Package reorganized to use roxygen2, markdown, testthat, code coverage, continuous testing, and other features.
The degrees of freedom for lme models were changed (again). Previously, it would make this determination from the number of data points, fixed effect parameters, and random effect parameters. Now it follows what the doBy package does, using internal data from lme objects and takes the smallest possible degrees of freedom if variables involved in the the contrast have variable df. The previous method for counting df did not align with the results inside of the lee objects for simple cases.
Maintainer email was changed
A typo was fixed in the vignette
lme models, Thorn Thaler found a bug in some models where the calculation of model degrees of freedom could not be determined and proposed a fix.If the sandwich estimate failed, the traditional covariance estimator is used instead.
Moved to a real NEWS file format.
A bug was fixed (spotted by Thorn Thaler) where normal (not t) distributions were being used with lme and gls models to calculate confidence intervals.
rms instead of DesignThe package was updated to work with the rms package instead of the Design package.
A bug was fixed for generalized linear models with no degrees of freedom, such as log-linear models.
The vignette was slightly reformatted.
An error check was added for lme models that will fail if the covariance matrix of the variance-covariance coefficients has issues (since lme counts the number of random effects when calculating degrees of freedom).
With lme, this problem is signaled by lmeObject$apVar being a character string (eg. “Non-positive definite approximate variance-covariance”)
nlme so that lme4 methods do not interfere.