2
% This file is a component of the package 'sn' for R
3
% copyright (C) 2002-2013 Adelchi Azzalini
6
Multivariate skew-\eqn{t} distribution
9
Probability density function, distribution function and random number
10
generation for the multivariate skew-\eqn{t} (MST) distribution.
12
\title{Multivariate skew-\eqn{t} distribution and skew-Cauchy distribution}
14
\description{Probability density function, distribution function and random
15
number generation for the multivariate skew-\eqn{t} (\acronym{ST}) and
16
skew-Cauchy (\acronym{SC}) distributions.}
13
dmst(x, xi=rep(0,length(alpha)), Omega, alpha, df=Inf, dp = NULL, log=FALSE)
14
pmst(x, xi=rep(0,length(alpha)), Omega, alpha, df=Inf, dp = NULL, ...)
15
rmst(n=1, xi=rep(0,length(alpha)), Omega, alpha, df=Inf, dp = NULL)
19
dmst(x, xi=rep(0,length(alpha)), Omega, alpha, nu=Inf, dp=NULL, log=FALSE)
20
pmst(x, xi=rep(0,length(alpha)), Omega, alpha, nu=Inf, dp=NULL, ...)
21
rmst(n=1, xi=rep(0,length(alpha)), Omega, alpha, nu=Inf, dp=NULL)
22
dmsc(x, xi=rep(0,length(alpha)), Omega, alpha, dp=NULL, log=FALSE)
23
pmsc(x, xi=rep(0,length(alpha)), Omega, alpha, dp=NULL, ...)
24
rmsc(n=1, xi=rep(0,length(alpha)), Omega, alpha, dp=NULL)
20
for \code{dmst}, this is either a vector of length \code{d},
21
where \code{d=length(alpha)}, or a matrix with \code{d} columns,
22
giving the coordinates of the point(s) where the density must be
23
avaluated; for \code{pmst}, only a vector of length \code{d} is allowed.
26
a numeric vector of lenght \code{d}, or a matrix with \code{d} columns,
27
representing the location parameter of the distribution; see Background.
28
If \code{xi} is a matrix, its dimensions must agree with those of \code{x}.
31
a symmetric positive-definite matrix of dimension \code{(d,d)}; see Background.
34
a numeric vector which regulates the shape of the density; see Background
37
degrees of freedom (scalar); default is \code{df=Inf} which corresponds
38
to the multivariate skew-normal distribution.
41
a list with three elements named \code{xi}, \code{Omega}, \code{alpha}
42
and \code{df}, containing quantities as described above.
43
If \code{dp} is specified, this overrides the individual parameter
47
a numeric value which represents the number of random vectors
51
logical (default value: \code{FALSE}); if TRUE, log-densities are returned.
54
additional parameters passed to \code{pmt}
57
A vector of density values (\code{dmst}) or a single probability
58
(\code{pmst}) or a matrix of random points (\code{rmst}).
29
\item{x}{for \code{dmst} and \code{dmsc}, this is either a vector of length
30
\code{d}, where \code{d=length(alpha)}, or a matrix with \code{d} columns,
31
representing the coordinates of the point(s) where the density must be
32
avaluated; for \code{pmst} and \code{pmsc}, only a vector of length
35
\item{xi}{a numeric vector of lenght \code{d}, or a matrix with \code{d}
36
columns, representing the location parameter of the distribution; see
37
\sQuote{Background}. If \code{xi} is a matrix, its dimensions must agree
38
with those of \code{x}.}
40
\item{Omega}{a symmetric positive-definite matrix of dimension \code{(d,d)};
41
see Section \sQuote{Background}.}
43
\item{alpha}{a numeric vector of length \code{d} which regulates the slant
44
of the density; see Section \sQuote{Background}.
45
\code{Inf} values in \code{alpha} are not allowed.}
47
\item{nu}{a positive value representing the degrees of freedom of
48
\acronym{ST} distribution; default value is \code{nu=Inf} which corresponds
49
to the multivariate skew-normal distribution.}
51
\item{dp}{a list with three elements named \code{xi}, \code{Omega},
52
\code{alpha} and \code{nu}, containing quantities as described above. If
53
\code{dp} is specified, this prevents specification of the individual
56
\item{n}{a numeric value which represents the number of random vectors to be
57
drawn; default value is \code{1}.}
59
\item{log}{logical (default value: \code{FALSE}); if \code{TRUE},
60
log-densities are returned.}
62
\item{...}{additional parameters passed to \code{pmt}.}
66
\value{A vector of density values (\code{dmst} and \code{dmsc}) or a single
67
probability (\code{pmst} and \code{pmsc}) or a matrix of random points
68
(\code{rmst} and \code{rmst}).}
60
70
\details{Typical usages are
62
dmst(x, xi=rep(0,length(alpha)), Omega, alpha, df=Inf, log=FALSE)
72
dmst(x, xi=rep(0,length(alpha)), Omega, alpha, nu=Inf, log=FALSE)
63
73
dmst(x, dp=, log=FALSE)
64
pmst(x, xi=rep(0,length(alpha)), Omega, alpha, df=Inf, ...)
