3# bridog
I think it is just the same as in linear models, because it is about the relationship among independent variables.
1. High correlation between any two independent variables suggests multicollinearity.
In SAS, the CORR procedure is used to computes Pearson correlation coefficients. PROC CORR statement invokes the CORR procedure and specifies the data set to be analyzed. The VAR statement identifies variables to correlate and their order in the correlation matrix.
proc corr data=one;
var Tmax hmin wmax wa ev p tmm ;
run;
2. We speak about strong multicollinearity if the largest VIF, variance inflation factor, is larger than 10 or if the mean of the VIF values is considerably larger then 1.
In SAS, the VIF option for MODEL statement in PROC REG requests the calculation of the variance inflation factors.
proc reg data= one;
model fa= Tmax hmin wmax wa ev p tmm /vif;
run;
3. The eigenvalues of the correlation matrix is another tool to detect multicollinearity. If there is perfect multicollinearity, some of the eigenvalues are zero. Near collinearity is associated with small eigenvalues. In SAS, the COLLINOINT option for MODEL statement in PROC REG provides analysis of the correlation matrix for the original data.
proc reg data= one;
model fa= Tmax hmin wmax wa ev p tmm /collinoint;
run;
扫码加好友，拉您进群
全部版块
我的主页
收藏
