Dear All,
I've been using step function to find me the best model.this basically works
by using AIC score fucntion that is implemented on step(). The problem I'm
facing with lots of variables on the model for example :
step(lm(x1~x2,x3,x4,......x13)) sometimes gives me a warning message which
is :
AIC=- inf
Coefficients:
(Intercept) wnt3.values wnt6.values wnt10b.values
wnt9a.values
2.3462 -0.4689 2.0730 1.2769
-0.2319
sfrp1.values wnt5b.values sfrp1.1.values sfrp5.values
fzd5.1.values
-0.2597 0.3150 0.3811 0.5926
-1.5567
fzd1.values fzd4.values fzd6.values fzd7.values
fzd7.1.values
0.6459 -2.3016 0.3636 NA
NA
fzd8.values
NA
Warning message:
attempting model selection on an essentially perfect fit is nonsense .
which stops the search.
Does this means that Residual Sum of Squares (RSS) equals to zero that
makes AIC goes to -inf .And how would I overcome this problem.Can I for
example find those that have strong correlation with x1 first and then use
AIC score to find me the best model among them,.
Regards
Adel,
[[alternative HTML version deleted]]