Statistics Seminar（2014-04）

Topic:Making Bayesian Hierarchical Models More Robust Through Benchmarking

Speaker:John Bryant, Senior Researcher. Statistics. New Zealand

Time:Thursday, 13 March, 14:00-15:00

Location:Room 217, Guanghua Building 2

Abstract：Benchmarking is a technique for making models more robust. Benchmark values are calculated for aggregate quantities such a overall means, and then disaggregated estimates are constrained to agree with these values.We have developed a Bayesian interpretation of benchmarking. In the presentation we will use analytic results from a simple model to illustrate the properties of benchmarked estimators. We will then present results from ongoing work investigating the effectiveness of Bayesian benchmarking using synthetic and real datasets. Results so far suggest that Bayesian benchmarking can indeed provide robustness. They also suggest that it is important to incorporate uncerrtainty in the benchmarks themselves.

About the speaker：John Bryant has a PhD in Demography from the Australian National University. He has worked at the New Zealand Treasury, and at universities in Thailand and New Zealand. He is currently a researcher at Statistics New Zealand. He is working on new Bayesian methods for demographic estimation and forecasting, and for exploiting administrative data.