Generating explanations can make you better at learning new patterns…even when they’re not there
Humans are remarkable at detecting patterns. For the most part, this is a good thing. If you can find a pattern relating interest rates to house prices, caffeine consumption to sleep disruption, or pollen counts to your allergy symptoms, you’ll be in a better position to predict and control your environment and your experience.
Unfortunately, patterns can sometimes be spurious or misleading, especially when they’re based on few observations. Even if the first three Australians you meet all hate peanut butter, it would be hasty to conclude that all (or even most) do. So the tendency to find patterns needs to be tempered against the danger of overgeneralization. Sometimes, you’re better off keeping track of individuals and their properties (e.g., Alan hates peanut butter, Bob hates peanut butter, but Carl does not…) than fitting them into patterns (e.g., Australians hate peanut butter).
In new research, Joseph Williams, Bob Rehder, and I find that trying to explain a particular observation (e.g., why Alan hates peanut butter) can make people more prone to overgeneralization when trying to learn about new kinds of objects or new groups of people.
To test this idea, we had people study descriptions of individual objects or people with the aim of learning how to classify them into two different groups: either “daxes” or “kezes” for the objects, or people who frequently or rarely give to charity for the people. As they studied, half the participants were instructed to explain to themselves why that individual might belong to its particular category (e.g., why Anna rarely gives to charity), and half were instructed to simply study or to think aloud while they did so. We measured how quickly and accurately the participants were able to learn to classify the objects and people.
Here’s where it gets a little complicated. We also varied whether there were perfect or misleading patterns differentiating the two novel groups that participants were learning to differentiate. For half the participants, there was a perfectly reliable pattern, such that (for example) all of the people who frequently gave to charity were described as having extroverted personalities, and all of the people who rarely gave to charity were described as having introverted personalities. For the other half of the participants, the pattern wasn’t perfectly reliable: there were exceptions. The only way to achieve perfect accuracy when classifying people as frequent or rare donors, for example, was to rely on individual features of those people instead of a pattern (e.g., Kevin does donate to charities, but Anna does not).
We found that when there were perfect patterns, participants who were prompted to explain learned very efficiently and accurately – at least as well as those in the other conditions. But when there wasn’t a perfect pattern, participants prompted to explain learned less efficiently and less accurately than those who weren’t prompted to explain. It’s as if the explainers kept looking for a perfect pattern despite encountering exceptions. They resisted the alternative strategy of simply memorizing individuals and their properties (e.g., Alan hates peanut butter, Bob hates peanut butter, but Carl does not…), and instead overgeneralized the pattern to cases that didn’t conform.
We hope that our findings not only shed light on why explaining is so often a good thing – pushing children, adults, and scientists to find reliable patterns in the world around them – but also when it can potentially lead people astray, whether it’s in educational contexts or beyond.
Williams, J.J., Lombrozo, T., & Rehder, B. (2013). The hazards of explanation: overgeneralization in the face of exceptions. Journal of Experimental Psychology: General. doi:10.1037/a0030996
This is fascinating work! Seems as though it offers a domain-general mechanism for lots of critical phenomena, not the least of which is out group stereotyping. One question I have is about individual differences here. I know the central finding here is a main effect of explaining on overgeneralization, but would you imagine that people who demonstrate this effect most strongly might also tend to engage in the most stereotyping of other groups?
Thanks for the great post!Reply
i find the original abstract more succinct and quite understandable.Reply
Yes, the original abstract was already great for people used to journal abstracts. But I think this version is more engaging, easier to read for people with less experience of academic speech, and contains a good intermediary amount of detail between an abstract and the full paper. Love it!Reply
Thanks for the comments so far!
Jamil Zaki: one prediction we've considered is that spontaneous explanation could play a role in initiating and maintaining stereotypes. I like your idea of also looking at how variation in the extent to which people overgeneralize relates to stereotyping.
Cathy Harris & Adona Iosif: glad to hear that the original abstract was comprehensible! I'm guessing that in fields with more jargon, or more opaque jargon, the gulf between a "scientific abstract" and a "pop abstract" will be wider.Reply
This finding -- that the process of explanation engages pattern detection -- is a deep and important one. I teach a class on how low-level pattern detection (even in noise) links all the way up to higher level meaning but I never considered the inverse direction. Fantastic.Reply