Skip to content Skip to navigation

While Birds of a Feather May Stick Together, People are More Complicated, Says New MS&E Research

While Birds of a Feather May Stick Together, People are More Complicated, Says New MS&E Research

March 26, 2018

group conversing

Photo credit: istockphoto/simonkr

Conventional wisdom suggests that we make friends with people who are similar to ourselves: same age, same gender, same political affiliation. This concept in social science is called homophily, Latin for “love of same.”

Some attributes among friends show strong homophily, such as political affiliation. For instance, political conservatives frequently befriend other political conservatives. But different attributes, like gender, can sometimes show weak homophily. Unlike the adage from the classic rom-com movie When Harry Met Sally, men and women are friends, and some of them have a strong preference for friends of the opposite gender.

This phenomenon now has a name—monophily—thanks to new research and a statistical framework for binary attributes designed by MS&E Professor Johan Ugander and third-year PhD student Kristen Altenburger.

Monophily is Latin for “love of one” and describes the situation where there are individuals in a social network with extreme preferences for a specific attribute, regardless of their own identity. In a network with gender monophily, we might see individuals who mostly friend men, for instance, but those individuals would be equally likely themselves to be male or female.

Ugander and Altenburger’s paper, “Monophily in social networks introduces similarity among friends-of-friends,” appears this month in Nature Human Behaviour.

According to Ugander, the lack of gender homophily in social networks has long stymied scientists. While some research shows tendencies to make friends along gender lines, many other studies of social networks, particularly online ones, show weak homophily.

Weak homophily has traditionally resulted in frustrating data analysis. According to Professor Ugander, “A lot of machine learning methods for figuring things out about people harness this phenomenon of homophily.” Weak homophily makes it difficult for researchers to make key inferences about groups of people.

But according to Kristen Altenburger, their model overcomes that challenge. “Our work not only describes monophily and how to measure it, but also how to create a statistical framework that accommodates homophily and monophily for better overall predictions.”

Ugander first identified the lack of gender homophily in online social networks in 2011 during his time on Facebook’s Data Science team. “I carried that curiosity with me to Stanford. I saw it as an interesting challenge question: how well can you predict gender on social networks?” At the time, predictions went poorly.

But with these statistical modeling techniques, Ugander and Altenburger were able to zoom in on individuals with super-strong preferences for a particular type of friend (for instance, male), giving a richer view compared to what one sees when just focusing on homophily. Using these people, they were able to look at extended social networks (not just friends, but friends of friends) and make surprisingly accurate gender predictions.

In addition to their analysis of gender in online social networks, they also studied political blogs and the Noordin Top Terrorist Network to confirm that their method could apply to other binary attributes beyond gender—like liberal or conservative, or suspicious or not.

According to Professor Ugander, the increasingly large amounts of data with online social media platforms has given researchers more opportunities for testing hypotheses. “The Internet is a digital microscope of human behavior,” said Ugander. “We can ask much more sophisticated questions and get much more detailed answers than 20 years ago simply because we didn’t have data on this scale before.”

While the ability to better study human behavior through large data sets has improved, lack of privacy may be one unintended consequence of making better predictions on social networks. “These people with strong preferences end up spilling the beans on their friends,” said Ugander.

This work could also lead to a better understanding of how people form social connections. You can learn more about Professor Ugander’s work on his research page.