Did Lightning Strike Twice in Florida?
November 18, 2004 2:35 PM   Subscribe

Florida is the New Florida Although many discussions of voting anomalies focused on Ohio, a statistical analysis of Florida voting patterns performed by sociologists at University of California, Berkeley suggests that electronic touch screen voting in Palm Beach, Broward, and Miami-Dade may have credited George Bush with up to 260,000 extra votes in Florida. The discrepancy is not enough to change who won Florida, but it could have narrowed Bush's lead to 90,000 votes instead of 350,000, highlighting the need for better auditing of elections with electronic voting.
posted by jonp72 (33 comments total)
This is from my own site, but I am going to post it again here:

In 10 minutes I came up with a near fool-proof method for tabulating votes with almost zero possibility of fraud. It is, basically:

Touch screen with both the candidate's names and faces listed for the voter to use for choice which then prints out a paper ballot, again with the choice's names and faces on said ballot as well as a unique, say, 12 digit alpha-numeric code which can be used to identify the ballot which can be matched to a receipt that is also printed out for the voter so the voter can, in the future, demand to see his/her vote. The vote counting is done electronically, but there is also the paper ballots to count in the event of a question about fraud. The voter's themselves can always, if they are concerned, go see if their vote is actually present on paper during a recount.

The voting machines would also print paper logs (a standard unix security procedure) of anything that is done to the machine - this is an easy way to make sure nothing gets hacked without leaving footprints.

Now, if I, a non-computer scientist, can come up with that foolproof of a system in 5 minutes, why can't diebold (or, in this case Election Systems & Software or Sequoia Voting Systems), who have been working on this problem for 5 years, come up with something equally as reliable?

The only answer that I can see: They can but they are not. And that really scares me.
posted by Yellowbeard at 2:47 PM on November 18, 2004

Oh lord. My first political thread. Gaia help me.

What I think all of these "have data, make your conclusions after the fact" studies are lacking is the psychological aspect of what might have dramatically swayed previous democratic strongholds.

Specifically, as trite as it sounds, the 9/11 effect. I know first hand the result of this effect, from an older relative. A through and through Democrat, I did not even think to discuss politics with her till the day of the election. In my utter horror, she told me confidently how she had voted for Bush.

But... but... but WHY?

9/11. She felt that since we were attacked, we needed to "kick some ass." Since the news showed her that we were indeed killing people somewhere, she felt that this was the proper action to take and she wanted to support the person doing it.

I'm not trying to objectify her solely on that, but think about that rationale. A woman who is pretty much a pro choice, relaxed, libertarian type, voted Republican because she felt that we were attacked and that we needed to strike back. Her face sort of dropped into suprise as I explained to her how Iraq was sort if a diversion, not a response, how this and that and Abu and yadda yada. But it didn't matter, because as someone who was not reading dailykos and news.google everyday, she was content to know that we were "doing something" about those who attacked us, and that's what counted.

Electronic voting. Come on. Paper receipt, impartial witnesses, done.
posted by cavalier at 2:52 PM on November 18, 2004

9/11. She felt that since we were attacked, we needed to "kick some ass."

I have seen this attitude as well among some of my own family members. There also seems to be a willful lack of analysis that goes along with it, however. I was all for kicking ass and taking names in the Al Qaede training camps of Afghanistan, but then redirecting our miltary and attacking a country that, while we didn't like, had nothing to do with 9/11 is psychotic.
posted by Flem Snopes at 3:00 PM on November 18, 2004

I don't know enough statistics to understand the paper. Are they actually trying to predict the vote results from independent variables, and then saying that the difference from the actual results are "extra votes"?
posted by smackfu at 3:00 PM on November 18, 2004

The voting machines would also print paper logs (a standard unix security procedure)

I've heard tell that a big anti-paper-trail argument is the risk of unreliable printers (running out of paper, jamming, and whatnot) -- but what of the once ubiquitous, noisy but simple, reliable and unlikely-to-jam dot-matrix printer, running on a seemingly endless stream of perforated paper?

