More on Interactions in Experimental Research

1. Interactions between factors, when there are only two, are easy to identify in two-way tables, as in Table 9.1, or even make visible in figures such as Figures 9.1 to 9.3.
2. The presence of interactions complicates the inter­pretation of the experimental results relative to the benefits or harmful effects of individual factors.
3. If there are no interactions, each factor may be treated as if it were independent of all other factors. That is when the main effects shine unobscured.
4. In most cases of experimental research, getting data like that in Figure 9.1 is rare. Almost always, some interactions may be expected, but all interac­tions observed may not be genuine in that some may simply be the effects of uncontrolled or uncontrollable “noise” factors. To make sure that this is not the case, statistical tests devised for the purpose, known as significance tests, need to be applied to the data.
5. Some interactions observed may be such that if the data—the values of dependent variable y—are transformed to other forms, such as yⁿ (where n is a real number) or log y, the interaction may disap­pear. It is worth making an effort, using mathe­matical methods available for the purpose, to get rid of the interactions.
6. In experiments with factors at three or more levels, interactions may exist only in certain segments, between levels of factors. If the factors are partly or fully qualitative, statements in words, such as cau­tions to be taken, are quite adequate. If, instead, the factors are quantitative, appropriate actions, in terms of controlling the quantities of factors, are necessary.
7. All interactions are not necessarily to be frowned at. Imagine, in a metallurgical experiment to enhance the hardness of steel, that a particular alloying element, Ap when used alone, is not effective and that another alloying element, A2, likewise is found to be ineffective. When both A1 and A2 are used together, however, the hardness is found to be much enhanced. The interaction in this case is found to be beneficial, prompting fur­ther experiments to optimize the percentage addi­tions of A₁ and A₂

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