Unpacking peer effects

Peer effects revolve around the idea that your peers - be they friends, colleagues, classmates or otherwise - might have an effect on your outcomes and behaviour. For example, the number of beers we drink on a Friday night might strongly depend on the drinking behaviour of our friends.

Economists have long been interested in these externalities in the area of education, focussing on how a student’s classroom peers could impact their grades. The motivation for this interest is easy to understand. Simply by reorganising students among classes and changing peer groups, it could be possible to increase overall student performance without using additional expensive inputs such as new computers or extra teachers.

Despite this long-held interest, attempts to harness peer effects to improve student’s grades have so far been unsuccessful. A central reason for this is the lack of evidence on the mechanisms driving peer effects. For example, if student-to-student interaction is crucial for peer effects, a policy aiming to generate these spill overs would fail if students are punished for chatting in class. Our aim is to unpack the mechanisms behind commonly observed peer effects, and specifically the degree to which they depend on social proximity between students.

'Do you think about who you are going to sit next to in the classroom? If not, perhaps you should.'

We study peer effects at a large European university, focusing on the impact of tutorial peers on student’s first year grades. To explore potential mechanisms we exploit the institutional manipulation of the social proximity between students and their classroom peers. Each student can divide her classroom peers into a group of close and distant peers. Interaction and bonding is encouraged between a student and her close peers via a series of informal meetings, while no social proximity is fostered between a student and her distant peers. By random assignment of a student’s close and distant peer group, and putting both of these groups in a tutorial group together, we experimentally test for the importance of social proximity for peer effects.

We find the existence of positive peer effects; the smarter your tutorial peers, the better your grades in the first year. Importantly, we show that these spill overs originate from students' socially proximate peers only. In other words, the only peers that are important for your grades are those that you are acquainted with and interact with. Supplementary data suggests that these interactions involve collaborative studying that occurs at university, outside of the classroom. While strongest in the first block of the year, the beneficial effect of students’ close peers diminishes over time, and is completely absent by the end of the first year. Using data on how students choose their peer groups, we argue that these vanishing peer effects can be explained by students gradually making new friends while at university, and drifting away from their initially assigned close-peer group. Students appear to have a preference to form groups with people who are similar to them in terms of gender, ethnicity, and other characteristics.

Apart from its importance for understanding the dynamics of peer effects, this analysis provides a rare insight into the degree to which friendship groups can be institutionally manipulated against the formation of homogeneous subgroups based on gender, ethnicity, and other characteristics. Our analysis suggests that, while shared characteristics are the most important factor in friendship formation, manipulated social proximity can, to some extent, encourage diversity on university campus.

Our findings carry both good and bad news for those wishing to improve student outcomes using peer effects in similar settings. Encouragingly, it appears that in the short term social proximity between students can be engineered in order to increase grades. On a less encouraging note, we also find that any gains are likely to be short lived, given that over time students increasingly sort themselves into more homogenous subgroups. Social interactions appear to be too powerful to be constrained by a one-time manipulation of peer groups, and thus it is doubtful that policies aiming to exploit peer effects will succeed in the longer term.