Thank you for your kind answer. Yes, the observations were obtained from the same bottle. I have two measurements per replicate bottle because I wanted to see how reproducible was my inoculation of the treatment. It was basically to just to visually compare if the values of the metabolites detected were the same and that I didn't make mistakes when injecting the treatment. Thus, I didn't want to add bottle as an additional factor because I was afraid that it would complicate my design. I appreciate that you pointed out that I can use the means instead. I have only missing data at the initial time point in 2 donors out of 8, so it may not make a difference, like you mentioned. I will definitely try this approach. Indeed donor is my replicate, that is the reason for using 8 people. My objective was to compare whether treatment 1 had and impact on the metabolites measured, in comparison with the control (trt2). I understand that it is a random effect, however I was interested in seeing the differences between each person, for each metabolite. For this reason, I added it in the model statement. I agree than then I would have a model only with fixed factors, but I cannot use GLM because it is a repeated measures design. Thus, I continued using PROC MIXED. I was actually not comfortable changin the ddf=, because I was actually not sure, and that is the reason for asking my question in the forum. I even continued thinking yesterday and I tried the following: proc mixed data=vfas covtest;
class trt donor tpt;
model acetate= trt tpt trt*tpt donor(trt tpt)/ddfm= kr;
repeated tpt/ subject=donor type=cs ;
lsmeans donor(trt tpt) trt*tpt /pdiff ;
run; Which gave me what I wanted, but I still have donor as fixed effect. So, please roast me by all means, I just ordered the books you recommended, and I would like to know whether the above is correct. It would be absolutely fantastic to have a knowledgeable colleague who could assist me in person with my questions. Unfortunately that is not the case in my Department, and/or Faculty. I even tried with the statistical support from the University, but they have only ONE GUY who has any idea of SAS. The last time I asked him about using an offset in PROC GLIMMIX he replied 1 month later, and I had a clearer and more interactive communication with Steve Denham in this forum. My posting history is so sparse because it takes me time to read and understand lots of sources, and the forum is literally my last resort. My background is clearly not Statistics, and as I do not have a person who can support me, the forum is my best bet. I could also quit doing this but I don't think my PI would be thrilled. Thus, I have to continue reading and learning, which is exciting anyway. Again, sincere thanks for taking your time to recommend me the books and for shedding some light on my code. Best wishes, Emma
... View more