In blangworthy/transCCA: Estimation of transelliptical CCA

knitr::opts_chunk$set(echo = TRUE)

transCCA: A package for estimation transelliptical CCA using the transformed Kendall's scatter matrix estimator

The transCCA package estimates the transelliptical CCA directions and correlations, as well as their asymptotic variances using the transformed Kendall's scatter matrix estimator.

Installation

devtools::install_github("blangworthy/transCCA")
library(transCCA)

Examples

library(transCCA)
library(mvtnorm)

sigmax <- diag(8)
sigmay <- diag(8)
sigmaxy <- diag(c(0.9,0.5,0.4,1/3,rep(0,4)))

sigma <- rbind(cbind(sigmax,sigmaxy),cbind(sigmaxy,sigmay))####Scatter matrix for simulated data

z <- mvtnorm::rmvt(n=1000,sigma = sigma,df = 5)####Simulate from multivariate t distribution with 5 degrees of freedom

transccout <- transCCA(z[,1:8],z[,9:16])####Estimate the transelliptical canonical directions and correlations

dir1var <- transCorVar(z[,1:8],z[,9:16],1)####Asymptotic Variance estimates of transelliptical canonical directions and correlations

####Upper and lower bound for asymptotic confidence intervals for first direction 

xdir1ub <- transccout$xcoef[1,] + qnorm(0.975)*sqrt(diag(dir1var$XCoefVar)/nrow(z))
xdir1lb <- transccout$xcoef[1,] - qnorm(0.975)*sqrt(diag(dir1var$XCoefVar)/nrow(z))

ydir1ub <- transccout$ycoef[1,] + qnorm(0.975)*sqrt(diag(dir1var$YCoefVar)/nrow(z))
ydir1lb <- transccout$ycoef[1,] - qnorm(0.975)*sqrt(diag(dir1var$YCoefVar)/nrow(z))

####Upper and lower bound for asymptotic confidence intervals for correlations
####Use bootstrap confidence intervals for transelliptical confidence correlations if sample size is small
corub <- min(transccout$cancor[1] + qnorm(0.975)*sqrt(dir1var$corVar/nrow(z)),1)
corlb <- max(transccout$cancor[1] - qnorm(0.975)*sqrt(dir1var$corVar/nrow(z)),0)