Listy Biornetryczne - Biometrical Letters Vol. 37(2000), No. 1, 21-40
|
LOGISTIC DISCRIMINANT FUNCTION — HOW MUCH ADVANTAGES OVER THE LINEAR ONE ? Anna Bartkowiak, Adam Szustalewicz, Joanna Zdziarek Institute of Computer Science, University of Wroclaw, Przesmyckiego 20, Wroclaw 51-151 Poland |
The logistic discriminant function receives much attention, especially in epidemiolo-gical research. It is thought as being more general than the ordinary linear Fisherian discriminant function. However, one has to pay for the assumed generality by applying more complicated iterative computations. Our aim is to issue a warning that the generality attained when using the logistic formula is very restricted. We show in more detail for what kind of data the logistic discriminant function is better, and when it is equivalent to the linear Fisherian one. We also show for what kind of data both the logistic and the linear Fisherian discriminant function are worse than the quadratic discriminant function.
two-group discriminant function, probability a posteriori, exponential family of distributions, shape of covariance matrices, long-tailed distributions.
