Identification Of Peer Effects

Author : Dean Eckles
Publisher :
ISBN 13 :
Total Pages : pages
Book Rating : 4.:/5 (795 download)

DOWNLOAD NOW!

Download or read book Identifying Peer Effects in Online Communication written by Dean Eckles and published by . This book was released on 2012 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: Understanding how the behavior of individuals is affected by the behavior of their peers is of central importance for social and behavioral science. Most of the credible evidence about peer effects currently comes from agent-based simulations and small experiments in constructed social environments, as until recently investigators have been largely unable to quantitatively study these processes at a large scale and in situ. Despite massive new data sources, credible identification and estimation of peer effects has remained a largely unsolved methodological problem. Peer effects can produce clustering of a behavior of interest in social networks, but so can other processes, including homophily and the consequences of prior peer influence. This presents major challenges for identifying, bounding, or even determining the sign of peer effects in observational studies. While experiments are an appealing alternative, it is not often possible to conduct field experiments that identify peer effects by directly manipulating peer behavior. Sometimes investigators can manipulate particular mechanisms of peer effects or manipulate network formation processes, but not only are these situations rare, but, these experiments do not always identify effects of primary interest. Thus, better understanding of the promise and limitations of quasi-experimental and observational methods are needed. This dissertation evaluates and advances the methods for identifying peer effects, with particular focus on applications to studying online communication behavior. We review of theories of peer effects multiple fields, highlighting their common interest in learning about individuals' exposure--adoption functions. We systematically review the assumptions required for identification of peer effects. We then use reanalysis of a large experiment that blocked a critical mechanism of peer effects as a constructed observational study; that is, we use methods for observational data to reanalyze the experiment while replacing the experimental control group with a nonexperimental control group. Treating the experimental results as a gold standard, we find that propensity score methods can eliminate most of the confounding bias, at least when good measures of closely related prior behaviors are available for use in the propensity score model. These results provide evidence that observational methods may sometimes provide informative estimates of peer effects. We also present the use of peer encouragement designs as a promising identification strategy for peer effects. In peer encouragement designs, experimenters manipulate assignment of an individuals' peers to an encouragement to the focal behavior. This peer encouragement is a shock to peer behavior that in turn affects ego behavior; we argue such encouragements often make valid and informative instrumental variables for peer effects. We illustrate these methods with a large encouragement design on Facebook in which individuals were randomly assigned to be prompted to express gratitude on Thanksgiving Day 2010. We find evidence of positive peer effects in expressions of gratitude via Facebook. A purely observational analysis produces smaller estimates of peer effects than the instrumental variables analysis using the peer encouragement design, likely reflecting differences in the population of individuals these analyses average over. We argue that encouragement designs are an underused research design for studying peer effects.