R help - Mar 2007 - Fwd: Re: [friday topic]: what exactly is statistical computing

This is a very fascinating discussion topic.  I find I run into
some fundamental differences in interpretation of the phrase "statistical
computing".  I think of it as writing programs or functions, such as R or
packages
in R, and of understanding the numerical analysis behind these functions.

I exclude USING computer programs, such as R, for data analysis from my
definition of statistical computing.  I see that as doing statistics.
I have had students, some sent by other faculty members, in my class
on statistical computing thinking they were going to learn how to do statistical
analysis using the computer.  There was a clash of expectations between what
they thought they were taking and what I had in the syllabus.

Rich

---- Original message ----
>Date: Fri, 2 Mar 2007 13:25:59 -0600
>From: "Lucke, Joseph F" <Joseph.F.Lucke at uth.tmc.edu>  
>Subject: Re: [R] [friday topic]: what exactly is statistical computing  
>To: "Roger D. Peng" <rdpeng at gmail.com>, "Wensui
Liu" <liuwensui at gmail.com>
>Cc: r-help at stat.math.ethz.ch, "Bos, Roger" <roger.bos at
us.rothschild.com>
>
>Statistical computing perhaps is not so much a single topic as a family
>of related topics (a la Wittgenstein) that share a lot in common but
>perhaps very little is common to all. For example,
>1. Statistical computing in contrast to statistical theory.
>6. Statistical computing as a supplement to statistical theory.
>2. Statistical computing as gaining insight through data visualization
>3. Statistical computing for asymptotic analyses
>4. Statistical computing for approximations to unknown distributions
>5. Statistical computing for teaching demonstrations
>7. Statistical computing for assessing behavior of statistics.
>
>and so on.
>
>Joe

On 02-Mar-07 Richard M. Heiberger wrote:
> This is a very fascinating discussion topic.  I find I run into
> some fundamental differences in interpretation of the phrase
> "statistical computing".  I think of it as writing programs or
> functions, such as R or packages in R, and of understanding the
> numerical analysis behind these functions.
> 
> I exclude USING computer programs, such as R, for data analysis
> from my definition of statistical computing.  I see that as doing
> statistics. I have had students, some sent by other faculty members,
> in my class on statistical computing thinking they were going to
> learn how to do statistical analysis using the computer.  There
> was a clash of expectations between what they thought they were
> taking and what I had in the syllabus.
It is indeed a fascinating topic, and I agree with the implications
of Richard's views above.

Though computing machinery (from Brunsvigas with manual crank-handles
upwards) has been used for doing the computations of statistics
since the year dot, statistical computing (in my view of it)
did not begin to develop until much later.

I think the first developments which could be recognised as
"statistical cmputing" (as opposed to using computers to do
statistics) were the pioneering GENSTAT and GLIM (1973-4,
though developed over some years previously). Possibly
what characterised them for this was the fact that their
programming language was recognisably statistical in flavour,
and the commands triggered computational procedures in which
statistical algorithms were implemented.

Then, as the science of computer programming developed, and
became more generalised, with "structures", "methods" and
all the rest, so these concepts were implemented for statistics.
The result of a computation was a data-structure, which could
be recognised by any method that was capable of dealing with

It's perhaps hard to say when S was actually born: perhaps
passage to the outside world began with its port to UNIX
in 1979, though it was conceived around 1975. But it must
be acknowledged that in its adaptation of advanced (for the
time) programming methods to statistics was a breakthrough
in statistical computing.

A quite early simple instance of this kind of programming
was SPIDA (Statistical Program for Interactive Data Analysis)
which was developed prior to 1988 -- since Dan Lunn & Don McNeil
issued a SPIDA User's Manual in 1988 (and perhaps grew out of
NcNeil's approaches described in his 1977 book "Interactive
Data Analysis: A Practical Primer"), followed by the book

  Computer-Interactive Data Aanalysis
  A.D. Lunn and D.R. McNeil (Wiley 1991)
  (with a couple of 5.25" DOS floppies with the SPIDA software)

which I remember using with pleasure!

So I see this confluence of the evolution of computational
concepts and techniques through the 1970's and 80's, with
the development of statistical modelling techniques and
their implementation in software, as the core of "statistical
computing".

Best wishes to all,
Ted.

--------------------------------------------------------------------
E-Mail: (Ted Harding) <ted.harding at nessie.mcc.ac.uk>
Fax-to-email: +44 (0)870 094 0861
Date: 02-Mar-07                                       Time: 22:53:33
------------------------------ XFMail ------------------------------