Abstract:
|
The Internet industry has advantages in the ease of conducting randomized experiments. But the nature of the networked products and services that decision-makers want to learn about introduces important challenges. In this talk, I present two experiments in networks with either a novel design or analysis method. The first implements a peer encouragement design, in which an individual's peers (i.e. network neighbors) are randomly assigned to an encouragement to interact with the individual. We use the assignment as an instrumental variable to estimate peer effects and inform understanding of "virtuous cycles" in social media. The second uses new randomization inference methods to detect the presence of spillovers (i.e. interference, exogenous peer effects) from a randomly assigned treatment. This particularly experiment provides further evidence for the role of social networks in political participation in the 2012 U.S. Presidential Election.
This talk summarizes three papers that are joint work with Susan Athey, Eytan Bakshy, Robert M. Bond, James H. Fowler, Guido W. Imbens, Jason J. Jones, René Kizilcec.
|