Seminars: Dept Seminar

Model-free Inference Problems with Dependent Data

Abstract:

This talk will give an overview of my research in developing nonparametric methods for dependent, or correlated, data collected over time or space.  In particular, there are three computer-intensive methods that generate “nonparametric likelihoods” to provide answers to a large class of statistical problems without stringent model assumptions on the data-generating process, these being 1. the bootstrap, 2. subsampling, and 3. empirical likelihood.  I have had an interest in investigating all three approaches over various inference scenarios with time series and spatial data.  I aim to compare the philosophical underpinnings of these methods as well as describe a research advancement for each.