1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
|
\name{mclustBootstrapLRT}
\alias{mclustBootstrapLRT}
\alias{print.mclustBootstrapLRT}
\alias{plot.mclustBootstrapLRT}
\title{Bootstrap Likelihood Ratio Test for the Number of Mixture Components}
\description{Perform the likelihood ratio test (LRT) for assessing the number of mixture components in a specific finite mixture model parameterisation. The observed significance is approximated by using the (parametric) bootstrap for the likelihood ratio test statistic (LRTS).}
\usage{
mclustBootstrapLRT(data, modelName = NULL, nboot = 999, level = 0.05, maxG = NULL,
verbose = interactive(), \dots)
\method{print}{mclustBootstrapLRT}(x, \dots)
\method{plot}{mclustBootstrapLRT}(x, G = 1, hist.col = "grey", hist.border = "lightgrey", breaks = "Scott",
col = "forestgreen", lwd = 2, lty = 3, main = NULL, \dots)
}
\arguments{
\item{data}{A numeric vector, matrix, or data frame of observations. Categorical
variables are not allowed. If a matrix or data frame, rows
correspond to observations and columns correspond to variables.}
\item{modelName}{A character string indicating the mixture model to be fitted.
The help file for \code{\link{mclustModelNames}} describes the available models.}
\item{nboot}{The number of bootstrap replications to use (by default 999).}
\item{level}{The significance level to be used to terminate the sequential bootstrap procedure.}
\item{maxG}{The maximum number of mixture components \eqn{G} to test. If not provided
the procedure is stopped when a test is not significant at the specified \code{level}.}
\item{verbose}{A logical controlling if a text progress bar is displayed during the bootstrap procedure. By default is \code{TRUE} if the session is interactive, and \code{FALSE} otherwise.}
\item{\dots}{Further arguments passed to or from other methods. In particular, see the optional arguments in \code{\link{mclustBIC}}.}
\item{x}{An \code{'mclustBootstrapLRT'} object.}
\item{G}{A value specifying the number of components for which to plot the
bootstrap distribution.}
\item{hist.col}{The colour to be used to fill the bars of the histogram.}
\item{hist.border}{The color of the border around the bars of the histogram.}
\item{breaks}{See the argument in function \code{\link[graphics]{hist}}.}
\item{col, lwd, lty}{The color, line width and line type to be used to represent the observed LRT statistic.}
\item{main}{The title for the graph.}
}
\details{The implemented algorithm for computing the LRT observed significance using the bootstrap is the following.
Let \eqn{G_0} be the number of mixture components under the null hypothesis versus \eqn{G_1 = G_0+1} under the alternative. Bootstrap samples are drawn by simulating data under the null hypothesis. Then, the p-value may be approximated using eq. (13) on McLachlan and Rathnayake (2014). Equivalently, using the notation of Davison and Hinkley (1997) it may be computed as
\deqn{\textnormal{p-value} = \frac{1 + \#\{LRT^*_b \ge LRTS_{obs}\}}{B+1}}{%
p-value = (1 + #{LRTS*_b \ge LRT_obs}) / (B+1)}
where \cr
\eqn{B} = number of bootstrap samples \cr
\eqn{LRT_{obs}}{LRT_obs} = LRTS computed on the observed data\cr
\eqn{LRT^*_b}{LRT*_b} = LRTS computed on the \eqn{b}th bootstrap sample.
}
\value{An object of class \code{'mclustBootstrapLRT'} with the following components:
\item{G}{A vector of number of components tested under the null hypothesis.}
\item{modelName}{A character string specifying the mixture model as provided
in the function call (see above).}
\item{obs}{The observed values of the LRTS.}
\item{boot}{A matrix of dimension \code{nboot} x the number of components tested
containing the bootstrap values of LRTS.}
\item{p.value}{A vector of p-values.}
}
\references{
Davison, A. and Hinkley, D. (1997) \emph{Bootstrap Methods and Their Applications}. Cambridge University Press.
McLachlan G.J. (1987) On bootstrapping the likelihood ratio test statistic for the number of components in a normal mixture. \emph{Applied Statistics}, 36, 318-324.
McLachlan, G.J. and Peel, D. (2000) \emph{Finite Mixture Models}. Wiley.
McLachlan, G.J. and Rathnayake, S. (2014) On the number of components in a Gaussian mixture model. \emph{Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery}, 4(5), pp. 341-355.
}
\seealso{\code{\link{mclustBIC}}, \code{\link{mclustICL}}, \code{\link{Mclust}}}
\examples{
\donttest{
data(faithful)
faithful.boot = mclustBootstrapLRT(faithful, model = "VVV")
faithful.boot
plot(faithful.boot, G = 1)
plot(faithful.boot, G = 2)
}
}
\keyword{htest}
\keyword{cluster}
|
