It's been a couple of days since I posted on the New Hampshire recount. At the time, I fully expected that I wouldn't do another post on the topic, but a couple of things that have happened since then changed my mind. First, Scibling Chris Chatham included me in a list of people who he thinks should get off their "soapboxes", stop "hurting America", and focus on the statistical anomaly he's identified. Second, and far more importantly, preliminary recount results are in from a number of precincts.
First, let's look at this "Diebold Effect" thing again. When I took my first look at the results, I decided that a detailed statistical analysis would not be appropriate, and I stand by that. My decision to avoid the potential pitfalls of an inappropriate statistical analysis is not merely because I "assume" that demographic factors don't always explain all of an election result, but because I think that there's a factor that should explain at least some of the results, but which isn't included in analyses that focus on the demographics: campaign effort. We do not have data about where the campaigns chose to focus their efforts during the period between Obama's win in Iowa and the New Hampshire primary.
In fairness to Chris, he did attempt to take this into account after I pointed this out the first time. He found a list of Clinton campaign offices, included their presence as a variable in his analysis, and found that the "Diebold Effect" was still significant. Unfortunately, I don't think that the variable he used as a proxy for campaign effort was remotely adequate. The presence of a campaign office might reflect decisions about where to focus effort that were made early in the campaign, but it's unlikely to reflect last-minute decisions about where to focus effort. (For example, did the campaign shift volunteers from office to office?)
Chris also chides me for not addressing a peer-reviewed paper that he cited which used statistical methods to examine similar accusations that were made in 2004. I've looked at the paper, but I don't think it's an apt comparison. First of all, although Chris claimed that the paper looked at the 2004 presidential primary in New Hampshire, it did not. It actually looked at the general election results there. There are major differences between the primary and the general elections in New Hampshire. The state, as we all know, holds its primary very early in the election season, and the results of the primary have an enormous amount of influence on how the media perceives the campaign for the rest of the primaries. In the general election, on the other hand, New Hampshire has a total of 4 electoral votes, and is not normally the focus of extensive campaign efforts. In addition, the paper in question did not focus on the demographic factors that Chris was looking at. Instead, it primarily looked at prior election data - something that is much easier to do in two-party general elections than it is in party primaries.
That having been said, I am happy that there is a partial recount going on right now. Like Mark, I think it's always a good idea to audit the ballots and make sure that there aren't substantial and systematic problems. Fortunately, that's happening right now in New Hampshire - and some of the precincts in one of the counties that the Kucinich camp was most concerned about have already been recounted.
I took a quick look at some of the results (excel), and was far from surprised by what I saw. In one precinct (Manchester Ward 5), there was a relatively large discrepancy between the machine count and the recounted ballots. In that precinct, the machines messed up and counted vice-presidential write-in choices as presidential votes. As a result, the machines gave more votes to all three of the major candidates than they actually received. Clinton lost 64 votes in the recount there, Obama lost 39, and Edwards 38. (Percentage-wise, Edwards had the largest decline there.) In the rest of the state, the results were entirely consistent with what was found during the 2004 partial recount discussed in the paper cited above: the machines tend to miss some votes, but not very many. As a result, all three major candidates have seen increases in the total number of votes that they received, but the increases are all small, both in terms of the number of votes (excluding Manchester 5, Clinton has received more votes from the recount than the other candidates, with a gain of 52) and in terms of percentage changes (under 1/2 of 1%).
It's entirely possible that statistical analyses will continue to show a significant "Diebold Effect" of some sort, and it might be interesting to try and find the cause. If the balance of the recount matches what was shown, though, I don't think that the "Diebold Effect" is particularly important, and I don't think it's extremely urgent to find the cause - particularly given the difficulties in quantifying things like the last-minute push and campaign strategies.
I am concerned about the habit that these machines seem to have of missing votes all together, but I'm going to hold off judgement on that until I have a better sense as to whether the accuracy of the machine counts is worse than the accuracy in hand-counted areas.