sxenv: Fit the scaled predictor envelope model

In Renvlp: Computing Envelope Estimators

Fit the scaled predictor envelope model

Description

Fit the scaled predictor envelope model in multivariate linear regression with dimension u. The scaled predictor envelope model is a scale-invariant version of the predictor envelope model.

Usage

sxenv(X, Y, u, R, asy = TRUE, init = NULL)

Arguments

X	Predictors. An n by p matrix, p is the number of predictors. The predictors can be univariate or multivariate, discrete or continuous.
Y	Multivariate responses. An n by r matrix, r is the number of responses and n is number of observations. The responses must be continuous variables.
u	Dimension of the scaled envelope in the predictor space. An integer between 0 and p.
R	The number of replications of the scales. A vector, the sum of all elements of R must be p.
asy	Flag for computing the asymptotic variance of the envelope estimator. The default is TRUE. When p and r are large, computing the asymptotic variance can take much time and memory. If only the envelope estimators are needed, the flag can be set to asy = FALSE.
init	The user-specified value of Gamma for the scaled envelope subspace in the predictor space. An p by u matrix. The default is the one generated by function sxenvMU.

Details

This function fits the scaled envelope model in the predictor space to the responses and predictors,

Y = \mu_{Y} + \eta'\Gamma'\Lambda^{-1} (X - \mu_{X}) + \varepsilon, \Sigma _{X} = \Lambda\Gamma\Omega\Gamma'\Lambda + \Lambda\Gamma_{0}\Omega_{0}\Gamma'_{0}\Lambda

using the maximum likelihood estimation. When the dimension of the scaled envelope in the predictor space is between 1 and p-1, the starting value and blockwise coordinate descent algorithm in Cook et al. (2016) is implemented. When the dimension is p, then the scaled envelope model in the predictor space degenerates to the standard multivariate linear regression. When the dimension is 0, it means that X and Y are uncorrelated, and the fitting is different.

Value

The output is a list that contains the following components:

beta	The scaled envelope estimator of the regression coefficients.
Sigma	The scaled envelope estimator of the error covariance matrix.
Lambda	The matrix of estimated scale.
Gamma	An orthonormal basis of the scaled envelope subspace.
Gamma0	An orthonormal basis of the complement of the scaled envelope subspace.
eta	The coordinates of beta with respect to Gamma.
Omega	The coordinates of Sigma with respect to Gamma.
Omega0	The coordinates of Sigma with respect to Gamma0.
muY	The mean of Y.
muX	The mean of X.
loglik	The maximized log likelihood function.
covMatrix	The asymptotic covariance of vec(beta). The covariance matrix returned are asymptotic. For the actual standard errors, multiply by 1 / n.
asySE	The asymptotic standard error for elements in beta under the envelope model. The standard errors returned are asymptotic, for actual standard errors, multiply by 1 / sqrt(n).
ratio	The asymptotic standard error ratio of the standard multivariate linear regression estimator over the envelope estimator, for each element in beta.
n	The number of observations in the data.

References

Cook, R. D., Su, Z. (2016). Scaled Predictor Envelopes and Partial Least Squares Regression. Technometrics 58, 155 - 165.

Cook, R. D., Forzani, L. and Su, Z. (2016) A Note on Fast Envelope Estimation. Journal of Multivariate Analysis. 150, 42-54.

Examples

data(sales)
Y <- sales[, 1:3]
X <- sales[, 4:7]
R <- rep(1, 4)

u <- u.sxenv(X, Y, R)
u

m <- sxenv(X, Y, 2, R)
m$beta

Renvlp documentation built on Oct. 11, 2023, 1:06 a.m.