Reply #69: I have the spookiest sense

[Febble]

that somewhere in meta-cyber-space, I just trod on somebody's
toe.

If I misunderstood a meaning somewhere, I apologise.

But let me clarify a few principles regarding my own approach
to data analysis:

1. I am more interested in hypothesis testing than data
mining.  If I have a theory about what caused something else,
I devise a hypthesis that can be tested statistically. 
Designing appropriate hypotheses can be difficult, and
sometimes requires imagination.  Often they come to me in the
night.  But the important thing is that, if possible, the
hypothesis should precede the analysis, i.e. be "a
priori".  Otherwise we need to use more stringent
criteria ("post hoc" tests for
"significance" and thus risk losing statistical
power (or being misleading).

2. While I think it is impossible to be completely objective
in data analysis, two guards against subjectivity are the a
priori hypothesis and the two-tailed test. Much as I deplore
the fact that Bush is your president, and dearly as I yearned
for President Kerry, in all the analyses I have done, whether
on my own, in collaboration with other people involved in
investigating the anomalies of the 2004 presidential election,
or for Warren Mitofsky, I have applied these principles as
stringently as it has been within my power to do.  Moreover,
any finding I have made I have subjected, as I am trained to
do as a scientist, to as much probing for alternative
interpretations as I have in my power. And I remain, as any
good data analyst must, always aware that findings are always
provisional - that new findings appear that cast my current
findings into a different light, or that some aspect of the
problem that you have ignored may turn out to be crucial. 
This has happened more than once during my analyses of the
election data, and is why I like working on blogs.  

3. All statistical inferences are made with a margin of error.
 This is as true of rules like the "incumbent approval
rating" rule as it is to survey findings.  We need to be
as rigorous about our inferences in cases where they suit our
a priori hypothesis as when they do not.

3. Human beings behave in complex ways.  This mean that it
means that obtaining a random sample is difficult; it means
that generalizing from a sample to a population is difficult,
as you cannot be sure you have sampled from the population you
meant to study;  it means that the data you collect from human
beings are subject to idiosyncrasies; and it means you can get
an apparently "significant" result from one study
and fail to replicate it in the next.  There are always more
factors affecting variance in your data than you will ever
know about.

4. If you get a finding that doesn't suit your hypothesis, you
look it squarely in the face.

Well, it is my provisional view, based on serious
consideration of a fair bit of evidence, and on serious
consideration of many of TIA's analyses (I cannot promise to
have considered them all) that his contention that the exit
polls demonstrate massive fraud is probably wrong.  I don't
doubt the accuracy of the calculations, which I am sure are
less error prone than mine, but I do question the validity of
the assumptions that underlie his inferences. And I also
question, at times, his methodology. I don't think that doing
either of these things is undermining American democracy. 

In fact, if I, and others who share my conclusions, are right,
and TIA, and those who share his, are wrong - in other words
if it is the case that while the election was corrupt, massive
vote-switching did not occur, then to have people seriously
believing that the Democrats are doomed and they might as well
not vote, or that there is no point analysing Kerry's campaign
because it was perfectly OK, then I think that is a problem.

Yes, I believe that American democracy is broken and needs
fixing.  But I also believe that Kerry lost, and that needs
fixing too.

I hope that has cleared the air.  It doesn't seem to have done
anything for my html which remains stubbornly in courier. 
Weird to be posting in TIA's typescript.  That guy is
definitely spooky.