Stats 306A: Exponential Families and Data Analysis Bradley Efron Winter 2005

A Final Message

Dear 306a students,

You did very well on the final, as a group and individually too. The only problem that caused very much difficulty was #4. Two points were forgotten: if using Poisson regression, rather than logistic, the sample size enters as an offset, not a covariate; and one should test for overdispersion before interpreting the slope coefficient.

The tests are upstairs, the grades marked being for the quarter.

Brad Efron

Course Description

Exponential families are the great connecting idea of 20th Century statistical theory. They have also become a force in modern data analysis, particularly through Generalized Linear Models. This course is devoted to understanding the theory in the context of its applications.

Topics:

• Review of basic ideas
• Familiar and some not so familiar exponential families
• Generalized Linear Models and the analysis of deviance
• Survival Analysis, especially as it relates to exponential family modelling
• Over-dispersion, quasi-likelihood, and double exponential families
• Curved exponential families, missing data, and the EM algorithm

Announcements

• Gill's office hours have moved to Wednesday 4-6pm.
• The handout from the R session is available here.
• The solutions to the first homework are now available.
• We may send out an occasional message via the class email list, so if you are auditing the class and wish to be added to the list, please contact Gill.
• Sourav's office hours have moved to Fridays, immediately after class.
• First Meeting: Wednesday Jan 5, Sequoia 200.

Resources

• Homework Solutions
• Data Sets

Teaching Staff

Office Hours:	Professor Efron	Friday 4.30-6 in Sequoia Hall 132
	Sourav Chatterjee	Friday 2.05-4.05 in Sequoia Hall 206
	Gill Ward	Wednesday 4-6 in Sequoia Hall 216

Contact Us:

You can contact all the teaching staff at: stats306a-win0405-staff@lists.stanford.edu or each of us separately at brad, souravc or gward, each at stat.stanford.edu.

Admin

Lecture: MWF 1:15-2:05, Sequoia Hall 200

Prerequisites: An advanced statistics course at the level of 300A or 305.

Suggested Text:

McCullagh and Nelder "Generalized Linear Models" (but I will lecture from my own notes)

Grading:

The grade will be based on problem sets and a 3 hour in-class final.

Class web page: http://stats306a.stanford.edu