Help using Base SAS procedures

Overdispersion in Glimmix Proc

Reply
N/A
Posts: 0

Overdispersion in Glimmix Proc

Hi to everyone.

I'm having problems to solve an overdispersion issue using the Glimmix proc.

This is the model I want to adjust


Proc Glimmix data=sasuser.Soller30032010frutosaflores IC=Q;

class Planta Tratamiento Rama;
model Frutos/Flores= Tratamiento Diametro
/ solution DDFM=BW dist=binomial link=probit;
random intercept /subject =planta*tratamiento;
lsmeans tratamiento / ilink cl plot=meanplot(ilink);
run;


that results overdispersed:

(Gener. Chi-Square / DF 4.21)

I’ve tried all kind of solutions as:


1- Changing the random structure, like:

random intercept /subject =planta;

or

random intercept /subject = tratamiento;




2- Adjusting an overdispersion term

Proc Glimmix data=sasuser.Soller30032010frutosaflores IC=Q;
class Planta Tratamiento Rama;
model Frutos/Flores= Tratamiento Diametro
/ solution DDFM=BW dist=binomial link=probit;
random intercept /subject =planta*tratamiento;
random _residual_;
lsmeans tratamiento / ilink cl plot=meanplot(ilink);
run;

I applied this option in several ways, and combined it with the changing of the random structure as well.


3- Indicated a inflated variance, and using a quasi-likelihood method for the estimation.

Proc Glimmix data=sasuser.Soller30032010frutosaflores plot=pearsonpanel;
class Planta Tratamiento Rama;
_variance_ = _mu_**2 * (1-_mu_)**2;
model Frutos/Flores= Tratamiento Diametro
/ solution DDFM=BW dist=binomial link=probit;
random intercept/subject=planta*tratamiento;
random _residual_;
lsmeans tratamiento;
run;

(With without random terms indicated)




And...after all this I can’t find a good solution.

Any suggestion?

Thank you very much and greetings from Spain.

Lucía Smiley Happy
Ask a Question
Discussion stats
  • 0 replies
  • 128 views
  • 0 likes
  • 1 in conversation