… and there is strong evidence of shenanigans on the part of Coleman supporters (or someone).

As I mentioned earlier, the idea is afoot that there will be more of a shift towards Al Franken in geographical regions that favor Franken than in Coleman-sympathetic regions, in the current US Senate recount for Minnesota.

I disagree, and in fact, I’m going to suggest the opposite.

There seem to be two reasons proposed as a basis for this assertion. One is this: The Democratic Party is the big tent party, with lots of immigrants from countries without bubbles, new voters, and people who otherwise might have a hard time with the voting process. Therefore the machines are going to chuck out more Democratic votes (=Franken votes) than Coleman votes, which will then be picked up again and added to Franken’s total during the recount.

Maybe, but this effect is going to be weak. Why? Because Minnesotans tend to vote more than other people, and while the turnout was higher this year, a big chunk of those new voters (about 20 thousand by my own thumb suck estimate) voted for Obama … that is what got them to the polls … but did not vote at all in the Senate race. Another couple hundred thousand or so voted for Obama but did not vote for Franken (they probably voted for the third party candidate who shall not be named here).

So I suspect that this effect is not strong to begin with, but yes, it may be greater in Democratic leaning areas. Though is isn’t’, and I’ll show you that in a moment.

The second reason, as far as I can infer, is that there is some sort of presumed mojo underlying the process, whereby votes that are about to change during the recount are more likely to fall from obscurity (such as over count votes not attributed by the machine to any candidate) disproportionately towards the candidate that is most liked in a particular precinct. Why would over votes (or any other ballot) do this? No reason that I know of, so I reject this vaguely implied hypothesis as woo.

I have a different, competing and opposite hypothesis. I hypothesize that a small fraction more Franken votes will appear in Coleman supporting areas because of a special group of Franken votes that have been hidden from us. For evil, nefarious reasons. Below, I’ll describe my hypothesis and provide a statistical test of it. The hypothesis is a bit complex, the test does not address necessary and sufficient conditions, but rather, is only suggestive, and the results are inconclusive. But interesting nonetheless.

One thing that can be concluded a this time: The idea that a simple and meanigful relationship between levels of Franken support (what I call “Frankenosity”) and more Franken votes pop up in the recount of a given precinct can be put aside. It just ain’t true.

Have a look at this graph:
i-0137779da3a9dac79819ce0a2d2c5674-frankenosity.jpg


The X-axis is “Frankenosity.” This is a number that represents how much a precinct went for Franken. It is Franken votes minus Coleman votes, in turn divided by the number of votes total for both candidates in that precinct, and that in turn is multiplied by 1000 to make it a number you can see. A precinct to the left of the zero line went for Coleman, a precinct to the right of the zero line went for Franken.

The Y-axis is the number of votes different by precinct towards Coleman (down on the graph) or Franken (up on the graph).

If there was a strong effect, you would see something other than a big blog of dots.

There are three things apparent on this graph. The first is that there is a handful of somewhat extreme cases in the upper right quadrant … these are cases where a bunch of votes went for Franken in a Franken-leaning precinct. The second thing is that even though there is a handful of such precincts, the data look pretty darn random. In fact, the data are random. There is no predictability here. Using Frankenosity to predict shift to Franken gets you nothing at all statistically. So even though there are a few dots floating way out there in the upper right, they probably mean nothing.

A third, somewhat subtle effect is also seen that actually explains the first effect as an artifact. The scatter in precincts with higher Frankenosity (above zero, really) is higher than in low Frankenosity precincts. This might be the Democrats As Sloppy Voter effect. But it probably isn’t. This is almost certainly an effect of something entirely different: Frankenosity ignores precinct size, but shift to Franken is in absolute numbers. What you are seeing here is the effect of small vs. large precincts. In other words, you are seeing an artifact of the fact that people clustered into larger precincts, in and near cities, tend to be Democrats, and thus more likely to vote for Franken.

The conclusion you can draw here is that there is not any really strong Democrats as Bad Voters effect that varies by precinct. The effect is probably there, but it happens in all the precincts to about the same degree, which is not surprising since this was a close race.

The Republican Effect Hypothesis

Now, on to the hypothesis I mentioned above, which I’ll call the Republican Effect Hypothesis. Let me first describe it as a real life scenario.

Every election judge is potentially biased, in that almost everyone in the state has an opinion about who they would like to win a given race. The judges, of course, are sworn and presumably trained to act in a non biased manner. However, there is abundant evidence that Republican (especially Coleman-supporting) judges have a slightly higher chance of acting in a biased manner, and, effectively, cheating. (I’m not going to supply the evidence here, but it is overwhelming.)

Given this, you can imagine this scenario:

A voter fills out a ballot. The ballot goes into the counting machine. The voter begins to walk away.

If there is an over vote on the ballot …. a stray mark has messed up a vote … then the machine spits the ballot out. At that point, the election judge can stop the voter and tell them that their ballot has at least one race which may now be invalid, and offers the voter an opportunity to re-vote.

Or, the election judge lets the voter walk away without stopping them.

My hypothesis is that in small communities (or even larger ones) there will be a small number of instances in which a Republican election judge knows the voter, and knows the voter’s likely preference is for Franken, sees an over vote happen, and lets the voter walk away. Of these, a small number are cases where a Franken vote is thusly missed. And this is more likely to happen in Coleman precincts.

