There has been a bit of a tizzy over a perceived inconsistency in the ballot counts between hand-counted and machine-counted votes in the New Hampshire primary. And today, Kucinich asked for a recount because of the "unexplained disparities between hand-counted ballots and machine-counted ballots."
I am constantly amazed at how people completely fail to understand statistics and just look at the bottom line. This web site does a nice job of presenting all the data, and breaks it down by type of ballot and small towns vs. large cities.
Overall, there were 228,307 machine ballots cast, and 59,157 hand ballots cast. Clinton: 39%, Obama: 36.4%. The supposed fraud is because if you look at the results of just hand-counted and just machine-counted votes, the results are different:
|New Hampshire - Hand vs. Machine|
So if you just count the hand votes, Obama wins!! There must be a scandal. Well, sorry to burst your bubble o' conspiracy, but it ain't so. Is it even remotely possible that there are demographic differences between the people who cast machine versus hand-counted votes? Well, that would hardly be surprising, if anyone bothered to look at where these votes are coming from.
Small towns are far more likely to use hand-counted ballots than cities. In small towns, 86.7% of the votes were counted by hand, whereas in cities a mere 2% were counted by hand. In fact, only 7% of all the hand-counted votes cast came from cities, and 63% of them came from small towns. What this means, is that the hand-counted votes are almost entirely representative of people from small and medium towns, whereas the machine-counted votes are almost entirely representative of people from big cities.
That is a major demographic difference. It is well established that rural areas vote more conservatively than urban areas. For example, see this analysis of the 2004 presidential election from the Carsey Institute which clearly shows urban areas voting for Kerry, and rural areas voting for Bush.
Each precinct uses only one method of vote counting. If you look at the town-by-town results, you will notice dramatic variation in the outcome from town to town. It's amazing how differently they each vote. But each county uses only one method of vote counting - either hand-counted or machine-counted. So, when you're comparing machine versus hand counts, you are not looking at a cross-section of all voters. You are looking at the results of a certain group of towns compared to the results of a different group of towns. This is very different than if all towns used both hand and machine counting, so there was an equal distribution of hand versus machine counting across the whole state. But this is not the case.
The machine-counted results favor Clinton in small towns, but Obama in large towns or cities. The vast majority of machine-counted votes (nearly 80%) came from larger towns or cities. Yet when only looking at large city votes, Obama did much better for machine counted versus hand counted votes - 36% verus 31% voted for him. Kind of an odd conspiracy - to boost the loser's numbers? And if the conspiracists think it was the hand-counted votes that were tinkered with, then that would be a HELL of a conspiracy - since the vast majority of them came from precincts with less than 750 votes, and Clinton actually lost to Obama when only looking at those small-town hand-counted votes! And finally, if you think it was the big-town hand-counted votes that were tinkered with, wrong again. Clinton kicked Obama's ass among those votes - unfortunately, there were only 4,281 votes in that category. Even if Obama had gotten every last one of them, he still would have lost.
Ironically, Kucinich, our man for the recount, is basically dead even for hand vs. machine statewide. Of course he only got about 4,000 votes, so who cares.
The bottom line: the disparity between hand-counted and machine-counted votes in New Hampshire is not a result of fraud - and in fact is entirely to be expected. There is absolutely no reason we would expect the outcome to be the same between them, because they do not represent a cross-section of voters.
Update 1/14/2008:I got the data for this post from this web site. Since last Friday, the summary data has been changed. He no longer breaks out small (<750 votes), medium (750<1500) and large (1500+). Now, medium and large are combined, and the cutoff for small has been changed to 700. I am not sure why he did this, but this reworking of his analysis means you can no longer see what's important -- that only a tiny fraction of the hand-counted votes came from the largest precincts, though overall 78% of all votes came from the largest precincts. The base summary data that I used, which used to be published there, is as follows:
|New Hampshire - Hand vs. Machine by Precint Size|
|Machine||Machine %||Hand||Hand %|
|Small (<750 votes)||5,799||13.5%||37,309||86.5%|
|Medium (750<1500 votes)||42,363||70.1%||59,930||29.3%|
|Large (1500+ votes)||180,145||97.7%||4,281||2.3%|
It is plain that the vast majority of hand-counted votes came from precincts under 1,500 votes, whereas overall the vast majority of votes overall came from 1,500+ precincts. There is a huge demographic difference. The way the medium and large precincts were combined on the CheckTheVotes web site now doesn't emphasize this any more, since they broke it down basically where the number of hand counted votes would be the same in the two sections (small vs. medium + large).