Pollsters have long known about the poor predictive power of asking respondents how they would treat members of an unfavored minority group, especially in politically polarized climates. In the 1930s, following a period, like today, of growing anti-immigrant sentiment, the Stanford researcher Richard LaPiere crisscrossed the country with a Chinese couple, visiting hundreds of hotels and restaurants. Nearly all of them welcomed the group as patrons.

But when he contacted the establishments months later asking them if they would serve Chinese people, over 90 percent said they would not. In an ensuing article, “Attitudes vs. Actions,” LaPiere concluded that polls about social attitudes often reflect how respondents feel rather than how they’ll actually behave.

Subsequent research has repeatedly confirmed this gulf between what people say they will do and what they actually do when it comes to treatment of certain groups. In the 1970s, surveys suggested that military officers would resign if women were admitted to the service academies. Those who opposed the change used the data to fight women’s inclusion, warning that the military would suffer a fatal blow. But when women were admitted anyway, virtually no one left as a result.