We give in this paper an easy to use and statistically sound way of ha
ndling replications in discrimination tests based on a concept called
overdispersion. The overdispersion is easily calculated by hand for an
y discrimination testing setup. And since the method simply amounts to
correcting the total number of observations, and similarly the overal
l response, by the overdispersion, we can use any standard technique,
table or software on the corrected numbers to perform a statistical te
st for difference, for similarity, power calculations, controlling ris
k I and risk II, etc. We also include a general discussion of the repl
icated discrimination testing situation, and some examples to illustra
te how to use the method. The SAS(R) macro REPRISKS that handles all c
omputations is available by e-mail: schlich@arome.dijon.inra.fr. (C) 1
998 Elsevier Science Ltd. All rights reserved