some practical considerations of good experimental design

1
Pestic. Sci. 1977, 8, 395 Some Practical Considerations of Good Experimental Design“ Percy F. Hunt Biosciences Laboratory, Sittingbourne Research Centre, Sittingbourne, Kent (Manuscript received 25 January 1977) Synopsis Theoretical requirements of a good design were reviewed and the success of the designs were discussed in relation to two styles of experimentation. Good experiment design is basically applied commonsense put into a formal framework, as opposed to a mystical set of rules that should be obeyed without question. They can be remembered as the 3 Rs of Randomisation, Replication and pRecision with the “s” stressing the need for Simplicity and the involvement of the Statistician at an early stage. An inexperienced experimenter is likely to gain more from the use of simple features in a design such as the use of blocks, factorial arrangements, than by copying complex designs from standard text-books on the subject, as he is unlikely to be aware of the associated pitfalls. The main developments of experimental design have been in an agricultural context, where there is need to develop agronomic methods including better varieties of crops and livestock, and chemicals to ensure both remained healthy, despite the variable background due to soil, climate, etc. Accordingly, the type of designs available are well suited to the various requirements of the experimental agronomist, an important feature being the accommodation of the seasonal time scale. This type of experimentation is essentially a batch operation. In contrast, the experimentalist in the laboratory does not have the seasonal constraints and his problems will be much less suited to the more complex conventional designs. To pursue such designs is often unsatisfactory leading to frustration for both the experimentalist and statistician. What is needed is a simple but flexible approach which enables the experimenter to maintain a better control over the results he is producing and allows him to modify his approach fairly quickly to respond to changing requirements outside his control. In this way the cycle, Hypothesis-Design- Experiment-Analysis is completed very frequently so that the output from this work program should be as pertinent as possible to the experimenter’s own needs. The experimenter and statistician should be engaged in a fruitful and mutually beneficial partnership as the work proceeds. a Synopsis of a paper presented at the symposium Experiment design for evaluating biologically active chemicals on 2 November 1976, organised by the Physicochemical and Biophysical Panel of the Pesticides Group, Society of Chemical Industry. 395

Upload: percy-f-hunt

Post on 06-Jul-2016

218 views

Category:

Documents


5 download

TRANSCRIPT

Page 1: Some practical considerations of good experimental design

Pestic. Sci. 1977, 8, 395

Some Practical Considerations of Good Experimental Design“ Percy F. Hunt

Biosciences Laboratory, Sittingbourne Research Centre, Sittingbourne, Kent

(Manuscript received 25 January 1977)

Synopsis

Theoretical requirements of a good design were reviewed and the success of the designs were discussed in relation to two styles of experimentation. Good experiment design is basically applied commonsense put into a formal framework, as opposed to a mystical set of rules that should be obeyed without question. They can be remembered as the 3 Rs of Randomisation, Replication and pRecision with the “s” stressing the need for Simplicity and the involvement of the Statistician at an early stage. An inexperienced experimenter is likely to gain more from the use of simple features in a design such as the use of blocks, factorial arrangements, than by copying complex designs from standard text-books on the subject, as he is unlikely to be aware of the associated pitfalls.

The main developments of experimental design have been in an agricultural context, where there is need to develop agronomic methods including better varieties of crops and livestock, and chemicals to ensure both remained healthy, despite the variable background due to soil, climate, etc. Accordingly, the type of designs available are well suited to the various requirements of the experimental agronomist, an important feature being the accommodation of the seasonal time scale. This type of experimentation is essentially a batch operation.

In contrast, the experimentalist in the laboratory does not have the seasonal constraints and his problems will be much less suited to the more complex conventional designs. To pursue such designs is often unsatisfactory leading to frustration for both the experimentalist and statistician. What is needed is a simple but flexible approach which enables the experimenter to maintain a better control over the results he is producing and allows him to modify his approach fairly quickly to respond to changing requirements outside his control. In this way the cycle, Hypothesis-Design- Experiment-Analysis is completed very frequently so that the output from this work program should be as pertinent as possible to the experimenter’s own needs. The experimenter and statistician should be engaged in a fruitful and mutually beneficial partnership as the work proceeds.

a Synopsis of a paper presented at the symposium Experiment design for evaluating biologically active chemicals on 2 November 1976, organised by the Physicochemical and Biophysical Panel of the Pesticides Group, Society of Chemical Industry.

395