Re: Interpreting the coefficient in a GLS binomial model

On Sun, 25 Feb 2007 16:55:55 -0600, David Winsemius
<doe_snot@xxxxxxxxxxx> wrote:

Serguei Kaniovski <kaniovsk@xxxxxxxxxx> wrote in
news:14612178.1172425327931.JavaMail.jakarta@xxxxxxxxxxxxxxxxxxxxxx:

The data is as follows. A jury of nine jurors vote together on N cases
(N very large, say 1000). I count the number of concurring votes "C"
(yes,yes, or no,no) between any two voters, say Mike and Jane, and the
number of opposite votes "O" (yes,no;no,yes) for them. I transfer the
counts into proportions. There are 36 pairs of jurors in total, and
thus so many observations in the proportion data.

I show by other means that the votes on a case are highly positively
correlated because the jurors talk to each other and presumably also
because they share similar ideologies, hence the index. Granted, the
index cannot be the only explanatory variable, but it is only one I
have.

I am a bit unclear on how the numbers work out in the design, but it is
very clear that the statistical model needs to represent the dependency
much better.

More questions/puzzles:
How do 9 jurors end up with 36 pairs? 8*7*(...)*1 is more than 36. Some
sort of Latin Square being used? How did that get constructed?

It looks to me like 9 Supreme Court justices as "jurors".
At least, that can serve as a model.

There are 9*8/2 or 36 pairs for any vote.

The overall correlation is positive because a number of
the decisions are heavily Yes or No.

- I would initially score up principal components and look for
differences on those. After that?
I don't know what the OP is looking for, or what his model
will give him.



Did the pairings get determined by your ideologic distance measure?
Only nine distinct jurors?
Just one jury panel of 9 jurors?
All test_cases voted on by each juror?

You are modeling the ratio of concordant to discordant votes on a paired
level? Why?

A yes:yes event is the same as a no:no event in the modeling? Why? Are
you really only interested in measure of agreement, and not interested in
predicting the vote by an individual?

When you say that the jurors talking together and having similar
ideologies, I had additional trouble figuring out how the "nine jurors"
gets represented and why you are pairing them.

--------------------

Here is how I am imagining the 9000 or so lines of data might be
arranged:

juror I-score juror_pair test_case vote_y-n

But then you seem to be expanding the data by creating more pairs than
individuals.

--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.

Relevant Pages

  • Re: Interpreting the coefficient in a GLS binomial model
    ... number of opposite votes "O" for them. ... correlated because the jurors talk to each other and presumably also ... because they share similar ideologies, ... You are modeling the ratio of concordant to discordant votes on a paired ...
    (sci.stat.math)
  • Re: Rueful Jackson Jurors
    ... > Juries often are not unanimous from the first. ... > surprised if all the jurors who change their votes are entirely ... Maybe the "rumors as fact" have caused them to second guess ...
    (alt.true-crime)
  • Re: Survivor 15 cast after 10th episode
    ... but the jury can see the votes so they ... The jurors can see the remaining players as the watch the bootee's torch ... smug grin on Todd's face and that fake look of 'shock' on Amanda's face. ... The only question is how the jury will interpret those looks. ...
    (alt.tv.survivor)
Quantcast