I use a dot when I refer to a macro variable. ... View more

PP stands for Per-Protocol analysis population. 🙂 ... View more

Thanks Paige, I understand the hesitation.  The program is pretty convoluted with macros embedded inside.  I actually figured out how to get the macro for the populations to work - I added two ampersands i.e.  if &&pop.="Y";  to each place within the program. This seems to work well now.  I am not completely sure why or how this works - but will figure that out. Thanks for your input. ... View more

Thank you for your suggestion. By groups wouldn’t work because of the structure of the datasets. There are variables to identify which observations (ie subjects) belong to which populations. Ie ITT=Y/N, PP=Y/N. The code is basically %let pop= ITT data a; Set b; If &pop.=“Y”; Run; This repeats a few times in the program bringing in and merging different datasets where the condition &pop.=“Y” is used. There are also some macros within the code. With a summary table at the end. I Want to produce the same summary table For the different analysis populations. I hope this clarifies the issue a bit ... View more

Thank you for your response. I agree that comparing variables of like type is better. This was not really doable for this particular situation. ... View more

Thank you - this seems to work. ... View more

Thanks,  That doesn't quite work with my dataset.  For example: data aa; input name $ Base $ Compare $ ; cards; x 123,456 123456 y 987654 987,654 date 16MAY2024 16/05/2024 match yes yes ; run;   The first two observations are working with the all_digits format.  But the dates are different.  The dates are values in the base and compare variables.    (you are correct the format of the date is day month year). ... View more

Thanks. The all_digits function with the compress works. Now I have to figure out how to deal with comparing the different date formats: i.e."16MAY2024" and "16/05/2024" when both of these are text values. Any ideas? ... View more

They are character because I transpose the output of the proc compare.  There are all different types of data that I am comparing - dates, numbers, text etc.  That is why the variables Base and Compare have both numbers (including sometimes dates) as well as text. I actually did see that when I did an ODS excel output, all the numbers with commas were read into excel without the commas.  This means that I can theoretically now write an "if" condition in excel to take these out (since now they are the same). But I would much rather do this up front in SAS.  Any suggestions? ... View more

Thanks for your response Steve. My model has actually has more terms in it than in my example (baseline parameters - some crossed with treatment) together with multiple treatment groups (with one control group). I ran the model as you suggested with site as a random effect (removing from model statement) and compared the ddf and the least square estimates and differences.  The models converged but with these notes in the log NOTE: Estimated G matrix is not positive definite. NOTE: A linear combination of covariance parameters is confounded with the residual variance. The second note did not appear in the model with emphirical option.   The ddfm was the same and high in the 2 models you suggested (removing ddfm=kr and removing ddfm=kr + emphirical option) for all the effects.  Type 3 test for fixed effects was different (understandably) for the cross effects with a higher "natural" df (i.e the trt*site and the trt*week).  Checking the lsmeans and differences, they were the same for the two models but pvalues were different due to the standard errors being smaller in the emphirical model (I'm not so clear exactly why - I need to read up more about that). However, the estimates themselves were very similar to the model with site as a fixed effect.   If I want to force the ddf would you suggest I use the degrees of freedom from the model without site as a random effect? Thanks again for your help and input.  I really appreciate it.   ... View more

Thanks Steve.   I ran your code which took a very long time to run- but again got a message about insufficient memory. Curious to what your thoughts are (or anyone else) about not using site as a random variable, but rather include as a fixed effect.   Often times site in included in the model statement and not as a random effect.   Actually the data with which I am analyzing now runs much better when I include site as a fixed effect and drop the random effect.  I do of course include the residual  effect (week).   Thanks. ... View more

Thank you so much for your reply.  I 1) removed site from the model statement 2) ran the G matrix random together with the R matrix random with  residual option statements on weekly summarized data with TOEP (instead of AR(1)) as follows RANDOM int / subject=site;  RANDOM week / subject=id(site) residual type=TOEP; After over an hour, SAS stopped processing because of insufficient memory.   When I run the model including site in the model statement without the RANDOM int/subject=site then the model runs successfully in around 15 seconds run time.   Using monthly summarized data with the RANDOM int/subject=site statement, the model converged but the G matrix was not positive definite - so something is wrong there as well.   I can't figure out what is the problem with using site as a random effect.  Any ideas? And how is the model affected if I use site as a fixed effect instead of a random effect?  Thanks for your insights. ... View more