November 07, 2004

Florida voting weirdness?

Posted by Matt

Updated again: Ryan Gabbard (of The Audhumlan Conspiracy) has referred me in the comments to this analysis at The Dead Parrot Society. I think they make a good case that this disparity is not really related to the type of voting machine in use -- it existed also in the 1996 and 2000 election results. There's a lot more there, but I think it's safe to say this was a red herring. (I still think it would probably good to have much more done in the future to ensure security of election systems, but it doesn't seem to have been a decisive factor in this election.)

Updated: I've changed the image I link to; this one is only for counties with over 100,000 registered voters. The effect is less striking but still statistically significant. I'm convinced that most of the small-county effect is due to rural voters registered as Democrat who tend to vote Republican in national elections. I'm hoping someone can suggest a similar distinction between the optical-scan and the touch screen larger counties.

Via Sean Carroll:

I also don't like conspiracy theories, and I hate to sound like an alarmist. But something is very weird about the Florida voting data. Unless there's a good reason why registered Democrats voting for Bush would be strongly correlated with the presence of optical-scan voting devices (an effect that, to a statistically significant level, happens in both large and small counties), it looks a lot like some votes could have been tampered with. Not that I'm claiming they were, I certainly don't have enough information for that, but.... Someone really should look into this.

See more explanation here. And check out the graph, for counties with over 100,000 registered voters only:

That looks pretty suspicious to me. Counties with many more Democrats than Republicans go strongly for Bush if and only if they use optical scanners?

(I should note that the percentages of registered democrats plus republicans tends to add up to over 90%, if you look at these data, so one can't explain this by large numbers of independents voting for Bush.)

Posted by Matt at November 7, 2004 10:31 PM

A few thoughts off the top of my head. It's late, my brain isn't working, and I haven't really look at things closely, so take them with a huge grain of salt.

(0) I'm a little nervous about them placing so much emphasis on registration data, since it is famously weird with respect to how people actually vote (I think something like 1/3 of the Democrats in KY vote consistently Republican), not to mention I've heard many places' voter rolls may deviate significantly from the actual population, depending on population churn and how good they are about updating them.

(1) There's a clear geographic difference between the counties using E-touchscreen and those using optical scan. Of the non-optical-scan counties, 8 are in south Florida, 6 are in central Florida, 1 is in north Florida, and 0 are in the Panhandle. Perhaps more importantly, just glancing quickly at the map (I'll try to systematically check this out tomorrow) they are nearly all counties containing or adoining large cities; checking a couple of opti-scan counties that contain large cities (Oceola, Volusia), they seem to act similar to the E-Touchscreen counties.

I don't know much about Florida, but I do remember from the 2000 mess that there are big regional differences in political behavior there, not to mention general urban/non-urban differences. So I think it's worth looking to see if there is some variable like region or population density which simultaneously predicts both vote/registration divergence and what type of voting system a county uses. Which I may do tomorrow, depending on how busy I am.

Posted by: Ryan Gabbard at November 8, 2004 12:42 AM

Thanks for the reply, Ryan. With regard to (at least part of) your second point: at the site I linked to there is another graph that restricts attention to counties with over 100,000 voters. The effect is less striking there, but still statistically significant. (In fact, it's probably the graph I should have displayed, since it seems to me to be much more meaningful.) If it weren't for that I wouldn't really be very suspicious.

But, you're right, there are other things to consider. As Sean Carroll says in his post, I hope that these data are looked into and that the result is that nothing is amiss. I'm not happy about the election results, but I would rather they stand and the vote be accurate than see that it is so easy to tamper with election results. (Looking at's information on the certification of voting machines, though, I worry that something really should be done to provide a more convincing security test of computerized voting systems.)

Posted by: Matt at November 8, 2004 01:24 AM

I'm skeptical about this, too, but I haven't had time to look into it much.

Here's some evidence against vote tampering. On his blog, Sean Carroll writes "sizeable number of overwhelmingly Democratic counties, mostly smaller ones, where Bush came away with a big victory. Franklin county, 77% Democrat, went for Bush 3,472 to 2,400. Hamilton county, 79% Democrat, went for Bush 2,786 to 2,252. Holmes county, 73% Democrat, went for Bush 6,410 to 1,810. Lafayette county, 83% Democrat, went for Bush 2,460 to 845. Liberty county, where Democrats outnumber Republicans by 88% to 8%, went for Bush 1,927 to 1,070. And on and on."

But you should be able to test this (at least somewhat) by looking at 2000 election results. (If it's so implausible that a majority Democratic county would vote for Bush, after all, then it would be odd if it did so two years in a row.) You can find 200 data here:

Here are some data from the last election (county name, Bush total, and Gore total):

Franklin 2,454 2,047
Hamilton 2,147 1,723
Holmes 5,012 2,177
Lafayete 1,670 789
Liberty 1,317 1,017

By my quick calculation, Bush got 54.2% of the two-party vote in Franklin County in 2000, 55.5% in Hamilton County, 69.7% in Holmes County, 67.9% in Lafayette County, and 56.4% in Liberty County.

Compare that to 2004, when Bush got 59.1% of the two-party vote in Franklin County, 55.3% in Hamilton County, 80.0% in Holmes County, 74.4% in Lafayette County, and 64.3% in Liberty County.

Could there have bene vote tampering in Florida this year? In theory, sure. But it looks a lot less likely if you compare the 2000 and 2004 data.

Posted by: Ed at November 8, 2004 08:40 AM

Of course, it's theoretically possible that someone tampered with the Florida vote in both 2000 and 2004. I think that's pretty unlikely, though, and I strongly suspect that the counties where Bush's vote totals are said to be suspiciously large have a high concentration of conservative rural Democrats who may still vote for their party in local or state elections, but who vote Republican in elections with greater national significance.

Posted by: Ed at November 8, 2004 08:50 AM

Ed -- I agree with you, but again, this is only looking at the smaller counties. I've altered the post a bit to reflect that there is still an effect when you consider only counties with over 100,000 registered voters. It is this that I haven't really seen any convincing explanation of.

Posted by: Matt at November 8, 2004 09:16 AM

DPS has data from 1996 and 2000 that shows basically the same pattern we see in the 2004 large county data; the differences between the touchscreen and op-scan counties in registration/voting difference were present before the touchscreen counties even started using touchscreens.

Posted by: Ryan Gabbard at November 9, 2004 02:13 PM

Thanks, Ryan. What is this DPS?

Posted by: Matt at November 9, 2004 03:23 PM

Oops - apaprently you can't make hyperlinks in comments (at least not the usual HTML way). DPS was supposed to to be a link to The Dead Parrot Society at

Posted by: Ryan Gabbard at November 9, 2004 05:31 PM
Post a comment

Remember personal info?