ABSTRACT
In the design of medical experiments two additional problems generally appear. The first is the ethical difficulty associated with the need to give a patient the treatment that is believed to be best for him or her as an individual rather than what the statistician’s random allocation might assign. The second problem relates to biases introduced by the patient’s or physician’s preconceptions of the merits of a treatment. Matching can be used in twin studies or, more generally, in paired comparison experiments to reduce unwanted variation. The principle can be extended for larger blocks of inherently similar experimental units and can thus increase the sensitivity of an experiment to detect a real effect if it exists. Medical experiments must, if possible, be designed to eliminate psychological biases in patient and physician. More complicated designs are commonly used, particularly in agricultural research, but the basic principles are the same.
