[R] The hidden costs of GPL software?

[John Fox]

Dear list members,

This has been a stimulating discussion, now spread over three lists.
Although I'd like to address issues that have been raised on all three
lists, I expect that more or less everyone reads r-help, so I'm just posting
these comments there.

(1) As everyone else, I've had experience with a number of other statistical
packages and programming environments in addition to R (including, more
years ago than I care to say, the mainframe predecessor of the MicrOsiris
package mentioned positively in the SCW article cited by Philippe in his
original message). I don't believe that extensive point-and-click GUIs for
broad statistical packages/programming environments such as Stata, R,
S-PLUS, or SAS are very helpful: They tend to be labyrinths that are
difficult to navigate. Some of the suggestions for other kinds of GUIs
(e.g., aids to command specification) seem to me more promising. Moreover, I
don't think that one should expect to learn an extensive system such as R or
SAS without doing some reading. My own experience is that S (i.e.,
encompassing R and S-PLUS) is easier, not harder, to learn than its true
competitors. 

(2) On the other hand, one can build quite nice graphical interfaces to more
limited packages. A couple of examples that I particularly like are SAS JMP
and Cook's and Weisberg's Arc (built on Lisp-Stat).

(3) Similarly, my Rcmdr package was meant to be a limited-purpose GUI,
useful for basic-statistics classes. Its range has grown somewhat to cover
linear and generalized-linear models, and I plan a few more modest
extensions (including the ability to incorporate other classes of
statistical models more easily). As a technical matter, I don't think that
it would be hard (although it would be time-consuming) to produce a much
broader extension, but the result (in my opinion) would be as dubiously
useful as the GUIs for SAS or S-PLUS. By the way, if there were something I
could wish for here it would be a slightly broader set of Tk widgets to be
included with the Tcl/Tk that installs with R for Windows, since using
widgets outside of this set creates installation obstacles for lower-level
users.

(4) Several people have pointed once more to the difficulty that novice
users experience in locating functions to perform particular tasks or in
figuring out how to use them once found. I suspect that even people who have
been using R for a while occasionally have a brain-cramp that leads to a
search through documentation. I know that I do. In my experience, the
various facilities for searching documentation in R work pretty well. 

(5) I think that examples in help files and vignettes can be useful, but are
not substitutes for text-books, manuals, and journal articles. It certainly
should not be the job of statistical software to teach the statistics,
although of course it can be used to help do that. I doubt that many list
members would look favourably on the statistical-methods "decision tree" in
MicrOsiris, for example. One solution is to include PDF "manuals" with
packages. I've done this, for example, with my effects and Rcmdr packages.
The introductory manual supplied with Thomas Lumley's survey package is
another, similar example. Maybe there's a better way of integrating such
non-vignette manuals with the help system -- something like
help(manual=package).

(6) As has been pointed out, e.g., by Duncan Murdoch, solving the
function-locating problem is best done by a method or methods that
automatically accommodate the growing and changing set of contributed
packages on CRAN.  Why not, as previously has been proposed, replace the
current static (and, in my view, not very useful) set of keywords in R
documentation with the requirement that package authors supply their own
keywords for each documented object? I believe that this is the intent of
the concept entries in Rd files, but their use certainly is not required or
even actively encouraged. (They're just mentioned in passing in the Writing
R Extensions manual.)

--------------------------------
John Fox
Department of Sociology
McMaster University
Hamilton, Ontario
Canada L8S 4M4
905-525-9140x23604
http://socserv.mcmaster.ca/jfox