As promised,
I have collated, in a fashion, those replies from the SAS versus R? question. Please forgive the haste, and therefore the rather rough collation, nonetheless still interesting I think.
Best regards and thanks once again to all those who replied.
Mike Griffiths
Collated answers
John Sorkin [log in to unmask]
Data manipulation is generally easier in SAS than R.
R syntax is more elegant that SAS syntax.
R is open source
SAS is proprietary
SAS handles large data sets somewhat better than R.
SAS runs somewhat faster than R for many jobs
R output is more concise that is SAS output.
R is free, SAS is not.
R is gaining popularity with academics, SAS has a strong foothold in
industry and academia.
Both are very powerful tools.
Neither R nor SAS are worth anything if the person doing the analyses
does not know what he, or she is doing.
BXC (Bendix Carstensen)" [log in to unmask]
SAS advantages:
- Human readable code for data handling
- Handles very large datasets with few problems
- Good tabulation facilities (proc tabulate)
SAS disadvantages
- Non-standard procedure takes forever to program
- Standard graphs are crap
- Custom graphs takes forever to program
- Costs a fortune
R advantages
- works a simple calculator
- Standard and non-standard analyses simple to implement.
- Versatile graphics
- New thing gets implemented fast
- Help list gets you answers quickly (< 1 hr, normally)
- Free
R disadvantages
- Code is difficult to read, particularly if data-manipulation
is involved
- Very large datasets can give problems
So if you run a bank or a telephone company you will probably want
SAS, otherwise you may get away with R.
Matz David [log in to unmask]
This is somewhat tricky question to answer in the absence of a specific
problem & context. A wide range of tasks could be done using either
software, dependent on the expertise of the user, but they are different
things.
R is a programming language - see http://www.r-project.org/
SAS is a statistical application package
http://www.statsadvice.bham.ac.uk/sas.htm
As such SAS [and similarly SPSS] have front end GUI interfaces as
alternatives to the method of programming code. Both SAS and SPSS have a
wide range of standard procedures either in the 'base' system or though add
on modules. They are both widely used in the social sciences and in
government, and for processing large volumes of data [SAS was originally
developed on a mainframe], and have copious manuals and training materials
available using datasets in a wide range of disciplines.
By contrast R is a language, so probably more flexible and suited to
developing new analytical methods. I've used the commercial version S-plus,
which is easy to use and has facilities such as defining matrices A and B,
and then readily understanding that AB means you want to do matrix
multiplication. The flexibility and ease of use is great, but comes at a
cost of speed that may be important in processing large volumes of data
[though C code can be linked to/from S-plus to overcome that] There are
likely to be more flexible individual functions in R for specific
statistical tasks [eg resampling], but the user may often need greater
expertise of statistical theory and programming ability to put them together
in a program, compared with using SAS or SPSS and the menu interface with
standard procedures, which suffice for very many purposes.
The best analogue I can think of is Visual Basic and Excel - a lot of what
Excel does can be done by programming in Visual Basic, and Visual Basic is
more flexible than Excel because VB is a 'language' not a 'package'. Which
is needed/chosen depends on the task in hand.
Darryl Bertolucci [log in to unmask]
I was going to response and say things like: this is a rather large
question, you'll probably recieve a lot of responses and would you please
summarize to the list, then add a few opinions and yadayaya (that's a very
technical term fournd only in conclusions of articles of same pretty sleezy
(uh, highly respectable) journals.
The are only a couple of things I might add but Dr. Sorkin from just up
I-95 (DC/Baltimore area roadway) has done a most excellent "nutshell"
presenation.
The last point is a very valuable thing, IMVHO, for all of his to keep in
mind.
Carl [log in to unmask]
I concur with John Sorkin on most points. I will emphasise though that
SAS can be quite a great deal faster than R for some things, and handles
large data much
better. Also,
- SAS is the gold standard for some clinical trials analysis i.e. the
results are not believed if not from SAS (patent nonsense, but a fact
nonetheless).
- SAS invariably gives you an answer - other packages may fail to
converge or return nonsense (but for good reasons), the SAS blackbox
however gives you numbers... note I consider this a fault rather than a
feature! This relates strongly to John Sorkins last point.
- SAS graphics are often hideous
- SAS help files are largely awful
- Contrary to John I personally think the data manipulation is much
easier in R than SAS.
