[R] SAS or R software
Frank E Harrell Jr
f.harrell at vanderbilt.edu
Sun Nov 21 18:04:01 CET 2004
Michael Friendly wrote:
> I can't resist dipping my oar in here:
> For me, some significant advantages of SAS are
> - Ability to input data in almost *any* conceivable form using the
> combination of features available through
> input/infile statements, SAS informats and formats, data step
> programming, etc. Dataset manipulation
> (merge, join, stack, subset, summarize, etc.) also favors SAS in more
> complex cases, IMHO.
I respectfully disagree Michael, in cases where the file sizes are not
enormous. And in many cases repetitive SAS data manipulation can be
done much easier using vectors (e.g., vectors of recodes) and matrices
in R, as shown in the examples in Alzola & Harrell
We are working on a project in which a client sends us 50 SAS datasets.
Not only can we do all the complex data manipulations needed in R, but
we can treat the 50 datasets as a array (actually a list( )) to handle
repetitive operations over many datasets.
> - Output delivery system (ODS): *Every* piece of SAS output is an
> output object that can be captured as
> a dataset, rendered in RTF, LaTeX, HTML, PDF, etc. with a relatively
> simple mechanism (including graphs)
> ods pdf file='mystuff.pdf'';
> << any sas stuff>>
> ods pdf close;
> - If you think the output tables are ugly, it is not difficult to define
> a template for *any* output to display
> it the way you like.
I would like to see SAS ODS duplicate Table 10 in
> - ODS Graphics (new in SAS 9.1) will extend much of this so that
> statistical procedures will produce many
> graphics themselves with ease.
> One significant disadvantage for me is the difficulty of composing
> multipanel graphic displays
> (trellis graphics, linked micromaps, etc.) due to the lack of an
> overall, top-down graphics environment.
> As well, there are a variety of kinds of graphics I've found
> extraordinarily frustrating to try to do
> in SAS because of lack of coherence or generality in the output
> available from procedures --- an
> example would be effect displays, such as implemented in R in the
> effects package.
> I can't agree, however, with Frank Harrell that SAS produces 'the worst
> graphics in the statistical software world.'
> One can get ugly graphs in R almost as easily in SAS just by accepting
> the 80-20 rule: You can typically get 80% of
> what you want with 20% of the effort. To get what you really want takes
> the remaining 80% of effort.
> On the other hand, the active hard work of many R developers has led to
> R graphics for which the *default*
> results for many graphs avoid many of the egregious design errors
> introduced in SAS in the days of pen-plotters
> (+ signs for points, cross-hatching for area fill).
Frank E Harrell Jr Professor and Chair School of Medicine
Department of Biostatistics Vanderbilt University
More information about the R-help