CSP 2017 Online Program

Estimating with Weights: Common Sense vs. Unbiasedness (303383)

Keywords: Importance sampling, weighted estimation, sensitivity analysis, bias correction

I'll talk about estimating with weights - how conventional wisdom and common sense conflict (the usual unbiased importance sampling estimate, and normalizing the weights, respectively), and a better estimate than both. This leads to an interesting connection between regression and weighted estimation. I'll talk about two other aspects of importance sampling - strategies for effective sampling when you can choose how to sample, and how to use weights for sensitivity analysis.