The Diagnostic Research And Influence Analysis Of The Multi-Collinear Relations Under The Complicated Data

The question of multi-collinearity under the complicated data that this work is studied is the main situation violating GM assumes the condition. Multi-collinearity refers to that there exists one or several approximate linear relations among the explainvariable X₁?X₂......X_k, XC=X₁C₁+X₂C₂+......X_KC_K≈0. When rank(X)1,......X_k, but the multi-collinear relations exist; a variable must beexpressed by other explained variables among X₁,...X_k. That if the other variables are in the model make the estimated property of model parameter bad. The model simulation and prediction can not object the true feature of thing in fact.The formulation and influence mechanism of the multi-collinear relations is very complicated. It is not only correlated with the relation of structure of explaining variables, but also with the form of the model, the variable distribution, lagging influence, the difference of the sample information, unusual value and the error perturbation way.This text has introduced six kinds of main diagnosis and measured method:â‘ Characterized analysisâ‘¡Condition indexâ‘¢Variance information factorâ‘£Diagnostic method of the determinant⑤Coefficient law of the judgementâ‘¥Informal method of object judgement and provided the real diagnosis case. The improvement of the estimate of the main composition is provided and the concept of Ridge Regression and applicated case are detailed explained. The text has also provided the inherent unity of several kinds of estimation methods, and the improvement and practice of k-value selection method in Ridge Regression by living examples.In fact, there are a lot of produce reasons of multi-collinear relations. The purpose of this text lies in the object explanation of the produce reasons of multi-collinear relations relative to model structure relations and data quality. The data nature is the inherent factors leading to the multi-collinear relations which break into the traditional idea of produce reasons of multi-collinear relations from information overlap among explanation variables. The warned person who make model should carefully analysis and portray the model from the model diagnose and influence analysis in micro angle etc. and should set up the true, believed, sane practical model finally. The "spurious regression problem" can be avoided.