Milad Eftekhar

Working alone, a typical learner has a low probability of completing an online course. When part of a team, however, his likelihood of success increases by a factor of 16. This makes teams an important tool for driving completion rates in online courses. But not all teams are created equal. Learners know this, which is why, when offered a choice, they are highly selective about whom they work with. My research with Farnaz Ronaghi and Amin Saberi sheds light on the selective biases of online learners, and attempts to identify strategies for optimizing team formation.

Our study focused on 11 NovoEd courses offered over the summer of 2014. Courses ranged in size from 200 to 25,000 learners, and lasted between four to six weeks. All courses took place online and were structured around team-based activities. Learners could ask to join a team or create their own. Team leaders, in turn, could use the learners’ profile data to invite other learners to join their team or to decide whether they should accept or reject requests for membership. Profiles included data about a learner’s level of education, location, age, and (depending on the course) skillset.

We found that learners tend to associate with learners similar to themselves. This tendency, known as homophily, led to groups formed along lines of similar age, educational achievement, and geographic proximity. Noteworthy, however, is that we did not find any statistically significant bias for gender.

The more interesting observation is that among successful organic teams (teams that possess a high fraction of members who acquired a statement of accomplishment at the end of the course), we also observed a high degree of diversity (heterophily) in terms of participants’ primary skill sets. In other words, multidisciplinary teams with more diverse skill sets were more successful than the rest: they sought to form multidisciplinary teams by seeking teammates with complementary skills. Homophily (across age, location, and education level) and heterophily (across skillset) proved to be an effective strategy, leading to more successful teams.

Our study concludes by offering several algorithms for efficiently allocating learners into high-performing groups based on their profile data. The findings are published in the paper Team Formation Dynamics: A Study Using Online Learning Data.

Milad Eftekhar is a PhD candidate in Computer Science at the University of Toronto. Milad’s research focuses on analyzing users interactions in online social networks in order to extract information that is helpful in various applications such as marketing and information dissemination.