Final Project: Due May 9th,1997

The project must be composed of three parts:

• Full description of a statistical method that uses the computer in an intensive way, why it is useful in a situation a classical method isn't etc.... This must be explanatory and well documented, with references to the literature in scientific format.

  Length: 6-10 pages.

• Program implementing the method written in either Splus or Matlab. The program must be tested on a simulated data set with known structure to show that it works.

• Full data analysis example, this must show a data set that you found on the web or in a book where the method you chose is relevant. The data must be documented and the output must be explained and interpreted.

Choice of method:
The method must not be already completely available in the software you choose to use. Here are some examples, if your choice is different, send me email so I can tell you in advance if it's OK.

1. Equivalent to CART building with Matlab.
2. Non parametric tests for anova and manova situations.
3. Approximate counting- I will not do it in class, but I can meet with you for an hour and explain it, this is useful for doing testing in computer intensive situations.
4. Moving Block Bootstrap for dependent Data.
5. Construction of clustering trees with Matlab. (Parsimony based or distance based).
6. Minimal Spanning Tree with Matlab. Friedman Rafsky test for comparing two multivariate populations.
7. Spatial analyses with Matlab, inspire yourself from Brian Ripley's spatial functions available in Splus.
8. Comparisons of neural nets (available in Matlab as its own toolbox) and non parametric regression as practised by statisticians.
9. Monitoring of convergence for Metropolis type algorithms.