Machine Votes, Hand-Counted Votes, and the Willie Sutton factor.

In the week since the New Hampshire voting, a number of people have become increasingly concerned about some of the things that they've seen in the results. Two things, in particular, have gotten a lot of attention. The first is the difference between the pre-election polling, which had Obama ahead by a considerable margin, and the final result, which was a clear victory for Clinton. The second is a difference in outcome when hand-counted precincts are compared to precincts where the ballots were counted using machines. Obama came out ahead in the hand-count areas, while Clinton came out ahead in the machine-counted regions.

Some people are concerned enough about this that they want to see a recount, and it looks like there will be at least a partial one. Dennis Kucinich came up with a little under half the money needed for a full statewide recount of Democratic ballots, and the state has agreed to count until his money runs out. I expect that they'll find some differences between the machine totals and the hand-count (it would be shocking if there was 100% accuracy), but I'd be surprised if there's a large difference, and even more shocked if the discrepancies between machine and hand tallies systematically favor one candidate.

Clinton did receive a much greater percentage of the vote in machine-counted precincts than she did in the hand-counted areas, but I think there's a single factor that can explain most (but not all) of the difference. Imagine that you are running a statewide campaign. Wining the state is very, very important to you, and with less than a week to go before the election, you are running way behind in the polls. Where do you focus your effort and resources in the time you have left?

When asked why he robbed banks, Willie Sutton famously replied, "because that's where the money is." If you are behind in the polls and want to win an election, you're going to focus your efforts (if you're smart) where the votes are.

I took some of the data that Chris Chatham has available through his blog post on the machine votes question, and looked at who was more likely to win in larger areas. Specifically, I took Mark Shauer's excel spreadsheet and modified it a bit. I deleted all of the candidates except for Clinton and Obama, and noted who had more votes in each precinct. I then took the precincts and categorized them based on the number of votes and whether the precincts was counted by hand or by machine.

I did a rough hand count, and found that Clinton beat Obama more often in precincts where more than 2000 people voted than she did when fewer than 2000 cast ballots. There were 25 precincts that fell into that category, and Hillary received more votes in 18 of those precincts. All 25 of those precincts were machine-counted. She also beat Obama more often than not in precincts where between 1000 and 2000 votes were cast, winning in 44 of 72 precincts. The majority of the ballots in this category were counted by machine, but Clinton won 2 of the 3 that were hand counted. Obama, on the other hand, won more than 2/3 of the precincts where fewer than 250 people voted (all of which were hand-counted).

There are some discrepancies in the precincts where between 250 and 1000 people voted. Here, Obama did win most of the hand-counted precincts, while Clinton won most of the machine-counted ones, but I think most of that is also explainable based on campaign effort. It's easier with these to look county by county. I'm not going to do all of them, but here are a few:

In Belknap County, both Clinton and Obama brought in 37% of the vote, with Clinton coming in very slightly ahead of Obama (4674 votes to 4633). Obama received more votes in the county's single hand-counted precinct (92 to Clinton's 89), and came out ahead in 9 of the 15 machine counted precincts. Most of the precincts were very close. Clinton pulled off the countywide-win there because she received more votes in 4 of Laconia's 6 precincts, and more votes in nearby Belmont. The rest of the county went to Obama.

In Carroll County, Obama came in solidly ahead of Clinton overall, and lead in 9 of 12 hand-counted and 4 of 6 machine-counted precincts. The area that went for Clinton here is also (relatively) clustered, with 4 of the 5 precincts that she took falling in an area centered on Wakefield. (The 5th precinct was Hale's Location, where she received 11 of the 21 votes cast.)

In Cheshire County, Obama again came in with more votes than Clinton. He lead in 13 of 17 hand-counted precincts here, but Clinton beat him in 4 of the 10 machine-counted precincts. Here, again, there's geographic clustering. 6 of the 8 precincts she took are south of Keene (Obama beat Clinton in all 5 of Keene's machine-counted precincts), and 5 of those 6 communities fall along Route 119. The two remaining precincts where Clinton won are small, hand-counting communities. In Marlow, she took 79 votes to Obama's 65; in Sullivan she received 74 to his 71.

I don't have the time to put in right now to do this sort of thing for the rest of the state, but based on what I've seen so far, I don't think Diebold decided that Clinton was the better candidate. I can understand why people are concerned about these results, but I think that when Kucinich's recount is done the results will be basically the same.

There's one final point that I should probably explain: my non-use of statistical methods. A lot of the people who are concerned about the results have supported their concerns with statistical analyses, and I have not. I chose not to attempt statistical analysis because I don't think it's appropriate under these circumstances. Demographic factors such as income level, community size, educational level, and so on, do correlate with people's voting choices to a certain degree, but they don't always explain all of an election result.

In this case, we are particularly interested in how a campaign managed to pull off a come-from-behind victory at the last minute. Unfortunately, we don't have data about where the campaign put its effort in the last days of the campaign. In particular, how much effort went into their campaign in each precinct? How many person-hours were spent in Manchester, and how many in Keene? How much money was spent on last-minute yard signs in Nashua? Those factors are probably important, too, and I certainly wouldn't feel comfortable saying that the "Diebold Effect" is inexplicable without looking at those.

Particularly when the bigger precincts are more likely to hand-count, and when candidates concentrate their efforts Willie Sutton style - where the votes is.

More like this

FWIW I reran the regression using your cut-off value of 2000 votes as a categorical covariate, along with education, income, pop density, age, "contestedness" of the precinct, and total votes as covariates. The diebold effect on Hillary's votes remains significant at p<.001 (F(1,228)=20.5) as does the effect on Obama's votes at p<.001 (F(1,228)=14.08).

"because that's where the money is."

Willie Sutton never actually said that. It was apparently made up by a reporter.

But I'm glad you don't see the need that so many others apparently do, which is to coddle Clinton by using her first name!