ABSTRACT
This chapter discusses non-inferiority based on the underlying means or medians when there are no missing data or censored observations. Means and medians are often used to describe the typical value or the central location of a distribution. Medians are preferred when the data are skewed. The outcomes may be continuous or discrete. For continuous outcomes where larger outcomes are more desirable and differences between outcomes have meaning (i.e., the data have an interval or ratio scale), the difference in the means of the experimental and control arms in a randomized trial represents the average benet across trial subjects from being randomized to the experimental arm instead of the control arm. A difference in the medians does not have any analogous interpretation unless additional assumptions are made on the underlying distributions (e.g., that the shapes of the underlying distributions are equal).
