| pdRankTests {pdSpecEst} | R Documentation |
Rank-based hypothesis tests for HPD matrices
Description
pdRankTests performs a number of generalized rank-based hypothesis tests in the metric space of HPD matrices equipped
with the affine-invariant Riemannian metric or Log-Euclidean metric for samples of HPD matrices or samples of sequences
(curves) of HPD matrices as described in Chapter 4 of (Chau 2018).
Usage
pdRankTests(data, sample_sizes, test = c("rank.sum", "krusk.wall",
"signed.rank", "bartels"), depth = c("gdd", "zonoid", "spatial"),
metric = c("Riemannian", "logEuclidean"))
Arguments
data |
either a |
sample_sizes |
a numeric vector specifying the individual sample sizes in the pooled sample |
test |
rank-based hypothesis testing procedure, one of |
depth |
data depth measure used in the rank-based tests, one of |
metric |
the metric that the space of HPD matrices is equipped with, either |
Details
For samples of (d,d)-dimensional HPD matrices with pooled sample size S, the argument
data is a (d,d,S)-dimensional array of (d,d)-dimensional HPD matrices, where the individual samples are
combined along the third array dimension. For samples of sequences of (d,d)-dimensional HPD matrices with pooled sample
size S, the argument data is a (d,d,n,S)-dimensional array of length n sequences
of (d,d)-dimensional HPD matrices, where the individual samples are combined along the fourth array dimension. The argument
sample_sizes specifies the sizes of the individual samples so that sum(sample_sizes) is equal to S.
The available generalized rank-based testing procedures (specified by the argument test) are:
"rank.sum"Intrinsic Wilcoxon rank-sum test to test for homogeneity of distributions of two independent samples of HPD matrices or samples of sequences of HPD matrices. The usual univariate ranks are replaced by data depth induced ranks obtained with
pdDepth."krusk.wall"Intrinsic Kruskal-Wallis test to test for homogeneity of distributions of more than two independent samples of HPD matrices or samples of sequences of HPD matrices. The usual univariate ranks are replaced by data depth induced ranks obtained with
pdDepth."signed.rank"Intrinsic signed-rank test to test for homogeneity of distributions of independent paired or matched samples of HPD matrices. The intrinsic signed-rank test is not based on data depth induced ranks, but on a specific difference score in the Riemannian manifold of HPD matrices equipped with either the affine-invariant Riemannian or Log-Euclidean metric.
"bartels"Intrinsic Bartels-von Neumann test to test for randomness (i.e., exchangeability) within a single independent sample of HPD matrices or a sample of sequences of HPD matrices. The usual univariate ranks are replaced by data depth induced ranks obtained with
pdDepth.
The function computes the generalized rank-based test statistics in the complete metric space of HPD matrices equipped with one of the following metrics: (i) the Riemannian metric (default) as detailed in e.g., (Bhatia 2009)[Chapter 6] or (Pennec et al. 2006); or (ii) the Log-Euclidean metric, the Euclidean inner product between matrix logarithms. The default Riemannian metric is invariant under congruence transformation by any invertible matrix, whereas the Log-Euclidean metric is only invariant under congruence transformation by unitary matrices, see (Chau 2018)[Chapter 4] for more details.
Value
The function returns a list with five components:
test |
name of the rank-based test |
p.value |
p-value of the test |
statistic |
computed test statistic |
null.distr |
distribution of the test statistic under the null hypothesis |
depth.values |
computed data depth values (if available) |
Note
The intrinsic signed-rank test also provides a valid test for equivalence of spectral matrices of two multivariate stationary time
series based on the HPD periodogram matrices obtained via pdPgram, see (Chau 2018)[Chapter 4] for the details.
The function does not check for positive definiteness of the input matrices, and may fail if matrices are close to being singular.
The data depth computations under the Riemannian metric are more involved than under the Log-Euclidean metric, and may therefore result in (significantly) higher computation times.
References
Bhatia R (2009).
Positive Definite Matrices.
Princeton University Press, New Jersey.
Chau J (2018).
Advances in Spectral Analysis for Multivariate, Nonstationary and Replicated Time Series.
phdthesis, Universite catholique de Louvain.
Pennec X, Fillard P, Ayache N (2006).
“A Riemannian framework for tensor computing.”
International Journal of Computer Vision, 66(1), 41–66.
See Also
Examples
## null hypothesis is true
data <- replicate(100, Expm(diag(2), H.coeff(rnorm(4), inverse = TRUE)))
pdRankTests(data, sample_sizes = c(50, 50), test = "rank.sum") ## homogeneity 2 samples
pdRankTests(data, sample_sizes = rep(25, 4), test = "krusk.wall") ## homogeneity 4 samples
pdRankTests(data, test = "bartels") ## randomness
## null hypothesis is false
data1 <- array(c(data, replicate(50, Expm(diag(2), H.coeff(0.5 * rnorm(4), inverse = TRUE)))),
dim = c(2,2,150))
pdRankTests(data1, sample_sizes = c(100, 50), test = "rank.sum")
pdRankTests(data1, sample_sizes = rep(50, 3), test = "krusk.wall")
pdRankTests(data1, test = "bartels")
## Not run:
## signed-rank test for equivalence of spectra of multivariate time series
## ARMA(1,1) process: Example 11.4.1 in (Brockwell and Davis, 1991)
Phi <- array(c(0.7, 0, 0, 0.6, rep(0, 4)), dim = c(2, 2, 2))
Theta <- array(c(0.5, -0.7, 0.6, 0.8, rep(0, 4)), dim = c(2, 2, 2))
Sigma <- matrix(c(1, 0.71, 0.71, 2), nrow = 2)
pgram <- function(Sigma) pdPgram(rARMA(2^8, 2, Phi, Theta, Sigma)$X)$P
## null is true
pdRankTests(array(c(pgram(Sigma), pgram(Sigma)), dim = c(2,2,2^8)), test = "signed.rank")
## null is false
pdRankTests(array(c(pgram(Sigma), pgram(0.5 * Sigma)), dim = c(2,2,2^8)), test = "signed.rank")
## End(Not run)