Advances in technology and scientific knowledge are often due to gains in the use of appropriate design of experiments. Understanding of this statistical concept allows researchers to assign cause-and-effect relationships and remove conjecture and preconceptions from analysis of results. Business and economic questions require as much care as scientific research in the design of experiments.
In the scientific design of experiments, the researcher attempts to prove the logical statement: If X, then Y. The opposite must also be demonstrated to establish a cause-and-effect relationship: If not X, then not Y. Intuitively, we understand, for example, that a plant needs water to live, and if the plant does not receive water, it will die. A causal relationship therefore exists between the plant’s needs and water.
The researcher attempts to prove both logical statements through the use of control groups. Ideally, the same research subjects experience the same experimental conditions simultaneously. When this is not possible, as is frequently the case in biological experiments, a second group of subjects is matched to the first group in as many factors as may affect the outcomes as possible. The efficacy of a diet, for example, may be tested by selection of a control group similar to the test group in age, income, activity level, and number of children. In more critical experiments, the design of experiments will incorporate the actual matching of individual subjects; that is, subject number 1A will be the same age, gender, activity level, and starting weight as subject number 1B, but will receive the test diet, while 1A will not.
Factorial designs allow more than one variable to be studied within the same experiment, yet with the same rigor as control groups, by applying the mathematics of probability. The breakthroughs in genetics accomplished by Mendel were due to factorial experiments and observations. In these experiments, two or more independent factors are tested at two or more levels. As an example, the subjects may be split among three independent variables: regular diet, Diet A, or Diet B. Each of these subgroups is split again, based on how long the diet is applied, either three weeks or six weeks.
Statistical methods are fairly easy to apply to the design of experiments in the realms of the natural sciences. In the social sciences, which include behavioral studies, they are more difficult. In the study of economics and business, the subjects are people and companies. These subjects do not easily lend themselves to study at all.
Marketing studies often depend on focus groups, whose care in selection is critical. Knowledge of the proper design of experiments is necessary to establish the selection criteria. Surveys, a common tool of product managers and political groups, also require this expertise in their design.
