OpenMx (version 2.2.2)

mxKalmanScores: Estimate Kalman scores and error covariance matrices

Description

This function creates the Kalman predicted, Kalman updated, and Rauch-Tung-Striebel smoothed latent state and error covariance estimates for an MxModel object that has an MxExpectationStateSpace object.

Usage

mxKalmanScores(model, data=NA)

Arguments

model

An MxModel object with an MxExpectationStateSpace.

data

An optional data.frame or matrix.

Value

• A list with components xPredicted, PPredicted, xUpdated, PUpdated, xSmoothed, PSmoothed, m2ll, and L. The rows of xPredicted, xUpdated, and xSmoothed correspond to different time points. The columns are the different latent variables. The third index of PPredicted, PUpdated, and PSmoothed corresponds to different times. This works nicely with the R default print method for arrays. At each time there is a covariance matrix of the latent variable scores. For all items listed below, the first element goes with the zeroth time point (See example). [object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object]

Details

This is a helper function that computes the results of the classical Kalman filter. In particular, for every row of data there is a predicted latent score, an error covariance matrix for the predicted latent scores that provides an estimate of the predictions precision, an updated latent score, and an updated error covariance matrix for the updated lated scores. Additionally, the Rauch-Tung-Striebel (RTS) smoothed latent scores and error covariance matrices are returned.

References

J. Durbin and S.J. Koopman. (2001). Time Series Analysis by State Space Methods. Oxford University Press.

R.E. Kalman (1960). A New Approach to Linear Filtering and Prediction Problems. Basic Engineering, 82, 35-45.

H.E. Rauch, F. Tung, C.T. Striebel. (1965). Maximum Likelihood Estimates of Linear Dynamic Systems. American Institute of Aeronautics and Astronautics Journal, 3, 1445-1450.

The OpenMx User's guide can be found at http://openmx.psyc.virginia.edu/documentation.

See Also

mxExpectationStateSpace

Examples

# Create and fit a model using mxMatrix, mxExpectationStateSpace, and mxFitFunctionML
require(OpenMx)
data(demoOneFactor)
# Use only first 50 rows, for speed of example
data <- demoOneFactor[1:50,]
nvar <- ncol(demoOneFactor)
varnames <- colnames(demoOneFactor)
ssModel <- mxModel(model="State Space Manual Example",
    mxMatrix("Full", 1, 1, TRUE, .3, name="A"),
    mxMatrix("Zero", 1, 1, name="B"),
    mxMatrix("Full", nvar, 1, TRUE, .6, name="C", dimnames=list(varnames, "F1")),
    mxMatrix("Zero", nvar, 1, name="D"),
    mxMatrix("Diag", 1, 1, FALSE, 1, name="Q"),
    mxMatrix("Diag", nvar, nvar, TRUE, .2, name="R"),
    mxMatrix("Zero", 1, 1, name="x0"),
    mxMatrix("Diag", 1, 1, FALSE, 1, name="P0"),
    mxMatrix("Zero", 1, 1, name="u"),
    mxData(observed=data, type="raw"),
    mxExpectationStateSpace("A", "B", "C", "D", "Q", "R", "x0", "P0", "u"),
    mxFitFunctionML()
)
ssRun <- mxRun(ssModel)
summary(ssRun)
# Note the freely estimated Autoregressive parameter (A matrix)
#  is near zero as it should be for the independent rows of data
#  from the factor model.

ssScores <- mxKalmanScores(ssRun)

cor(cbind(ssScores$xPredicted[,1], ssScores$xUpdated[,1], ssScores$xSmoothed[,1]))
# Because the autoregressive dynamics are near zero, the predicted and updated scores
# correlate minimally, and the updated and smoothed latent state estimates
# are extremely close.

# The first few latent predicted scores
head(ssScores$xPredicted)

# The predicted latent score for time 10
ssScores$xPredicted[10+1,]

# The error covariance of the predicted score at time 10
ssScores$PPredicted[,,10+1]