Hello,
I am needing some help to determine if a covariate is needed in some of my data. Also need help with correct phrasing of experimental design and analysis.
The project evaluated two products that were injected into cattle and they grazed pasture all summer. Cattle were stratified to one of 8 pastures in which both treatments (products) were found equally within each pasture. Individual animal weights were recorded at initiation of grazing, mid-point, and final. There were 280 animals in the study with 1/2 receiving product A and 1/2 receiving product B. The 280 head were split into 8 pastures that ranged in number from 24-40 head, once again with equal number of each product injected in animals per pasture. When using Proc Mixed and evaluating initial weight by treatment there was a tendency P=0.09 for initial weight to be different within treatment groups. At this level do I need to include initial weight as a covariate in the model? I am evaluating total gain (final minus initial for each animal), mid-weight gain (mid-point weight minus initial for each animal), and average daily gain for each (final and mid).
I am assuming this is a randomized complete design with pasture as experimental unit and animal as subsample. Product is fixed, animal and pasture are random effects.
The model utilized was
proc mixed data=cattle;
class past id ;
model total_gain = product/ ddfm=kr;
random past id past*id;
lsmeans product ;
run;
past=pasture
id=individual animal identification (new line for each animal in dataset)
Lines include (with 8 animals as example):
Id Past product in_weight weight_45d out_weight ADG_45 ADG_4592 ADG_total gain_45 total_gain gain_2half;
1 1 A 520 620 740 4.44 2.67 2.5 100 220 120
2 1 B 540 610 795 1.71 4.11 2.90 70 255 185
3 1 B 570 715 855 3.54 3.11 3.24 145 285 140
4 1 A 585 754 870 3.90 2.77 3.24 160 285 125
1 2 B 565 680 750 2.80 1.56 2.10 115 185 70
2 2 A 615 770 875 3.78 2.33 2.95 155 260 105
3 2 A 610 735 890 3.05 3.44 3.18 125 280 155
4 2 B 440 560 670 2.93 2.44 2.61 120 230 110
