Social learning helps humans navigate the world by learning the consequences of actions through the behavior of others. Humans are also quite sensitive to group membership, in that we tend to place ourselves within social groups with which we identify (e.g., gender, race, nationality) and membership in these groups can motivate our beliefs and behaviors. New research published in Psychological Science suggests that people are more likely to use social learning to copy members of their in-group compared to members of their out-group.
“In recent years, there’s been a surprising proliferation of misinformation, fake news, and opposition to vaccination, particularly on social media,” said study author Marcel Montrey, a PhD candidate in McGill University’s Department of Psychology.
“Why these beliefs and behaviors have spread so virulently is puzzling because they often seem harmful to the people who adopt them. One potential clue is the emphasis placed by anti-vax communities and state-sponsored disinformation campaigns alike on group identity and ‘us versus them’ language. This got us interested in how social groups influence people’s copying. Are we more likely to spread information (and potentially misinformation) simply because it comes from ‘one of us’?”
Montrey and his colleague Thomas R. Shultz conducted two experiments using the “minimal-group paradigm,” which is a strategy of arbitrarily assigning people to groups, to establish ingroups and outgroups in both samples. “This prevents groups from having any shared history or unifying features beyond membership and ensures that participants have no personal stake in their group’s success or reputation. Surprisingly, meta-analysis reveals that such groups often elicit in-group favoritism on par with that found in preexisting groups,” wrote the researchers. Thus, participants were randomly assigned to one of two color coded groups to create the in-groups (same color) and out-groups (different color) in both studies.
Participants then played a guessing game where a simulated rabbit chooses one of two nests to inhabit, and participants guessed which nest the rabbit would choose. In each round, participants randomly chose 3 previous participants to observe and decided whether to use a rabbit-finding machine, which was described to have a two-thirds likelihood of correctly guessing which nest the rabbit would choose.
Next, participants were shown which nest the 3 previous participants chose (social learning) and which nest the machine chose (individual learning) before they made their guess. After every 5 rounds of guessing, participants saw their score and how it compared to the choices of previous participants. Importantly, no information was given about the accuracy of the score. Before and after the game, participants compared the reliability of both color groups.
The researchers assessed intergroup copying and attentional biases by measuring which group participants agreed with or observed more often. In other words, participants who agreed more often with in-group members were considered biased toward copying the in-group. If participants agreed equally with each group, they were considered unbiased. These same criteria were used for attentional biases.
“Participants showed an in-group bias in copying. On average, 62% of agreements were with in-group members, whereas only 38% were with out-group members. Likewise, participants showed an in-group bias in attention. On average, they observed in-group members 61% of the time, but out-group members only 39% of the time,” researchers wrote.
Interestingly, despite these in-group bias observations, most participants did not rate the in-group to be more reliable either before or after the game than the out-group. This suggests that participants were either not aware of their biases or were responding in such a way to appear less biased.
To address limitations of this experiment, Experiment 2 differed in a few key ways. First, Montrey and Shultz randomly assigned participants to choose either one, two, three, or four previous participants to observe. “One possible explanation for the in-group biases in copying and attention is that participants repeatedly observed an odd number of individuals. This forced them to over-represent either the in-group or the out-group within rounds, which could elicit a preference for in-group members,” researchers posited.
Second, instead of having participants rate the reliability of each color group, they rated each group for warmth and competence. To create an overall intergroup bias score, researchers subtracted participants’ out-group warmth or competence ratings from their in-group warmth or competence ratings. Thus, positive values reflect seeing the in-group as warmer or more competent, and negative values indicate the out-group was seen as warmer or more competent.
Results indicated an in-group copying bias, which is consistent with Experiment 1. Interestingly, in-group copying bias weakened with the increasing number of previous participants that was observed. In other words, in-group copying bias was the strongest when participants observed one individual (71%), weaker for two (65%), even weaker for three (60%), and weakest for four (56%). “This suggests that the in-group-copying bias was not contingent on observing an odd number of individuals but instead grew with the scarcity of social information,” concluded the researchers.
“Researchers have known for some time that people prefer to copy members of their own social group (e.g., political affiliation, race, religion, etc.), but have often assumed that this is because group members are more familiar with or similar to each other,” Montrey told PsyPost. “However, our research suggests that people are more likely to copy members of their own group even when they have nothing in common. Simply belonging to the same random group seems to be enough. Surprisingly, we found that even people who rated their own group as less competent still preferred to copy its members.”
Montrey and colleagues were careful to address potential limitations in their studies by limiting similarity and familiarity with random assignment, by ensuring there was no advantage to copying either group, and by deterring intergroup competition by using an assessment of individual performance. But the study, like all research, includes some caveats.
“Human social learning is a complex and multifaceted phenomenon, where many factors other than group membership play a role,” Montrey explained. “For example, we know that people also prefer to copy successful, popular, or prestigious individuals, which is why companies advertise through endorsements. How do people’s various learning biases interact, and which ones are most important? Because these questions have only recently begun to be explored, the real-world relevance of our findings is still up in the air. A key question to answer is whether people prefer to copy their own group’s members even when doing so is riskier or more costly.”
The researchers suggest these findings can illuminate the proliferation of misinformation on the Internet and social media. “By offering such fine-grained control over whom users observe, these platforms may spur the creation of homogeneous social networks, in which individuals are more inclined to copy others because they belong to the same social group,” Montrey and Shultz wrote.
Future research could also examine the potential benefits of the in-group copying bias.
“While it’s easy to see the potential downsides of people preferring to copy their own group’s members, there are likely advantages as well,” Montrey explained. “This bias could help groups coordinate their behavior and enhance cooperation. It could also have helped our species maintain cultural diversity, which is thought to be a key driver of technological evolution.
The study, “Copy the In-group Group Membership Trumps Perceived Reliability, Warmth, and Competence in a Social-Learning Task“, was published December 23, 2021 .