The state police will carry the ballot box during the Wyndham recount in 2021. GEOFF FORESTER / monitoring file

Under the gaze of many people-including video files-cyber security expert Harri Hursti (right) and Windham police chief Michael Caron (center) register for the forensic election audit in the town's electronic voting machine on May 11, 2021 Started on Tuesday at Edward Cross Training Center in Pembroke. GEOFF FORESTER

The machines that count paper ballots in each election in New Hampshire are too old. They run the obsolete Windows XP 20 years ago, which is why the state seeks to replace them. This is why some people doubt its accuracy.

Nowadays, there is a lot of heated debate about voting accuracy, but at Granite Geek, we prefer numbers to noise, so we seek objective information about how many errors these AccuVote machines actually make.

I am happy to say that there is such information thanks to natural experiments that have been running since the first AccuVote machine was inserted in the late 1990s.

It comes from a manual recount by electors in a sweltering office in Concord. They collect paper ballots run by AccuVote machines and check the results one by one.

Candidates can request a recount in a close election (there is usually less than 1 percentage point between winners and losers). Six or more such recounts are seen every year, including party primary elections and general elections.

All recounts are listed in the biennial general court manual, which is the red book. In order to test the accuracy of AccuVote, all I have to do is to compare the before and after results and see how many machine errors were found through manual recounting.

I looked at 37 recounts of state representative races or primaries in 2020, 2018, and 2016, involving 150 seats. (Why so many seats? The three recounts involve Derry, which has 10 seats on its own. Most recounts involve regions with three or four seats.)

I didn't look at the other 17 recounts that occurred during this period because they involved at least one town that was manually counting votes on election night instead of using AccuVote machines.

According to my calculations, a total of 476,203 votes were cast in these contests; that is, 476,203 times someone filled in the small circles next to the candidate's name 476,203 times, and the AccuVote optical reader should record it.

How many of these ballots changed during the recount? I calculated 2,849 changes, which means that 0.6% of the circle-slightly more than one-half of 1%-is either up or down. This is a 99.4% success rate.

Hardly any changes had any effect: out of the 150 seats I checked, only 5 recounts changed the winner.

By the way, almost half of the vote changes came from the 2020 Wyndham contest, which led to a well-known audit. The audit found that the way hundreds of ballots were folded changed the way the machine counted them. (The results of the Wyndham election have not changed due to the recount in the region.)

Without that election night messed up and its 1,368 changes, in 36 multi-candidate elections, your vote change rate would be 0.33% or one-third of 1%.

All in all, I want to say that the AccuVote machine passed with excellent results. If anyone doubts the accuracy of our ballot table, please tell them to look for other things to worry about. There must be enough possibilities.

By the way, please note that there are two complications in calculating the percentage.

First of all, I did not count the votes, but the total votes of all candidates. Since most state representative seats have multiple members, and voters are required to vote for more than one candidate, my count of any recount is greater than the number of votes cast, sometimes much larger. In your opinion, this may be considered an underestimation of the error rate.

Second, I double-calculated some errors, which seems to exaggerate the error rate.

Suppose the machine casts a vote for candidate A by mistake, and then gives it to candidate B. When the votes are recounted, it will be shown in the red book as A one less vote and B one more vote. I have no way of knowing that both of these are from a single vote error, so I count them as two separate errors.

In the Windham audit, some ballots completely missed multiple ballots because they were in the folds of paper ballots. This issue affected approximately 400 paper ballots, but these multiple vote conversions meant that it had a total of 1,368 changes in the count of different candidates, and the apparent error rate had more than tripled.

The question is how these two imprecise methods interact. Do they cause an exaggerated error rate, or do they downplay the error rate? I don’t know. Either way, the interest rate is still insignificant.

You can get a more accurate count by looking at each individual town ballot that is recounted each time, but this is impractical-it is not for me anyway. Maybe some multi-science graduates who are looking for a thesis topic want to give it a try.

Despite supply issues, rising costs, and pandemic concerns, the roadside home, which is celebrating its 50th anniversary, has had its best year...

On Saturday, Mason held a Christmas celebration in the town pavilion in the center of town. The event includes suppliers, "Polar Express" trains,...

The US$500,000 federal grant will eventually allow the opening of a safe house for victims of human trafficking. This is the first transitional...

The machines that count our paper ballots in each election in New Hampshire are too old. They run the obsolete Windows XP 20 years ago. This is...