SPHERING AND ITS PROPERTIES

Guoying Li, Jian Zhang, S Academia

Sankhya. Series A · 1998 · 19 citations · 5 references

Concepts

Abstract

SUMMARY. Centering and sphering is an intuitive approach to remove, in a sense, loca tion, scale and correlation structure in data sets and to force us to examine other aspects of them. It is frequently applied in data analyses. This paper is intended to discuss properties of sphering procedures, such as affine (including orthogonal and lower triangular as special cases) equivariance/invariance, application to projection pursuit (PP) and asymptotic behav ior. In particular, the three commonly used sphering procedures, named LTS, SRS and JFS, are studied. It is shown that all sphering methods in a PP-after-sphering procedure results in the same optimal projections. It is also shown that the sphering matrix of JFS is inconsistent, whereas, those of SRS and LTS not only are consistent but also have asymptotic distributions.

References

5