[R-pkgs] new version 2.0-0 of the sem package
jfox at mcmaster.ca
Mon Nov 7 20:05:27 CET 2011
Dear R users,
Jarrett Byrnes and I would like to announce version 2.0-0 of the sem package
for fitting observed- and latent-variable structural equation models. This
is a general reworking of the original sem package (which is still available
on R-Forge as package sem1).
Some highlights of sem 2.0-0 include:
o More convenient and compact model specification, including the default
automatic generation of error variances for endogenous variables and more
compact specification of error covariances. In the near future, we
anticipate releasing an update that permits equation-style specification of
o The ability to update model specifications.
o Soft-coded objective functions ("fit" functions in SEM jargon) and
optimizers. Two objective functions are provided, for multinormal
full-information maximum likelihood and for generalized least squares; and
three optimizers are provided, based on the standard R nlm(), optim(), and
nlminb() optimizers. The user can add objective functions and optimizers.
o Analytic standard errors are provided by default for the FIML estimator
(standard errors based on the numeric Hessian are now optional), and robust
standard errors and tests are optionally available.
o The ability to fit a model to a data frame, as a preferred alternative to
computing a covariance or moment matrix in an intermediate step. The
original data are required to obtain robust standard errors and tests, and
are optional otherwise.
o Correctly computed "modification indices" (score tests for fixed
o Enhanced output, including R^2s for endogenous variables, additional
information criteria, and the computation of indirect effects.
o Executable examples in the help pages for the package, along with (as
before) non-executable examples in which model specifications and
correlation/covariance/moment matrices appear in the input stream.
The new sem package is designed to be upwards compatible with the old one,
so that scripts that worked previously should still work.
Although we have tested the new sem package, this is a major update and it's
possible that bugs will surface. Please let me know if you encounter
Senator William McMaster
Professor of Social Statistics
Department of Sociology
Hamilton, Ontario, Canada
More information about the R-packages