VARIANCE-COVARIANCE PERTURBATION ANALYSIS OF A LINEAR MODEL
Fengxiang Jin, B. Merminod, H. Dupraz, P-Y. Gillieron · Survey Review · 2003
In this paper the perturbation influence properties of the stochastic model (variance-covariance) in linear models is discussed in detail. Some very useful formulae are established about the variance-covariance perturbation influence on the model parameter's estimation. Then the influence properties are analyzed in different cases and illustrated by two examples. A very important fact is proved that the non-influence cases exist under some conditions. These conditions are carefully studied. The conclusions obtained are very helpful for people to well understand the function of stochastic model in the linear models, and to vividly work with them. The paper offers some very important theoretical results that should be considered for the weight function design in Robust Estimation.