Compute standard errors in OpenMx
This function allows you to obtain standard errors for arbitrary expressions, named entities, and algebras.
mxSE( x, model, details = FALSE, cov, forceName = FALSE, silent = FALSE, ..., defvar.row = as.integer(NA), data = "data" )
x |
the parameter to get SEs on (reference or expression) |
model |
the |
details |
logical. Whether to provide further details, e.g. the full sampling covariance matrix of x. |
cov |
optional matrix of covariances among the free parameters. If missing, the inverse Hessian from the fitted model is used. |
forceName |
logical; defaults to |
silent |
logical; defaults to |
... |
further named arguments passed to |
defvar.row |
which row to load for any definition variables |
data |
name of data from which to load definition variables |
x can be the name of an algebra, a bracket address, named entity
or arbitrary expression.
When the details argument is TRUE, the full
sampling covariance matrix of x is also returned as part of a list.
The square root of the diagonals of this sampling covariance matrix are
the standard errors.
When supplying the cov argument, take care that the free parameter
covariance matrix is given, not the information matrix. These
two are inverses of one another.
This function uses the delta method to compute the standard error of arbitrary and possibly nonlinear functions of the free parameters. The delta method makes a first-order Taylor approximation of the nonlinear function. The nonlinear function is a map from all the free parameters to some transformed subset of parameters: the linearization of this map is given by the Jacobian J. In equation form, the delta method computes standard errors by the following:
J^T C J
where J is the Jacobian of the nonlinear parameter transformation and C is the covariance matrix of the free parameters (e.g., two times the inverse of the Hessian of the minus two log likelihood function).
SE value(s) returned as a matrix when details is FALSE.
When details is TRUE, a list of the SE value(s) and the full
sampling covariance matrix.
- mxCI
library(OpenMx)
data(demoOneFactor)
# ===============================
# = Make and run a 1-factor CFA =
# ===============================
latents = c("G") # the latent factor
manifests = names(demoOneFactor) # manifest variables to be modeled
# ===========================
# = Make and run the model! =
# ===========================
m1 <- mxModel("One Factor", type = "RAM",
manifestVars = manifests, latentVars = latents,
mxPath(from = latents, to = manifests, labels=paste0('lambda', 1:5)),
mxPath(from = manifests, arrows = 2),
mxPath(from = latents, arrows = 2, free = FALSE, values = 1),
mxData(cov(demoOneFactor), type = "cov", numObs = 500)
)
m1 = mxRun(m1)
mxSE(lambda5, model = m1)
mxSE(lambda1^2, model = m1)Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.