Wednesday, November 05, 2008

Election Day Recap

I spent a long, busy day running the polling station for Forks Western District #1 at the Forks Community Center. It was my first presidential election as a poll worker and an interesting day. I learned a lot and continue to be dismayed about how fragile the voting process is. It's easy for individual votes to be irrevocably lost despite the best efforts of the election board. The lines and registration process make it hard for people who want to vote to actually vote, regardless of whether there votes are recorded.

Depsite all the problems, I think we had a good day. Voter turnout was about 65%, more than 1,200 people voted either in person or by absentee ballot. We had long lines, ranging from 45 minutes first thing in the morning to an hour and fifteen minutes at midday. We had no serious problems, the vast majority of people were patient and friendly.

My location was two different polling stations in the same building. This causes lots of confusion for voters, since there are two different lines and they need to be in the right line to vote. The line for our precinct was much shorter than the other line. (Some people waited almost two hours in the other line.) The two election boards worked out a deal where someone who waited in the wrong line would be taken to the front of the other line. This minimized the actual harm and kept people happy, but meant I spent a lot of the day walking back and forth between the two lines.

We had two poll watchers / observes who were form the Obama campaign. They never identified themselves as Obama supporters, but I knew one of them from the neighborhood. The other came from New York. They were incredibly helpful and were essential to keeping the lines moving smoothly. They spent the day answering questions and helping people find the right line. The staff at the community center also helped out a lot.

My biggest concern was the lost votes. The tally of voters who signed in didn't match the tally of votes recorded on the electronic voting machines. Our precinct lost 26 votes out of 1186 cast, a 2.2% loss rate. The other polling station lost 19 out of 1746 votes, a 1.1% loss rate. I didn't think to check the counters on the machine until late in the day, so I don't know what went wrong early. After we started paying extra attention, we cut the error rate to less than one percent. It was probably a combination of human factors and usability issues that caused the problems, but I can't rule out some early glitch with the machine itself. I've got more data about lost votes at the end of the post.

A Princeton report concluded that the voting machines we use were too insecure to be used in New Jersey. My experience is consistent with their complaint that: "Design flaws in the user interface of the AVC Advantage disenfranchise voters."

We had seven provisional ballots cast. In two or three of those cases, I think the people simply weren't registered correctly and the ballot with not be counted. A few others will probably work out fine. I wish there were a way for me to check.

Forks Township might need to make big changes.
  • The boundaries among the four precincts are very hard to understand. We seem to have cases where neighbors vote in different precincts without being on obvious bounaries.
  • The precincts are vastly different in size. West #2 has at least 50% more people than West #1.
  • The precincts are too big. Several smaller precincts would mean shorter lines for voting.
  • We have to produce a single numbered list of voters, which seems like a central bottleneck, and we have two books that contain the entries for voters to sign. These physical lists need to be sharded more aggressively if we the precincts stay the way they are.
  • I'm not sure whether it's a good idea to have two precincts vote in one location, because people get confused about which precinct to vote in. I'd estimate we had 100 people wait in our line when they were registered in the other precinct.
My choice would be to have twice as many polling stations. There are about 10,000 people registered in Forks Township. It seems reasonable to have five to ten polling stations for them instead of just four

Here is the raw data on missing votes. I didn't think to check the machine counts until late in the day, by then 23 votes were missing. One possibility is that the machine suffered some software error early in the day that caused it to drop or not record some votes. I think that's unlikely, but I can't rule it out. I'm surprised that the operators missed so many failures. Once we started checking the counters regularly, the error rate dropped substantially (3% to less than 1%). That probably means that increased vigilance caught problems while they could still be corrected. We were also much less busy late in the day.

I recorded the total number of voters signed in and the counters on the two voting machines several times starting around 4pm. The number of missing votes goes up and down slightly, because I didn't keep careful track of people in the voting booth or waiting to get in a booth when I checked the counters. If a person was in the voting booth, they'd counted as missing according to my procedure.

4:15pm, 860 voters, 428 + 409 = 837, 23 missing
4:40pm, 903 voters, 448 + 431 = 879, 24 missing
5:08pm, 943 voters, 448 + 470 = 918, 25 missing
5:45pm, 1,012 voters, 481 + 505 = 986, 26 missing
6:05pm, 1,037 voters, 494 + 515 = 1,009, 26 missing
6:45pm, 1,097 voters, 523 + 546 = 1,069, 28 missing
7:22pm, 1,146 voters, 548 + 570 = 1,118, 28 missing
7:32pm, 1,157 voters, 553 + 575 = 1,128, 29 missing
8:30pm, 1,186 voters, 569 + 591 = 1,160, 26 missing

In the final four hours of voting, we lost three more votes, an error rate of less than 1%.

