Studies in which there were less than five randomised clusters were excluded because of the interpretational difficulties caused by the total confounding of two sources of variation: the variation in response due to the effect of intervention, and natural variation that exists between the clusters even in the absence of an intervention effect. Measuring and adjusting for baseline differences can help reduce such confounding, but the inherent problem that such trials can only be analysed at the level of the individual remains. To analyse at the individual level, one would have to assume that there is no clustering of individual responses within the community, which is almost always untenable (Donner 2000).