trace data provide the items to be explained in a diagnostic argument, NTD provide everything else needed for the explanatory task. This comes primarily in two forms: (a) trace-signaling NTD and (b) explanatory resources. The first, trace-signaling, help us see what needs explaining. The second, resources, tell us what is available to help in the explaining.
In familiar cases, we usually know pretty well what needs explaining and have little use for trace-signaling NTD. But sometimes a particular audience needs help to see just what the traces are, and trace-signaling NTD will provide what's required. In the missing person case, for instance, those of us unfamiliar with cell phones may not realize that leaving it at home is worth noting in the argument. (I mean, I leave home without a phone all the time, but a cell phone “user” rarely does). A conversation with any cell phone addict provides some trace-signaling NTD: no one who has a cell hone would ever leave home without it. We can now see that finding the man’s cell phone at home is a trace, something to explain, a clue in the mystery.
Explanatory resource NTD are far more common. Even when we know what needs explaining, we may still require help in discovering plausible explanations. Consider again the argument about my wilting plants. If you did not know I watered my plants a lot, root rot would not be any more plausible than a number of other explanations of their wilting, perhaps much less. The soil might be too dry, for instance, or aphids might be lowering fluid pressure, or it might be too much sun. But when we add to the support the NTD that I water the plants frequently, that makes it easier for the root rot rival to explain the wilting. At the same time it makes it harder for dry soil (another rival) to explain wilting. This is the diagnostic articulation of its relevance. None of the rivals is even trying to explain why I water my plants frequently. So frequent watering cannot gain relevance through getting explained or being a hurdle. Its relevance lies in its helping one rival explain the TD and hurting another in its attempt. This is how it affects the serious rivals. NTD is relevant if it either shows that something needs explaining or helps/hurts a rival’s ability to explain a trace.
PLAUSIBILITY RANKING
To articulate ranking we must say how much better or worse the various rivals work as explanations of the traces. This is a complex judgment, requiring us to keep in mind the struggle the rivals have in explaining each trace and how much the NTD help or hurt each one. But in practice, the task is not overwhelming. Consider again the Flight 800 case, for instance. We can say that although the spark rival had a slight edge on the others, because it would easily explain patterns in the debris, it was not very far ahead given what else we knew. It is nevertheless still ahead of the others because they have explanatory troubles too. The collision and missile rivals both have trouble explaining the complete absence of signs of external penetration on the aircraft, for instance. This is an explanatory hurdle for each of them. They also have trouble with NTD. The collision rival was made less plausible by the failure to find any other missing aircraft in the entire northeast of the United States that afternoon and evening. And the missile rival's plausibility as the explanation was damaged by the absence of military exercises in the area.
STRENGTH
To show that an argument is plausible we must say why its conclusion is the best of the rivals. Sleep, for instance, is the most plausible explanation of the tire marks (and hence of the crash) to begin with in the truck crash because the other two (suicide and heart attack) are relatively rare among truck drivers. If we had to mention this in the argument it would be (General) NTD, but we typically do not: it is just part of the vast general understanding we must take for granted in reasoning. The additional NTD that the driver had been up all night would make sleep an even better account of the crash. So we could say then that the argument (with sleep as its conclusion) was not just plausible but strong. That is, the gap between the best and the next best rival is large.
CAUSE AND CORRELATION
Caution: Dropping out of school may be hazardous to your health. That message was delivered yesterday in the government's annual report charting the nation's health. The National Center for Health Statistics said the death rate among adults ages 25 to 64 who failed to finish high school was 30 percent higher than those who got their diplomas.
We are constantly bombarded by news of correlations between things such as sunburn and skin cancer, national savings rates and economic growth. And these correlations are almost always offered as evidence that the two correlates are connected in some way or other. Sunburn is connected to skin cancer, cigarette smoking to lung cancer. A problem arises, however, because sometimes this is done so quickly and naturally that we hardly see any reasoning going on: we read the correlation and simply see the connection. In such cases, we can say that a “False Cause” fallacy has been committed. But all such cases actually involve an argument (even if an easy one), with the correlation as support and the connection as conclusion.
Premise: CORRELATION (support ↓)
Conclusion: CONNECTION (explains ↑)
To see this - and to see that the argument is always diagnostic - we must first understand that the correlation is not the same as the connection: it is at most evidence for a connection. Giving a correlation between two things by itself does not say whether or how the two are connected. So we must find a way of talking about correlations that does not prejudge that question. Correlations must be neutral on the question of a connection. The most natural way to describe correlations in general is to say that they are two things that "go together." A and B are correlated with one another when they "go together" in some way. This covers many different relationships, but the simplest is one in which A and B are events, and whenever A occurs, B does too. The light in my office comes on whenever I flip the light switch, so we have in this sense a correlation between flipping the switch and the light coming on. If I say to my mechanic that there's a terrible grinding noise every time I step on the brake, I have described a correlation between pressing the brake and the grinding noise. Other examples would be the correlation between bad weather and a pain in uncle Charlie’s elbow and the correlation between eclipses and high tides.