Peer Effects in a Randomized Setting

Dartmouth College, Hanover NH

Investigators

Abstract

9979260 Sacerdote We all rely heavily on our peers and colleagues in every aspect of our economic life. We use our peers to gather information, make labor market decisions, investment decisions, locational choices, and perhaps even to help determine our underlying preferences. But how can one begin to isolate and measure these peer effects? After all, along with making all of our basic economic choices, we also choose our friends. How can one separate out the effect of our peers on a given labor market choice from those aspects of our underlying preference function that made us choose the peers? This project takes a novel approach to this problem. Freshmen entering Dartmouth are assigned a dorm and 1-2 roommates using a random process. The random assignment effectively removes the selection problem which normally prevents researchers from separating out the actual treatment effect of the peer group from selection into that peer group. By construction, there is no correlation between students' pre-treatment characteristics and their roommates' characteristics. The hypothesis of the research is that in spite of this random assignment, a student's roommates and other dorm peers are an important influence on the student's behavior. The research focuses on peer effects on outcomes such as Freshman year GPA, future housing choices, membership in a fraternity, choices in foreign study programs, and athletic participation. Dartmouth College has agreed to provide access to its extensive database on its students. Dartmouth has a myriad of data on every student ranging from admissions data, to academic performance, to post-graduation and career choices. This will enable the project to match students together with their freshman year (randomly assigned) roommates and determine the correlation between students and their roommates. For example, if X is assigned a roommate with an unusually high SAT scores and high school GPA, is X more likely to have a high freshman year GPA ? Student confidentiality will be protected by stripping identifiers such as names, college IDs, and home addresses. Instead of student names and dorm names, codes will be substituted which will allow the PI to match roommates and dorm mates.