I always assume this is the kind of printer people mean when they talk about paper trails, but perhaps I'm being presumptuous. I certainly hope these are not references to the gas station/cash register type of printer, as those are terribly, terribly unreliable and difficult to load (I was a cashier at a grocery store once upon a time.)

impartial witnesses

If you manage to find a group of those that all sides are willing to trust, well, you get my vote for president next time 'round.
posted by davejay at 3:00 PM on November 18, 2004

smackfu - yes, the authors of the statistical study were trying to mathematically (model) predict (explain), for each of the 67 counties in Florida, what percentage of votes George W. Bush would have been expected to get, and to see what variables explained (predicted) the changes in votes for George W. Bush between 2000 and 2004.

Some variables used: change in total voter turnout between 2000 and 2004; median income for those living in the county; the hispanic population (percent), and the size (population) of the county.

From the Wired article: "They discovered that in the 15 counties using touch-screen voting systems, the number of votes granted to Bush far exceeded the number of votes Bush should have received -- given all of the other variables -- while the number of votes that Bush received in counties using other types of voting equipment lined up perfectly with what the variables would have predicted for those counties."

Correlation isn't causation, of course. But it is reasonable to ask skeptics of the statistical study to either (a) explain why the chosen statisical approach was wrong, and create a revised model that is less wrong, or (b) offer additional, logical variables (no data fishing allowed!) that can be added to the model which will reduce the explanatory/predictive power of having/not-having touch-screen voting machines.

And if (a) or (b) don't appear, then (c) finding out as much as possible about the systems that were used, and (d) adding safeguards, including much more thorough audit trails; would appear to be called for.
posted by WestCoaster at 3:54 PM on November 18, 2004

Speaking as a computer geek, the very idea of conducting voting without a paper trail is either a) an example of utter stupidity/insanity, or b) a good way to steal an election. Computer records can be changed quite easily, that's one of the nice things about computers, but its also one reason to keep hardcopy backups of important things.

More important is that the paper trail be human readable. It does no good at all for the vote machine to spit out a slip of paper covered in incomprehensible code. That does nothing to assure the voter that the paper trail accurately recorded his vote. After verifying that the paper accurately recorded his vote, the voter drops it into the standard locked box where it will probably never be used. But the operative word there is "probably".

On the same subject, the notion that we can allow closed source vote machines to exist is preposterous. The only way to be sure that the machines are not programmed to give votes to people who didn't earn them is to allow the source code to be analyzed by anyone who wants to. I certainly don't trust any for profit corporation not to cheat and I doubt I'm alone on that. Openness is the only way known to be effective against cheating.
posted by sotonohito at 3:59 PM on November 18, 2004

As a grad student in sociology at Berkeley (although not one who authored the research paper), I hope I can make some of the research more understandable to the layperson. Whether I succeed is another story. The dependent variable (i.e., the variable that the equation tries to predict) is not the Bush vote, but the change in the Bush vote from 2000 to 2004. By modeling their equations this way, the Berkeley team was able to account for outside factors, like 9/11, that would have increased Bush's vote across the board. (So while I acknowledge the many post-9/11 anecdotes on the board, they can't be used as proof against the study at hand.)

What the sociologists' mathematical models actually detect is whether Bush's vote increases or decreases more than what can be attributed to random chance in comparison to other counties. What the researchers found is that if a county had a high Gore vote and electronic voting, then the increase of the Bush vote was significantly much higher than it was in counties that didn't have both of these things.
posted by jonp72 at 4:03 PM on November 18, 2004

If you manage to find a group of those that all sides are willing to trust, well, you get my vote for president next time 'round.

This guy doesn't seem to care who will be the president of the US, maybe he can lend some of his people?
posted by kika at 4:20 PM on November 18, 2004

This isn't a "problem" Diebold needs to work on; they've been making similar systems that require unique ID, a PIN, and provide a paper trail for years.
posted by eustacescrubb at 4:36 PM on November 18, 2004

Whenever people begin talking about e-voting, it seems that someone always talks about how hard it is to secure them, and how new processes need to be built from scratch. Then, they usually go into a list of requirements to be met. The interesting thing is that very many of these same requirements have already been met by government agencies in the past. They are rigorous, well-thought-out, and comprehensive. They don't even require "open-source" of the code. I'm talking about slot-machines, and the validation of the software which runs them. I would still love to see a paper trail, but for all the "background" stuff, such as validating the software to ensure no cheating, and that the validated software is actually installed on any given machine, we could hardly find a better place to start.

