The Stone is the new section of the New York Times devoted to philosophy and this week it contains an interesting piece called “Stories vs. Statistics” by John Allen Paulos. It is worth reading in its entirety, but for my money the most important point he makes is this:
The more details there are about them in a story, the more plausible the account often seems. More plausible, but less probable. In fact, the more details there are in a story, the less likely it is that the conjunction of all of them is true.Our tendency to confuse plausibility with probability is also at the heart of a short essay of mine (forthcoming in the journal Think), called “Beware the Convincing Explanation.” Paulos clarifies the excerpt above by reference to the ‘conjunction fallacy,’ which I discussed in an earlier post. In my essay I try to get at it from a different angle, by distinguishing the respective functions of argument and explanation.
Here is the basic idea: Normally, when we ask for an argument we are asking for evidence, which is to say the grounds for believing some claim to be true. An explanation, on the other hand, is not meant to provide grounds for belief; rather it tells us why something we already believe is so. Almost everyone understands this distinction at an intuitive level. For example, suppose you and I were to have this conversation about our mutual friend Toni.
Me: Boy, Toni is seriously upset.You can tell immediately that we aren’t communicating. You asked for an explanation, the reason Toni is upset. What I gave you is an argument, my reasons for believing she is upset. But now consider a conversation in which the converse error occurs:
You: Really? Why?
Me: She’s out in the street screaming and throwing things at Jake.
Me: Boy, Toni is seriously upset.This time my response actually begs the question. Jake blowing off the date would certainly explain why Toni is upset, but an explanation is only appropriate if we agree that she is. Since your question was a request for evidence, it is clear that you are not yet convinced of this and I’ve jumped the gun by explaining what caused it.
You: Really? How do you know that?
Me: Jake forgot their date tonight and went drinking with his pals.
What’s interesting is that people do not notice this so readily. In other words, we often let clearly explanatory locutions pass for arguments. This little fact turns out to be extremely important, as it makes us vulnerable to people who know how to exploit it. For example, chiropractic medicine, homeopathy, faith healing -- not to mention lots of mainstream diagnostic techniques and treatments -- are well known to provide little or no benefit to the consumer. Yet their practitioners produce legions of loyal customers on the strength of their ability to provide convincing explanations of how their methods work. If we were optimally designed for detecting nonsense, we would be highly sensitive to people explaining non-existent facts. We aren’t.
Now, to be fair, there is a sense in which causes can satisfy evidential requirements. After all, Jake blowing off the date can be construed as evidence that Toni will be upset when she finds out. However, it is quite weak evidence compared to actually watching Toni go off on him. So, we can put the point a bit more carefully by saying that what people don’t typically understand is how weak the evidence often is when an explanation gets repurposed as an argument.
Following Paulos, we can say that the convincing explanations succeed in spite of their evidential impotence because they are good stories that give us a satisfying feeling of understanding a complex situation. Importantly, this is a feeling that could not be sustained if we were to remain skeptical of the claim in question, as it is now integral to the story.
Belief in the absence of evidence is not the only epistemic mischief that explanations can produce. The presence or absence of an explanation can also inhibit belief formation in spite of strong supporting evidence. The inhibitory effect of explanation was demonstrated in a classic study by Anderson, Lepper and Ross which showed that people are more likely to persist in believing discredited information if they had previously produced hypotheses attempting to explain that information. Robyn Dawes has documented a substantial body of evidence for the claim that most of us are unmoved by statistical evidence unless it is accompanied by a good causal story. Of particular note are studies by Nancy Pennington and Reid Hastie which demonstrate a preference for stories over statistics in the decisions of juries.
Sherlock Holmes once warned Watson of the danger of the convincing explanation: “It is a capital mistake to theorize before one has data. Insensibly one begins to twist facts to suit theories, instead of theories to suit facts.” Damn good advice from one of the greatest story-tellers of all.