Statement corr in Proc power

[Clg]

Hello everyone,

I try to use the PROC POWER in SAS to retrieve a number of subject in a study (it is to understand this procedure). I use test=EQUIV_RATIO and I use statements alpha, CV, meanratio, power, lower, upper and npairs=.

This procedure requires the statement CORR and I am wondering how we know what value is the most appropriated ?

My study is a crossover 3-periods.

Thank you in advance for your answer.

Clemence

[Rick_SAS]

Please post your SAS program.

[Clg]

Hi,

My SAS Program is:

proc power;
pairedmeans test=equiv_ratio alpha=0.05 /*because I want an IC to 90%*/
npairs=.
meanratio=0.95
power=0.9
lower=0.8
upper=1.25
corr=-0.04 0
cv=0.35;
run;

When I put corr=0 I obtain 69 pairs necessary and when I test corr=-0.04 , the number of pairs is 72 (it is the number that I want retrieve). But how know what value put for CORR= ?

Thank you.

Clg

[Rick_SAS]

> But how know what value put for CORR= ?

The estimates (such as the correlation) are usually based on small pilot studies. For example, you might choose 20 participants for a preliminary study and use the correlation in the preliminary study to plan the larger study.

[Clg]

OK thank you for you answer. And, is it the correlation between subjects ? so, we can determined it with a linear mixed model ?

Clg

[SteveDenham]

The correlation is the within-subject correlation over periods in the cross-over design.  A mixed model approach is the best way to incorporate this (period as a repeated effect).

SteveDenham

[Clg]

Thank you both for your answers.

Clg

[Clg]

For example, I can do this following model ?

proc mixed data=DATA method=reml ;
class usubjid TRT TRTSEQ PERIOD;
model PARAM= TRT TRTSEQ PERIOD / s cl ddfm=kr ;
repeated PERIOD / sub=usubjid type=CS;
run;

 Will the within-subject correlationbe the "Residual" estimates in table Covariance Parameter Estimates ?

Thank you

Clg

[SteveDenham]

To get correlations, add an R option after the slash in the REPEATED statement.  The values in the Covariance Parameter Estimates are not within subject correlations.  Note that a CS structure results in equal covariances (and consequently correlations) between any two time points and equal variances at each time point.

SteveDenham

[Clg]

Thank you. And, I have a last question, what is the difference between R and Rcorr option ?

Clg

[SteveDenham]

R = blocks of the estimated R matrix (covariances between residuals)

RCORR = Correlation matrix corresponding to blocks of the estimated R matrix

The default is for the first subject.  You can get more (see the MIXED documentation).

SteveDenham

[Rick_SAS]

As Steve indicates, the doc is your friend. Another great tip is to find an example in the doc that is related to your question and modify it to see if you can answer your own question. For example, use the data for the repeated measure example and run this:

proc mixed data=pr method=ml covtest;
   class Person Gender;
   model y = Gender Age Gender*Age / s;
   repeated / type=cs subject=Person r rcorr;
   ods select R RCorr;
run;

You will see that the R matrix does not have 1s on the diagonal. In contrast, the RCORR option scales the R estimates so that the resulting matrix has 1s on the diagonal.