Model redefining conformity excels against real-world data

Imagine you poll your friends on how many minutes per pound to roast a turkey. Five respond with 15 minutes; one answers 33 minutes.

The most popular model of conformity, the French-Harary-DeGroot model (or commonly, Degroot Model), assumes that you would carefully weigh all six recommendations, calculating a cooking time of 18 minutes per pound. But under a model of conformity previously published by SFI Complexity Postdoctoral Fellow Kaleda Denton and colleagues, you would disregard the outlier and move ahead with 15.

In a paper published in [month TK] in Proceeding of the National Academy of Sciences, Denton, SFI External Professor Marcus Feldman, and independent researcher Jonathan F. Johannemann build on this previous theoretical modeling work by testing the model against real-world data.

Feldman says the model’s fit to the data that Denton achieved in this paper “really proves that this model that she’s developed is much better than the previous ones that are based on averaging.”

In this new work, the researchers updated the model to account for both personal beliefs and social information, or what an individual observes about the preferences of others. They then tested it against the DeGroot model on an empirical data set containing five different scenarios. The conformity model often produced a fit that was superior to the DeGroot model—even when two of the model’s four variables were fixed for a more even comparison.

Finally, to determine whether the new model would produce useful theoretical predictions, the group tested it on various population structures, such as complete networks, where everyone knows everyone else; static networks, where each person initially has an equal chance of connecting with another person, and those connections do not change; and adaptive networks, where people update their connections based on who they observe to have the lowest error. 

Johannemann says he is most excited about how well the model works in cases where fewer observations are available. This ability stems in part from how it de-emphasizes outliers compared to trait averaging. “The DeGroot model is really affected by outliers,” Denton explains, “but the person who gave the outlying answer may have just misunderstood the question.”

Since much theoretical work on opinion dynamics relies on the DeGroot model, revisiting the conclusions using the new conformity model may yield interesting theoretical insights in areas such as information diffusion, emergency decision-making, and consensus formation. 

In the study of cultural dynamics, Feldman says conformity isn’t always given the emphasis it deserves. “These papers place conformity in a central position in studies of cultural change,” he says.