Event
Pioneer Centre for AI Talk: Shane Lubold
Location
Date
Title
Network Inference from Sampled Data
Abstract
Networks play a key role in many social science applications, yet collecting full network data is often too expensive or time-consuming or can’t be collected due to privacy reasons. A cheaper and easier type of network data called aggregated relational data (ARD) collects responses to questions of the form “How many people do you know with trait X” for various pre-chosen traits. ARD does not collect individual links between individuals, only the number of links between the respondent and pre-chosen groups. In this talk, I will discuss applications of ARD to two important network problems. The first problem deals with a common class of network formation models known as latent space models, in which nodes lie on an unobserved latent space and the probability of connection decreases as the distance between the nodes increases. We show that ARD is sufficient to estimate the properties of the latent space from ARD about the underlying graph. The second problem deals with running randomized controlled trials (RCTs) in the presence of network-based spillovers. I will show how to use ARD to effectively conduct RCTs on networks and provide ARD-based estimators of important causal quantities, such as the global average treatment effect. I will then conclude by discussing future directions where ARD might be helpful. This is joint work with Tyler McCormick (University of Washington statistics) and Arun Chandrasekhar (Stanford economics).
Bio
I am a PhD candidate in the statistics department at the University of Washington. I am advised by Tyler McCormick and Arun Chandrasekhar. Previously, I received my BS in math from Arizona State University. I have worked with Yen-Chi Chen in the UW stats department and Clark Taylor at the Air Force Research Lab. I am supported by an ARCS fellowship.