Data can help us make decisions to move society forward, but we have to make sure we’re coming to the right conclusions.
Christina Yu, assistant professor of operations research and information engineering at Cornell University, examines this.
Christina Lee Yu is an assistant professor at Cornell University in the School of Operations Research and Information Engineering. Prior to Cornell, she was a postdoc at Microsoft Research New England. She received her Ph.D. in electrical engineering and computer science from Massachusetts Institute of Technology and B.S. in computer science from the California Institute of Technology. A recipient of the 2021 Intel Rising Stars Award and a JPMorgan Faculty Research Award Yu’s research interests include algorithm design and analysis, high dimensional statistics, inference over networks, sequential decision making under uncertainty, online learning, and network causal inference.
Taming Network Interference in the Digital Age
In today’s digital landscape, data serves as the driving force behind decision-making across various sectors. Randomized control trials are often hailed as the “gold standard” for evidence-based insights.
However, the intricacies of our interconnected world have given rise to a phenomenon known as network interference. Picture this: an app developer conducts an A/B test to assess the influence of a new texting feature on phone usage. Seems straightforward, right? But what if this feature not only changes your texting habits but also has a ripple effect on your friend’s behavior? This complex interdependence between the test and control groups can result in misleading conclusions.
Data scientists are actively exploring ways to monitor and account for network interference. A ride-sharing company may test a new feature in one city and compare the results with those from a city miles away. Or scientists may use sophisticated mathematical models of contagion to estimate how a new vaccine’s efficacy might be influenced by the vaccination status of one’s neighbors.
In my own research, I’m dedicated to developing robust solutions for flexible models that accommodate personalized network effects, and providing practical recommendations for optimizing the experimental design and estimator. However, it’s crucial to acknowledge that these solutions rely on assumptions that are challenging to validate.
