This is a quick web page to describe why I cannot support Kathy Dopp's
statistical analysis of the Florida voting discrepancies. I want
to apologize to Kathy for this. She's worked very hard on her
numbers, and definitely has her heart in the right place, but I don't
believe that she's looked at the numbers from the angle presented below.
One of the things that I have heard repeatedly when presenting Kathy's
information to people is "but people in those counties have always
voted that way". With this in mind, I decided to perform a
statistical comparison of the 2000 voting information with the 2004
voting information. This is a description of the process I went
through to develop my numbers. If you would like to go straight
to the numerical analysis, that can be found on this page.
The quick description is that for each county in Florida, I started
with the number of people who voted in 2004 in the county, and then
multipled that result by the percentage that voted Republican and
Democratic in the 2000 election. I then adjusted the result based
on changes in voter registration, multiplying by a factor of 101.4271%
to compensate for the increase in the number of people who registered
independent but voted Rep or Dem this year. The resulting value
told how many people we could have expected to vote Republican and
Democratic this year in each county, assuming no shifts in allegiance.
I then took the resulting numbers and compared them to actual numbers,
individually and in sums of E-Voting and Optical Scan groups.
What I found was that there was a shift of +3% (std dev 16%) for
Republicans, and +1% (std dev 8%) for Democrats in the Optical Scan
group. In the E-Voting group, however, there was a shift of 11%
(std dev 9%) for Republicans, and a shift of -1.2% (std dev 6%) for
Democrats.
The conclusion that I have to come to is that, if there is a problem
with the voting in Florida, it probably isn't in the Optical Scan
voting method. The larger standard deviation for optical scan is
expected because those counties are smaller and more numerous.
I'll leave further speculation to others.
Here's a more detailed explaination of how I derived these
numbers. I'm going to use Broward County as my example because
it's the first one on the list.
1. I started with the percentage of each party registered in 2004, and
divided that by the percentage of each party registered in 2000 to
determine the "registration adjustment"
Republicans in 2000: 30.5%
Repulbicans in 2004: 26.8%
Republican Registration Adjustment (2004/2000): 87.9%
Democrats in 2000: 51.4%
Democrats in 2004: 50.5%
Democratic Registration Adjustment (2004/2000): 98.2%
2. I multiplied the registration adjustments by a uniform 101.4271% to
account for the increase in people registering as independent, but
voting Rep or Dem.
Final Republican Registration Adjustment: 87.9% * 1.014271 = 89.2%
Final Democratic Registration Adjustment: 98.2% * 1.014271 =
99.56%
I set this number aside for step 4.
3. I determined the % of votes that went to each party in 2000 and
applied those percentages to the 2004 total votes.
2004 Total Votes: 686,715
2000 % of votes for Republicans: 30.93%
2000 % of votes for Democrats: 67.41%
Expected votes for Republicans: 686,715 * 30.93% = 212,413
Expected votes for Democrats: 686,715 * 67.41% = 462,914
4. I then pulled out my Final Registration Adjustments from step 2 and
multiplied the expected votes by this to come up with my projection of
what the votes should be.
Republican: 212,413 * 89.2% = 189,467 projected votes, 2004
Democratic: 462,914 * 99.56% = 460,890 projected votes, 2004
5. I then divided the actual votes by this number to find a variance
from the projected value.
Republican: 236,794 / 189,467 = 124.98%
Democratic: 441,733 / 460,890 = 95.84%
From there it was a simple process of summing and calculation of a
standard deviation. The full set of calculations can be found on the statistics page. If you
have any questions or comments, please send them to politics@rapplean.net.