02-27-2018 10:57 AM
Lamb feeding trial designed as follows:
*Lambs placed into individual pens (experimental unit = lamb ID) and remained in the pen throughout the trial
*Lambs fed for 84 day, using 2 feeding periods (diet changed going into Period 2). Period 1 (days 0 to 28); Period 2 (days 29 to 84)
*On day 0 and multiple times during each period (e.g., d 0, 14, 28; 29, 50, & 84), various things are measured; e.g., body weight (BW), BW gain...
1. Can data be analyzed by Period without worrying about/checking for a trt x period interaction (evaluate by period even if interaction is not signif.)? I like this method because of the differences we see at times, where 1 trt does poorly during Period 1 but then "rebounds" in Period 2 and catches up to the others.
PROC MIXED; by Period;
CLASS ID TRT DAY;
MODEL BW = TRT day TRT*day;
REPEATED DAY / SUBJECT=ID TYPE = XXX;
2. Or, can I just analyze without including period anywhere in the model; just use day? Will the meaning of the results be confounding if I examine effects over different Periods? For example, lambs with less BW gain early in the feeding period may go through accelerated gain in Period 2 and that greater ADG may not be because of the feeding period but because of the previous ADG.
CLASS TRT DAY ID;
MODEL BW = TRT DAY TRTxDay / XXX;
REPEATED DAY/SUBJECT=ID TYPE = XXX
02-27-2018 01:41 PM - edited 02-27-2018 01:41 PM
We don't know your data. We don't know if the interaction is important or significant. If you think it might be important or significant, you need to include it in the model.
If I were you, I would put PERIOD and TRT*PERIOD into the model initially, and not use the BY statement. You also want to have day nested within period