The EM algorithm constitutes the following steps: Start with an initial parameter vector. Predict the missing data to form a completed data model. Optimize the completed data model to obtain a new parameter vector. Repeat these steps until convergence criteria are met.
mxComputeEM(expectation = NULL, predict = NA_character_, mstep,
observedFit = "fitfunction", ..., maxIter = 500L,
tolerance = 1e-09, verbose = 0L, freeSet = NA_character_,
accel = "varadhan2008", information = NA_character_,
infoArgs = list(), estep = NULL)a vector of expectation names (DEPRECATED)
what to predict from the observed data (DEPRECATED)
a compute plan to optimize the completed data model
the name of the observed data fit function (defaults to "fitfunction")
Not used. Forces remaining arguments to be specified by name.
maximum number of iterations
optimization is considered converged when the maximum relative change in fit is less than tolerance
level of diagnostic output
names of matrices containing free variables
name of acceleration method ("varadhan2008" or "ramsay1975")
name of information matrix approximation method
arguments to control the information matrix method
a compute plan to perform the expectation step
The arguments to this function have evolved. The old style
mxComputeEM(e,p,mstep=m) is equivalent to the new style
mxComputeEM(estep=mxComputeOnce(e,p), mstep=m). This change
allows the API to more closely match the literature on the E-M
method. You might use mxAlgebra(..., fixed=TRUE) to
contain the results of the E-step and then cause this algebra to
be recomputed using mxComputeOnce.
This compute plan does not work with any and all expectations. It requires a special kind of expectation that can predict its missing data to create a completed data model.
The EM algorithm does not produce a parameter covariance matrix for standard errors. The Oakes (1999) direct method and S-EM, an implementation of Meng & Rubin (1991), are included.
Ramsay (1975) was recommended in Bock, Gibbons, & Muraki (1988).
Bock, R. D., Gibbons, R., & Muraki, E. (1988). Full-information item factor analysis. Applied Psychological Measurement, 6(4), 431-444.
Dempster, A. P., Laird, N. M., & Rubin, D. B. (1977). Maximum likelihood from incomplete data via the EM algorithm. Journal of the Royal Statistical Society. Series B (Methodological), 1-38.
Meng, X.-L. & Rubin, D. B. (1991). Using EM to obtain asymptotic variance-covariance matrices: The SEM algorithm. Journal of the American Statistical Association, 86 (416), 899-909.
Oakes, D. (1999). Direct calculation of the information matrix via the EM algorithm. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 61(2), 479-482.
Ramsay, J. O. (1975). Solving implicit equations in psychometric data analysis. Psychometrika, 40 (3), 337-360.
Varadhan, R. & Roland, C. (2008). Simple and globally convergent methods for accelerating the convergence of any EM algorithm. Scandinavian Journal of Statistics, 35, 335-353.