Choice of Statistical Tests — Practice
A community trial randomises 20 villages, rather than individuals, to receive either a piped chlorination intervention or usual water sources, and diarrhoeal incidence is compared. When analysing such a design, the principal statistical concern that must be accounted for is
ABerksonian bias from hospital-based selection
BRegression to the mean in the control arm
CIntra-cluster correlation inflating the effective sample size requirement
DLoss of blinding among participants