To Volunteers and Friends of the Election Transparency Project
Thank you for helping make this another successful election. This email recaps what we accomplished together and outlines our next steps. Please contact us with questions.
Our goals are
To collect images of all ballots, making them available to voters and other interested parties after the election.
To work collaboratively with the Elections Department during the canvas period to scan ballots and compare counts. If discrepancies are discovered, to assist Elections in identifying the issue and correcting it before the election is final.
Within the limits of our volunteer bandwidth, to independently corroborate the official counts of the votes in the ballots.
Archived Images
With our volunteer's generously-offered time, the new donation-funded scanner (Thank you, David , and Scan, George WinterSun's new program, we scanned the last ballots the same day Elections made them available. The digitally signed ballot images are archived on an external drive. Anyone who is interested in receiving a copy should contact Kevin or Wes.
We processed 161,636 images, which, divided by 4, would be 40,459 ballots. The official reported total is 40,461; we are off by 2. We believe the issue is in our software or an error in the way we handled a batch or dealt with an error. David, George, and Wes have been going over the data and the manual logs but we haven't been able to nail down the cause of the difference. Before the upcoming General Election in November, we will modify the programs and procedures to provide more information if the issue recurs.
In the last General Election we were unable to analyze the vote counts in 1073 images (0.4 percent of the total). The problem images were tilted or folded or had other defects. Thanks to the new procedures you used and George's new program, this year we had zero!
Corroborating Vote Totals
Our final goal is to go beyond counting the ballots to corroborating the votes. There is a reason we say "corroborating" rather than "counting". The official vote-counting process, although computerized, requires a lot of supporting staff work. This includes interpreting write-in votes, interpreting mail-in ballots where the voter crossed out an initial choice and entered a correction, and deciding whether a light mark is an attempt to enter a vote or just an accidental touch. We have decided not to duplicate all that work considering the amount of volunteer time that would be required.
Instead, we have a method to match the counts where we can and highlight any races that are so tight that the judgement calls might make a difference. David's program, Pipeline, counts all the votes that can be unequivocally decided by the computer and then it makes two alternate totals to account for the ballots that would require a person to make the judgement. For each choice in a contest it computes a "best case total" - all the votes that would have been counted if every one of those judgement-based interpretations had fallen in favor of the choice. It also computes a "worst case" total for the choice, assuming the opposite. We then compare the official election total with the worst case ("ETP Lower") and best case ("ETP Upper") totals as shown in this example
Underfilled Squares
One source of inconsistency between our results and the official counts has to do with voters who do not completely fill in a square. The goal is to as accurately as possible interpret the voter's intent. The official software has its own method of deciding the cutoff between an incidental mark and a vote and we have ours. In most cases we agree but there are edge cases where we might differ. In a close contest someone might look at whether there was a difference in the undercounts to decide whether to get the ballot images and count them. Here are two examples:
Write-Ins
There is special treatment for write-ins. The Elections staff has identified the intention of the voter's writing and, if it names a registered write-in candidate, the system counts it for that candidate.Otherwise it tallies the vote as an "unassigned Write-in". Because we don't have a person interpret the hand-written text we just count the number of ballots where the write-in box was filled in and there was not an overvote. In the pattern below you see that the official system found 326 write-ins whereas we found 325. However we found an extra undervote. It is likely there was a ballot with a light mark for the write-in line and the official software counted it as a vote whereas Pipeline took it for an incidental mark and called it an undervote.
The important match up is the total line at the bottom. For each of our systems the totals of votes, undervotes, overvotes, assigned write-ins, unassigned write-ins should be the same, and that number should be the number of ballots for the contest.
Thanks … and See You Soon!
We really appreciate your help and assistance in improving our accuracy and efficiency. Attached to this email you will find a link to a PDF of our report, contest by contest. It could not have happened without you. We look forward to working with you again in the General Election in just 83 days!
Download the Report
