Here’s the abstract for my talk tomorrow for the 50th anniversary conference of the Harvard statistics department.

Some Open Problems in Hierarchical Models

Andrew Gelman, Department of Statistics and Department of Political Science, Columbia University

Hierarchical models have changed statistical practice and also how we think about statistics, sweeping aside various concerns both classical (for example, unbiased estimation) and Bayesian (for example, reference prior distributions). We discuss some ways in which hierarchical ideas unify seemingly opposing views of statistics and then consider some open questions involving how to construct families of models that are structured enough to be able to learn from data but not be so strong as to overwhelm the data. There’s lots of room for more discoveries in the Harvard statistics department’s next 50 years.