4.3.3 Blocking
If the individual units are homogeneous, respond in a consistent way to the various
treatments, then the experimental error variance will be small. Therefore, the experiment
will have high precision and it will be relatively easy to detect differences between the
individual treatment population means. However, Fisher noticed that in agricultural
research, the individual plots in a field were anything but homogeneous, they differed
with respect to fertility, drainage etc. Furthermore, it was felt that such heterogeneity
between units was desirable, in order to give the experiment wide coverage, so that it
would produce results applicable to the real world. Thus, Fisher was faced with two
apparently irreconcilable requirements—to design high-precision experiments using
widely heterogeneous units.
Fisher introduced the idea of blocking to overcome this problem. Individual
experimental units are grouped together into blocks in such a way that within each block,
the units are as homogeneous as possible. Each block usually contains a single replicate
of all the treatments, with the allocation of units to treatments within each block being
randomized. In the subsequent analysis, the experimental error variance is estimated from
units within the same block, so is relatively small, preserving the precision and power of
the experiment. However, the blocks themselves may be as heterogeneous as the
experimenter wishes, thus giving the experiment as wide a coverage as may be desired.
For example, in a clinical trial it may be that a precise comparison could be effected by
restricting the age, sex, clinical condition and other features of the patient, but these
restrictions make it difficult to generalize the results. Therefore, the factors should be
used as blocks.
4.3.4 Statement of Objectives
The goal of experimental design is statistical efficiency and the economizing of
resources. Most experienced practitioners of experimental design have a sequence of
steps to follow. The first is that the objective of the experiment should be clearly and
fully stated. Having chosen the appropriate response variables, the factors which might
influence them should be identified, and a decision taken as to which of these are to be
studied in the current experiment. The statement of objectives should also include a
description of the target population. The most common faults in setting out the objectives
of an experiment are excessive vagueness about the hypotheses to be tested and the
effects to be estimated, and excessive ambition!
4.3.5 Sample Size
An experiment is only worthwhile if it is powerful enough to detect differences between
the various treatment population means of a magnitude that the scientist deems to be
important. The ability of an experiment to discriminate depends upon:
1 The critical difference (the size of the effect it is desired to detect).
2 The significance level (the probability of an effect being detected when none exists).
4.3.3 BlockingIf the individual units are homogeneous, respond in a consistent way to the varioustreatments, then the experimental error variance will be small. Therefore, the experimentwill have high precision and it will be relatively easy to detect differences between theindividual treatment population means. However, Fisher noticed that in agriculturalresearch, the individual plots in a field were anything but homogeneous, they differedwith respect to fertility, drainage etc. Furthermore, it was felt that such heterogeneitybetween units was desirable, in order to give the experiment wide coverage, so that itwould produce results applicable to the real world. Thus, Fisher was faced with twoapparently irreconcilable requirements—to design high-precision experiments usingwidely heterogeneous units.Fisher introduced the idea of blocking to overcome this problem. Individualexperimental units are grouped together into blocks in such a way that within each block,the units are as homogeneous as possible. Each block usually contains a single replicateof all the treatments, with the allocation of units to treatments within each block beingrandomized. In the subsequent analysis, the experimental error variance is estimated fromunits within the same block, so is relatively small, preserving the precision and power ofthe experiment. However, the blocks themselves may be as heterogeneous as theexperimenter wishes, thus giving the experiment as wide a coverage as may be desired.For example, in a clinical trial it may be that a precise comparison could be effected byrestricting the age, sex, clinical condition and other features of the patient, but theserestrictions make it difficult to generalize the results. Therefore, the factors should beused as blocks.4.3.4 Statement of ObjectivesThe goal of experimental design is statistical efficiency and the economizing ofresources. Most experienced practitioners of experimental design have a sequence ofsteps to follow. The first is that the objective of the experiment should be clearly andfully stated. Having chosen the appropriate response variables, the factors which mightinfluence them should be identified, and a decision taken as to which of these are to bestudied in the current experiment. The statement of objectives should also include adescription of the target population. The most common faults in setting out the objectivesof an experiment are excessive vagueness about the hypotheses to be tested and theeffects to be estimated, and excessive ambition!4.3.5 Sample SizeAn experiment is only worthwhile if it is powerful enough to detect differences betweenthe various treatment population means of a magnitude that the scientist deems to beimportant. The ability of an experiment to discriminate depends upon:1 The critical difference (the size of the effect it is desired to detect).2 The significance level (the probability of an effect being detected when none exists).
การแปล กรุณารอสักครู่..
