Proc. Natl Acad. Sci. USA 115, 9714–9719 (2018)

When people start a group discussion, this ideally leads to individuals learning from more knowledgeable members of the group, until the collective outsmarts individuals. However, research has also shown that group interaction, for example, on social media, can lead to formation of echo chambers and drive people’s opinions away from the ground truth, particularly on partisan issues. This poses the question under which conditions social interaction about strong partisan issues can lead to more people having better-informed opinions.

Credit: Smeagorl/Alamy Stock Photo

Douglas Guilbeault, at the University of Pennsylvania, and colleagues conducted an experiment to establish under which conditions groups of liberal and conservative participants would perform best in estimating trends in climate change, following structured online interaction. The researchers found that the best performing group was made up of liberals and conservatives who did not know about each other’s political leanings. If groups only contained subscribers to one ideology, or if people in mixed groups knew the political leanings of other members, then partisan bias led groups to perform significantly worse.

This study represents an advance into understanding the conditions that make human social learning successful, even when people initially show partisan bias. These results for climate change suggest that it will be important to see whether similar mechanisms can improve decision-making on other partisan issues.