My question is exactly the title : to whom can we report a problem with SAS ?
Below is an example. This problem is not really severe but somewhat dangerous (in fact I have just updated my example below after Aniko's comment; there was a confusion in the first version of this post).
Consider such a dataset:
> dat
tube position y
1 1 top 0.25602779
2 1 top 2.99327392
3 1 top 0.03673459
4 1 top -0.94515391
5 1 bottom 9.12947343
6 1 bottom 5.96666893
7 1 bottom 6.65291454
8 2 top -2.32616858
9 2 top -1.61491564
10 2 top -2.88930533
11 2 top -1.48685691
12 2 bottom 0.03474644
13 2 bottom 4.23073725
14 2 bottom 1.43776713
15 3 top 3.04525229
16 3 top -1.06611380
17 3 top 0.64097731
18 3 bottom 5.63571519
19 3 bottom 5.96779074
20 3 bottom 2.14091389
21 3 bottom 5.46937089
22 4 top 7.00724734
23 4 top 4.33632991
24 4 top 1.90765886
25 4 top 1.91688415
26 4 bottom 9.54251973
27 4 bottom 6.88220097
28 4 bottom 3.62175779
29 5 top 6.38900310
30 5 top 7.19216388
31 5 top 8.29793550
32 5 bottom 9.46722783
33 5 bottom 9.11261143
34 5 bottom 11.08097843
35 6 top -1.05244281
36 6 top -0.86450352
37 6 top -0.66251724
38 6 top -1.29278055
39 6 bottom 4.99175539
40 6 bottom 3.92459045
41 6 bottom 6.90398638
This SAS model
PROC MIXED DATA=dat ;
CLASS POSITION TUBE ;
MODEL y = POSITION / cl ;
RANDOM POSITION / type=CS subject=TUBE ;
RUN; QUIT;
is theoretically equivalent to this other SAS model (the marginal models are the same):
PROC MIXED DATA=dat ;
CLASS POSITION TUBE ;
MODEL y = POSITION / cl ;
RANDOM TUBE TUBE*POSITION;
RUN; QUIT;
However the two models yield the same estimates and standard errors but they yield completely different degrees of freedom for the estimates (with the default option).