STA 376: Advanced Modelling and Scientific Computing

Instructor: Merlise Clyde

Office Hours

TBA

Prerequisites

Basic statistical theory, linear models, and Bayesian modelling at the level of STA214, STA215 and STA244. Some experience in a lower level programming language such as C or FORTRAN and familiarity with UNIX workstations is assummed.

Course Description

An introduction to advanced statistical modelling and modern numerical methods useful in data analysis, model exploration, inference, and prediction, with particular emphasis on Bayesian analysis. Topics include: approximate methods of integration (Normal approximations, Laplace), random variable generation, Monte Carlo methods of integration, methods of obtaining samples from posterior distributions (Gibbs sampling and other MCMC methods), diagnostics, variance reduction techniques, and optimization methods. Methods will be illustrated on examples from hierarchical models, linear and generalized linear (additive) models, variable selection, outliers, missing data, nonconjugate problems, and experimental design. Students are expected to develop analyses using high level languages such as S-plus and BUGS, as well as low level languages such as C and Fortran using functions available in standand libraries.

References

  1. Tools for Statistical Inference, (3rd Edition) Tanner
  2. Markov Chain Monte Carlo in Practice, Gilks, Richardson, and Spiegelhalter
  3. Stochastic Simulation, Ripley
  4. Numerical Computation Using C Glassey In addition to these books, material will be drawn from the current literature as this is a rapidly developing field.

Grading

The course grade is based on periodic assignments and a project. By the end of Fall Break, students should have identified a topic for the course project. The project should use techniques developed in the course applied to a problem of scientific interest, using real data (of course). A classroom presentation and a written report on the project is due on December (dates TBA).


clyde@stat.duke.edu - 9/1/97