- R contains more cutting edge tools as it is commonly used in academic
Stats methods research (however the volunteer nature of the developers
means intermittent support and somewhat "beta" type releases)
Peter Flom [log in to unmask]
This is a question of general interest really, I hear an awful lot about the SAS package. Indeed, when job adverts appear connected with clinical data analysis (CRO) this package is nearly always cited as a definite prerequisite. My question is this, If one were to compare R with SAS what would be the strengths and weaknesses of R compared to SAS, or the other way round if easier.
[log in to unmask]
Another point about SAS that should be mentioned is that the procedures
are largely self-contained and do not communicate very easily with each
other. The net result is that to do some programming you will probably
have to learn SAS macro, which is just like learning another language and
in fact you probably have to use proc IML as well whcih is yet another
language. Thus you have to learn 3 languages to use one. R is much more
integrated in this respect.
In fact I think that Peter's comment
'SAS is more consistent, procedure to procedure, because it's written by
one set of people in a central location.'
is very misleading. In fact you can find baffling differences from
procedure to procedure although many of these are being eliminated. For
example, at one time you could define factors withing pro glm but not
within many other procedures. And for example, proc nlmixed (although very
powerful) is very different from many other stats procedures.
However, if you want something that will stand up to very big data sets
then SAS is a good tool.
Personally I use GenStat.
"Stas Kolenikov" [log in to unmask]
You might want to look at a bunch of software reviews -- The American
Staitistician has a regular column, I am pretty sure you can find it
somewhere on the Islands, too. I don't know about JRSS, it might have
something, as well.
"Macgregor, Lachlan" [log in to unmask]
I suggest you also consider Stata. It is flexible, has excellent graphics capability, and seems more user-friendly (in my view) than SAS. Specifically (i) the syntax is more natural (ii) the `help' facility is easier to use (iii) very wide range of statistical applications, enhanced by the hundreds of user-written add-ons you can download from the Stata website (for free). There is also a version of Stata for `large' datasets.
Clearly, SAS is still used by major pharmaceutical companies. I don't know the reason for this, but curious to find out.
I've never used R.
Brian G Miller [log in to unmask]
Lachlan McGregor said
Clearly, SAS is still used by major pharmaceutical companies. I don't
know the reason for this, but curious to find out.
As I understand, presumably in a bid to ensure quality of statistical
analysis in drug trials, the US FDA specified that all statistical
analyses of data from trials were to be done in SAS. I don't know if
that still applies, but it set an industry standard. I've been told
that, if you did analyse data using another package, you had to redo it
in SAS to satisfy the regulators.
"Maynard, Trevor" [log in to unmask]
I did a survey a while back and R came out as a clear winner.
Generally people felt that R had much better graphics.
R suffers when you want to use very large files (I understand)
Michael Griffiths, Ph.D.
Chemometrician
Training, Quality and Statistics Group
LGC Limited
Queens Road
Teddington
Middlesex, TW11 0LY, UK
Tel: +44 (0)20 8943 7352
Fax: +44 (0)20 8943 2767
e-mail: [log in to unmask]
>>> "Macgregor, Lachlan" <[log in to unmask]> 04/20/06 11:05 pm >>>
Dear Mike
I suggest you also consider Stata. It is flexible, has excellent graphics capability, and seems more user-friendly (in my view) than SAS. Specifically (i) the syntax is more natural (ii) the `help' facility is easier to use (iii) very wide range of statistical applications, enhanced by the hundreds of user-written add-ons you can download from the Stata website (for free). There is also a version of Stata for `large' datasets.
Clearly, SAS is still used by major pharmaceutical companies. I don't know the reason for this, but curious to find out.
I've never used R.
I hope this helps you on your quest.
Dr Lachlan MacGregor, Biostatistics Fellow
Clinical Epidemiology & Health Service Evaluation Unit
Royal Melbourne Hospital
Parkville, Victoria 3050
Australia
*******************************************************************
This email and any attachments are confidential. Any use, copying or
disclosure other than by the intended recipient is unauthorised. If
you have received this message in error, please notify the sender
immediately via +44(0)20 8943 7000 or notify [log in to unmask]
and delete this message and any copies from your computer and network.
LGC Limited. Registered in England 2991879.
Registered office: Queens Road, Teddington, Middlesex, TW11 0LY, UK
|