There are several possible sources of error. The least likely case is a voter who signed the book, but never actually entered the voting booth. I think this case is the least likely. Why would someone wait in line for almost an hour and then leave without trying to vote?

It's possible that the voting machine did not properly record a vote. We use Sequoia AVC Advantage voting machines. They update a counter after each recorded vote. The counter is displayed on the operator panel. The machine operator could check that the counter goes up after each vote. Perhaps the machine logs more information that might catch cases where the counter fails to behave correctly.

The most likely cause of problems is a combination of usability issues and operator errors. A user unfamiliar with the machine might leave the voting booth without pressing the right combination of buttons needed to record a vote. If the operator doesn't notice, then the intended vote is lost. We had one operator for two machines used by almost 1,200 people. It would be easy to miss a few cases where people left thinking they had voted but without actually recording a vote. Here's the process in a nusthell, along with a bunch of failure modes.
  1. The operator activates the machine for a specific voter. The machine makes a chirping noise.
  2. The voter touchs the square to the right of the candidate's name. A lighted arrow appears next to the square.
  3. After at least one candidate is selected, a red vote button in the lower right of the voting booth is enabled. It lights up when it is enabled.
  4. The voter presses the red vote button to record his or her vote. The machines makes another chirping noise.
  5. The voting machine returns to inactive mode.
There are many failure cases:
  • If the voting machine isn't activated, the touch screen still responds by turning on the lighted arrow for a second or two. The red vote button is never activated. A voter could believe that they are voting because the touch screen provides some feedback. Since the vote button never lights up, they'd never see it. In this failure mode, the voter presses a bunch of names and thinks a vote has been cast. The machine stays in the inactive state through the whole process. Since the machine ends up back in the inactive state after a successful vote, the operator can't distinguish between a succesful vote and this failure mode after the voter leaves the booth. The chirping noises should help catch this failure, but voters probably don't know to listen for it. In a busy polling station with several machines, it's easy to get confused about whether you heard something or didn't.
  • A voter can entered an activated machine and fail to touch the right parts of the screen. If you touch the candidate's name instead of the box to the right of the name, nothing happens. Again, the red vote button doesn't light up. So a voter could touch each of the candidates names and then walk out. They don't know they're supposed to see lighted green arrows to the right of the name, so they don't realize they've done something wrong. The operator could catch this case, because the machine stays in active mode after the voter leaves. But if the operator is busy and doesn't check the machine after a voter leaves, the next voter could enter the booth and vote before anyone notices.
  • The voter could press some of the squares and get some of the lights to turn on, but not press the red vote button. If they walk away before anyone notices, their vote can't be cast. We noticed this happening a couple of times. If another voter entered before we noticed, they might be surprised that some of the lights are on. Then again, the new voter might overlook it entirely. The lighted arrow is a small indicator, so they might not notice it. Or they might think it normal for some arrows to be lighted before a vote is cast.
I think the key usability problem with the Sequoia machine is that it's not clear enough what has to be done to record a vote. The old, mechanical lever machines had a strong feedback system. You had to pull the lever to close the curtain, and pull it again to open it. It was actually hard to get out of the voting booth without pulling the lever. You'd get tangled in the curtain. There's no multi-level system like that here. It's physically easy to leave without recording your vote. The cues are some small lights and a high pitched chirp. These are cues easy for an elderly voter to miss entirely.

2 comments:

Unknown said...

I've thought for a while now that the best "you have voted" feedback would be for the machine to print out an actual paper ballot, which the voter would then VERIFY, stick in a privacy envelope, and carry over to the ballot box. The electronic record on the voting machine would be submitted at the end of the day, but if a recount were asked for, the paper ballots would be used for the recount.

Advantages: the ballots are all printed by the machine, so there's no chance of "hanging chads", incompletely filled-in circles, or illegible write-ins. Any recounts will find unambiguous votes. And there's an actual paper record so that recounts can happen.

With this system, the only way of messing with votes is: 1) chicanery with the ballot boxes, which is a well-known method of attack with well establish defenses; or 2) steal someone else's identity (or make up your own) and register twice or twenty times so you can vote twice or twenty times. The latter would best be defeated by requiring ID, which is a whole 'nother rant. (There's some problems with requiring ID to vote, but none that I've found insoluble -- for instance, you'd need to ensure that ID's are available for free so you don't create an unconstitutional poll tax).

Basically, we really, really need to get back to leaving a paper trail behind our voting machines.

Jeremy Hylton said...

The problem is not the number of people. I was thrilled to see so many voters. The problem is the long lines. If everyone has to wait more than an hour to vote, some people won't be able to vote. People who have to get to work may not have the time. I heard several reports of elderly voters who left because they didn't think they could stand that long.