Thank you to everyone who has offered suggestions! To answer some of the above questions- I am getting some fractional values because I am attempting to analyze medical billing data. For some procedures (like those involving anesthesia), billing is done in 15 minute increments (thus, we could end up with counts at 1.5 as a hour and a half of anesthetic administration). This is not generally the case, however. If I additionally log transformed these counts, would a lognormal distribution be better fit? In regards to the concerns of zero inflation-This was absolutely a concern initially. However, I have filtered for a complete case analysis where there is observed data in all the time points. I have tried fitting the data using proc genmod, glimmix, and mixed and am now confused about the relative merits of each. The following are examples of models which now successfully converge for my data set (using type=UN was preventing my model from converging so I reverted to AR(1)): proc glimmix data=procdata14 plots=residualpanel(conditional marginal); class code; model proc_count=code period_count code*period_count / dist=lognormal solution ddfm=betwithin ; random period_count / sub=code type=AR(1); run; proc genmod data=procdata14; class code; model proc_count= period_count code code*period_count / dist=normal link=log ; repeated subject=code / type=ar(1); output out = Residuals resraw = Resraw stdreschi = Stdreschi pred = Pred; run; [The website prevented me from attaching the plots of these residuals]
... View more