I guess I don't understand the question.  If you have a known sample size and known power, do you wish to solve for the minimum detectable difference?  You have specified what the difference is, so that seems out.  More info would help here. Thanks, Steve Denham ... View more

So long as the clustering variable is explicitly included in the CLASS statement, it should be OK.  I worry somewhat about the 'repeated' statement, but I believe you will be using PHYSICIAN as in SUBJECT=PHYSICIAN part.  If not, then careful consideration needs to be made as to the possible covariance structures and the effect of coding on them. Steve Denham ... View more

Check out PROC POWER, and look at the syntax for TWOSAMPLESURVIVAL.  I am pretty sure that you just need to plug in the values, and exponentiate the hazard rate for the reference group.  There is an example 70.6 in the SAS/STAT 9.3 documentation to get started. Steve Denham ... View more

Yes, there is.  In your lsmeans statement, add the ilink option: lsmeans ecotype nitrogen/diff ilink: This will give the estimates and standard errors on the inverse linked (back-transformed scale).  Note well that the DIFFERENCES are not the differences between the back-transformed values, but are the back-transformed values of the differences on the logit scale--a very different thing. Steve Denham ... View more

Bingo!  Full credit for discovering what was happening. Steve Denham ... View more

I'm horrible at this--that's why I always use the /e terminology to write estimates.  I would also take a look at SAS System for Mixed Models by Littell et al. in the section on analysis of covariance.  George Milliken has a TON of examples using ESTIMATE statements in there. Here would be my workaround--shift to PROC GLM or MIXED, which do support the LSMeans statement, and use the /e option there.  From those L matrix printouts, work out the code for the estimates, then drop those back into SURVEYREG. Also, my version of 9.2 (9.22 for the documentation, 9.2 (TS2M3)) does support LSMEANS.  This is a simple upgrade/hotfix and may be well worth your time.  Besides the LSMEANS statement, you will have access to the LSMESTIMATE statement, which may be of use in this project. Good luck. Steve Denham ... View more

For the continuous variable dropoutofschool, you need to provide some value at which you wish to estimate the function.  A good starting point would be to add lsmeans dropoutofschool*year2 move*year2 dropoutofschool*move*year2/e; The 'e' option will print the L matrix associated with each of the least squares means, including the slope coefficient for dropoutofschool.  From this, the estimates can be correctly constructed. Good luck. Steve Denham ... View more

I think it all comes down to the granularity of the interpolation, so yes the precision differs depending on the option.  Rather than NPTS=, you might specify STEP= in the plot statement, for spacing on the X-axis for the interpolation. Steve Denham ... View more

The standard graphics associated with SURVEYREG require an ESTIMATE, LSMEANS, LSMESTIMATE or SLICE statement.  You can get these by adding on ODS graphics on; statement prior to invoking surveyreg (include an ODS graphics off; after it for neatness), and adding the appropriate statement to proc surveyreg.  For regression type variables (estimated linear predictors and their standard errors, residuals, confidence bounds), use the OUTPUT out=SAS-data-set statement.  You can then use this dataset with any of the graphics procs that give what you need--SGPLOT and SGPANEL kind of come to mind. Steve Denham ... View more

I think we need more information.  What is the purpose of combining the variables?  It could be as simple as calculating a sum or product--that is a combination.  It may be to find a principal component, sort of an underlying single variable that explains the bulk of the variation and covariance between the two variables.  Without knowing the purpose, it is very difficult to offer a suggestion. Steve Denham ... View more