R help - Apr 2004 - randomization test

Hello all,

 

I am conducting a randomization test on a given dataset. One of the
covariates, gender, is randomly assigned 1000 times to create a
randomization dataset (rand.data). To these 1000 datasets, I fit a full
model (see below) and the aim is to generate a distribution of LRT statistic
under the null. Here are some of the questions I have:

 

1. I am using for() loop to fit the dataset grouped according to
rand.id(1:1000). If there is an error in fitting one dataset, the loop
terminates prematurely. My question is how to make the loop to continue? I
tried using try() in SPLUS but did not work for me. May be I did not
understand the implementation properly. 

 

2. Is there any other efficient way to do this?

 

The Loop I am using-

 

library(nlme)

for (i in 1:nsets)

            {

            od.fit<-nlme(MODEL,                  # NLME MODEL gender as a
covariate

            data=rand.data,                         # Data to be used for
fitting

            fixed=......~1,                    #Three fixed effects
parameters

            random=...~1|id,                    #Random variability on two
parameters with subjects grouped by "id"

            start=c(..,..,..),                    #Initial estimates for 3
fixed effects

subset=trial==i)                         #Fit the NLME function by trial

            rand.loglik[i,]<-matrix(od.fit$logLik)          #Extract
loglikelihood

            }                           

 

I don't mind if the solution is applicable in R or Splus.

 

Thanks,

 

Pravin

 

Pravin Jadhav

 


	[[alternative HTML version deleted]]