Conformity model outperforms opinion averaging in real-world tests
Researchers say a model built around popular clusters of opinion fit five empirical scenarios better than a widely used averaging approach.
By Priya Raghavan · Science Reporter
3 min read
A new study reports that people may be better modeled as shifting toward popular clusters of opinion rather than blending every view into an average. The finding matters because many theories of how beliefs spread still rely on an averaging model that can be strongly affected by outliers, according to the researchers.
Kaleda Denton, a Complexity Postdoctoral Fellow at the Santa Fe Institute, worked with SFI External Professor Marcus Feldman and independent researcher Jonathan F. Johannemann on the study, published in Proceedings of the National Academy of Sciences. The work tests a conformity model that Denton and colleagues had previously developed against real-world data.
The older and widely used French-Harary-DeGroot model, commonly called the DeGroot model, treats social influence as a weighted average of the opinions people encounter. In contrast, Denton’s model assumes people often give more weight to common positions and less weight to unusual responses.
The Santa Fe Institute described the difference with a cooking example: if five people recommend 15 minutes per pound for roasting a turkey and one recommends 33 minutes, the DeGroot approach would pull the answer toward the average. Denton’s conformity model would place more weight on the shared answer from the larger group.
Testing conformity against averaging
In the new paper, the researchers revised the model so it could account for both a person’s own beliefs and social information, meaning what that person sees about other people’s preferences, according to the Santa Fe Institute.
They compared the updated model with the DeGroot model using an empirical data set that included five different scenarios. The conformity model frequently fit the data better than the DeGroot model, the institute said, including in tests where two of the conformity model’s four variables were fixed to make the comparison more even.
Feldman said Denton’s results showed the new model fits the tested data better than earlier approaches based on averaging. Johannemann said he was most encouraged by the model’s performance when fewer observations were available.
Denton said the improvement is partly tied to how the model reduces the influence of outliers. She noted that an unusual response in a data set may come from someone who misunderstood a question, according to the Santa Fe Institute.
Implications for opinion dynamics
The researchers also used the model to generate theoretical predictions across several population structures. Those included complete networks, where every person is connected to every other person; static networks, where people start with equal chances of forming connections that then remain unchanged; and adaptive networks, where people change ties based on who they observe to have the lowest error.
The Santa Fe Institute said the findings could prompt researchers to revisit theoretical work in opinion dynamics that has relied on the DeGroot model. Possible areas include information diffusion, emergency decision-making and the formation of consensus.
Feldman said conformity has not always received enough attention in studies of cultural dynamics. He said the new papers put conformity in a central role in research on cultural change.
The paper is titled “Conformity to popular, not average, opinions: Models, data, and evolution.” It was published in Proceedings of the National Academy of Sciences with the DOI 10.1073/pnas.2530712123.
This story draws on original reporting from Phys.org.