In survey sciences and global health, raking is the problem of adjusting estimates to match marginal constraints and observations. We present raking from an optimization perspective, highlighting the important role that entropic distance plays in the classic problem and algorithms. We then show how the optimization lens makes it easy to add long-sought-after functionality to the framework, including a unified approach for raking data across multiple dimensions; incorporating importance weighting so estimates are adjusted differentially; efficiently propagating uncertainty; raking bounded estimates such as prevalence; and imputing missing data.

Refreshments will be served preceding the talk, starting at 2:45.