CATMOD for ANOVA

[Ksharp]

1)

I think you want this ?

data operate;
input Hospital Treatment $ Severity  wt @@;
datalines;
1 a  1 23 1 a 2 7 1 a 3 2
1 b  1 23 1 b 2 10 1 b 3 5
1 c  1 20 1 c 2 13 1 c 3 5
1 d  1 24 1 d 2 10 1 d 3 6
2 a  1 18 2 a 2 6 2 a 3 1
2 b  1 18 2 b 2 6 2 b 3 2
2 c  1 13 2 c 2 13 2 c 3 2
2 d  1 9 2 d 2 15 2 d 3 2
3 a  1 8 3 a 2 6 3 a 3 3
3 b  1 12 3 b 2 4 3 b 3 4
3 c  1 11 3 c 2 6 3 c 3 2
3 d  1 7 3 d 2 7 3 d 3 4
4 a  1 12 4 a 2 9 4 a 3 1
4 b  1 15 4 b 2 3 4 b 3 2
4 c  1 14 4 c 2 8 4 c 3 3
4 d  1 13 4 d 2 6 4 d 3 4
;

proc catmod data=operate order=data ;
weight wt;
response mean;
model Severity=Treatment|Hospital ;
quit;

If you want do multiple dimension of ANOVA ,you could use PROC GLM or GENMOD, since ANOVA is under framework of General Linear Model.

If you are talking about non-parameter version of multiple dimension of ANOVA, I have no idea about it. You could try  @StatDave 's method.

2)

If there are no weight varible(a.k.a frequency of obs), you do not need WEIGHT statement.

[palolix]

Thank you very much for your feedback Ksharp! I tried StatDave approach. 

Thanks

Caroline