[ThePythonicCow]

Forty posts later one person reads the paper

Some of what Torie saw in the paper differs from what I saw. The first part of what Torie writes of the article I agree with. But then he writes:

The counties are not listed, so I don't know if there is another logical explanation.

They list all the Florida (and Ohio) counties, with individual stats, in the data appendices, county by county.

From http://ucdata.berkeley.edu/new_web/VOTE2004/index.html:

VOTE, 2004
• Working Paper: The Effect of Electronic Voting Machines on Change in Support for Bush in the 2004 Florida Elections
  by Michael Hout, Laura Mangels, Jennifer Carlson, and Rachel Best
  Summary of findings
  DATA
  Variable Descriptions
  Election Data - delimited text file
  Election Data - Excel file
  Election Data - stata file
  Election Data - SPSS file

Torie also wrote:

The study did not also consider that maybe the electronic voting counties had more population growth than the state as a whole from 2000 to 2004.

I disagree. In the above data, they account for the change in each counties size.

Torie wrote:

The correlation could also be random noise.

The correlation they find has a rather remote chance of being random.

There really does seem to be something that caused Bush to make greater gains in Broward and Palm Beach, as a percentage of the votes cast, then he made in the other Florida counties.

The only serious weakness I see in the paper is that it completely misses the alternative hypothesis, that the e-voting machines suppressed 'Rat fraud with paper ballots, rather than increasing Republican fraud using e-voting machines.

[Torie]

Thanks for your post. You make some good and interesting points, and the counties are listed in the data files I see. You will note that in model 2, whatever that is, there is no significant T stat, as the authors admit. I played with the excel file. Of the 13 counties with Etouch, Bush gained 2.4% over 2000, on average. Of the balance of 54 counties, Bush gained 4.1% on average.

Apparently what the authors did was flog the data, by more heavily weighting size. And do you have a comment to my post number 101, where the authors admit that there was a negative correlation with the manufacture of votes for Bush via etouch in counties where Bush did his best in 2000? So what we have on net, is the authors weighting for size, and thus apparently claiming that size had something to do with the etouch miscounting or some odd statistical anomaly, and then admitting that that only where there were lots of Gore voters, did the correlation pop up. Isn't a better explanation that something was going on in the Dem belt in Broward, Palm Beach and Dade (namely Jewish voters), and maybe a slight tendency for counties to regress to the mean because the play of issues cut less than in 2000 by geography?

If you just look at the excel spread sheet, you will see that the whole thing does not pass the smell test. Just excise the non etouch high population counties Duval and Orange from the data base, and I suspect you will see the T stat drop down towards nothing.

Oh yes, and the thesis that Gore stole votes last time, and Kerry didn't this time, I find absurd.

[Torie]

Oh yes, one other thing. The average Bush percentage gain in Dade, Broward and Palm Beach was 2%, the lowest percentage gain of all. Odd. The percentage is, granted, dragged down by Dade, where Cubans tended to abandon Bush vis a vis 2000. But Palm Beach and Broward still have a lower percentage gain for Bush than his average in the non etouch counties.