Errors in Skagit vote count point to statewide problem
By JAMES GELUSO   Skagit Valley News   03 December 2004

Mistakes by counting machines could affect outcome of election
The machines used by Skagit County to tally votes made at least 36 mistakes in the count for the governor's race, which was decided by just 42 votes statewide.

It's impossible to tell whether some of the mistakes occurred on the first or the second count, according to county election officials.

The race between Republican Dino Rossi and Democrat Christine Gregoire was so close that there was an automatic machine recount of the votes.

The recount narrowed the gap even further, to the point where the standard error rate of the machines used by Skagit and other counties could affect the outcome of the governor's race.

Skagit County uses optical-scan ballots, as do most of Washington's 39 counties. Errors of a similar scale throughout the state would mean that the count's margin of error substantially exceeds the narrow gap produced in the recount.

Rossi was officially declared governor-elect on Tuesday, when Secretary of State Sam Reed certified the election results.

State law required an automatic recount, by machine, because the original margin of victory for Rossi was less than one-half of one percent. Had it been less than 150 votes, a hand recount would have been mandatory.
  
The recount narrowed Rossi's margin to just 42 votes, but if either party seeks a recount by hand, they will have to pay for it. Both campaigns have indicated they will announce their plans for a possible hand count on Friday.

Errors in Skagit County's system raise the possibility that neither the original count nor the recount were accurate.

So it's possible that more people voted for Gregoire than voted for Rossi - or that Rossi really won by a larger margin.

In Skagit County, errors occurred when the machines either counted a ballot twice in the first count, or missed counting a ballot in the second count, said Skagit County Elections Supervisor Erika Kubischta.

It's impossible to tell which happened in each case, but the result is the same - a candidate getting one less vote in the second count than in the first. It happened 13 times to Rossi and 19 times to Gregoire.
In four more cases, the number of ballots counted in a precinct decreased, but the number of votes received by each candidate didn't change.

The differences are detectable only when the precinct-by-precinct vote counts are analyzed. A review of the results by the Skagit Valley Herald found the 36 mistakes, as well as the possibility for many more.

In all, the first and second counts matched in only 38 of the county's 113 voting precincts. (The county has 116 precincts, but two have no voters and one has only one voter, who is given a ballot from an adjacent precinct to protect her privacy.)

The mistakes are caused when the machine jams, Kubischta said. The machine takes ballots from a stack, moves them across a scanner, and then sends them around a drum to a pile at the bottom of the machine. That's done at such a high speed that the ballots often get jammed in the machine - especially absentee ballots, which have been folded. Absentees make up 58 percent of the ballots cast.

When the machine jams, it displays an error message, Kubischta said, instructing the operator to either re-scan the ballot because it wasn't counted, or to remove the ballot because it was counted before the jam.

But sometimes the machine gets it wrong. Maybe two ballots got fed in and it only recognized one. Maybe it counted a ballot but didn't think it did. It's hard to say exactly what went wrong, only that the machine isn't perfect.

Kubischta said she was told by Election Systems and Software, the machine's Nebraska-based manufacturer, that even when all the ballots are flat, the machines have an error rate of about one for every 10,000 ovals it must read.

With four ovals in the governor race - for Rossi, Gregoire, Libertarian Ruth Bennett and write-ins - that means one per 2,500 ballots. Skagit County had 52,724 ballots counted, which means about 21 mistakes can be expected in each count of the governor race.

That's not normally enough to make a difference, but the Washington governor's race is believed to be the closest race of its magnitude in American history.

There might be more mistakes that the machines made, but differences between the first count and the recount make it impossible to prove they happened.

Aside from the mistakes, there were two reasons for differences between the first count and the recount:

n In 57 precincts, more ballots were counted the second time than the first time. Most of those are probably because a batch of ballots wasn't counted the first time after an election worker put them in the wrong basket.

n Another 20 votes were added to the mix through the enhancement process. On the second count, any ballot with no vote for governor caused the counting machine to stop. Out of 944 such ballots, 20 were considered to have votes that were too light for the machine, and election workers darkened the oval.

A mistake that resulted in a higher count for a candidate could be one of those factors. Or it could be a machine mistake - a vote missed on the first count, or counted twice on the second count. Elections workers didn't keep proper records during the first count, Kubischta admitted, so it's unknown exactly how many of those ballots were missed.

Kubischta was out of state during most of the first count due to a family emergency, so the counting and record-keeping was done without her supervision.

If a hand recount is done in the county, each precinct will be essentially counted twice. A hand recount involves separating the ballots in each precinct into stacks by candidate, then having two people count each stack until they come up with the same number.

The machines counted each precinct twice. And in two-thirds of the precincts, the machines came up with different numbers.

Kubischta said the current machines will be replaced before next fall's elections. The manufacturer doesn't plan to have the county's current machines certified for compliance with federal standards that will go into effect in 2006, so Skagit County will look for new machines early next year. The machine's design will be one of the factors considered, Kubischta said, in an attempt to reduce the number of jam-related errors.