Guiding Principles

Before building, whether a bridge or a machine, planning is nec­essary. Likewise, before conducting it, an experiment needs to be planned. A well-planned experiment should be the shortest means (both in terms of time and money) to the ends, answering specific questions asked in the form of hypotheses. Planning for an experiment with quantitative parameters requires an analysis in terms of probability and statistics. This is occasioned by the fact that experimental findings, in general, and measurements, in particular, are not certainties: they are probabilistic in nature. Designing a highly sophisticated experiment requires the help of trained statisticians. This chapter is meant to present the basic outlines involved in the design for students and experimenters who are not specially trained in statistics. The methods of analy­sis dealt with here are elementary. Nonetheless, they are adequate for planning the simple, comparative experiments required of undergraduate and graduate students, as well as most investiga­tors in industrial research and development. The essential charac­teristics of designed experiments follow:

1. An experiment should be considered a project, and like an engineered project, it should have well-defined goals. For instance, the phrase “to accomplish as much as possible” is contrary to planning. The quantum of accomplishment should be clearly stated.
2. The planning done at the beginning, middle, and end of the project should have different objectives. In the beginning, it should aim to identify all the possible variables, in the middle, to identify all the significant ones, and at the end, to test for the effect of the most significant ones.
3. The range of each variable should be carefully con­sidered. In the initial stage, testing the effect of an independent variable at more than one level may be unnecessary; in the middle stages, testing at two levels may be adequate; only at the final level can multilevel experiments (with a limited number of variables) be justified.
4. When several independent variables are found sig­nificant, some variables may “interact” synergisti- cally with others to produce either “good” or “bad” effects on the outcome, the dependent vari­ables. Such effects should be identified in the early stages, and planning should be done for specific experimentation in the final stages. The experi­menter should specify how much information is adequate at different stages; to say “all that can be known” is contrary to the idea of planning.

5. Each item of information should be assigned a specific level of confidence with the understanding that to expect a 100 percent confidence is to ask for the (statistically) impossible.
6. The experimenter should anticipate the kind of data to be harvested as a result of the experiment. These data should be considered as a means to answering the questions posed by the hypothesis.

Source: Srinagesh K (2005), The Principles of Experimental Research, Butterworth-Heinemann; 1st edition.