Blocks
In practice, experiments must often be conducted with experimental units that are quite variable.
The inaccuracy that results from varying experimental units in a completely randomised experimental design can be reduced if the units can be grouped into blocks of similar units.
This may be done based on measurements from the units before the experiment is conducted, or may be based on their location or other known characteristics.
Example
An experiment was conducted on a dairy farm owned by Massey University in New Zealand to assess three different pasture management strategies (early, late and very late severe grazing).
For operational reasons, it was decided to use paddocks numbered 6, 9, 11, 16 and 17 in the experiment and each paddock was divided into three parts (the experimental units) — the minimum size to apply a different management strategy.
It is fairly obvious that the five paddocks are not identical; creeks affect some and they probably have different histories of fertiliser use. However the three experimental units in each paddock are likely to be relatively similar to each other, so we can consider each paddock to be a block of three experimental units.