There’s a very nice post by the guy at Developing Intelligence exploring the “Diebold Effect,” or the tendency where Hillary Clinton’s surprise win in New Hampshire seems to come from precincts using Diebold machines. He does not think that any way of explaining away the effect has appeared so far. The discussion is quite careful, and it doesn’t seem as if the author has an axe to grind. Here’s a latter post explaining more details of the analysis, here’s a complaint about people not discussing the issue very seriously. It’s certainly worth watching him if you’re concerned about the issue.
What do I think? I think I don’t really understand statistics. Which is a shame. However, my comments are that my prior probability for vote fraud in the democratic primaries are quite low. This is both because the costs of vote fraud are higher here (since you would have to get democrats to undermine other democrats), and because voting fraud is such a sore subject with democrats (many of whom view the 2000 and 2004 elections as having been stolen). A more nebulous Diebold effect not happening via fraud is more credible to me since the machines are so bad, however it’s entirely unclear what the causal mechanism would be.
Re-Updated: Via Marc Ambinder, there are counter-accusations. I won’t say I know who to believe. On the other hand, it does seem like the caucuses are partially to blame. The system should be open (except for secret-voting) and simple. Caucuses are neither–there are a number of byzantine rules like viability, and since they’re held at one specific time during the day, there’s no way for everyone to participate.