topic Re: Standard deviation with PROC NLMIXED in Statistical Procedures

Standard deviation with PROC NLMIXED

Clemence — Thu, 20 Jun 2019 11:01:10 GMT

Hello,

I use PROC NLMIXED to estimate a mean. And I wish to compare the result at the PROC MEANS.

My code is :

proc nlmixed data=lbmain_inclus_final;
parms sigsq0=0.2 sigsqe=0.3 beta0=0.3851 beta1=0.1433 beta2=0.5633 beta3=0.01331 beta4=0.01 beta5=0.02 ;
bounds sigsq0 sigsqe >= 0;

pi=2*arsin(1);

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

ll=(1/(sqrt(2*pi*sigsqe)))*exp(-(LB_qSTAanum_LOG_cop_cm-mu)**2/(2*sigsqe));

L=log(ll);

model LB_qSTAanum_LOG_cop_cm ~ general(L);
random a_i ~ normal(0,sigsq0) subject=SUBJID;

/**Estimates**/

moy_v1_prod=beta0+beta1+beta2+beta4;
estimate "moy_v1 produit" moy_v1_prod;

run;

With this code, I obtain the same mean that PROC MEANS but how obtain the standard deviation ? Because proc nlmixed give the standard error but when I calculate : sqrt(n)*standard_error, I don't obtain the same standard deviation that proc means. Is it normal ? Or, how obtain the good standard deviation with proc nlmixed ?

Thanks for your answer.

Clémence

Re: Standard deviation with PROC NLMIXED

Rick_SAS — Thu, 20 Jun 2019 11:48:34 GMT

The mean of what? Please show your PROC MEANS code.

The NLMIXED code is computing the MLE of regression coefficients for a regression of LB_qSTAanum_LOG_cop_cm onto main and interaction effects of produit and visite1 and visite2. You cannot duplicate that analysis by using PROC MEANS, which is a univariate analysis. You can try to use PROC GLM with a MODEL statement

proc glm data=lbmain_inclus_final;

MODEL LB_qSTAanum_LOG_cop_cm = produit visite1 visite2 produit*visite1 produit*visite2;

/* optionally include an ESTIMATE statement */

run;

Re: Standard deviation with PROC NLMIXED

Clemence — Thu, 20 Jun 2019 11:55:47 GMT

My code is :

proc means data=lbmain_inclus_final;
var LB_qSTAanum_LOG_cop_cm;
by prdt tv_visit;
run;

I thought I could find the same results with proc nlmixed but yes, you are right , with proc nlmixed it's a multivariable analysis.

Thanks