4.5  A Comment on Discriminant Analysis  Versus' Logit Analysis

This note contrasts discriminant analysis with logit analysis.  In
causal models, it is seen that forecasting leads to classification
problems based on selection probabilities.  The posterior
distributions implied by the selection probabilities and prior
distribution may provide a useful starting point for estimation of the
selection probability parameters in a discriminant-type analysis, but
this procedure does not tend to be robust with respect to
misspecification of the prior.  In conjoint models, on the other hand,
the posterior distributions and selection probabilities are
alternative conditional distributions characterizing the joint
distribution.  In these models, it is generally not meaningful to
examine the effects of shifts in explanatory variables.