HEAD 2000, November 2000
Session 10. On Beyond $\chi ^2$ (and Bevington): Making the Most of Your Poisson Data
Workshop, Monday, November 6, 2000, 7:30-9:30pm, 8:30-10:00pm, Pago Pago Ballroom

## [10.01] Data Augmentation, Hierarchical Models, and Markov chain Monte Carlo

D. A. van Dyk (Statistis Department, Harvard University), Harvard Astrostatistics Working Group

In this tutorial we introduce several state-of-the-art statistical methods which can be used to solve numerous outstanding data analytic challenges in high-energy astrophysics. These methods are especially useful for high-resolution low-count data for which methods in common use (e.g., \chi2 fitting) are not appropriate. In particular these methods allow us to directly model the Poisson character of count data and avoid unjustifiable Gaussian assumptions. Thus, there is not need to sacrifice information by binning data to obtain a minimum count per bin or to subtract off background, thus avoiding the potential for negative counts and unpredictable results. The tutorial is designed to be accessible to statistical novices.

[Previous] | [Session 10] | [Next]