Proc nlmixed type3

[Clemence]

Hello, 

I use PROC NLMIXED and I would like know if it's possible to have the p-value for fixe effect with the method of type3 in proc nlmixed ?

My code is :

proc nlmixed data=simul ;
parms sigsq0=0.2 sigsqe=0.3 beta0=1 beta1=0.15 beta2=0.55 beta3=0.04 beta4=-0.8 beta5=0.2;

bounds sigsq0 sigsqe >= 0;
pi=2*arsin(1);

mu=beta0+beta1*produit_bis+beta2*visite1+beta3*visite2+beta4*produit_bis*visite1+beta5*produit_bis*visite2 + a_i;

if Censure_aureus=0 then ll=(1/(sqrt(2*pi*sigsqe)))*exp(-(LB_qSTAanum_LOG_cop_cm-mu)**2/(2*sigsqe));
if Censure_aureus=1 then ll=probnorm((LB_qSTAanum_LOG_cop_cm-mu)/sqrt(sigsqe));

L=log(ll);

model LB_qSTAanum_LOG_cop_cm ~ general(L);
random a_i ~ normal(0,sigsq0) subject=id_bis;
by simul;
ods output ParameterEstimates=param_mle ;
run;

With this code, I obtain p-values for beta4 and beta5 which represent the product effect at visite1 and visite2 but, I would like obtain the global pvalue for the interaction. According you, it's possible with proc nlmixed ?

Thanks for your answer.

Clémence

[Rick_SAS]

I mostly see the type 3 tests for fixed effects being used for linear models and generalized linear models (eg, GLM and GENMOD). I don't think these statistics are supported in PROC NLMIXED, which fits nonlinear regression models. If it's possible for you to fit your model in PROC MIXED, you can also test the fixed effects there.

[Clemence]

Thanks for you answer. However I can't fit my model with PROC MIXED because I defined the likelihood for this model. I thought fitting the same model without interaction then calculate the difference between -2log likelihood (model without interaction) and -2log likelihood (model with interactions) with a chi square distribution. Do you think that it's rigth ?

[Rick_SAS]

Ah, so you want a likelihood ratio test similar to the Type 3 Analysis described in the GENMOD documentation?

There is a SAS Usage note about comparing models by using a LR test.

SAS distributes a macro that compares two nested models by using the likelihood ratio test.

See the PROC NLMIXED example "Failure Time and Frailty Model" for an example.

[Clemence]

Thank you. I'm going to read that.

Clémence

[Clemence]

Hello,

I have another question. Is it possible to add an option in proc nlmixed to adjust the alpha risk ? 

Clémence