Variance Covariance matrix of for g(n)ls and (n)lme models
Extracts the variance covariance matrix (residuals, random or all)
var_cov(
object,
type = c("residual", "random", "all"),
aug = FALSE,
sparse = FALSE,
data = NULL
)object |
|
type |
“residual” for the variance-covariance for the residuals, “random” for the variance-covariance of the random effects or “all” for the sum of both. |
aug |
whether to augment the matrix of the random effects to the dimensions of the data |
sparse |
whether to return a sparse matrix (default is FALSE) |
data |
optional passing of ‘data’, probably needed when using this function inside other functions. |
Variance Covariance matrix for (non)linear mixed models
See Chapter 5 of Pinheiro and Bates. This returns potentially a very large
matrix of N x N, where N is the number of rows in the data.frame.
The function getVarCov only works well for
lme objects.
The equivalence is more or less:
getVarCov type = “random.effects” equivalent to var_cov type = “random”.
getVarCov type = “conditional” equivalent to var_cov type = “residual”.
getVarCov type = “marginal” equivalent to var_cov type = “all”.
The difference is that getVarCov has an argument that specifies the individual
for which the matrix is being retrieved and var_cov returns the full matrix only.
require(graphics)
require(nlme)
data(ChickWeight)
## First a linear model
flm <- lm(weight ~ Time, data = ChickWeight)
vlm <- var_cov(flm)
## First model with no modeling of the Variance-Covariance
fit0 <- gls(weight ~ Time, data = ChickWeight)
v0 <- var_cov(fit0)
## Only modeling the diagonal (weights)
fit1 <- gls(weight ~ Time, data = ChickWeight, weights = varPower())
v1 <- var_cov(fit1)
## Only the correlation structure is defined and there are no groups
fit2 <- gls(weight ~ Time, data = ChickWeight, correlation = corAR1())
v2 <- var_cov(fit2)
## The correlation structure is defined and there are groups present
fit3 <- gls(weight ~ Time, data = ChickWeight, correlation = corCAR1(form = ~ Time | Chick))
v3 <- var_cov(fit3)
## There are both weights and correlations
fit4 <- gls(weight ~ Time, data = ChickWeight,
weights = varPower(),
correlation = corCAR1(form = ~ Time | Chick))
v4 <- var_cov(fit4)
## Tip: you can visualize these matrices using
image(log(v4[,ncol(v4):1]))Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.