Data Files

SERIES B
Statistical Methodology

An adaptive estimation of dimension reduction space
Y. Xia, H. Tong, W. K. Li and L.-X. Zhu
J. R. Statist. Soc. B, 64 (2002), 363 - 410

Files iMAVE.m, rOPG.m, rMAVE.m (and CVdn.m) are Matlab codes for the
methods proposed to estimate the dimension reduction space spanned by
the columns of B_0 in the following model

y = g(B_0^T X) + e.

1. Installation:
Please copy the file to a directory in your computer, e.g. c:\dimension\
2. Open Matlab.
3. Set the path to the directory in which the files are copied to, e.g.
>>addpath c:\dimension\ [enter]
4. Test: type "help iMAVE" or "help rMAVE" in Matlab prompt, e.g.
>> help rMAVE [enter]
(the following message should be displayed on your screen)

Searching the effective dimension reduction subspace of model
y = g(B^TX) + e
Useage: directions(x, y, h, d, g)
Input:
x --- expanaltory variables
y --- response
h --- bandwidth
d --- working dimension of the space
Output:
Directions B

-------------------------------------------------------------
Example
n = 200;
x = randn(n,4);
beta1 = [1 2 3 0]';
beta1 = beta1/sqrt(beta1'*beta1);
beta2 = [-2 1 0 1]';
beta2 = beta2/sqrt(beta2'*beta2);
y = (x*beta1).^2 + x*beta2 + 0.2*randn(n,1);
B = rOPG(x, y, 0.5, 2)
% estimation errors
B0 = [beta1 beta2];
error = B'*(eye(4)-B0*B0')*B

5. You can use these program now.

Howell Tong
Department of Statistics
London School of Economics and Political Science
Houghton Street
London
WC2A 2AE
UK

E-mail: h.tong@lse.ac.uk

• Dataset (dataset.zip, size - 36.7kb)

SERIES A
Statistics in Society

SERIES B
Statistical Methodology

SERIES C
Applied Statistics

SERIES D
The Statistician