Of the nearly three million votes case in this race, about a million plus were cast for Franken among several hundred precincts. The total number of votes being recovered from the “other” bin seems to be in the range of about 500 to 1000, and of these, there is a bias towards Franken votes that were lost to the over count phenomenon (or in some other way).

So let’s look at this again, statistically.

I have divided all the data from the first three days of counting into four categories (and you can refer to the graph above to understand this … each category is a quadrant in that graph) given here with the number of precincts in each category:

1) Positive Frankenosity, and a shift towards Franken after recount (161).
2) Positive Frankenosity, and a shift towards Coleman after the recount (177).
3) Negative Frankenosity, and a shift towards Coleman after the recount (160).
4) Negative Frankenosity and a shift towards Franken after the recount (234).

Wow. All those numbers are the same except one! That one number, 234, that number of net shifts towards Franken occurring in Coleman-supporting counties (just a vote or two here, a vote or two there, as would be expected if my hypothesis is correct) stands out a bit, doesn’t it?

A two by two chi-square test shows that p=0.0553. Typical freaken’ p-value, just out of the range of “significant.”

That’s my only statistical test. I’m sure if I played around with the data and the tests I could come up with a statistical result that was significant, but I’m not going to do that. Play the ball where it lies. At the moment, there is a strong reason to believe that Franken votes were hidden by an over vote effect, or something similar, differentially in Coleman-leaning precincts.

Is this enough for a law suit? Not yet. And I don’t know what kind of suit it would be. But something seems to be going on here and it …. well, it smells like … Norm.

Ick.

Comments

  1. #1 jake
    November 22, 2008

    Thank you Greg, for all that you’re doing with this.

  2. #2 Stephanie Z
    November 22, 2008

    I’m not sure one has to go as far as judges disenfranchising people who are likely to disagree with them to explain this result. You could simply be seeing judges who are comfortable disenfranchising people who “don’t count.” Slightly more likely to happen where views on who counts and who doesn’t aren’t challenged. Same effect but easier for the judges to rationalize their actions as not politically biased.

    I wonder whether the pattern is repeated in other recounts.

  3. #3 Notagod
    November 22, 2008

    So why isn’t the over count voter required to either acknowledge that they were told about the problem or required to re-vote? That may also have inherent problems but, in any case, the election judge should not be in a position of making the choice.

    The United States is infested with christians and everyone knows an elevated number of christians can’t be trusted.

  4. #4 dave
    November 22, 2008

    The second reason, as far as I can infer, is that there is some sort of presumed mojo underlying the process, whereby votes that are about to change during the recount are more likely to fall from obscurity (such as over count votes not attributed by the machine to any candidate) disproportionately towards the candidate that is most liked in a particular precinct. Why would over votes (or any other ballot) do this? No reason that I know of, so I reject this vaguely implied hypothesis as woo.

    Depending on what exactly is meant by “disproportionately”, there may be real reasons to expect this.
    Suppose Alice gets 2/3 of the votes and Bob gets 1/3 of them, and the chance of the machine misreading the ballot is essentially random. Then of the valid-but-misread votes, you would expect to see twice as many for Alice as for Bob, unless something else was skewing the numbers.
    So it would make sense to see numbers go up more for a candidate with more support, by approximately the same proportion as the total vote preferred them. (This is the same skew that makes “terrorist recognition” mechanisms with a nonzero false-positive rate useless.)
    But unless you’re seeing really big numbers of valid-but-misread ballots, I would be surprised if this actually produced enough signal to make it through the small-sample noise.

  5. #5 Greg Laden
    November 22, 2008

    Dave: Right, … the vote was very very close, and the variation around the kind of changes we might expect because of the changes simply representing the status quo are very large. So it’s like the effect of O2 molecules made of 18-O vs 16-O. You should find the heavier Oxygen molecules in the room you are in down by your feet and the lighter ones up by your head, but you don’t.

  6. #6 Andrew
    November 22, 2008

    There must be a hundred way in which a vote can be tilted a tiny amount. Instead of a recount for a close vote, perhaps there should be a new campaign and election. Total do-over.

    (…ducking….)

  7. #7 Elizabeth
    November 22, 2008

    It will be very interesting to see what the Saturday vote recount shows.

  8. #8 Greg Laden
    November 22, 2008

    Stephanie: I wonder whether the pattern is repeated in other recounts.

    I imagine that could be predicted to work by a number of demographic factors such as mix of religion, age, race, etc. One might even see this sort of passive aggressive behavior occur in some parts of the country more than others. This is why we need more diversity among the election judges.

  9. #9 Pierce R. Butler
    November 22, 2008

    … there is abundant evidence that Republican (especially Coleman-supporting) judges have a slightly higher chance of acting in a biased manner, and, effectively, cheating. (I’m not going to supply the evidence here, but it is overwhelming.)

    As a Floridian, I’d be very interested in (a link to) this purported evidence – or is the Sunshine State itself the evidence?

  10. #10 Monado in Toronto
    November 22, 2008

    In our _municipal_ elections, which have optical card readers counting the vote, there are observers from all major parties at each polling station–both at the tables where people pick up their votes and at the machine that counts them. Everyone stands around while the vote is fed in and the voter is asked to wait. If the machine spits it out, the voter gets another ballot and their first vote is discarded. It works very well and I’d love to see the U.S. get something like it.

    In federal elections we still do hand-counting with observers from all the parties examining each vote.

Current ye@r *