Skip to main content
Indiana University Bloomington

Events & News

Conditions for Identifying Causal Peer Effects in Social Networks Using Genes as Instrumental Variables

Seminar: Department Colloquium
Speaker: Felix Elwert, Ph.D. ()
Department of Sociology University of Wisconsin–Madison
Date: From 3:00 PM to 4:00 PM on 1 April 2013
Location: IMU Walnut Room

Abstract
Joint work with James O’Malley, J. Niels Rosenquist, Alan M. Zaslavsky, and
Nicholas A. Christakis (Harvard Medical School)


The identification of causal peer effects from observational data in social networks is challenged by latent homophily, which generically biases interpersonal associations (Shalizi and Thomas 2011). This paper investigates the range of structural conditions (i.e., the data generating models) under which it is possible to identify causal peer effects in social networks by using genes as instrumental variables (IV). In order to intelligibly catalogue and communicate the large set of models supporting IV identification of peer effects, we use Pearl’s (1995, 2009) directed acyclic graphs (DAGs). We employ two different IV strategies and report three main findings. First, using a single gene (or allele) as IV will generally fail to identify peer effects if the gene affects past values of the treatment. Second, multiple genes can identify peer effects using Brito and Pearl’s (2002) IV-set strategy if we instrument exclusion violations as well as treatment. Third, IV-set identification remains valid even under pleiotropy on observables, homophily on phenotype, inter-phenotype peer effects, and various types of population stratification. We apply our results to estimating peer effects of body mass index in the Framingham Heart Study

 


 

Calendar Options: