A simple model illustrates this idea. Suppose we have an experimental
study with r farms, each with a plot of treatment A and a plot of treatment B.
The plot-to-plot variation (within farms) in yield is cry. Each plot is taken as a
single stratum, and divided into a large number of unit area quadrats. A
sample of a quadrats is taken in each plot to measure the yield, so the total
sampled area is a. The variation in yield between quadrats within plots is a~.
Then the standard error of the difference in mean crop yields for the two
treatments is