[B5JMS] statistics not Joethras' skill

[b5jms at mail.fsl.cs.sunysb.edu]

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
From: "crawf" <crawf at optonline.net>
Date: Mon, 28 Nov 2005 13:54:41 +0000 (UTC)
Lines: 67

As Joe told us long ago, mathematics not Joethras' skill...

As with many of you, I have been reading Volume 1 of the script book,
now that the hamster express has finally delivered it.  On page 50, in
the footnote discussing the origins of the term "mundane" (and yes, I
am aware of the irony of  what is to follow :-), Joe writes:

"Yes, a given individual may not know who Philip K. Dick was, may not
know offfhand the three rules of robotics, or how to perform a chi
square analysis of variance, but that shouldn't mitigate against that
person, who may know Shakespeare and be able to design a really nifty
art deco dining room."

Joe, we love you to bits, but a "chi square analysis of variance" is
akin to saying a critter is both a bird and a fish.  There may be
something slightly fishy about your fowl, or something vaguely foul
about your fish, but it can't be both by the rules of zoology.

If I claim that 20% of the people on my block don't like spoo, you
might take issue with how I tried to measure that or how accurate my
numbers are, but with such a small group of interest, I probably would
try to ask each of the neighbors about their tastes.  That would make
it a descriptive statistic (a simple summary number of everyone asked).

If I make a similar claim about the people in my *state*, well even I
don't have enough time on my hands to ring that many doorbells.  I'd
have to take some kind of sample, then extrapolate  out to the larger
group of interest.  In general terms, this is what you're doing when
you use inferential statistics.

Because of potential errors from sampling and other "noise" sources,
you are usually trying to demonstrate a high probability that your test
drug really *did* work better than the placebo, or that more people
like spoo toasted rather than fried.  Different kinds of  inferential
statistics make different assumptions about what kind of numbers you
have or what the underlying distributions are and so forth.

"Chi square" is a nonparametric test meant to assess observed versus
expected *frequencies* of something (e.g., do more men than women
attend sf cons?).  "Analysis of variance" (ANOVA) is a parametric test
designed to be used with numbers that are on a constant scale (e.g.,
the 10 gram difference between 200g and 210g is the same as the 10g
difference between 400g and 410g).  Parametric tests are more powerful,
but have more stringent assumptions than nonparametrics.

Yes, Virginia, there *are* nonparametric versions of  ANOVA
(Kruskal-Wallis for one-way analysis of ranks and Freidman for
two-way), but even these are not designed for the frequency data of the
chi square.

Thus ended the rant.

Joe was a psychology major in school.  Typically they are forced to
endure the torture of a required statistics course.  Most of them would
rather gnaw off both of their own arms than take this course.  Once
they no longer need to confront this demon they set about repressing as
much of the material as possible.  These two marginally related terms
obviously popped up in one of Joe's many nightmares, and...

That's my story, and I'm sticking to it. :-)

Larry Crawford
who now ranks third in the pack under the dog and the Significant
Overunit




=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
From: jmsatb5 at aol.com
Date: Mon, 28 Nov 2005 22:25:13 +0000 (UTC)
Lines: 5

Put your face in the book....

jms