Bounding causal effects for ordinal outcomes: partial identification and objective inference
We consider the problem of causal inference for ordinal outcomes, in which the estimands are often unidentifiable. Instead of invoking strong modeling assumptions, we take the objective approach of partial identification and derive closed form expressions for the sharp bounds of various estimands. Our results are assumption free, scientifically meaningful and extremely easy to compute and interpret. We illustrate our results through simulation studies and a real data example.
Title: Transelliptical Graphical Models: Theory and Computation
Abstract: We introduce the theory and computation of a semiparametric approach for estimating high dimensional graphical models. There are two main themes of this talk: semiparametric sparsity and statistical optimization. We illustrate the concept of semiparametric sparsity through the transelliptical graphical modeling, and illustrate the concept of statistical optimization through the theoretical analysis of a pathwise coordinate optimization algorithm named PICASA.
(Random Matrix and Probability Theory seminar: flyer)