The aim of equivalence testing is to verify that two parameters are sufficiently close or, alternatively, that the parameter of interest lies within two pre-specified limits. The two one-sided tests procedure is arguably the most widely known approach for assessing equivalence in the pharmaceutical field. Using a model that accounts for missing data, it is shown analytically that the type I error rate may exceed the nominal significance level. A refined estimate of this error is also obtained. For the standard 2x2 crossover design, a method is proposed that enables control of the type I error in the presence of missing data.
