variables failing tolerance test - why?
ari
I am starting my studies in discriminant analysis using SPSS. I have
a small sample with 22 individuals and 50 variables. I used the option
"enter independent togheter" and many variables were eliminated
("variables failing tolerance test"). My doubts are:
1) why did SPPS do it?
2) what was the test that SPSS used?
3) which equations that SPSS did it use?
I thank in advance any help.
ari
Rich Ulrich
SPSS is doing fine. Your modeling is at fault.
Read up on "Variance inflation factor".
In the social sciences, you usually want to have 10 individuals
(a popular rule of thumb) for every variable that you try to
use in a discriminant function or regression. [A two-group
discriminant function is exactly the same as regression
with a 0/1 criterion, after you re-scale the coefficients.]
Consider: One continuous variable gives perfect prediction
when there are just two cases. You can add single cases
and single continuous predictors, and keep the same relation,
so that 21 variables ought to predict 22 individuals perfectly.
If there are tied values, then it might not work out -- but these
tests assume no-ties. So, 50 bombs. 22 would bomb.
You don't end up with any tests at all, in any case,
unless you have SOME degrees of freedom left over.
--
Rich Ulrich, wpi...@pitt.edu
http://www.pitt.edu/~wpilib/index.html