R help - Jan 2009 - Ashlee Vance's article on R in the New York Times

Ashlee Vance's article on R in the New York Times.

This is typical of the New York Times. Because they get to coast on the
prestige and reputation of their brand , they have a history of just this
sort of journalistic sloppiness. Whether it's the author or the editor at
fault doesn't really matter, they do this screw-up all the time.

Look, if you write an article on the first page of the business section,
you're not just presenting yourself as a writer or entertainer, you're
presenting yourself as a journalist, and that implies two commitments:

1: I believe that my writing is true , and as fair and balanced as
appropriate in the context.
2: I've invested the time in research and fact-checking so that point #1
actually has credibility.

Vance clearly fails on point #2. He just didn't do his homework. And as
I've
seen over the years, this is typical for NYT contributors. That's
complacency. A bit like SAS Institute - NYT is overly reliant on it's brand
name.

First of all, the third paragraph is a falsehood. I'm not saying Vance is
lying. I'm saying he's lazy. A couple of hours of research, and he could
have corrected that.

If you find computer programming to be tedious, unpleasant, or quite
difficult, then R is the wrong software for you. R has a reputation for
having a tougher learning curve than the SAS programming language. Even if
you disagree, neither is appropriate for people who don't have the time and
patience to study programming languages.

Vance's article is also deeply misleading , he gives the wrong impression of
where R actually came from, and who deserves credit for what. It's
especially glaring given that he does briefly mention R's precursor, S. Yet,
funny that, he neglects to mention that S and R basically use the same user
interface ( the same programming language ). Hey Vance, um, that's a big
oversight.

R is a quality software package, with years of development and debugging,
and substantial documentation, and diverse and reliable statistical function
libraries. The R project team deserves a great deal of credit for this. But
they don't deserve all of the credit. A great deal of the R software product
was already achieved before the R team ever came along. There is a tendency
to poo-poo the blood and sweat that go into the design of the user
interface.

The choices made when designing the user interface of any data analysis tool
are critical, whether GUI or language. Assuming the CPU is not overloaded,
which is often the case, it is the user interface that makes the difference
between a piece of cake , and hours lost coding what should have been a
routine task.

Well, Gentleman and Ihaka did not design the user interface for R. AT&T
researchers did, during the cold war. It's possible that a few employees at
proprietary software companies also contributed. It might have been largely
financed by American taxpayers, because there were a lot of backroom deals
during the cold war, and AT&T was typically in the thick of it. The user
interface for R, otherwise known as the S programming language has the same
origins as C and Unix.

Some R promoters point out that R has lexical scope and lots of Scheme
goodness. ( and what widespread programming language today does not have
lexical scope? ). But other R promoters point out that programs in S-Plus
usually work in R, and vice-versa. Well, in that case, then it's the same
damn programming language!

Quite likely, the R founders were careful to point this out in their
interviews with Vance. Even if they forgot, minutes of research on Vance's
part would have told him that. The New York Times - sloppy as usual. More
like an advertisement than a bona fide article.

And the upshot of this , in the outlook for statistical software, is that
regarding the strengths and (considerable) limitations of the three
classical statistical programming languages ( S, SAS, SPSS) , R really
doesn't change anything at all. I definitely like the pricetag though. And
that does not mean that R cannot achieve a quality and reliability
comparable to S-Plus and SAS, not withstanding Milley's snide comment. But
if you want to attack the chronic and painful productivity problems with
data preparation and statistical table production, you need to go beyond R
and SAS. You have to develop new user interfaces, and that is very risky,
and takes years of technical work and marketing.

And, to be honest, that is not what open source developers are willing to
do. In the majority of software categories, including specialized languages(
such as statistical), open source developers are not motivated to develop
user interfaces that make a ground-breaking difference in the user's
productivity level. One big, and crucial exception is the category of
all-purpose programming languages. Thousands of open source developers go to
bed dreaming of being the next Larry Wall. Thankfully, we have Ruby and
Python as a result.

	[[alternative HTML version deleted]]

On Thu, 08-Jan-2009 at 01:26AM -0500, Robert Wilkins wrote:

[....]

|> Some R promoters point out that R has lexical scope and lots of
|> Scheme goodness. ( and what widespread programming language today
|> does not have lexical scope? ). But other R promoters point out
|> that programs in S-Plus usually work in R, and vice-versa. Well, in
|> that case, then it's the same damn programming language!

Same language, but a different 'interface'.

The effort to get them to work in the other interface relates mostly
to dealing with the the presence or absence of lexical scoping.  But
even if Splus did have lexical scoping, I'd probably have changed to R
anyway.

The main reason why I changed was because it was such a hassle getting
anything new added.  Bean-counters had to be satisfied that the admin
work was value for money before it was permissable to have an IT guy
spend time setting up a different function library.  It could take
months getting that through.

By contrast, because of no bean-counter issues, with R, I could
install everying myself and simply add packages that looked
interesting.  We're comparing seconds with months.  The crucial
difference in the interfaces is the difference between proprietary and
open.

[....]


|> One big, and crucial exception is the category of all-purpose
|> programming languages. Thousands of open source developers go to
|> bed dreaming of being the next Larry Wall. Thankfully, we have Ruby
|> and Python as a result.

So your point is that the problem with data organisation is that it is
not all-purpose enough, so we have to wait until somebody produces a
proprietary product?

Your list of Ruby and Python reminds me of the scene in 'The Life of
Brian' where the John Cleese character asks the rhetorical question:
"What have the Romans ever done for us?"  

http://www.youtube.com/watch?v=ExWfh6sGyso


-- 
~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.   
   ___    Patrick Connolly   
 {~._.~}                   Great minds discuss ideas    
 _( Y )_  	         Average minds discuss events 
(:_~*~_:)                  Small minds discuss people  
 (_)-(_)  	                      ..... Eleanor Roosevelt
	  
~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.

> comparable to S-Plus and SAS, not withstanding Milley's snide comment.
But
> if you want to attack the chronic and painful productivity problems with
> data preparation and statistical table production, you need to go beyond R
> and SAS.
What are these problems?

Hadley

-- 
http://had.co.nz/

On Thu, Jan 8, 2009 at 1:26 AM, Robert Wilkins
<irishhacker@gmail.com>wrote:
> ...The user interface for R, otherwise known as the S programming language
> has the same origins as C and Unix....

We could take this one step further, and note that C's design (its
"user
interface"?) was based on BCPL, which was developed at Cambridge University
and MIT (which was in turn loosely based on CPL).  But BCPL declined into
obscurity, while we're still stuck witxxx benefitting from C.

As for Unix, most of its ideas came from Multics (developed mostly at MIT,
but with Bell Labs and GE). And some of the core ideas and utilities of
Multics came from CTSS (also MIT) -- Unix roff/nroff/troff was a knock-off
of CTSS's runoff, for example.  But Multics failed and Unix succeeded.

Fast forward a few years, and Unix itself was (alas) fading into marginality
until GNU/Linux* came around. (I had a front-row seat at this decline as a
staffer at OSF.) As with R/S, the specification was of course important, but
the dynamics were completely changed by the development of a free and open
version.  The NYT reporter correctly focussed on the success of R.

I do agree that the core of R could use renewal and rethinking, and that
many free/open projects have been reimplementations of existing designs. But
there is also innovation within the framework of R, such as Hadley Wickham's
ggplot2.

Is it enough?  It never is....

            -s

* Speaking of giving credit where credit is due, Stallman is absolutely
right when he insists on recognition for the huge contribution of the GNU
project to GNU/Linux, though it's no doubt to late to insist on the full
name....

	[[alternative HTML version deleted]]