Hi,
I would like to ask the following set of equations concerning NLMIXED and %GLIMMIX.
In the NLMIXED procedure if we have two explanatory variables
And we consider eta, the linear equation, as
Model 1 one explanatory variable (X1 is a dummy variable)
Eta =beta1+beta2*X1+u
And
Model 2 two explanatory variables (X1 and X2 are two dummy variables)
Eta =beta1+beta2*X1+beta3*X3+u
The subject is the same
1/There is any difference for u between the two models
2/ What is the importance for the starting points ?
I have remarked that I can also not use the option of starting points
3/ I have considered a continuous variable Z= age as independent variable in model 1, but I did not obtain convergence.
When I have considered log(age), I have obtained convergence of the model and the parameters, there is any explanation for that?
I have used the GLIM
%Include "c:\GLIMMIX.sas";
%GLIMMIX(data=mydata,procopt=NOCLPRINT,
STMTS=%STR(
class X;
MODEL Y=X Z/solution;
RANDOM INTERCEPT ;),
ERROR=binomial,LINK=logit);
RUN;
To obtain the started points from work table _soln (the parameters are obtain in a work table called _soln) and I have used them in the NLMIXED
I have remarked that the problem of age mentioned above disappears
proc nlmixed qpoints=50 gconv=1e-10 data=mydata;
parms a= -1.1220 b= 0.4343 c= -0.02782
s2u=1;
eta=beta1+beta2*X1+beta3*age+u; /*X1 is the dummy of X in GLIMMIX */
beta=u;
expeta=exp(eta);
p=expeta/(1+expeta);
model UNEM ~ binary(p);
random u~normal(0,s2u)
subject=AREAP;
predict beta out=mu ;/*The predicted u*/
predict p out=probability ;/*the predited probability*/
estimate 'exp(beta2)' exp(beta2);
run;
4/ It is possible to specify the reference class in GLIMMIX, and if is is possible to have the quantity u In eta ,as a SAS table like for NLMIXED with predict beta out=mu
5/ There is any problem in NLMIXED if the data contains more than 500 000 records? And the number of explanatory variables more than 10?
Thanks
Adel
---------------------------------
Découvrez le nouveau Yahoo! Mail : 250 Mo d'espace de stockage pour vos mails !
Créez votre Yahoo! Mail
|