|
Shital,
I am having same the problem in GLIMMIX procedure. My sas code is similar to
yours as well. I don't think your code is wrong but you could still try
putting more iteration using the following code:
nloptions miniter = 100; * It did not work for me;
Here is my code:
proc glimmix data=A ic=PQ ;
class c1 c2;
model y =x1 x2 x3 c2/noint s solution dist=binomial link=logit oddsratios
(at x1= 40 unit x1=10) ;
random c1;
run;
I have used ~40-45 different models with almost the same code (but diff
dependent variables and diff. independent variables using lag values) like
above but i have had problem with around 7-8 models statements which in fact
did not converge.
I have also tried same model using correlated data library in Splus using
'glme' method: it did converge.
The problem with Splus: everything comes up significant in terms of p-val
compared to SAS. In Splus, ~90% of the models are significant. In SAS only
~10% of the models are significant.
good luck!
SW
On Wed, 24 Jan 2007 11:01:54 -0500, Sheetal Kamble <skamble@GMAIL.COM> wrote:
>Thank you, Dale. I tried the following syntax-
>
> proc glimmix data = one;
>
>class female agevar racevar insurance ch_ren_fail copd cvd chf diabet sAMI
>pAMI hospadm hv4 pv6 MD Hospital;
>
> model hem (descending) = female agevar racevar insurance ch_ren_fail copd
>cvd chf diabet pAMI sAMI hospadm hv4|pv6 / dist= binary link= logit s chisq
>CL ODDSRATIO;
>
>random hospital md_operating ;
>
>run;
>
>However, the model did not converge. I am not sure what is wrong with the
>syntax.
>
>Could you or anyone on this listserve guide me with the same?
>
>Thanks,
>Shital
>
>
>On 1/23/07, Dale McLerran <stringplayer_2@yahoo.com> wrote:
>>
>> --- Shital Kamble <skamble@GMAIL.COM> wrote:
>>
>> > Hi,
>> >
>> > I am using proc glimmix procedure in SAS 9.1 to model the interaction
>> > effect of physician volume (pv6) and hospital volume (hv4) on adverse
>> > dichotomous outcome "hem". The data structure is cross classified
>> > (patients
>> > nested within hospitals, patients nested within physicians, but
>> > approximately 25% of physicians in my dataset are not nested within
>> > hospitals). I am using proc glimmix syntax for cross classified
>> > random
>> > effects, however, my model does not converge for some reason. I would
>> > like
>> > to seek your advise pertaining to the same.
>> >
>> > Following is the proc glimmix syntax-
>> >
>> >
>> > proc glimmix data = one;
>> >
>> > class hem female agevar racevar insurance ch_ren_fail copd cvd chf
>> > diabet
>> > sAMI pAMI hospadm hv4 pv6 MD Hospital;
>> >
>> > model hem (descending) = female agevar racevar insurance ch_ren_fail
>> > copd
>> > cvd chf diabet pAMI sAMI hospadm hv4|pv6 / dist= binary link= logit s
>> > chisq
>> > CL ODDSRATIO;
>> >
>> > random int hospital md_operating / s ;
>> >
>> > run;
>> >
>> > *Note- hem = 0,1 (dichotomous variable) ; Primary regressor -
>> > interaction
>> > of physician vol (pv6 = 0=low, 1=high) and hospital volume (hv4 =
>> > 0=low,
>> > 1=high); hospital = hospital id and MD = physician id. All other
>> > independent variables are dichotomous or categorical variables.
>> >
>> > Please guide me peratining the correct syntax for this SAS procedure.
>> >
>> > Thanks,
>> > Shital
>> >
>> > Shital Kamble
>> > Ph.D. Student
>> >
>>
>> Shital,
>>
>> Generally, an intercept term is specified as a random effect only
>> when you also specify a SUBJECT on the RANDOM statement. You have
>> not named a SUBJECT for your RANDOM statement (and rightly so).
>>
>> You might take a look at Example 2: Mating Experiment with Crossed
>> Random Effects in the GLIMMIX documentation. Note that their
>> RANDOM statement is
>>
>> RANDOM FPOP*FNUM MPOP*MNUM;
>>
>> and that FPOP*FNUM exactly specifies a single female salamander -
>> salamander FNUM in either the FPOP='rb' or FPOP='ws' salamander
>> population. Similarly, mpop*mnum exactly specifies a single male
>> salamander. This is, then, analogous to specifying
>>
>> RANDOM HOSPITAL MD_OPERATING;
>>
>>
>> HTH,
>>
>> Dale
>>
>>
>> ---------------------------------------
>> Dale McLerran
>> Fred Hutchinson Cancer Research Center
>> mailto: dmclerra@NO_SPAMfhcrc.org
>> Ph: (206) 667-2926
>> Fax: (206) 667-5977
>> ---------------------------------------
>>
>>
>>
>>
>>
____________________________________________________________________________________
>> Sucker-punch spam with award-winning protection.
>> Try the free Yahoo! Mail Beta.
>> http://advision.webevents.yahoo.com/mailbeta/features_spam.html
>>
|
