those rules can be easily constructed. The safe approach is to retain all data but to seek to understand the
effects of causative factors and to identify different groups of sites showing different patterns of results.

5. Choice of Experimental Treatments

There are four main types of experimental treatment structure. First, the essentially unstructured set of
alternative treatments, such as a set of varieties or herbicides, where each treatment is of equal potential
importance. Second, the complete factorial structure including, equally, each combination of levels of the
factors included in the experiment. The third type could be defined simply as "the rest" between these two
extremes. Typical examples are control treatments in the unstructured set; stepwise combinations, where
each treatment is a particular modification of the previous treatment; subsets of a factorial structure
omitting inappropriate combinations. The fourth type, which may also be within types two or three, is for
levels of a quantitative factor where the particular levels are not chosen primarily for their direct interest
but rather as representatives of the range of interesting levels of the quantitative factor.

The only general rule is that the selected set of treatments shall provide the best possible information about
the questions which the experiment is purposed to answer. This statement,of course, assumes that the
questions precede the choice of treatments rather than the reverse (which is not scientifically justifiable).

There is relatively little to say about type 1 structures except that the number of treatments should be
determined by the number of interesting alternatives, always within the limitations of resources. The range
of incomplete block design structures now available for testing large numbers of varieties, using any
appropriate block sizes, is so comprehensive that there is no excuse for tailoring the number of varieties to
suit a particular design structure.

5.1 Complete Factorials

Complete factorial structures, possibly omitting one or two unsuitable combinations, always provide a very
powerful method of acquiring information because of their twin advantages: first they allow us to
investigate whether there are important interactions, and second, whether or not there are interactions, the
information about each separate factor effect will be more precise with factorial structure.

For initial experiments within a research programme factorial structured treatments will be suitable because
they provide information about the existence of interactions between factors, thus allowing those
interactions which are found to be unimportant to be ignored in subsequent experiments. The second
advantage of factorial structures in giving more efficient information about main effects and two.factor
interactions through the use of every combination in each effect estimate will, of course, also be beneficial
in initial experiments.

However this second advantage becomes much more important in subsequent experiments where it is more
efficient to continue to ask several questions in each experiment rather than relapsing to the classical
scientific approach of asking only a single question in each experiment.

5.2 Quantitative Factor Levels

For the choice of levels of a quantitative factor we should be concerned to maximise the information about
the pattern of response as a whole or about a particular characteristic of the response, such as the position
of the maximum. General statistical theory shows clearly that both forms of information are maximised by
(1) choosing as wide a range of values of the quantitative factor as is consistent with the assumption that
 the general pattern of response over that range can be summarised by a simple form of response
 curve.