The standard item on every list of this kind is to put yourself in the other person’s shoes. Read what they read, imagine their situation, try to see it from where they stand. It is the most recommended technique in the category, and it has been tested more carefully than most advice ever is.
It did not do what people assume it does.
Twenty-five experiments on imagining another person’s view
Tal Eyal, Mary Steffel and Nicholas Epley published Perspective Mistaking in the Journal of Personality and Social Psychology in 2018. It reports 25 experiments with roughly 2,608 participants.
The design is straightforward. Participants had to predict what another person thought, felt or preferred, and the researchers could check the answers against what that person actually reported. Some participants were instructed to take the other person’s perspective before predicting. Others were given no such instruction.
Across the first 15 experiments, the perspective-taking instruction made people less accurate, with an effect size of -0.26. Not neutral. Worse. The pattern held in more naturalistic tests as well.
The second half of the result is the part that stops it being a simple debunk. Perspective-taking also lowered confidence, by a similar margin. Participants who had been told to imagine the other person’s view thought they had predicted fewer answers correctly, and they were right to think so.
So the technique did not produce the confident misreading you might expect. It produced a quieter, less accurate, appropriately uncertain reader of other people.
What actually improved accuracy
The same paper tested an alternative the authors call getting perspective, which is less a technique than an admission. Instead of imagining what the other person thought, participants were given the opportunity to ask them.
That worked.
The authors’ summary of why is the useful sentence in the paper: improving how accurately you understand someone appears to require gaining new information rather than making better use of what you already have. Imagination operates on your existing model of the person. If that model is wrong, thinking harder about it does not fix it, and may elaborate the error.
This has an uncomfortable implication for the usual list. Reading a newspaper you disagree with, following commentators outside your usual set, watching a documentary about a group you know nothing about: none of these are perspective-getting in the sense the research supports. They are all still one person constructing a model from material, without any correction arriving from the person being modelled.
They may be worth doing for other reasons. They are not the thing that was shown to work.
The ten-minute conversation
The strongest evidence that a real exchange can shift a view comes from a field experiment rather than a laboratory. David Broockman and Joshua Kalla published Durably Reducing Transphobia in Science in 2016.
The design is unusually good. It was randomised at household level and placebo-controlled, with 913 voters assigned to the treatment and 912 to a placebo conversation on a different subject, so the comparison is not against nothing but against being canvassed about something else.
The intervention was a doorstep conversation averaging about ten minutes, in which canvassers asked people to recall a time they had been judged negatively for being different, and then invited them to connect that to the experience of transgender people.
Attitudes moved by roughly ten points on a feeling thermometer. The authors note that this single conversation produced a larger shift than the average change in American warmth towards gay men and lesbians across the fourteen years from 1998 to 2012.
The effects were still measurable at three months, through every follow-up wave. Support for a nondiscrimination law rose by 0.36 scale points and remained 0.30 points higher three months later, and it survived exposure to attack advertising on the other side.
What the two findings do together
Read side by side, they point at the same thing from different directions. Accuracy about another person improves when information arrives from that person. Attitudes shift when a real exchange happens with someone who has a stake in it. Neither result rewards the solitary version of the exercise.
That makes the genuinely actionable list shorter and less comfortable than the usual one. It contains conversations with specific people, in which you ask things and are told things, and it does not contain much reading.
It also suggests what to do inside such a conversation. If accuracy improves through new information and not through better inference, then the useful move is to ask about specifics you could not have guessed rather than to check a general impression you already hold. Asking someone what they think about an issue invites them to confirm or deny your existing model. Asking what happened to them, when, and what they did about it supplies material the model did not contain.
The canvassing conversation had a similar shape, though it ran in the other direction. It did not begin by asking people to consider transgender lives in the abstract. It began by asking them to retrieve something specific from their own experience, and only then made the connection.
Where both studies stop
Eyal and colleagues are careful to say their experiments may not have found the conditions under which perspective-taking does work. Twenty-five experiments failing to produce a benefit is strong evidence that the effect is not robust in these paradigms. It is not proof that the instruction is useless everywhere.
Broockman and Kalla are similarly careful about mechanism. They state that they cannot be certain perspective-taking is responsible for their result, because the study was designed for external validity rather than to isolate a single ingredient. Something in a ten-minute conversation with a canvasser worked. Precisely which part is not established by this design.
It is also one canvassing programme, on one issue, in one place, and the wider literature on intergroup contact has known weaknesses of its own, including a shortage of well-controlled field experiments of exactly this quality. The reason this study is cited so often is that it is unusual, not that it is typical.
What survives both sets of caveats is narrow and worth keeping. The version of exposure to other perspectives that has evidence behind it involves another person supplying information you did not have. The version most commonly recommended, in which you supply it yourself by imagining harder, has been tested at length and did not deliver.