Slot Machines are More Trustworthy than Electronic Voting.
posted by Invoke at 4:56 PM on November 18, 2004

I like the effort; I'd be happy to find out that Bush only got in this time on a fubar too. But.

explain why the chosen statisical approach was wrong

Modeling county vote percentages isn't very smart, IME. What they should be doing is building a good, comprehensive individual-level vote model and then adding voting method to see if it's significant.

A county level analysis here is particularly bad because the counties with e-voting are pretty distinctive. The e-voting counties are mostly heavily urbanized areas -- Miami, Fort Lauderdale, Palm Beach, Tampa, St. Petersburg, Sarasota, Naples if you want to count that. If some other factor related to that is what's driving their vote, it's going to contaminate their estimates.

That said, they almost certainly can't do a good individual level analysis until survey data come out; I don't know when the National Election Study will release its data for this year.

Minimally, though, they ought to check their model. A simple way to do this would be to run their model for the 2000 or 1996 elections, keeping in place the voting method for 2004. If the model for 2000 shows that e-voting matters even when it's not being used, that's a clue that it's something else.

(b) offer additional, logical variables (no data fishing allowed!)

They haven't paid much attention to a vast literature on voting models. Off the top of my head,

The only variable they have to tap race is %hispanic. It's certainly wrong to leave out the percentage of black voters; race is a strong predictor of votes. It might well also be wrong to have a single "hispanic" variable for Florida, where Cubans != Puerto Ricans != Nicaraguans != Hondurans != Dominicans != Mexicans. Minimally, they really should distinguish between %Cuban and %Other-Hispanic, since Cubans tend to be rather dramatically more conservative than other Latinos.

They should add variables for religion and religiosity.

They should add variables for contacts by each campaign. Either appearances by each candidate, or money spent in the county.
posted by ROU_Xenophobe at 5:06 PM on November 18, 2004

this analysis seems really damning to me. i read the whole PDF, but i have to admit i probably know less about statistics than smackfu.

can anyone with a solid grasp of ordinary-least-squares regression models debunk this? if not, this research should be big news.

on preview: thanks, ROU_Xenophobe!
posted by mrgrimm at 5:12 PM on November 18, 2004


Technically, the fubar was not necessary for Bush to win. The 260,000 maximum estimate of excess votes was still less than Bush's margin of victory in Florida

Regarding the statistical method used, individual-level data will not really be much of an improvement, because what you're analyzing is vote totals, not what influences an individual's choice of candidates. Votes can be "added" or "subtracted" in one county without regard to the characteristics of the voters who live there. As for the effect of urbanism, the statistical team's researchers tested models that include a variable for county size, but this did not effect the results attributed to e-touch voting.

Regarding the addition of other logical variables, it is true that the voting model does not have as many variables as those you might find in academic political science journals. However, the model is designed to explain as much of the variance as possible by using the Bush vote in 2000 to predict the change in the Bush vote from 2000 to 2004. Any variable you suggest adding to the model would have to be something that changed significantly from 2000 to 2004, because the model focuses on discrepancies in 2000 vs. 2004 vote totals instead of just the 2004 vote total itself. In fact, I was the grad student who suggested to the research team that they include %hispanic in their models, because of the possible effect of Senate candidate Mel Martinez in increasing support for Bush. We would have liked to have a more detailed ethnic breakdown of the Hispanics, but we don't have it for the voting population. We can't use the raw Hispanic numbers on different ethnicities, because different Hispanic ethnicities would have different rates of voting, due to nonrandom differences in percentage of noncitizens in each group.

Time constraints may explain the omission of other variables, although other more methodologically legitimate reasons such as the desire to reduce collinearity are probably also be involved. In addition, the study was vetted by several somewhat skeptical professors before being released, including Henry Brady, who is one of the world's foremost experts on punch cards, optical scan machines, and Votomatics.
posted by jonp72 at 5:35 PM on November 18, 2004

Regarding the statistical method used, individual-level data will not really be much of an improvement, because what you're analyzing is vote totals, not what influences an individual's choice of candidates.

