Professional Documents
Culture Documents
10.1
Concept
The objective of discriminant analysis is to determine group membership
of samples from a group of predictors by finding linear combinations of the
variables which maximize the differences between the populations being
studied, with the objective of establishing a model to sort objects into their
appropriate populations with minimal error.
10.2
Definitions
Discriminant Functions. A linear combination of weighting coefficients and
standardized values of discriminating variables. It is one less than the number of
groups being compared.
Centroid. Mean discriminant scores for each group on each function.
Canonical Correlation. Correlation between a discriminant function and the
groups.
Wilks'Lambda. A calculation used to determine if amounts of variance accoun-
ted for by discriminant variables are significant. A problem that arises quite
often in science is to discriminate between two groups of individuals or objects
on the basis of several properties of those individuals or samples.
10.3
Overview
In geohydrology, for example, a hydrologist may want to classify a water sample
into one of two classes based on measured chemical properties. When two or more
variables are used to predict membership in categories or groups, the method is
known as multiple discriminant analysis. The degree to which members and
different groups can be differentiated in terms of an array of discriminator
variables is the essence of this technique. It may be very difficult in some instances
to find a discriminating index number if the two samples have almost identical
properties.
then represents a plane in three dimensions passing through the origin and
having direction numbers AI' ~, and -1. The geometry of this discriminating
process is shown in Fig. IO.l.
As another example, consider an archeologist who wishes to determine which
of two possible indigenous groups created a particular statue found in an ex-
ploratory dig. Measurements are taken of several characteristics of the statue. It
must now be decided whether these measurements are more likely to have come