[FoRK] Shooting dice with Zogby, or, Poll-less pollstering

Jeff Bone jbone at place.org
Thu Nov 4 07:21:55 PST 2004


Well, although my model got it wrong in the end, the delta was 
surprisingly small.  And what's even more surprising about that was the 
method I used to come up with the numbers, and what that (might) mean 
about America.

W/o going into too much detail, the model assumed as a base that 
American preferences were Gaussian distributed;  it then randomly 
selected a number of samples (i.e., modeling voter turnout 
independently of preference;  I mistakenly correlated those in my 
intuitive "runaway" model) for each state (2x the electoral total for 
that state) with a base 50-50 split to decide where they were in the 
distribution, with a small proportional correction for previous 
performance of the state.  (Basically, the normal curve was divided 
into 6 slices, from "strong Kerry" on the left to "strong Bush" on the 
right.  2d6 were rolled 2x as many times as the states electoral tally 
to pick among the slices.  An positive or negative factor from -3 to +3 
was added to each roll based on the margin of victory in the popular 
vote in 2000.)

That method beat Zogby's comprehensive polling and analysis;  indeed, 
it called 'em all (at this point) except for Ohio.  (Iowa and NM may 
fall out as well, I suppose, but overall --- score one for randomness.)

What does this tell us?  I dunno.  At a minimum it should make us all 
question the validity of any polling technique.  With a nation as 
perfectly polarized as ours, and with the margins as narrow as they 
are, it should make us question whether any sample set much smaller 
than "everybody" is in any way representative.

It should also make us question the meaning of the popular vote victory 
margin;  we should be cautious what conclusions we draw from that.

Lastly:  I'm sick of hearing about the "values" issue.  The same 
freakin' talking heads can on the one hand lament how badly astray they 
were lead early in the night by the exit polls and then, immediately, 
turn around and start talking about how the exit polls said "values" 
were a deciding factor...  now look:  either the polls were meaningful 
or they weren't.  Make up your minds, pundits!  ;-)

jb



More information about the FoRK mailing list