Report from Elsa Crick to VoteAllegheny
Parallel Voting Machine Test
During 11-3-2015 General Election
This test was conducted by Baker Tilly and included their voting on test machines during the same times as the election voting (7 AM through 8 PM) plus pre-voting setup/preparation and post-voting results/conclusion. Our primary VoteAllegheny observer was Geoffrey Grabowski who was to witness the entire long day, except for a brief respite midday when I took his place. These are my observations based on the time period 11:00 AM – 2:45 PM.
When I arrived, no voting was in progress. I spoke briefly with one of the Baker Tilly representatives, to orient myself (trusting that Geoff would have gotten the basic information during the pre-voting phase, and not wanting to burden them unnecessarily). 3 voting machines had been selected “at random”, and a precinct had been chosen “at random”. Parallel machine test voting during the day had been planned to imitate, based on summaries from past elections, voter behavior at that precinct. I asked about how many people were in the precinct and how many ballots were being tested. She told me she “thought there were around” 500 people in the precinct, with a voter turnout of “around 30%” which was being imitated. Despite their qualifications, her statements suggested that around 150 ballots would be voted in the test. During my 2 ¾ hours I observed 25 ballots. They had a break for pizza during the time I was there, so it seems reasonable that during the 21% of the “voting” time I was there, I observed 16-17% of 150 ballots being voted. So, although the Baker Tilly representative had given me what sounded like a very roughly approximate answer, it was plausibly consistent with my observations.
I want to state my limitations as an observer frankly: it was a new type of assignment for me; I had no training; I did not take written notes (but conclude it would have been better to do so); I considered myself just filling in for Geoff (but conclude I should have been more careful in getting and recording general test information despite my expectation that he had taken care of that); and I had some difficulty remaining fully alert: specifically, due to the heat of the room and the monotony of the test procedure, I twice caught myself almost dozing off!
However, apart from a few seconds on those two occasions, I believe I was fairly alert and observant. The machines were being tested at one end of room; I was required to stay on the opposite half of the room, behind an imagined middle line (which was clearly imagined by the auditors – I was once instructed to step back behind it). The three voting machines were open toward me, and while I was there only the leftmost of the three, from my perspective, was being voted. The test voter stood in front of the voting machine, somewhat to the left side: directly in front of the voting machine, a video camera had been set up, and camera and operator were between me and the ballot screen. I moved to the right side of the room (to my right as I stood watching) and could see -- but not read, due to distance -- the actual voting most of the time; however, at the end of each ballot (the vote button and choice to confirm), the camera operator relaxed the bent posture he had assumed over the camera during the voting of contests, when he had to make sure each vote was clearly videotaped, and as he straightened and stepped back, I could not see the vote and confirm being pressed; I tried adjusting my posture, but I couldn’t manage to see it.
The voting procedure was as follows: Between the voting of ballots, a general ballot number was announced by the test voter. (This ballot count was at 52 before the first ballot I saw voted and at 77 after the last ballot I saw voted.) The test voter announced the ballot number and her insertion and removal of each personal electronic ballot cartridge at the beginning of voting a ballot. A prompter stood on the other side of the machine being voted and held in his hand a stapled bunch of papers which told him what was to be voted on each ballot. (I had the impression that the paper ballot plan he held could have well been made easier to read. On one occasion he had to ask the test voter to change a vote he had just prompted, because the guide paper he was holding to keep his reading straight across the page had slipped to the next line and he had prompted the wrong vote. The correction was easily made. And, incidentally, this was the ONLY vote-change I observed in the 25 ballots/2¾ hours.) The prompter read the name of the contest, e.g., “Justice of the Supreme Court”, followed by the name(s) to be voted for that contest. The test voter voted. They proceeded through each contest of the ballot. After all these contests had been voted, the test-voter went to the screen summarizing results, and from there she read aloud through the results for the prompter’s verification: she named the contest and then the names voted for the contest (to which the prompter confirmed “yes”), and this was repeated for contest after contest to the end of the ballot. She announced her pressing “vote” and “confirm”. Then the next general ballot number was announced. All of this was being captured by camera. (At the end of the day, should there be a discrepancy between the expected and reported totals, the videotape could be reviewed to see if the discrepancy resulted from not pressing the intended choice. This had happened, I was told, once, a number of years ago. They had had to go back and read the actual voting from the videotape.)
The voting of ballots had been planned as groups of ballots (like the group of 25 ballots I observed) to be voted by a test-voter-and-prompter team on a specified machine (or machines – before I left, they had said that the next group of ballots would include 7 more on the machine I had seen used and the rest on another machine) during a specified clock time period) … the timing assignments specified in the test were reasonable in allowing some flexibility in the exact timing of the test voting within an assignment block, and yet they would serve to control the general spacing of numbers of ballots throughout the day according to whatever plan they had devised.
I did not observe, among the 25 ballots, ANY choice of a one-party option, ANY write-in, or ANY omitted contest. (In one contest in one ballot, I observed the casting of 2 out of a permitted 3 votes; otherwise, as many votes as allowed were always voted.) I wondered whether these were typical of the precinct; however, I was only seeing about 1/6 of the test ballots being voted. (Ron Bandes, hearing of this from me, commented that, within the part of the testing I observed, some paths of code were not being executed.)
During the test (the monotonous procedure in the overly warm room), I aided my own alertness by thinking about whether there might be a way that hacking software could tell that this was a test (as in the recent Volkswagen fiasco). I noted that there was a time consistency in the voting from ballot to ballot that did not seem to me to characterize what I thought normal based on my personal experience in voting … where around me some people were in and out of the booth in an extremely short time, whereas others who were working away at their voting before I started were still working at it after I left. Also, I imagine that some people would check the summary of their votes carefully while others would neglect to check at all.
It took approximately the same amount of time to vote every single ballot by the test procedure. It also took approximately the same amount of time to review the summary of votes on every single test ballot, and that was close to (but, I’d guess, slightly less than) the time it took to vote the ballot. This might be an opportunity for hacking software to detect a test condition…if it had a way to compare from one ballot to another.
November 9, 2015