74
pmst(x, xi=rep(0,length(alpha)), Omega, alpha, nu=Inf, ...)
66
rmst(n=1, xi=rep(0,length(alpha)), Omega, alpha, df=Inf)
76
rmst(n=1, xi=rep(0,length(alpha)), Omega, alpha, nu=Inf)
78
dmsc(x, xi=rep(0,length(alpha)), Omega, alpha, log=FALSE)
79
dmsc(x, dp=, log=FALSE)
80
pmsc(x, xi=rep(0,length(alpha)), Omega, alpha, ...)
82
rmsc(n=1, xi=rep(0,length(alpha)), Omega, alpha)
70
The positive-definiteness of \code{Omega} is not tested for efficiency
71
reasons. Function \code{pmst} requires \code{pmt} from package \code{mnormt};
72
the accuracy of its computation can be controlled via use of \code{...}
86
Function \code{pmst} requires \code{\link[mnormt]{dmt}} from package
87
\pkg{mnormt}; the accuracy of its computation can be controlled via
88
argument \code{\dots}.}
75
90
\section{Background}{
76
The family of multivariate skew-\eqn{t} distributions is an extension of the
77
multivariate Student's \eqn{t} family, via the introduction of a \code{shape}
78
parameter which regulates skewness; when \code{shape=0}, the skew-\eqn{t}
79
distribution reduces to the usual \eqn{t} distribution.
80
When \code{df=Inf} the distribution reduces to the multivariate skew-normal
91
The family of multivariate \acronym{ST} distributions is an extension of the
92
multivariate Student's \eqn{t} family, via the introduction of a \code{alpha}
93
parameter which regulates asymmetry; when \code{alpha=0}, the skew-\eqn{t}
94
distribution reduces to the commonly used form of multivariate Student's
95
\eqn{t}. Further, location is regulated by \code{xi} and scale by
96
\code{Omega}, when its diagonal terms are not all 1's.
97
When \code{nu=Inf} the distribution reduces to the multivariate skew-normal
81
98
one; see \code{dmsn}. Notice that the location vector \code{xi}
82
99
does not represent the mean vector of the distribution (which in fact
83
may not even exist if \code{df <= 1}), and similarly
84
\code{Omega} is not \emph{the} covariance matrix of the distribution,
85
although it is \emph{a} covariance matrix.
86
For additional information, see the reference below.
100
may not even exist if \code{nu <= 1}), and similarly \code{Omega} is not
101
\emph{the} covariance matrix of the distribution, although it is \emph{a}
103
For additional information, see Section 6.2 of the reference below.
105
The family of multivariate \acronym{SC} distributions is the subset of the
106
\acronym{ST} family, obtained when \code{nu=1}. While in the univariate case
107
there are specialized functions for the \acronym{SC} distribution,
108
\code{dmsc}, \code{pmsc} and \code{rmsc} simply make a call to \code{dmst,
109
pmst, rmst} with argument \code{nu} set equal to 1.}
90
Azzalini, A. and Capitanio, A. (2003).
91
Distributions generated by perturbation of symmetry
92
with emphasis on a multivariate skew \emph{t} distribution.
93
\emph{J.Roy. Statist. Soc. B} \bold{65}, 367--389.
112
% Azzalini, A. and Capitanio, A. (2003).
113
% Distributions generated by perturbation of symmetry
114
% with emphasis on a multivariate skew \emph{t} distribution.
115
% \emph{J.Roy. Statist. Soc. B} \bold{65}, 367--389.
117
Azzalini, A. with the collaboration of Capitanio, A. (2014).
118
\emph{The Skew-Normal and Related Families}.
119
Cambridge University Press, IMS Monograph series.
96
\code{\link{dst}}, \code{\link{dmsn}}, \code{\link[mnormt]{dmt}}
122
\code{\link{dst}}, \code{\link{dsc}}, \code{\link{dmsn}},
123
\code{\link[mnormt]{dmt}}, \code{\link{makeSECdistr}}
99
126
x <- seq(-4,4,length=15)