Dramatically new versions of the Hmisc and Design libraries
may be obtained from
http://hesweb1.med.virginia.edu/biostat/s/splus.html
Those desiring to install the new versions should completely
remove the old versions first.
These versions of the libraries are the versions that are being
shipped with S-Plus 6.0 for Windows. The web site above does not
have the binaries for S-Plus 6.0 for Windows but those will be
placed there soon.
The Design library uses a new and safer method for adding
model design attributes to fit objects. Now Design will not
interfere with other fitting functions that are not in the
Design library, and the design attributes are stored in a
different place in the fit object. But fit objects made
from previous versions will still be usable. To use glm
with Design you now have to use glmD.
Some of the biggest changes to Hmisc are:
- adapted sas.get for SAS version 8
- fixed bugs in xYplot and Dotplot and added color-filled bands
thanks to Madeline Bauer
- new function upData - a safe and consise way to change characteristics
of variables in data frames (labels, levels, dropping variables, etc.)
- improved cleanup.import
- new function matchCases for "closest on a covariate" case matching
- put latex-related functions in print.display library of Heiberger
and Harrell inside Hmisc
- added some basic functions for interactive drawing (clicking
data points to add to a graph, bezier curves, etc.)
I owe some of these changes and several bug corrections
to other users, who are acknowledged in the change history
log for Hmisc on the web site.
Biggest changes to Design are:
- not overriding model.frame.default to add design attributes
to fit object; fitting functions now must explicitly invoke
Design()
- added glmD
- fixed bug in calculation of omnibus goodness of fit statistic
for logistic regression models (thanks to Jorge Sirgo) in
the residuals.lrm function
- added "perim=function definition" argument to plot.Design so
that users can "white out" sections of plots (2-D and 3-D
image plots) where data are too sparse to trust the model;
the user can specify a general function used to suppress
plotting of sections of the fitted curves
For both libraries I made a multitude of changes to allow
them to work with Version 4 of the S language (for S-Plus
5.1 and 6.0 although only tested on 6.0). This was
incredibly painful because of non-downward-compatible
changes to the S language. Users will get warnings
when using the libraries on S-Plus 6 but these can
usually be ignored. The new language has as many
disadvantages (especially removal of support for multiple
inheritance) as it has advantages, in my humble opinion.
I was unable to use the new class mechanism to advantage
in any of my functions; they use oldClass'es exclusively.
[A major problem I had with new classes is their inability,
for example, to allow latex methods for different objects
to have different arguments.]
--
Frank E Harrell Jr Prof. of Biostatistics & Statistics
Div. of Biostatistics & Epidem. Dept. of Health Evaluation Sciences
U. Virginia School of Medicine http://hesweb1.med.virginia.edu/biostat
|