You're trying to find whether or not there were consistently more votes for Bush than you should expect in the e-voting counties.

To find this out, you first need to know how many votes you should expect in 2004.

You can do this with county characteristics, but you'll hit ecological inference problems. A far superior way to construct an expected vote is to build an individual-level voting model and aggregate up to the county level.

As for the effect of urbanism, the statistical team's researchers tested models that include a variable for county size, but this did not effect the results attributed to e-touch voting

But urbanism is not size. Florida has populous counties without classically urban cores.

Any variable you suggest adding to the model would have to be something that changed significantly from 2000 to 2004

That's simply not true. It would be enough for the import of the variable to change between 2000 and 2004. Even in a county where the proportion of white evangelicals who attend church weekly didn't change between 2000 and 2004, we'd expect to see a change in evangelical voting patterns because religiosity seems to have been substantially more active in 2004 than in 2000.

As a more general version of the point, the model ignores the effects of campaigning. You don't know whether Bush did better than expected in those counties because of e-voting, or because Bush did a lot of campaigning there get the white evangelical vote out, or what.

As an aside, what are you using for votes in 2000? Actual votes as recorded? Or intended votes estimated one way or another? I'd have to look stuff up to be sure, but I'd bet that e-voting in 2004 is correlated with punch-cards in 2000. Which means that we can't be all that certain about 2000 voting levels to start with!
posted by ROU_Xenophobe at 7:04 PM on November 18, 2004

jonp, ROU - touchscreen eVoting is only one of a number of approaches to vote fraud.

For example : ".....A secret document obtained from inside Bush campaign headquarters in Florida suggests a plan - possibly in violation of US law - to disrupt voting in the state's African-American voting districts, a BBC Newsnight investigation reveals.

Election supervisor Ion Sancho believes some voters are being intimidated

Two e-mails, prepared for the executive director of the Bush campaign in Florida and the campaign's national research director in Washington DC, contain a 15-page so-called "caging list".

It lists 1,886 names and addresses of voters in predominantly black and traditionally Democrat areas of Jacksonville, Florida.

An elections supervisor in Tallahassee, when shown the list, told Newsnight: "The only possible reason why they would keep such a thing is to challenge voters on election day."

Ion Sancho, a Democrat, noted that Florida law allows political party operatives inside polling stations to stop voters from obtaining a ballot.

Mass challenges

They may then only vote "provisionally" after signing an affidavit attesting to their legal voting status.

Mass challenges have never occurred in Florida. Indeed, says Mr Sancho, not one challenge has been made to a voter "in the 16 years I've been supervisor of elections."

"Quite frankly, this process can be used to slow down the voting process and cause chaos on election day; and discourage voters from voting."

Sancho calls it "intimidation." And it may be illegal.

A Republican spokeswoman did not deny that voters would be challenged at polling stations

In Washington, well-known civil rights attorney, Ralph Neas, noted that US federal law prohibits targeting challenges to voters, even if there is a basis for the challenge, if race is a factor in targeting the voters.

The list of Jacksonville voters covers an area with a majority of black residents."

Hell - I've read metafilter members, college students - and likely white - who've talked about being disenfranchised from voting in 2004 due to absentee balloting process "screw ups" - but there were an absurd number of such screw-ups, it seems...


There are multiple, convergent methods which can be used for hacking elections.
posted by troutfishing at 8:48 PM on November 18, 2004

The higher the voter turn out, the harder it is to conceal pushing the numbers... I think that we have a shot at getting fraud exposed now, regardless of media control. Scouring the net for these kind of links takes hours and people get burned out, go have a beer and move on to the next spectator sport. I have found a few sites that are up to date on this stuff.: DemocraticUnderground.com, BBVforum but I think that we need more sites like fugw.org that consolidate the current actions needed to get people off the couch.
posted by 0of1 at 9:20 PM on November 18, 2004

OK - so, we've got the motive (4 more years of increasing aggression, bare-faced lies and mass looting); the means (paperless, unauditable, closed source, proprietary voting and vote counting machines, along with the partisan apparatchiks who own, control, and operate them); and now we're starting to see some well-documented evidence of non-random anomolies (all of which just happened to favor Bush). For those in the "faith-based community" this is no doubt strong evidence of divine intervention - JAHEEZUZ entered the voting machines and changed the votes from Kerry to Bush, hallelujah it's a miracle!

