Statistical thinking in practice: Handling variability in experimental situations

In all experimental situations, handling variability properly is important in order to achieve reliable and informative experimental results. The benefits from 'design of experiments' are well known to most experimentalists, but this article emphasizes the more practical matters of planning and performing experiments. Poor control of the variation sources in the experimental situation will generally lead to poor results. Three important statistical tools for handling variability in experiments are presented--replication, blocking and randomization. These tools are discussed in examples of real experiments, to illustrate the practical aspects of statistical thinking. In this context, statistical thinking may be viewed as utilizing common sense before it becomes belated wisdom. The illustrative examples may be viewed as failures, and not worthy of publishing, but in this context they are useful for illustrating commonly occurring obstacles and mistakes in experimental work.