This sure is an interesting challenge. To follow on to the previous comment by Kathleen, it appears that cov and animnum are confounded in this dataset. Although cov is a continuous variable it only takes on 8 values, a unique one for each of the 8 levels of animnum. Easy to see with: proc freq data=one(where=(param=700)); table animnum*cov / nopercents norow nocol; run;  I became suspicious when I looked at the solution table. The last level of animnum was 'estimated' as 0, as it must be with the matrix sweep of the overparameterized model. However, the next-to-last level is also 0 (with no SE). Suggests that animnum is confounded with something else.    ... View more

It is an interesting problem. I did not get your error message with GLIMMIX, but I did duplicate your problem in the output. I tried this with PROC MIXED (changing the last statement to REPEATED) and got the message (with one of the orders for the covariate, can't remember which): WARNING: Stopped because of infinite likelihood This typically means with repeated measures that you have more than one observation per time/subject combination, which is not allowed as you know. Quoting from the great Kiernan et al. SASGF (2012) article: "PROC MIXED requires that you have one observation per repeated effect per subject. This note occurs when your PROC MIXED step includes a REPEATED statement and the input data set has multiple observations at a given time point for that particular subject." You said you had two observations per subject/time level.  But I am surprised that you did not get this error message with GLIMMIX. I think you need to define animal as the subject, or is that what is happening with your animnum(day grp_no) subject?  Other things can give this error as well.   I also think that Kathleen identified the other (maybe the first) problem. When I look at frequencies of the levels, using  proc freq data=one(where=(param=700)); table grp_no*day*animnum / nopercents norow nocol; run; I see mostly zeroes, suggesting a possible coding issue with grp_no.  But I don't really know your data, so I can't say. I hope this gives you some ideas. ... View more

As others have correctly pointed out, there are a few ways to fit models to data with a beta distribution. GLIMMIX is the easiest way. However, since the original question dealt with HPGENSELECT, one would assume that they were trying to do variable selection from a large number of potential predictor variables. That cannot be done in an automated way with GLIMMIX or NLMIXED.   One should always be careful with the beta distribution: it is defined for 0 < y < 1. This means that all values of y equal to 0 or 1 will become missing values in GLIMMIX. My experience is that datasets with continuous proportions usually have 0s and 1s. ... View more

If you serarch SAS Communities, you will find several older postings about rank-based analysis of factorials. Brunner and several colleagues have devleloped methods that can be done in SAS MIXED (with the right options, after ranking the data). Don't confuse this with rank transformation method (where one just ranks data and proceeds with GLM). You need to choose other options for unequal variances. Old postings, including by me, give some code. ... View more

Unfortunately, this procedure cannot handle the beta distribution. As an approximation, you could use PROC GLMSELECT. You could use the weight statement to account for unequal variances for Y. ... View more

The side IDs should be unique for the nesting you describe. Easy to handle. First, trial should be listed in the class statement (I can't imagine how trial is a continuous variable). then in the random statement, use subject=siteid(trial); ... View more