Internetworkers: http://www.ibiblio.org/internetworkers/

[Ron Thigpen <rthigpen AT nc.rr.com>]

Well, that Palm Beach data point is the textbook definition of an outlier.  
It really jumps
out at you when charted like that.  To extend this analysis a bit (probably 
too far), let's
do a little math.

I'll make the assumption that the numbers from the other counties are 
somewhat predictive
of the Palm Beach result.  Using these numbers we can get to an estimate of 
the expected
Buchanan totals and compare that to the actual.  A big discrepancy indicates 
that something
might be amiss.  I'm breaking all sorts of statistical guidelines here, but a 
rough look
could be fun.

The frequency distribution peaks between 0.004 and 0.005.  Let's call it 
0.0045.
We'll assume that this indicates the most likely ration in Palm Beach as 
well, ignoring
local differences.
The actual ratio was .02229 of the 156,253 votes.
Using my 0.0045 number implies that we expect Buchanan to get only 703 votes.
His tallies read 3,407.
A difference of 2,704 votes.
Even if only a fraction of these were in error, it could have been enough to 
swing the
state based on the differential between the Bush and Gore tallies in the 
ongoing recount.

It's a rough model, but you get the idea.  What a day for democracy.  Could 
we maybe do a
little usability review of the ballot design next time around?

Regards,

Ron Thigpen





Steven Champeon wrote:

> Here's some great info from Jason Kottke WRT the real math behind the
> Palm Beach/Buchanan debacle.
>
> ----- Forwarded message from Jason Kottke <jason AT moreover.com> -----
>
> From: "Jason Kottke" <jason AT moreover.com>
> To: <schampeo AT hesketh.com>, ...
> Subject: proper statistical analysis of FL election data
> Date: Thu, 9 Nov 2000 10:24:39 -0800
>
> Found this on Slashdot today:
>
> http://web.mit.edu/norstadt/Public/election.pdf
>
> It's a proper analysis involving standard deviations of the FL election
> data. It's a very strong mathematical argument that a discrepancy exists in
> the Palm Beach data. This analysis is much more accurate look at the data
> than something like this:
>
> http://tropics.eastwindgraphics.com/uploaded/newwinter/bv.gif
>
> -jason