S. Sajid Ali The discussion on what is the experimental unit in poultry trials has a very long history.

Basically you have two schools; the first one sees the animal as the experimental unit and as a result has many degrees of freedom to test against due to the large number of experimental unit. Thus rather small differences in growth can be declared to be significant.The second considers the pen to be the experimental unit, as the animals within one pen eat from the same feed trough and also otherwise interact with each other and thus data on individual animals can not be regarded to be independent. Furthermore feed intake and thus feed conversion will always be on a pen basis.

We belonged to the second school as you can also deduct from the attached paper.

C.A. (Kees) Kan Thank you Sir, Ya as we help the poultry scientist for analysis of their research data regularly, for body weight we used individual bird as an experimental unit, but for rest parameters like feed intake, FCR, etc it is not possible as they maintained in the flock. My point of interest was if both have no difference statistically, then for body wt too, we may take the average of each pen as one unit instead of taking individual birds wt.

I am not an statistician so forgive me for not using statistical language. As Dr. Kan mentioned, animals in the same experimental unit (in that case the pen) suffer from common environmental situation (including feeder, drinker, litter, etc.) and so I understand cannot be considered as true replications. Whenever you collect their information on an individual basis, you can use it in ANOVA but the information on their respective pen number would have to be included in the analysis so the right error (differences between pens on the same treatment) is used for the F test and not simply difference between individuals. On the observation that using individual birds as a replication would lead to finding very small differences as significant, my experience in comparing both methods of analysis is that this doesn't happen, one of the reasons being that the standard deviation of the individual measurements is quite higher than in the pooled observations. One potential advantage would be that degrees of freedom for the error would be substantially increased and so would tend to diminish the F-value for any probability test. Since this is a value that diminish asymptotically (tends to a plateau as the error degrees of freedom increase) this would be only important for trials with small number of replications. Hope it helps.