May 24, 2024

Expensive and ineffective recounts in Los Angeles County

Part 3 of a 4-part series

In a recent article I wrote about the recount of a very close tax-rate referendum in the city of Long Beach, California.  The referendum passed by 16 votes out of 100,000 ballots; the opponents of the measure requested a recount, as they are entitled to do by California law—provided that they pay the costs.

Recounts or audits of paper ballots can be useful to detect and correct many different kinds of problems:

  • Were optical-scan voting machines inaccurately reading the voters’ marks?
  • Were the voting machines so misconfigured that they were producing meaningless results?
  • Were election practices so sloppy that it’s hard to trust the results?
  • Were the voting machines hacked to be deliberately cheating?

Recounts and audits are typically done with several small teams.  In each team, one person reads the paper and calls out the votes, watched by one or two other people; two people independently tally the votes called out.  If there are 4 people on a team, paid $20 per hour each, and they can count one ballot every 10 seconds, that works out to about 22 cents per ballot for a recount.  Audits are more expensive per ballot— in part because they may cover several different contests (not just one race), in part because the checking is more complicated—but they only have to cover a small fraction of all the ballots.

Based on such guidelines published in 2020 by Los Angeles County, the citizens’ group asking for a recount of the 100,000 Long Beach ballots expected to pay about $20,000 (more or less) for a recount.  But the 2.1 million LA County ballots (about half-and-half from vote centers and from mail-in/dropbox envelopes) were all mixed together, not sorted by city.  So the County told the citizens’ group it would cost $187,000 just to find the Long Beach ballots within all the LA County ballots even before a single ballot could be recounted.

In Part 1 of this series I explained that the recount would have been a lot faster and cheaper if LA County had sorted their mail-in ballots by precinct, before opening the envelopes.  In Part 2 I explained how some jurisdictions use high-speed envelope sorters; LA County could have done this too.  But they didn’t.

The Los Angeles elections office offered instead a recount of the virtual ballot images.  That is, all those 2.1 million ballots had been scanned by optical-scan voting machines, as they were counted, and it was easy to sort the computer file of pictures of ballots.  So, they claimed, they could recount those pictures for a reasonable price.  This turned out not be true, as I’ll explain.

Can recounts of pictures of ballots really tell us anything useful?

  • Were optical-scan voting machines inaccurately reading the voters’ marks  Inspection of the pictures can sometimes be useful to answer that question.  I’ll show examples below where LA County’s equipment mistakenly read the voter’s mark.  LA County’s equipment seems to have produced high-enough quality scans for this purpose.  But whether “looking at pictures” is good enough, depends a lot on the quality of the images.  Dominion’s optical scanners, as configured in Georgia, produce 200-dpi b&w images, which do not clearly represent voters’ marks on the paper.  And some scanners can’t “see” certain colors well—a black&white image might completely leave out some votes.
Here, the intent of the voter is clear, but the machine records an undervote (BLANK CONTEST) in the two out of three of these races (Fulton County, GA, August 2020).
  • Were the voting machines so misconfigured that they were producing meaningless resultsRecounting the pictures may or may not be useful in that case, because a misconfiguration might mean that you can’t even find the right pictures.
  • Were election practices so sloppy that it’s hard to trust the results?  When an election office can’t even keep track of how many voters cast ballots in which precincts, examining the paper ballots can be “ground truth” to help diagnose the situation.  Computer files of images, produced by a chaotically disorganized process, can’t be trusted to tell us much.
  • Were the voting machines hacked to be deliberately cheating?  If so, then hacked voting machines can make up completely fraudulent pictures, and a virtual-ballot-image recount will just be “security theater.”

The Long Beach citizens’ group decided to settle for a virtual-ballot-image recount.  In that recount, they discovered that the voting machines were indeed inaccurately reading the voters’ marks—only rarely, but with a 16-vote margin out of 100,000 votes, rarely could make a real difference.

The Los Angeles VSAP optical-scan voting machines are so eager to treat a mark as a vote, that they treat stray marks of the kind illustrated here as overvotes—so the voter was disenfranchised by the machine count.  (This was observed by the recount teams.  Marks shown here are not actual voter marks, but just illustrate the observed phenomenon.)

Unfortunately, LA County’s ballot-image recount method was so clunky and slow and expensive that it was costing about $1.50 per ballot!  The citizens’ group was being charged about $10,000 per day, and after 8 days only 12,000 ballots had been recounted.  They were on track to finish for $150,000, but they ran out of money and abandoned the recount.

The clunky recount method displayed the ballot image on screens (one screen for each worker on the team).  But the networked computer system had such time lags that the ballots didn’t display simultaneously, so one team member would be looking at ballot 376 while the other teammate would be looking at ballot 375; to prevent this, they had to slow down and wait.  There were other kinds of clunkiness as well, and things went very slowly compared to handling actual sheets of paper.

When the Los Angeles County’s election office designed their new VSAP voting system, they gave no thought at all to how recounts would be done (the word “recount” never appears in their 59-page Request For Proposals, nor in any meeting of their technical advisers).  When the Long Beach citizens’ group asked for a recount, LA County had to scramble to cobble something together.  The result was the slow and clunky system that failed to even complete a recount of a single contest covering 5% of the county’s voters.  Who would have imagined that experienced election officials would forget that recounts are a thing?

California permits arbitrary and capricious charges for recounts

Ten years ago, Brad Friedman documented how California county election officials can charge arbitrary amounts for recounts, with Fresno County charging $1/ballot while Orange County charged only 14 cents per ballot (and Stanislaus County charged $20 per ballot). 

If we are to count paper ballots by computer (as I think we should), then we must be able to recount ballots by hand.  Overcharging for recounts does not lead to trust in the election process.

Part 4 of this series: Willful disregard of voter intent in Los Angeles