Social Network Endogeneity and Its Implications for Statistical and Causal Inferences

A broad of spectrum of disciplines have adopted social network data to examine relevant contextual issues in a wide array of fields. Yet, statistical methods to provide valid inference within the confines of endogeneity, or interlocking relationships between subjects within the social network, are woefully lacking. Traditional statistical models define relationships between variables, or measures of attributes within each subject, and as such are fundamentally flawed when used to model such endogenous relationships between subjects. Although various attempts have been made to address this problem such as exponential random graph and influence models, none has been successful in addressing this critical issue.  In this talk, we discuss a new approach for modeling between-subject attributes and utilize on the functional response models and the theory of theories of U-statistics and U-statistics based generalized estimating equations to provide valid inference about such models.  We illustrate the approach with models for social network densities and data from a real study on interpersonal interactions from a hard to reach study population involving lesbian, gay, bisexual, transgender and questioning (LGBTQ) youth.