the correct philosophy is to try to limit the potential for the choice of unsuitable blocks. This can probably
be achieved by setting an upper limit of 8 plots per block and requiring that the blocks be made compact.

However I shall assume that the experimenter does have the opportunity to assess variation within the
proposed experimental area and that (s)he, in collaboration with the farmer, can make judgements
identifying those plots likely to produce similar results. To avoid the suggestion that a "Block" must be
rectangular in nature, I propose to call sets of plots judged to be similar "Groups". It is crucial that this
identification of groups be separate from decisions about the choice of treatments. These two aspects of
design inevitably interact later but we must first try to ensure that we have the best possible grouping of
plots (as well as the most relevant set of treatments.

Let us consider an example from one of the on-farm experiments in the Poza Rica area. There were 24
plots in three rows of eight plots, as shown:


 Row 3 17 18 19 20 21 22 23 24
 Row 2 9 10 11 12 13 14 15 16
 Row 1 2 3 4 5 6 7 8


Row I is at the bottom of the hill, row 2 above row 1, and row three above both.

The correct grouping system, identified by the experimenter, was three groups

 (1,2,3,4,5,6,7,8) (9,10,11,12,13,14,15,16)
 (17,18,19,20,21,22,23,24).

The experiment was actually designed in four blocks of six plots


(1,2,3,4,5,6)
(17,18,19,20,21,22)


(9.10,11,12,13,14)
(7,8,15,16,23,24)


You can probably guess how many treatments there were!

The correct design would have three blocks of eight plots each with each block containing all six
treatments plus extra plots for two of the treatments, the extra two being different in each block. The irony
of this particular situation is that the actual allocation of treatments to blocks was inevitably equivalent to
the correct design (can you see why?) and the analysis that should be applied to the results should treat the
data as if the design had been intended to be the correct design (i.e. 3 blocks of eight plots, not 4 blocks of
six plots). However there is one difference between the allocation of treatments to plots in the ideal design
and in that actually used and this is that the randomisation of the design in three blocks would include all
possible allocations of treatments to the eight plots (see design example 1, section B).

Naturally experimenters cannot always expect to achieve the ideal design when trying to use an
inappropriate block structure. Another example in the Poza Rica farms had 24 plots as shown:


Row 5 19 20 21 22 23 24
Row 4 13 14 15 16 17 18
Row 3 7 8 9 10 11 12
Row 2 4 5 6
Row 1 1 2 3