Us poor saps in the "reality based community" have some serious soul-searching to do.

posted by dinsdale at 10:04 PM on November 18, 2004

I'm sure the Berkeley researchers are TOTALLY objective and have NO axe to grind.

Those damn little gremlins in those machines.


You lost by 3.5 MILLION votes.

If you keep throwing Michael Moore at us... if you keep equating morality to economics... if you continue with your zany Diebold conspiracy theories... if your media keeps up its racist attacks on Dr. Rice and every other minority who doesn't high-step to Berkeley fight song... you will loose by 7 million votes when H-Rod runs in 2008.

Just trying to help... now go see a counselor for your post-election blues before you act on your suicidal or homicidal ideations.

BTW I voted for Bush.... I'm a skeptical Christian... don't give a crap about gay marriage but am laughing at the stupidity of the Massachusettes judge who ramned it down the red-states throat... support stem cell research... am personally against abortion but do not want to overturn Roe V Wade... don't think W is a genius but is not as stupid as you think... am against a good part of the patriot act... think NO presidents have done us right in environmental policy and fossil fuel dependency... and agree with the neocons that the fasciism of radical islam needs to be stopped (don't call me racist... I said radical islam)... and think that the modern democratic party is the most racist organization on the planet and depend it to cling to what little power they have left.
posted by dancingbaptist at 10:50 PM on November 18, 2004

d.b. - wow - sounds like you've got a regular swiss-army axe to grind there, best take that beam out of yer eye before trying to find the mote in mine. One more time - with 35+ million votes cast on mystery machines, *no-one* knows what happened behind the curtain - and no-one ever will. Why has every effort to provide a minimum of accountability to the process of recording and counting votes been stymied at every turn? The whole point of having free and fair elections is so that the *losers* accept defeat. If it's not free or fair then the bargain is off. The present reaction of those without your "faith" is not only predictable, but *was predicted* by many observers. Those responsible for the massive deployment of highly insecure and unauditable voting machines were either incredibly, brainlessly incompetent, or they were leaving the door open to manipulate the system. Unfortunately, I share your opinion the dubya's not as dumb as he seems...
posted by dinsdale at 11:11 PM on November 18, 2004

Nice entrance there , DB.
posted by PareidoliaticBoy at 12:00 AM on November 19, 2004

Does anyone know if there is an organized campaign yet to very vocally demand transparency in voting systems? I feel that that's a cause I'm really willing to put myself behind as we prepare for '06 and '08.
posted by rustcellar at 1:37 AM on November 19, 2004

I see that verifiedvoting.org in the post is probably what I'm looking for.
posted by rustcellar at 2:33 AM on November 19, 2004

This has been covered in detail, and while an interesting technical article on the nuts and bolts of election fraud, I fear those who liked the election results will say "but there is no proof of fraud!!!!" and those who didn't like the election results will say "This is about Democracy!!!!" and no one will get anywhere.

As for me, either voting is transparent and as fraudproof as possible, or we have a monkey trained by the highest bidder pull names out of a hat. Your choice America.
posted by Freen at 5:01 AM on November 19, 2004

"if you keep equating morality to economics.....I'm a skeptical Christian... don't give a crap about gay marriage but am laughing at the stupidity of the Massachusettes judge who ramned it down the red-states throat" - most professing Christians I know are more judicious in their use of language, and quite a few do equate morality to economics.

It goes with the ethic...... the Christian ethic, from Christ - remember ?

Cracked your Bible in the last decade or two ?
posted by troutfishing at 7:19 AM on November 19, 2004

"As for me, either voting is transparent and as fraudproof as possible, or we have a monkey trained by the highest bidder pull names out of a hat. Your choice America." - I think the mass media has made that choice for us. The media blackout on election irregularities demonstrates it's almost total subservience and fellatio to power."

DB - Oh, so you buy into the myth of race, eh ? That's OK, most people who don't bother to keep up with the latest scientific thought on the matter - of the last decade or so - buy into the race myth too. Genetics researchers have dispensed with "race" though, as a meaningful concept.

