Mixtures of independent and paired data arise in many areas of health research.
For example, in perinatal trials, women may give birth to a single infant or a
set of twins. Similarly, in ophthalmology, one or both eyes may be affected by
the condition of interest and included in the trial. Since outcomes of pairs are
likely to be correlated, this should be taken into account when designing the
trial. The sample size can be calculated by assuming all observations are
independent and then inflating the resultant sample size by an appropriate
design effect. The design effect depends on a number of factors, including the
strength of the correlation between outcomes from the same pair, the planned
analysis approach and whether members of the same pair will be randomised to
receive the same treatment, different treatments or independently. Using design
effects will enable more accurate sample size calculations to be performed for
trials involving both independent and paired data.