CMSA Member Seminar: An Introduction to Statistical Inference for Partially Identified Models

Classical statistical inference often relies on asymptotic approximations that are valid for each fixed data-generating process but may fail to hold uniformly over a class of relevant distributions. Starting with the familiar t-test, I will illustrate why this distinction matters and discuss the additional challenges that arise when the parameter of interest is only partially identified, i.e., when the observed data and maintained assumptions do not determine a unique parameter value.

Lunch will be provided.