But as for "racism", or discrimination against African Americans.... well, I think the Republican Party has got that field well covered, and - to the extent that you fail to acknowledge that, you are demonstrating your own "racism" :

Voter Suppression : "Republicans Pressed To Halt Voter-Suppression Efforts

by Jim Lobe

WASHINGTON – With political analysts agreeing that voter turnout, especially of minority and youth voters, will likely determine the outcome of next Tuesday’s presidential election, civil and human rights groups are pressing the Republican National Committee (RNC) to call off plans aimed at discouraging people from casting ballots.

At a press conference held in front of RNC headquarters here Thursday, the Leadership Conference on Civil Rights (LCCR), the country’s largest civil and human-rights coalition, demanded that RNC chairman Ed Gillespie ensure that the party does nothing to suppress the vote or try to intimidate voters, particularly in minority communities.

“In state after state, Republican officials and operatives are working to deny American citizens the right to vote,” charged LCCR executive director Wade Henderson. “We’re today to ask the RNC Chairman to put a stop to these activities.” "

Welcome to Metafilter.
posted by troutfishing at 7:33 AM on November 19, 2004

I'll say it again. Next time:


They've worked fine for centuries. If it ain't broke, don't fix it, especially with dubious technology controlled by partisan hands.

It it really that @#$%ing hard for people to grok?
posted by zoogleplex at 10:04 AM on November 19, 2004

A few thoughts.

ROU_Xenophobe noted it, but it needs to be emphasized: the extraordinary ignorance of Florida political conditions betrayed by the treatment of the "Hispanic" vote as a uniform block makes me not only question the analysis but make me wonder what other basic facts they got wrong. Cubans are mostly Republicans. The growing population of immigrant Mexican, South and Central American elites are pretty well split between the parties. Non-Cuban working class Hispanics are virtually all Democrats.

Several groups of voters shifted by material extents to Bush: older voters, Jewish voters, richer voters. Guess what three demographics are particularly well represented in the more populous South Florida counties, relative to other places in Florida.

That their method didn't show variances in Ohio wouldn't be surprising, given that Hispanics in Ohio are monolithically Democratic and Jewish, older, and richer voters aren't particularly concentrated (that I know of).

(And ... to toss in a little bit of Free Republic tin-foil-hat stuff, which of course I don't endorse, who's to say that electronic voting didn't simply disable traditional Democratic paper voting fraud schemes.)
posted by MattD at 10:47 AM on November 19, 2004

ROU_Xenophobe noted it, but it needs to be emphasized: the extraordinary ignorance of Florida political conditions betrayed by the treatment of the "Hispanic" vote as a uniform block makes me not only question the analysis but make me wonder what other basic facts they got wrong.

First, you need to look again at the model in the actual academic paper. The paper used Bush's 2000 vote share to predict the increase/decrease in George W. Bush's vote from 2000 to 2004. You'd run into methodological problems if you added controls for all these other additional variables (e.g., black vote, religiosity, urbanism), because they are so highly correlated with George W. Bush's vote from 2000. (Check the link on collinearity in my previous post for the methodological reasons behind this.) So much of the variance in Bush's support in 2004 is explained by his level of support in 2000 that black vote, religiosity, urbanism, and other variables don't add that much to the model.

Second, if you run the model developed by the Berkeley research team, it is extremely accurate in predicting the vote (without controlling for race, income, religiosity, urbanism etc.) in almost all Florida counties with the sole exception of urban counties that are both heavily anti-Bush and have e-touch voting machines (e.g., Palm Beach, Broward, Miami-Dade). As Occam's Razor says, a parsimonious explanation is to be preferred over a less parsimonious explanation. Most counties fit into the standard model. For the counties that don't fit, the easiest way to resolve the discrepancies is simply to add the effect of e-touch voting without any need for additional control variables. If you want to argue that a much more complex model with controls for 5 or more extra independent variables is necessary, the burden of proof is on you to argue why that would be superior to the much more parsimonious model.

Third, the paper was vetted by an immigration scholar Irene Bloemraed. It would be ideal to have data on the size of the voting population for Cubans, Mexicans, Nicaraguans, Guatemalans etc. in Florida at the county level, but it's not available. If anybody on Metafilter has collected that data, I can personally forward it to the research team that did the report and I'm sure they'll be happy to re-do the analysis with those values. The Hispanic vote was mainly added to the model in order to test whether atypical increases in the Bush vote could be explained by Latinos voting for Senate candidate Mel Martinez who also decided to vote for Bush.

