Let's not get too personal here, OK?  I mean, shouldn't we all genuflect toward
Redmond each morning in appreciation for Excel and Windows?  We send our money
($, C$, pounds, marks, yen, etc.) already.

:)   :)    :)

I agree, the default settings for Excel charts are rarely suitable for
presentations.  A beginning student who generates a lot of such charts, hoping
that the instructor (or supervisor) will select out the useful one, should be
strongly encouraged to think a minute and communicate more clearly.  Very
strongly.

but what was Excel designed for?  There are many different 'optimum' displays,
and in any case, the analyst must have some idea what they are looking for, in
order to find or make it.

the help window problem is semi-serious in my view.  It provides an opportunity
for the instructor to show that the emperor has few clothes.  Most 'help'
sections don't, anyway.  Unless you think exactly like the author did, which I
rarely do.

If the erroneous words in a help section come from or lead to erroneous
equations, then there is a serious issue.  I understand that the Excel code
writers have not always listened to those pointing out errors in the past, but
would-be users should be alerted in any case.

But when we get to negative sums of squares, I think we have a very serious issue
on our hands.  Mathematically negative sums of squares are impossible, so we have
an issue of digital execution to deal with.  This is what Excel is _supposed_ to
do well.

Would someone post a sample dataset that produces this error?  Can we have a
discussion of what properties of this dataset lead to the errors?  What should a
user, forced or allowed to live with Excel, look out for to avoid falling into
the same trap?

NIST posts some datasets used for numerical validation of computer algorithms.  I
downloaded one that showed me that Excel's precision in a one way AoV _can_ go
only to 2 significant figures.  I mostly understand why it did that, and what I
need to look for to assure that beginning students and others are alerted to the
potential for this error.

Show me some data that gets to negative sums of squares.

And BTW, if the 'computational' form of the variance is so poor, what
alternatives are available?  The definition form
is subject to truncation error.  Many of my datasets exhibit more truncation
error than anything else, even with my Excel's (apparent) 14 digit precision.  If
we all went to 64 bit chips & 64 bit code, would this issue 'go away'?

Cheers,
Jay

"Robert J. MacG. Dawson" wrote:

> David Heiser wrote:
>
> > Much of Cryer's complaints, just don't hold up after deeply looking into
> > them. I havre spent a lot of time on the issue of Excel's statistical
> > capabilities, and am not able to support many of Cryer's complaints.
>
>         Please explain...  it looked to me as if he provided adequate evidence
> for many of his complaints.  Which of them can you not support?
>
>         I don't consider the examples of graphics that he gives to be
> acceptable - do you? No doubt one can produce better graphics by
> changing the settings - but why should one have to?
>
>         Cryer gives several examples of help windows that I would say betray a
> deep ignorance of statistics on the part of the writer. Do you disagree?
> Or do you think that users have no right to expect better?
>
>         Most authorities do not consider the so-called "computational formula"
> to be numerically well-conditioned, and hold that in extreme cases it
> can even lead to negative sums of squares (often followed shortly by a
> program crash as the computer attempts to take the square root of a
> negative number.) Are they wrong?
>
>         [The title "computational formula" dates back to a time in which the
> use of very little memory was a virtue outweighing accuracy, either on
> mechanical calculators in which "memory" was a pencil and scratchpad or
> in primitive scientific calculators with only a few storage registers.
> Sadly, this has misled many into thinking that it is a good formula
> computationally in any other regard. It isn't.]
>
>           As you pointed out, "the IT revolution is here."  This does NOT mean
> that everybody will soon have to code their own kludge in BASIC (Visual
> or otherwise) - that was the 1960's.  It means that reliable, mature
> statistical software is now available.
>         If you want to use an outdated formula "out on the surface" to see how
> it works, most stats packages will permit you to do this. But the
> criteria for choosing algorithms for serious work have moved beyond how
> easy they are for amateurs (or specialists in other fields) to code
> themselves.
>
>         -Robert Dawson
> .
> .
> =================================================================
> Instructions for joining and leaving this list, remarks about the
> problem of INAPPROPRIATE MESSAGES, and archives are available at:
> .                  http://jse.stat.ncsu.edu/                    .
> =================================================================

--
Jay Warner
Principal Scientist
Warner Consulting, Inc.
4444 North Green Bay Road
Racine, WI 53404-1216
USA

Ph: (262) 634-9100
FAX: (262) 681-1133
email: [EMAIL PROTECTED]
web: http://www.a2q.com

The A2Q Method (tm) -- What do you want to improve today?




.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to