Representing Data in the Science of Emotion
A recent study published in PNAS has been floating on my radar for a while — this one, proclaiming a physical reaction to emotional content that is universal across cultures and languages. There’s some interesting data here, and a couple issues I have about how they represented their data, and how it’s been picked up by the public as far as I have seen.
The Good: More Evidence that Feelings are, in fact, Felt
This was a careful study that really does provide solid evidence that there is a consistent pattern of physical feelings associated with emotional content. Their strongest result, in my opinion, is that stories, films, faces, and words that carry emotional content — whatever that emotion is — provoke physical responses different from the physical responses provoked by stories, films, faces, or words without emotional content. In other words: feelings are felt.
As perhaps a side note (as in, it is independent of the quality of the data or the strength of the results, although not my emotional response to either) — I think that any data which contradicts mind-body dualism is a good thing, especially if it takes off in the public imagination. Not only is mind-body dualism a relic of an idea inconsistent with many modern scientific findings, it’s a rather pernicious one at that. It potently feeds into the stigmatization of mental illness — but if feelings are physical, then there is no more shame in depression than there is in the flu. It creates a contrast between “hormonal”, “female” emotions which are somehow more effected by the physical body than the “default”, “male” emotions — which is a load of crock.
The Dubious: Does “Happy” Always Feel the Same?
I wasn’t completely sold on the universality of physical ramifications of emotion. Mostly this is because of a figure that hasn’t taken off in the public eye the way the heat-map-on-a-body graphic does (ala the featured image, more on that next).
This is a clustergram. It’s basically a condensed family tree. Items that are closer together are more closely related — they’re more similar. And most importantly, the length of the line vertically between an element and a branch point represents the “distance” between the two elements in that branch point. I’ve highlighted a couple things I found interesting below.
This clustergram shows, rather strongly, that things with emotional content provoke a response different than things without emotional content. Most of the vertical space is taken up by a single branch: between the “neutral” cluster and the rest of them.
However, the physical sensations provoked by emotional content do not seem to cluster by emotion. Which is to say, for instance: “happy” films provoke physical sensations most similar to “sad” films, stories, or the word “sadness”, while the word “happiness” provokes a physical sensation most similar to a scary story. So while sadness and disgust seem to have relatively constant reported physical sensations no matter how they are conveyed — films, stories, images, or just the word itself — that is barely true of anger or fear, and not at all true of happiness or surprise.
Why might this be? There are several possibilities. One is that the researchers were not actually that great at provoking a given emotional response with a story or film. (Does the “happy” film cluster in the middle of “sadness” because it actually made people sad?) A second possibility is that the rather simplistic request — paint what is “stronger or more active” versus “weaker or less active” — left out some of the content of a feeling, for instance. A third is that certain commonalities among all emotions (such as an “activation” in the head — presumably facial muscles?) threw this off.
That said, the overarching result: that emotions have physical ramifications and that these ramifications are to a certain extent universal, not linguistic or cultural, is relatively strongly demonstrated. And that’s pretty cool.
The Misrepresented: How Do You Show Happiness in a Study on Happiness?
I can empathize with the strength of the images that have been going around — this heatmap superimposed over a silhouette of a person. But there’s a fundamental representational error that gets into a certain difficulty in how we represent biological data in general.
Here’s the figure:
What are those heatmaps? You might think, to look at them, that they’re average “activation” patterns over all the participants: that warmer colors are where people, on average, reported more activation and cooler colors are where people, on average, reported more weakness. But that’s not quite true.
What the colors represent, instead, is p-value. Which is to say: the probability that the activation observed, at that pixel, in their samples occurred just due to chance. A lower p-value and positive activation resulted in a warmer color (red-yellow); a lower p-value and a negative activation resulted in a cooler color (blue-cyan).
These two things are similar, but there’s one important difference, and that’s the fact that a slight, consistent activation would show up at least as brightly as a strong, inconsistent one. In other words, the brightness refers to the universality of a feeling, rather than the intensity necessarily.
This conflation of intensity and statistical significance is something that I see again and again in biology. It’s easy to do, and perhaps it’s a minor enough point that it’s not the end of the world (reading that chart as “The vast majority of people feel the sensation of ‘love’ in their hearts” rather than “You feel ‘love’ most strongly in your heart” isn’t really that big a deal, and it’s a relatively minor point in the paper). But there are some situations in which it’s a far more dangerous contrast, perhaps where something like “almost everyone with this particular allele have a slightly increased risk of cancer” turns into “this allele is strongly linked to cancer” which turns into “this allele will give you cancer!”
Moral of the story: we have a lot of figures in biology that are appealing because they seem to say something clearly, but half the time that something isn’t what they are actually saying. That’s dangerous.