Online social networks have become key tools in the daily existence of most Americans: we use Twitter to get updates on breaking news, LinkedIn to search for job opportunities and Facebook to reconnect with old friends. But what if social networks could help with something fundamental to human health: our physical fitness?
That’s what Greg Ver Steeg, a research assistant professor in the USC Viterbi School’s Information Sciences Institute and Department of Computer Science, and his research team started wondering when a 2007 study suggested that obesity might be contagious.
“If we can ask whether bad things like obesity can be contagious, can we ask the same questions about good traits, like physical fitness? Is it just a matter of designing the right social network and incentives?” Ver Steeg said.
After partnering with Google and Evidation Health, a company that aims to quantify health outcomes using digital technology, Ver Steeg had the opportunity to dive deeper into the issue. The collaboration’s recent paper, “The Spread of Physical Activity Through Social Networks,” was presented April 5 at the international World Wide Web Conference in Australia.
The study, which monitored 44,468 Fitbit users, found that people with larger and more active social networks have higher levels of physical activity. Specifically, for each additional social tie, participants walked an average of 6.5 more steps.
Ver Steeg’s role in the study was to help determine whether the correlations observed were also causal. The results were mixed—while the group found some evidence for a causal link, they couldn’t rule out all other possible explanations. In particular, it was difficult to discount correlations related to the changing environment; for example, maybe good weather causes people to spend more time outside and get fitter as a result.
“There’s this mathematical question about causality. It’s especially complicated in this case because we’re looking at human behavior, and there are so many factors that influence human behavior,” said Ver Steeg.
His goal was to create a mathematical model that would rule out homophily—the idea that people who are active will tend to be friends with other active individuals—as the main reason for the observed correlation.
The challenge is to rule out hidden variables that the study’s authors could never measure or anticipate, which could theoretically cause two users to become friends and simultaneously impact their physical fitness. Ver Steeg uses the example of two participants joining a secret pie-eating club: it’s something the researchers wouldn’t think to ask about, but it could be an external factor influencing the correlation between friends and fitness level.
Ver Steeg is the first researcher of online social networks to create an algorithm that can rule out hidden homophily. He says he drew inspiration for the model from his Ph.D. work in quantum physics.
“Einstein didn’t believe in the property of entanglement, which states that distantly separated quantum particles can interact, preferring to believe that hidden factors must be involved. But then John Bell developed an inequality that ruled out hidden factors, proving that quantum entanglement is real and responsible for the correlations we observe,” he said.
“I took the idea from quantum physics and changed it, so the latent variables were instead about hidden factors that might affect human behavior,” said Ver Steeg.
In addition to exploring the question of causality, the study also looked at the relationship between social networks and physical activity in the context of chronic conditions. One of the key findings was that the increase in activity among those with larger social networks was even more significant for users with chronic diseases, specifically depression and diabetes. These users walked 36 additional steps for each new tie, compared with 6.5 steps for users overall.
“When you’re burdened by a chronic condition, your social network may have an especially high influence on your behavior,” said Luca Foschini, cofounder and chief data scientist at Evidation. One possible explanation for the enhanced effect is that for users with chronic diseases, there may be more overlap between online and offline friend groups. Foschini says the next step is to validate the results with a prospective study.
The study can also help inform design choices for social features in mobile apps.
“A thing that you don’t see much covered by research in behavior change is the importance of the user experience. How are people engaging with their friends? How is that interaction coded in the app?” Foschini said. “Those are the real drivers of behavior.”
Collaborating with computer scientists such as Ver Steeg is one way Foschini hopes to turn results into technology that can positively impact people’s health.
“When I started at USC, I got into machine learning, and the number of interesting problems that you can have an impact on immediately is huge,” Ver Steeg said. “If you have the right set of mathematical tools, you can really have an impact right away.”
Published on April 21st, 2017
Last updated on October 26th, 2018