Despite the hysterical claims of ballot box fraud, it has become clear -- at least to those with open minds -- that the New Hampshire primary was well-run and produced a highly accurate result. There is no evidence of hacking of the Diebold scanners; there are no issues with chain of custody; there is no significant shift in votes as a result of the recount. There have been errors found, however, and they were all caused by the same group of culprits: humans.
Let's go over the results so far, at the New Hampshire Secretary of State's recount page. About 80000 votes have been recounted, or just under 30% of the total. Hillary Clinton's totals have increased by 0.136%; Barack Obama's by 0.361%; John Edwards', by .0185% (if you're following along at home, make sure you subtract Weare's totals, as that part of the recount is not finished). There is now no chance that the results will vary significantly from those originally reported.
That's not to say that everything went perfectly; no election ever does, despite our best efforts. Here's a rundown of the errors that have been found, how they were handled, and their materiality.
About five hundred votes in Stratham (a machine-count town) didn't register, because the wrong pens were given to voters. Election officials immediately caught the problem, and instituted a hand count of the ballots, prior to reporting results to the State. Materiality: none.
Then there was Manchester Ward 5:
The one major exception has been Ward 5 in Manchester, where votes for all the major candidates dropped significantly after the recount. Clinton's total went from 683 to 619, Obama's went from 404 to 365, and other candidates saw similar drops.
But state officials said the problem wasn't with the machine that did the counting. The disparity appeared when the ward clerk filling out the official tally sheet accidentally added vice presidential votes to the presidential votes.
Since the votes were closely proportional to the correct ones, the materiality is miniscule.
A number of vote totals changed by small amounts. In most cases the difference was single digits. The best way to explain this is the programmer's old hoary cliche: GIGO, or Garbage In, Garbage Out. The machine will sometimes read an extraneous mark by a voter as a vote; it will also throw out votes when the voter doesn't follow instructions, and circles the candidate's name instead of filling the little box. Those who think hand counts would produce better results might want to consider this story:
Bergeron isn't convinced. He points to his experience running for state representative two decades ago, when he lived in Hudson and ballots were all counted by hand.
He lost the race until another candidate requested a recount; when it was over, results had changed for six people – and he had won.
...
So the actual errors found so far are all human error, not machine. Now let's look at two other issues raised: chain of custody and the so-called "Hursti hack."
Since the recount results are so close, we can already say that the New Hampshire machines were not hacked. That possibility was already unlikely, because all ballots in New Hampshire are completed by hand, so they can be compared with machine totals, unlike with touch-screen voting. In addition, there are pre-election machine audits performed:
Nashua City Clerk Paul Bergeron said this test is why he thinks Hursti's hack, in which a change in the software on the memory card hid some votes, wouldn't work in reality. The pre-test, he said, would spot it.
A week before the primary, Bergeron's office ran dozens of marked ballots through each of the 10 AccuVote machines owned by the city (one for each of the nine wards, and a backup).
Bergeron said he filled in the oval next to the name of each of the 51 candidates for president and vice president in the two parties, to ensure the cards were programmed to spot every candidate. When it was over, he hand-counted them to see if the AccuVote result agreed.
"In 13 years, I think there have been two times (when the pre-test showed problems)," he said.
Much has been made of chain of custody of the memory cards used by the Diebold Accuvote-OS tabulators used in the primary. It's all smoke and mirrors. I have spoken to local election officials and to the Secretary of State's office. The memory cards remain in each ward until either the time to request a recount has passed, or any requested recount is completed. They are not missing, and everyone I spoke to finds the whole question funny. The memory card can tell you only two things: the same totals it has already given you, and evidence of defective software. Pre-audits already check for software problems. In addition, we still have the paper ballots, and the machine manufacturers and servicers are paid by the state for their work. One election where a recount shows a hack or defective software, and their gravy train is gone. Why, with the paper ballots to compare to, would they chance it? The ballots themselves remain in their local wards until a recount is requested or no recount is requested; at the end of either circumstance, they are archived centrally. While kept locally, procedures state that they should be locked up as confidential records. They are kept under lock and key during the recount.
New Hampshire elections produce a verifiable paper trail that has now been shown, within the margin of error produced by the humans voting in and running the election, to be as accurate as the machines used locally to tabulate the results. I know there are many here who simply want to believe otherwise, but we just ran a very clean primary election here in New Hampshire. If you're looking for a way to make it better, I can see only one: eliminate all humans. Since, alas, we can't have elections without them, New Hampshire's primary is about as clean as we can get it.
If you actually want to know how Hillary Clinton won, you'll find a good analysis here.
UPDATE: Elwood's excellent analysis here on Blue Hampshire.