Evaluating the Risk in Sample Size Determination

This article considers experimental costs, besides power evaluation, in order to determine the sample size of an experiment. We focus on the use of standard tools of decision theory in the context of sample size determination. The loss function is defined, from the perspective of an experimenter which adopts the classical frequentist approach, and the risk function is computed. Then, we show the behavior of the risk function in the two-sample t-test, for a small sample experimental setting, with a medium-sized sample, and with large samples. Moreover, an objective criterion for a convenient sample size choice is introduced. Finally, a practical example of sample size determination, which also considers risk computation, is shown.