STEPWISE DISCRIMINANT-ANALYSIS IN ARCHAEOMETRY - A CRITIQUE

Discriminant analysis is widely used in archaeometry and the use of stepwise selection procedures is common. It is also common to judge the success of an analysis by the ability to correctly allocate specimens to groups using the resubstitution procedure. This paper argues, by way of example, that such procedures may often mislead. In particular, much smaller subsets than those selected by automatic procedures can often be found that perform as well or better; selection of subsets may be method dependent; and resubstitution can mislead as to the success of an analysis, with the less biased cross-validation procedure producing more realistic results.