In article
<a8dedb7b-1f42-4817-8fc0-aec4b22bd403@[EMAIL PROTECTED]
>,
Francogrex <franco@[EMAIL PROTECTED]
> wrote:
>While reading a review, I saw this interesting paragraph:
>"However, the accessibility of statistical computer programs may
>provide an excuse
>for not consulting a statistician, to the subsequent detriment of the
>scientific quality of
>the research. A reliance on statistical software without enough
>statistical knowledge
>could result in incorrect statistical treatment of data (****mada,
>2001; Jolliffe, 2001)"
>True, but more generally, do you think people (those with stat/math
>degrees and those who want to dabble here and there) are relying too
>much on computers, softwares, packages and programs, instead of really
>being experts in pure mathematical statistics? I think that
>anecdotally the evolution might go something like this:
>Novice: use Excel
>Advanced: use SAS/SPLUS...
>Expert: use Mathematica, Maple
>Professional: use a pen and a paper!
>Though the above is just a caricatural illustration but after all,
>those great mathematicians who set the foundations (Gauss, Pascal etc)
>had only a pen and a paper (and a brain).
>Your thoughts?
Gauss was a rapid calculator, and avoided using logarithmic
tables by mental arithmetic which few can match. Pencil and
paper have their limitations; a good computer operator using
a good electromechanical calculator could invert a symmetric
10x10 matrix in one day if no mistakes were made. Many refined
computations were not made; from the 17th to early and mid 20th
century, we used massive books of tables of logarithms and
other functions, and interpolated.
The problem with someone who does not really understand
statistical THEORY is not that poor computational
procedures will be followed, but WRONG procedures, and even
totally wrong. The applier of statistics needs to know how
to formulate a probability model, and what assumptions are
needed. We have all the ignoramuses who convert their data
to normal, destroying the relation****ps which are there
initially. It would be well if in our beginning courses we
taught probability modeling and carefully eliminated all
statistical methodology.
Even with this, one has to be skeptical of computer output.
A computer is a superfast subimbecile, and the output is
not likely to be better than the programming. As for
Mathematica and Maple, see the corresponding newsgroups
and sci.math and sci.math.symbolic for their failures.
Computers can give answers if properly used. I probably
do not use them often enough, given my calculating speed
and that I go back to BC (before computers). Gauss had
to use Bessel function expansions and tables of Bessel
functions, because he did not have computers which could
do the job.
--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
hrubin@[EMAIL PROTECTED]
Phone: (765)494-6054 FAX: (765)494-0558
|
