The summary method for the discrim object.

DESCRIPTION:

Computes summary statistics for a discrim object and statistical tests of the p-variate training data used to compute the discriminant function.

USAGE:

summary.discrim(object, ...)

REQUIRED ARGUMENTS:

object
a discrim object constructed by the discrim function.

OPTIONAL ARGUMENTS:

...
arguments to the various methods called by this summary method.

VALUE:

a summary.discrim object.

METHODS:

Objects of this class have the following methods:

ARGUMENTS:

print
the print method.

STRUCTURE:

The structure of a summary.discrim object is a list with the following data members.

ARGUMENTS:

discrim
the discrim object, object.
classify
the g x g classification table, where g is the number of groups. The method used to classify observations of the training data can be any of the method options of the predict method, where the default is "plug-in". See predict.discrim for details. The error rate estimates are discussed under the CV data member.
rule.MSE
the mean squared error of the Bayes Rule (McLachlan, 1992, p.20). This is only computed if the plug-in method is used in computing the classification table contained in the classify data member.
CV
the classification table using leave-one-out cross-validation. This is only computed if object is fitted without weights using the spherical, homoscedastic, group spherical, or heteroscedastic covariance models. See crossvalidate.discrim for details on cross-validation. Nonparametric estimates of the conditional error rates are estimated based on miss classified counts and base on posterior probabilities. If the prior probabilities are not proportional to the sample sizes, then stratified posterior error rates are estimated also. As seen in the example below, it is possible for the posterior error rate estimates to be negative.
means.tests
contains the multivariate statistics computed by anova.discrim for the training data used to compute the discriminant function. These statistics are only computed if the spherical or homoscedastic covariance models are used. See anova.discrim for details.
hotelling.T2
contains the statistics computed by multicomp.discrim. These include Hotelling's T squared statistics for all pairs of groups in the training data. For each significant T squared statistic confidence intervals are computed for the differences between means of the p-variate feature vectors of the two groups.
homogeneity.tests
contains Box's M statistic and the adjusted M statistic for testing homogeneity of covariances (Seber, 1984, p.448-451). Both statistics are computed if the homoscedastic or the heteroscedastic covariance models are used. Box's M statistic is computed if the spherical or group spherical covariance models are used. Box's M statistic is the likelihood ratio test between the homoscedastic and heterosedastic models (see anova.discrim for details).
KS.tests
contains the Kolmogorov-Smirnov test for normality. See ks.gof for details.

DETAILS:

Computes various statistics for the discriminant function and for the training data used to compute the discriminant function. Most statistics computed are implementations of methods of the discrim object. See the STRUCTURE section above for the various implementations.

REFERENCES:



McLachlan, G. J. (1992). Discriminant Analysis and Statistical Pattern Recognition, John Wiley & Sons.

Seber, G.A.F. (1984). Multivariate Observations, John Wiley & Sons.

SEE ALSO:

, , , , , , .

EXAMPLES:

# Iris data
iris.quad <- discrim(Species~Sepal.L. + Sepal.W. + Petal.L. + Petal.W.,
    data=iris.mm, family=Classical("heter"))
summary(iris.quad)