discrim
object.
multicomp.discrim(x, alpha=0.05, significant.only=T, ...)
discrim
object constructed by the
discrim
function.
alpha
,
multiple comparisons are computed between the means of the p-variate feature
vectors for those two groups.
multicomp.discrim
object.
Objects of this class have the following methods:
The structure of a
multicomp.discrim
object is a list with the following data members.
multicomp
objects. A
multicomp
object will exist for each significant
T squared statistic in the
hotellings.T2
data member.
Computes Hotelling's T squared statistics for all pairs of groups. For T squared
tests that have a significant F statistic, confidence intervals are computed
for the difference between the p-variate means of the feature data for those
two groups using multiple comparison techniques to control the
family-wise error rate (Hsu, 1996). See
multicomp.default
for details.
When the covariance matrices for each group are not equal the covariance
for the difference between the two p-variate mean vectors is taken to be
the sum of the two covariance matrices weighted by the inverse of the group
sample size. Yao's approximation to the second
degrees of freedom for the corresponding F statistic is used (Seber, 1984, p.115).
This covariance and degrees of freedom are also used in computing the pairwise
confidence intervals for the p means if the F statistic is significant. These
correspond to the
vmat
and
df.residual
in
multicomp.default
.
Hsu, J. C. (1996).
Multiple Comparisons Theory and Methods,
Chapman & Hall.
Seber, G.A.F. (1984).
Multivariate Observations,
John Wiley & Sons.
# Flea beetle data from Seber (1984) flea.lin <- discrim(species ~ x1 + x2, data=flea.beetles, family=Classical("lin"), prior="none") flea.spher <- discrim(species ~ x1 + x2, data=flea.beetles, family=Classical("spher"), prior="none") multicomp(flea.spher)