Dear list members,
I look for a way (or alternative) to specify initial values when estimating
linear mixed models in R, and to avoid iterative estimation.
This is a way to control specific parameter values (eg. variance parameter
values) such that the result (F-value) is based on them. This result can
then be used for power analyses that uses the non-central F-distribution, as
is done with SAS using the -PARMS- and -noiter- statements, indicated in the
following SAS example:
after using/constructing a data set, with actual scores changed into
predicted scores (no error)
proc mixed data = dataSet;
class treat group;
model y = treat*time ;
random intercept / subject=group(treat);
PARMS (2.1) (1.2) / NOITER;
contrast 'slopes' treat*time 1 -1 0,treat*time 1 0 -1;
ods output contrasts=c;
run;
data dataSet;
set c;
alpha=0.05;
ncparm=numdf*fvalue;
fc=finv(1-alpha,numdf,dendf,0);
power=1-probf(fc,numdf,dendf,ncparm);
run;
proc print;run;
cheers,
Wilfried
Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm
