ISyE Seminar Series: Sean Taylor
"Dynamic, Personalized Surveys via Adaptive Matrix Sampling"
Presentation by Dr. Sean Taylor
Wednesday, October 3
Surveys represent the best chance online platforms have to efficiently gather feedback from users. In an era of declining response rates, we seek to efficiently gather the most meaningful data possible from respondents who are often unwilling to answer long surveys. One way to improve efficiency is to leverage rich sets of covariates from large data sources to estimate response models and borrow information across respondents and questions. We show how survey response models can be used shorten surveys without decreasing information, via an adaptive matrix sampling procedure that downsamples questions into the most informative subset for each respondent. Our proposed question selection method optimizes for variance reduction while incorporating side information about respondents. As a respondent answers questions, their posterior response distributions are updated in an online fashion. The efficiency gains of our approach enable the researcher to terminate sampling early when desired precision is reached or evolve a survey over time by adding and removing questions.
Sean J. Taylor is a social scientist and applied statistician on Facebook’s Core Data Science team. Prior to Facebook, he earned his PhD in Information Systems from NYU’s Stern School of Business. He specializes in using machine learning methods and randomized experiments for measurement, prediction, and policy decisions. Sean’s research ranges from studying online social influence, viral marketing, and social networks to measuring how sports fans behave and the impact of data science on decision making in organizations. He is also an avid engineer who enjoys putting academic research into practice by building open source software for data scientists.