Fourth, no matter how complex or airtight a statistical model may be, it is still not a "smoking gun" when it comes to investigating voting irregularities. It's not as widely reported, but the Berkeley research team performed similar analyses for other battleground states, including Ohio, and they did not find irregularities in the other states. The only place the models detect irregularities is in Florida. As a practical consideration, when you have statutes that impose strict deadlines on when you can raise questions about electoral irregularities, you can't afford to waste time by gilding the lily in the search for a mathematical model that doesn't give you any more traction than the simpler model you started with. Instead of a "smoking gun," a study like this works as a "smoke detector." It provides sufficient evidence that something about the Florida vote results are not kosher, but it does nothing to prove whether that's due to fraud or other factors such as mechanical problems, incompetence, or mathematical errors in vote tallies.

Fifth, the analysis argues that the discrepancy in votes is not sufficient to change the outcome in Florida, let alone for the presidential election. The major import of this analysis is that it shows how statistical analysis can be used to detect electoral irregularities and provide investigators with guidance about the most likely places where voter fraud might occur. As far as the analysis is concerned, what's important is not who won the election, but the possible untrustworthiness of e-touch voting and the need to investigate the soundness of e-touch machines.
posted by jonp72 at 2:55 PM on November 19, 2004

You'd run into methodological problems if you added controls for all these other additional variables (e.g., black vote, religiosity, urbanism), because they are so highly correlated with George W. Bush's vote from 2000

There really need to be variables that we should expect to correlate with an increase in Bush support or an increase in Bush turnout; variables that either changed between 2000 and 2004, or the substantive significance of which changed between 2000 and 2004. Religiosity is one, since it seems plain that evangelicals were substantially more activated this year than in 2004. The other main one would be the campaign intensity in the various counties, which you'd probably have to find a proxy for.

it is extremely accurate in predicting the vote

Pleeeeeeeaaaaaaaase tell me you don't mean by this that the model has a high R2.

the burden of proof is on you to argue why that would be superior to the much more parsimonious model

No no no nonononononono... That's a quick roadmap to crappy inference. You should include the variables that are theoretically relevant (and available), all of them, and you shouldn't include ones that aren't. This isn't an exercise in maximizing R2 or otherwise fitting the data. It's about having a theory, deriving empirical statements from that theory, and testing to see to what extent the observable implications of your theory are borne out.

What you basically have is a finding that Bush did better than expected-with-a-naive-model in the major urban centers of central and south Florida, which almost exactly the same thing as counties using e-voting (less Orlando).

Why might that be? It might be that e-voting was skewing things. It might also be that these counties had lots of campaign action on one or both sides. It might be that these counties had large numbers of evangelicals who were more activated in 2004. It might be that what we're really seeing are the fuckups in the punch-card counties in 2000. There are lots of reasons, including hijinks with e-voting, why the large urban counties of central and south Florida might have had higher Bush voting than a naive model predicts. If you want to claim one reason, you need to show that the other contenders aren't the reason.
posted by ROU_Xenophobe at 9:38 PM on November 19, 2004

If history has taught us anything it is that elections have always been rigged and will always be rigged. Somehow. Someway. Only the techniques change, but the principle remains the same.

It would fly in the face of everything we know about humans as political animals to suppose that there was not some ex-facto massaging of the vote in the last US election. The only real question is where, when, by how much, and by whom?
posted by meehawl at 1:37 PM on November 20, 2004

I participated in the New Hampshire recount on Thursday, November 18th. From what I could see, the recount validated the machine counted totals to a very high degree. I have posted photos and .mov files from the recounting process, if anyone is interested.
posted by notmtwain at 4:33 PM on November 21, 2004

notmtwain - NH didn't produce the same sort of avalanche of electoral complaints as did Ohio or Florida.
posted by troutfishing at 9:43 PM on November 21, 2004

« Older The Scrawls of War   |   just in time for thanksgiving! Newer »

This thread has been archived and is closed to new comments