omxParallelCI {OpenMx} | R Documentation |
Calculate confidence intervals without re-doing the primary optimization.
Description
OpenMx provides two functions to calculate confidence intervals for already-run MxModel objects that contain an MxInterval object (i.e., an mxCI()
statement), without recalculating point estimates, fitfunction derivatives, or expectations.
The primary function is omxRunCI()
. This is a wrapper for omxParallelCI()
with arguments run=TRUE
and independentSubmodels=FALSE
, and is the recommended interface.
omxParallelCI()
does the work of calculating confidence intervals. The "parallel" in the function's name refers to the not-yet-implemented feature of running independent submodels in parallel.
Usage
omxRunCI(model, verbose = 0, optimizer = "SLSQP")
omxParallelCI(model, run = TRUE, verbose = 0, independentSubmodels = TRUE,
optimizer = mxOption(NULL, "Default optimizer"))
Arguments
model |
An MxModel object that contains an MxInterval object (i.e., an |
run |
Logical; For |
verbose |
Integer; defaults to zero; verbosity level passed to MxCompute* objects. |
independentSubmodels |
Logical; For |
optimizer |
Character string selecting the gradient-descent optimizer to be used to find confidence limits; one of "NPSOL", "CSOLNP", or "SLSQP". The default for |
Details
When independentSubmodels=TRUE
, omxParallelCI()
creates an independent MxModel object for each quantity specified in the 'reference' slot of model
's MxInterval object, and places these independent MxModels inside model
. Each of these independent submodels calculates the confidence limits of its own quantity when the container model is run. When independentSubmodels=FALSE
, no submodels are added to model
. Instead, model
is provided with a dedicated compute plan consisting only of an MxComputeConfidenceInterval step. Note that using independentSubmodels=FALSE
will overwrite any compute plan already inside model
.
Value
The functions return model
, augmented with independent submodels (if independentSubmodels=TRUE
) or with a non-default compute plan (if independentSubmodels=FALSE
), and possibly having been passed through mxRun()
(if run=TRUE
). Naturally, if run=FALSE
, the user can subsequently run the returned model to obtain confidence intervals. Users are cautioned that the returned model may not be very amenable to being further modified and re-fitted (e.g., having some free parameters fixed via omxSetParameters()
and passed through mxRun()
to get new point estimates) unless the added submodels or the non-default compute plan are eliminated. The exception is if run=TRUE
and independentSubmodels=TRUE
(which is always the case with omxRunCI()
), since the non-default compute plan is set to be non-persistent, and will automatically be replaced with a default compute plan the next time the model is passed to mxRun()
.
See Also
mxCI()
, MxInterval, mxComputeConfidenceInterval()
Examples
require(OpenMx)
# 1. Build and run a model, don't compute intervals yet
data(demoOneFactor)
manifests <- names(demoOneFactor)
latents <- c("G")
factorModel <- mxModel("One Factor", type="RAM",
manifestVars=manifests,
latentVars=latents,
mxPath(from=latents, to=manifests),
mxPath(from=manifests, arrows=2),
mxPath(from=latents, arrows=2, free=FALSE, values=1.0),
mxData(observed=cov(demoOneFactor), type="cov", numObs=500),
# Add confidence intervals for (free) params in A and S matrices.
mxCI(c('A', 'S'))
)
factorRun <- mxRun(factorModel)
# 2. Compute the CIs on factorRun, and view summary
factorCI1 <- omxRunCI(factorRun)
summary(factorCI1)$CI
# 3. Use low-level omxParallelCI interface
factorCI2 <- omxParallelCI(factorRun)
# 4. Build, but don't run the newly-created model
factorCI3 <- omxParallelCI(factorRun, run= FALSE)