sample from multiple farms When sampling from animals in different farms, first, think about what your unit of replication is. This is not a trivial as it sounds (a farm with 3000 animals in it is not necessarily n = 3000!). A common mistake is to have all members of Group 1 in one farm, and all members of Group 2 in another. The unit of replication in this case would be farm not individual (i.e., one farm would constitute n = 1!). The reason this is important is because the variability within each farm will usually be lower than the variability between each farm. So, any effect that you observe might be an effect of farm, rather than an effect of your treatment.

One way to deal with this is to employ a phase design:

AB
ABA
ABAB

where 'A' is baseline and 'B' is your treatment (x).

These types of design are very useful as the allow you to return to baseline to ensure that any effects of treatment are not due to uncontrolled factors. Also, you can randomize the order of 'A' and 'B' between the different enclosures:

Week 1 Week 2
 		Week 3
 		Week 4
Farm 1
 		A
 		B
 		A
 		B
Farm 2
 		B
 		A
 		B
 		A
Farm 3
 		A
 		B
 		A
 		B
  etc....

If you sample animals from a small number of farms (2 or 3, for example), you could treat farm as a variable and use it in the analysis to see if there are any systematic differences evident. If you are sampling from many farms, it may be possible to treat farm as a random effect - we will deal with this in the data analysis sections.