Almost every analysis needs some basic assumptions. In our article,1 we assumed that the "wash-out" period was long enough between the different treatments. This implies that the effect of correlation is negligible. We have no reason to believe that this assumption is fundamentally wrong. It is obvious that if each subject received only 1 treatment, then the interindividual variations can be expected to increase, forcing us to increase the number of subjects considerably. In this case, a pairwise analysis was possible because the response variable was the intraocular pressure in the drug-treated eye minus the intraocular pressure in the placebo-treated eye. We believe that the design of our experiment is efficient and that the analysis is appropriate. We agree that it may also be appropriate to use generalized linear models, but it should then be recognized that an analysis of variance in this situation is a special case of generalized linear models when normal distribution can be assumed.
Lindén C, Alm A. Latanoprost and Physostigmine in Human Eyes—Reply. Arch Ophthalmol. 1998;116(3):398-399. doi: