About 250 years ago, Adam Smith famously described the way observers might feel watching a tightrope walker. Even while standing on solid ground, our palms sweat and our hearts race as someone wobbles hundreds of feet in the air (you can test this out here). In essence, we experience this person’s state as our own.

Centuries later, this definition does a surprisingly good job at capturing scientific models of empathy. Evidence from across the social and natural sciences suggests that we take on others’ facial expressions, postures, moods, and even patterns of brain activity. This type of empathy is largely automatic. For instance, people imitate others’ facial expressions after just a fraction of a second, often without realizing they’re doing so. Mood contagion likewise operates under the surface. Therapists often report that, despite their best efforts, they take on patients’ moods, consistent with evidence from a number of studies.

One tempting conclusion about automatic behaviors is that are also “dumb:” occurring whenever the right stimulus comes along. On this view, empathy is the emotional equivalent of a patellar reflex: while observing someone’s emotions, you can’t help but take those emotions on yourself. Intuitive as it may be, a “reflex model” glosses a vital feature of empathy: it is often a choice. Even if others’ emotions rub off on us automatically, this process is only set in motion if we decide to put ourselves in a position for empathy to occur. And that decision is anything but automatic. Instead, people frequently make deliberate choices to avoid others’ emotions, in attempts to stave off the discomfort or costs of empathy.

One of my favorite studies on this topic–a long forgotten gem from 1979–measured empathy by circumference. Mark Pancer and his colleagues set up a table in a busy tunnel at the University of Saskatchewan, and secretly measured the distance people kept from the table while walking past. They manipulated two features of the situation. The first was whether or not the table had a box placed on it requesting charitable donations. The second was who was manning the table: (i) no one, (ii) an undergraduate, or (iii) an undergraduate sitting in a wheelchair. Both the request to donate and the presence of a handicapped person were considered triggers to empathy. Instead of approaching these triggers, however, students avoided them: walking a wider arc around the table in the presence of either trigger, and keeping the greatest distance in the face of both the handicapped student and donation box.

In a more recent study along the same lines, Daryl Cameron and Keith Payne examined the well-known “collapse of compassion.” Cameron and Payne told participants about the suffering of children in the wake of Darfur’s civil war, and showed them pictures of either one or eight of these children. Critically, they told some participants that–after viewing these pictures–they would have a chance to donate money to help these children. Participants who believed they would be put on the spot to donate felt less empathy for eight children than for one, consistent with the idea that they purposefully “turned down” their empathy when empathizing could prove costly.

Together, these studies suggest that instead of automatically taking on others’ emotions, people make choices about whether and how much to engage in empathy. Pancer and Cameron’s observations at first appear bleak–people shut down empathy when it might cost them–but I think they paint a more encouraging picture. For instance, Paul Bloom recently argued that empathy is a bad guide for decision-making, precisely because it is a slave to triggers such as images of others’ suffering. On Bloom’s reasoning, this means that empathy will often drive irrational choices based on emotions: for instance, helping a single suffering child we see on television while ignoring countless others who receive less press. Although Bloom is right in many cases, if empathy is a choice, then people can presumably learn to use it when they know it is most important. For instance, people could decide to “turn up” empathy for victims with whom they might not immediately connect (a suggestion made earlier by Daryl Cameron as well). Broadly speaking, empathy we can control is empathy we can co-opt to help others as much as possible.