Proc logistic gone crazy, follow-up

[dglassbrenner]

Thank you for the post entitled “Proc Logistics gone CRAZY!”  This post noted that Proc GLIMMIX correctly identifies x1-x5 as insignificant in the following example:

data temp;
input Y N;
cards;
97870      270000
12890      55000
120071    313000
43446      150000
1405102   1903000
125402     254000
79192       109000
14087       29000
10714       9000
983775     1587000
316543     654000
8592        29000
76061      130000
217492     501000
132423     354000
29163      127000
57013      161000
82747      192000
101778     344000
44258      77000
;
run;
 
data temp;
call streaminit(123);
set temp;
x1=1-rand('BETA',3,0.1);
x2=rand('CAUCHY');
x3=rand('CHISQ',22);
x4=rand('ERLANG', 7);
x5=rand('EXPO');
x6=rand('F',12,322);
trial=_n_;
run;
 
proc glimmix data=temp method=laplace;
class trial;
model y/n=x1 x2 x3 x4 x5/solution chisq;
random intercept/subject=trial;
run;

My question is: Why are the df for x1-x5 all zero?  It seems disconcerting to have zero df, but should it (be disconcerting)? 

P.S. I'm using SAS 9.4 TS Level 1M3.

[ballardw]

x1 has an issue with the way you build it:

I get this when I run your code:

NOTE: Argument 3 to function RAND('BETA',3,0.1) at line 419 column 6 is invalid.

So x1 is missing for every record.

And

432  proc glimmix data=temp method=laplace;
433  class trial;
434  model y/n=x1 x2 x3 x4 x5/solution chisq;
435  random intercept/subject=trial;
436  run;




ERROR: Invalid or missing data.

So I don't see how you get any DF.

 I'm using SAS 9.4 TS Level 1M4.

[dglassbrenner]

That's funny - it works fine for me.  If you run it without trying to create x1 and taking x1 out of the predictors, do you get 0 df for x2-x5?

data temp;
input Y N;
cards;
97870      270000
12890      55000
120071    313000
43446      150000
1405102   1903000
125402     254000
79192       109000
14087       29000
10714       9000
983775     1587000
316543     654000
8592        29000
76061      130000
217492     501000
132423     354000
29163      127000
57013      161000
82747      192000
101778     344000
44258      77000
;
run;
 
data temp;
call streaminit(123);
set temp;
*x1=1-rand('BETA',3,0.1);
x2=rand('CAUCHY');
x3=rand('CHISQ',22);
x4=rand('ERLANG', 7);
x5=rand('EXPO');
x6=rand('F',12,322);
trial=_n_;
run;
 
proc glimmix data=temp method=laplace;
class trial;
model y/n= /*x1*/ x2 x3 x4 x5/solution chisq;
random intercept/subject=trial;
run;

[SAS_Rob]

I think the issue is that you have over-parameterized your model.  With only 1 observation per subject, the random intercepts essentially exhausts your DF.

[dglassbrenner]

That explanation makes sense to me, and over-parameterization sounds bad.  But this was the recommended solution to the earlier post (https://communities.sas.com/t5/SAS-Statistical-Procedures/Proc-Logistics-gone-CRAZY/td-p/146717). Was this solution maybe not so great an idea after all?