Controlled for demographics, the voting anomaly continues to be present

[solarbird]

People at ScienceBlog and a variety of other statistics-literate sources have been doing demographic analysis on the unpredicted/unpolled Clinton/Obama vote swings in machine-counted vs. hand-counted balloting in Vermont. Controlling for a large number of demographic variables - including such outliers as geography - the variation against polling and against hand-counted ballot remains pretty constant. The calculated statistical probability of this being a random effect is p<.001, which is to say, around 1000:1 against. Further analysis is ongoing.

I would like to see similar analysis applied to the Romney surge, which appears to be comparable at the top level.

ETA: I was in a hurry before and forgot to credit cafiorello for the link. Thanks!

Comments

[solarbird]

The results were not within margin of error. They also disagreed with exit polling, as I heard it reported at the time. The results require that essentially all of the large number of undecided voters voted for Senator Clinton, and that Clinton voters were less likely to respond to pollsters than Obama voters. That's possible. It does not help that the same theories were floated by the GOP in Florida in 2000, and those theories, on later study, turned out to be completely false, but it's possible.

any fraudulent Diebold counting conspiracy
It is also possible that something else is going on, either with or without fraud. And election fraud long predates electronic counting, in any system.

A statistical analysis is only as good as the underlying data
The underlying data is public; the demographic information is public. A variety of people have been going over that data; more people are going over it as I type. We'll see what happens.

speculation about voting fraud is not an honest question, it's groundless conspiracy theory. In the absence of evidence--not half-assed statistical analysis--the grander claims of voting fraud are dangerously close to libel.
Why was this not true, then, in 2000 and 2004? Why was it reasonable to see smaller statistical indications of problems in those elections and want to investigate, but not reasonable to do that here? Why is Florida 2000 and Ohio 2004 sensible public oversight and outrage and this "a swiftboating of Clinton by her rivals"?

Is my desire to see a similar analysis of the strange Romney numbers also "a swiftboating of Romney by his rivals?"

For the record, I plan to have very little if anything to do with either party's nominations process. The candidate to whom I donated and who I planned to support in our caucuses pulled 1% in Iowa and dropped out of the race. The other candidate I have considered supporting continues to poll 1% and is barely acknowledged as running at all. I will consider this again once our caucus date approaches, but right now, my plan is to stay home and see who the parties nominate. Arguably, as I'm not a Democrat or a Republican, it's really none of my business anyway. But if something looks funny, I'm still likely to look at it and go, "Hey - that looks funny. What the hell?"

[llachglin.livejournal.com]

Obama got all of the votes the polls said he would get (not counting the party internals or exit polls, which both use demographic weighting factors that are prone to mistakes--in this case they assumed turnout overall and for independents on par with Iowa, which did not happen.) Clinton got what she polled, plus the undecideds, plus or minus the margin of error. The results were consistent with a late surge by Clinton that was too late to show up in the polls, combined with a lack of turnout by the demographics that favored Obama in Iowa. In other words, they're not the least bit suspicious until you start counting Diebold counted vs. hand-counted precincts, and that could easily be explained by something else.

You point out that election fraud existed before electronic voting. That's true, but the alleged fraud in this case is electronic. No one is suggesting ballot box stuffing, the old way of fixing elections, because if that happened you wouldn't see a correlation with Diebold counted precincts. All precincts use the same optical scanning paper forms so it's no easier to stuff Diebold-counted ballots than other ballots.

Serious statisticians seem to be saying that you can't draw any conclusions about fraud from a discrepancy like this because you can't account for all the other factors that might produce the discrepancy. On the other hand, if there was counting fraud, any audit of suspect counties would quickly reveal the fraud. That's the advantage of having a paper trail.

I have always had problems with the analyses from 2000 and 2004 that use similar statistical techniques to suggest fraud without corroborating evidence (I think I've called them conspiracy theorists). On the other hand, there are differences. In 2000 the butterfly ballot and hanging chads resulted in a lot of missed votes. A hand count showed that the intention of voters was to deliver a majority to Gore, and had the whole state been recounted, Gore would have won. There was also serious evidence of vote suppression (as there always is by Republicans). Finally, the Supreme Court judged the case on political grounds. That's all quite different from a statistical anomaly in a few precincts. 2004 was all about voter suppression in Ohio, electronic voting without a paper trail, and the failure of the GOP Secretary of State in Ohio to do anything about it. Again, none of that is relevant in the case of the New Hampshire primary.

I suspect that this claim originated with the knee-jerk anti-Clinton right, which has a history of smearing the Clintons without evidence. That Andrew Sullivan is one of the people who got this started only confirms that. I basically think Sullivan is full of shit even when I agree with him. The other people pushing it are anti-Clinton liberals and Democrats who have bought into right-wing propaganda, if they're not open misogynists themselves. Again, I'm not a Clinton supporter, but I know the history of her enemies and I have much more respect for her than for them.

I have no problem with your requests that this and Romney's numbers be investigated. The problem I have is with the tone of your posts, which seem to assume that something fraudulent is happening even though there's zero evidence of fraud to date. (And no, statistical discrepancies are not evidence.)