I am running a Proc mixed to look at the effects of a procedure on 11 animals on 7 different behaviors over 4 days (day is the repeated variable). I am working with proportion data and performing an arcsine sq. root transformation on the data since some of the values lie outside the .3-.7 range. I run each of my dependent variables in separate mixed models. When I plot my residuals everything looks great, however when I plot my residual vs. predicted values for 3 of my behaviors I am getting predicted values that are the same for each day. The raw data looks fine, no missing values. I am wondering why SAS calculates the same predicted values for each day for the particular behaviors? I also get in the log statement that the G matrix is not positive definite. I have changed the covariance structure from ar(1) to cs, un, and simple, however the predicted values are still the same. Does anyone have any suggestions for why the predicted values might be the same for each day and what type of covariance structure might help to correct this?

Thanks! Kristen