A Simple Lesson From Strep Throat
What strep throat and ear infections teach us about the nature of simplicity
Posted May 23, 2012
Together with Michael Pacer, a graduate student at UC Berkeley, I’ve been studying whether and why people prefer simple explanations.
But what exactly makes an explanation simple?
To answer this question, we teach people about diseases on an alien planet and then ask them to identify the best explanation for an alien’s symptoms. By varying the properties of the diseases and their frequency in the alien population, we can see which kinds of explanations people prefer. If people systematically prefer an explanation that isn’t very likely based on the frequencies of the diseases, that’s pretty good evidence that an explanatory preference—such as a bias for simpler explanations—is influencing people’s judgments.
Using this strategy, we’ve investigated which of two metrics people use to evaluate an explanation’s simplicity. We call these metrics “count simplicity” and “root simplicity.” Fortunately, I can illustrate without the help of alien diseases. Unfortunately, that’s because I’ve spent most of the last two weeks with a fever and sore throat in an uncanny imitation of one of our experiments.
Here’s how it started. My one-year-old daughter caught what seemed to be a cold, and a few days later I was also home with a sore throat and fever. We both seemed to get better, but then our fevers went up again. Surely the simplest explanation was that we’d caught the same virus. Right?
After a week of poor sleep and bleary days with no more signs of improvement, we went to our respective doctors and learned that my daughter had an ear infection and I had strep throat. Two entirely different bacterial infections.
My initial suspicion (a shared virus) and the actual diagnosis (two different bacterial infections) vary in count simplicity: the number of causes they invoke. The virus counts as one cause, while the ear infection plus strep throat count as two. So according to count simplicity, the shared virus was the simpler explanation. It just happened to be wrong.
If not for the pull of the simpler explanation, might we have seen our doctors sooner? Perhaps. But the idea that explanations involving fewer causes are better has some pretty compelling advocates. Consider Newton’s first rule of reasoning from the Principia Mathematica: “We are to admit no more causes of natural things than such as are both true and sufficient to explain their appearances.” And in an earlier post, I presented a study showing that even preschool-aged children prefer explanations with one cause over two.
But now that we know my daughter had an ear infection and that I had strep throat, consider these two explanations:
Explanation A: My daughter developed an ear infection and I developed strep throat independently—that is, there was no common cause.
Explanation B: My daughter and I had the same viral infection, which for her developed into ear infections and left me susceptible to strep throat. In other words, our respective bacterial infections had a common cause in the form of a viral infection.
According to count simplicity, explanation A trumps explanation B: it invokes 2 causes (ear infection + strep throat) rather than three (ear infection + strep throat + a nasty virus).
But there's something compelling about explanation B: it all boils down to one cause, really, the initial virus. These explanations also vary in root simplicity: the number of unexplained causes invoked. While explanation B has three causes total, only one is unexplained (since the initial virus can explain the subsequent bacterial infections). And though explanation A invokes only two causes, both are unexplained.
So A beats B on count simplicity, but B beats A on root simplicity. And when we tested cases like this in the lab—only with alien diseases—we found that people preferred explanations with low root simplicity, with no independent evidence for count simplicity. Even when frequency information pointed towards an alternative explanation, people were drawn towards explanations with lower root simplicity. In our current experiments we’re trying to pin down why.
I still don’t know whether explanation A or B is right. But either provides a perfectly reasonable explanation for why I’ve failed to write a new blog post for the last few weeks, so please accept this simple apology!