Re: Help with VIF calculation in regression

VIFk = 1/(1-Rk**2)

where Rk**2 or Rk square is the R-square of Xk on all other variables.

So it should not be a negative number.

VIF=10 which implies R-square = 0.9.

Another measure is condition number that is defined as

C.N.=square root of (max eigenvalue/min eigenvalue) of moment
matrix (X'X)
Usually C.N >20 indicates some collinear problems.

If you have SAS software, here is an example.

data test;
do j=1 to 100;
x1=rannor(1);
x2=x1 + rannor(1);
x3=6*x1 + rannor(1);
y=x1+x2++x3+3*rannor(1);
output;
end;
keep j x: y;
run;

proc reg data=test;
model y=x1 x2 x3/vif COLLIN;
quit;


HTH

.