Minimal Assumption Tests for Contagion in Observational Social Network Studies

Greg Ver Steeg
University of Southern California (USC)
Information Sciences Institute

Starting from the wrong assumptions, anything can be proved. Refusing to make any assumptions leads to nihilism. Scientists constantly seek balance between these extremes by making the minimal, reasonable assumptions that are nevertheless powerful enough to lead to interesting, falsifiable conclusions. The realm of social network studies presents a challenging case because human behavior is complex and assumptions are often difficult to justify. What are all the relevant variables that affect human behavior and the formation of relationships? Answering this question incorrectly can lead current tests for contagion to fail. I will discuss an alternate framework that allows us to test for contagion even in the presence of arbitrary, unobserved individual traits. Extending these tests reveals the more critical obstacle to identifying social contagion: non-Markovianity in human behavior.

Back to Mathematics of Social Learning