group selfie photo by AllGo on Unsplash

Social Influence Online: Do Friends Make Friends Pay?

Do friends influence one another to buy products, embrace beliefs, and adopt behaviors? Whether it’s among family members, peer groups, or social circles, influence is both naturally occurring and something people try to harness and use. It can be incredibly powerful and equally elusive.

The Study

If two friends buy the same product a week apart, did one friend influence the other’s decision to buy? Or do they simply have similar taste? Or did they both see a promotion for the product?

Teasing out cause-and-effect evidence of social influence online is challenging when so many alternative explanations are possible. To tackle this question, University of Minnesota researchers Ravi Bapna and Akhmed Umyarov conducted a randomized experiment with Last.fm, a music-listening social network. 

The experiment used a multi-threaded data crawler to track four million users over two years, resulting in a dynamic dataset with more than a billion rows. Testing included a control and treatment group to determine whether or not peer influence was present in the network. 

The study showed that when a person becomes a paid subscriber of the Last.fm service, the odds her friends will buy a subscription increase by 50% due to peer influence. 

man wearing headphones

Average users were 50% more likely to buy a subscription when an online friend was a paid subscriber, showing peer influence in the social network.

Bapna & Umyarov paper

Impact

This research can help “freemium” businesses convert more users into paid subscribers. The same techniques and findings can help any kind of business develop more effective digital marketing strategies. Beyond product marketing, greater understanding of influence online may change political campaigns, public health education, and societal movements. 

Methods & Tools

  • Large scale randomized trial with no self-selection of subjects for more valid, generalizable findings
  • Analysis using SAS and R for visualization and logistic regression