The controversy over the AB/BA design.
Stephen Senn
University College, London
Abstract:
Standard advice in texbooks of medical statistics is that you should
analyse the AB/BA cross-over using the two-stage procedure: testing for
carry-over first and using first period data only if it is 'found'.
Standard advice is wrong. No medical statistician would accept a
physician's claim that screening for a disease is obviously beneficial.
(S)He would ask after senitivity, specificity, prevalence and losses
first. Testing for carry-over is a form of screening. When understood in
these terms the two-stage procedure is seen to be nonsense. It is
concluded that adequate washout is the only defence against carry-over.