C

Fig. 3.9. A complete randomized block design: (a) the blocks arranged according to the gradient; and (b) according to various levels of infestation, depicted by the intensity of shading.

Fig. 3.9. A complete randomized block design: (a) the blocks arranged according to the gradient; and (b) according to various levels of infestation, depicted by the intensity of shading.

cific locations within the trial area. One way of dealing with this is to use a randomized block design, which is based on the principle that patches of ground that are close together tend to be similar, while more distant patches differ. The blocks in the trial (consisting of a small number of plots) and the plots in each block are considered to be experiencing similar soil, pest and environmental conditions. Within each block, each treatment is randomly allocated a plot position. The conditions may differ between blocks (Fig. 3.9a). The layout and shape of blocks will depend on the conditions in each experimental area; although the ideal shape for a block is a square (this reduces the edge length relative to the area occupied) the experimenter should use his knowledge of the differences in soil fertility, yield uniformity, drainage etc. to determine the exact shape of the blocks. A pre-count of pest numbers may reveal differences in the relative size of pest infestation over the experimental area, and blocks could be allocated to take this distribution into account (Fig. 3.9b). The blocks act as replication for each treatment.

Thus, the randomization of treatment plots within a block ensures that localized variability in conditions does not cause bias and the blocking ensures that any differences over the experimental area do not bias treatment differences. However, when variation gradients within a site are not known, the use of blocks may result in their being positioned across gradients, in which case the assumptions made in any subsequent analysis would be incorrect. In such situations a complete randomized design would be more appropriate. The complete randomized block design is fairly common in experiments where the number of treatments varies between five and 20 with fewer replicates than treatments (Simmonds, 1979).

t

0 0

Post a comment