UPDATE: 11:20 p.m. -- Perhaps the most convincing evidence so far that human error explains the initial omission of Brookfield's results comes from our colleagues at the Brookfield Patch. On election night, they reported a vote total for Brookfield that exactly matches the vote total Kathy Nickolaus did not include in the County level count until Thursday. As Joe Petrie and Lisa Sink of the Brookfield Patch reported on Thursday (via Mickey Kaus):
On election night, the City of Brookfield reported that Prosser received 10,859 votes from city residents, or 76 percent of the vote, compared to the 3,456 votes cast for challenger JoAnne Kloppenburg. The Brookfield Patch reported those numbers in a story with chart posted about 12:30 a.m. election night.
[Brookfield City Clerk Kristine] Schmidt said her office also posted the results on the city's web site before going home on election night.
WASHINGTON -- Faith in America's electoral-tabulation processes took a hit late this week when a Wisconsin county clerk who announced a bombshell correction: nearly 15,000 missed votes, which dramatically upended the state's supreme court race.
The clerk described her mistake as "human error ... which is common in this process." While an ongoing review by state election officials will ultimately provide the best evidence, both turnout statistics and historical precedent generally support her claim.
Unofficial tallies for Tuesday's special election gave Democratic-backed challenger JoAnne Kloppenberg an extremely slim lead over incumbent Wisconsin Justice David Prosser -- just 204 votes out of nearly 1.5 million cast. But on Thursday, Waukesha County Clerk Kathy Nickolaus announced that the early counts had missed roughly 15,000 votes, mostly due to the omission of the entire city of Brookfield. Kloppenberg's revision put Prosser ahead by 6,744 votes, leaving Kloppenberg hard-pressed to catch up.
Reviews of the election turnout statistics by Craig Gilbert of the Milwaukee Journal Sentinel and by FiveThirtyEight's Nate Silver show that the revised Waukesha results appear more plausible than those initially reported. As Gilbert wrote, the addition of nearly 15,000 votes "puts that county's turnout rate more in line with the neighboring GOP strongholds of Ozaukee and Washington counties."
The Huffington Post created the following chart, which compares the turnout in last year's general gubernatorial election to that from last week:
Statewide turnout was lower for last week's special election than in November's election, but the level of turnout correlates strongly within individual counties. That's why the circles in the chart form a diagonal line (Each circle represents an individual county, and the size of each circle represents its number of registered voters).
But the originally reported Tuesday turnout for Waukesha County was lower than the its 2010 vote totals would have suggested. The new votes added in the revised count brings the county's turnout number for this week more closely into line with the statewide pattern.
Wisconsin Democrats have expressed doubt about Nickolaus' claims, given her longstanding ties to the state GOP. She worked for the Wisconsin Republican Caucus for 13 years, including the period during which Prosser served as minority leader and speaker in the state Assembly.
State election officials have questioned the delay in reporting the error and are reviewing the Waukesha County results. Ultimately, the best evidence will come from the poll books that poll workers use to record whether a voter has cast a ballot and, in the event of a recount, from the actual ballots.
University of Wisconsin political scientist Charles Franklin (who co-founded Pollster.com, the forerunner of HuffPost Pollster) lent credence to Nickolaus' amended results, however, arguing that Waukesha County's vote "appears to be corroborated by reports from the city on election night, by a Democratic election official and by the plausibility of the corrected data."
So just how common is this the sort of error that Nickolaus claims she made on election night? Michael McDonald, a George Mason University professor who studies voter turnout and closely monitors election results, said they are "not uncommon ... we see these sorts of errors happen all the time."
Recent history bears him out:
- In the 1988 Democratic primary in Delaware, a keypunch error gave Samuel Beard 2,800 votes more than he received in a race against incumbent Lt. Gov. S.B. Woo. Although Beard had claimed victory, the error was not discovered until two days after the primary and reversed its result, putting Woo ahead by just 71 votes.
- The initial count in New York's 2001 Democratic mayoral primary gave Mark Green a roughly 30,000-vote lead over Fernando Ferrar, but an "official machine count" conducted a week later reduced that margin by 18,029 votes after election officials discovered 42,000 votes that had been counted twice. The final official count narrowed Green's margin to just 15,981 votes.
- Unofficial returns in the 2002 Alabama governor's race showed incumbent Democrat Don Siegelman narrowly defeating Republican challenger Bob Riley. But a late-night recount conducted by Republican officials in Baldwin County revealed what they described as a "computer glitch" that gave Riley a statewide lead of 3,195 votes. Although Democrats suspected foul play -- the recount had occurred after Democratic party observers had gone home for the night -- Siegelman ultimately dropped efforts to obtain a statewide recount.
- Folllowing New York's 2008 presidential primary, a New York Times review found "about 80 election districts among the city's 6,106 where Mr. Obama supposedly did not receive even one vote, including cases where he ran a respectable race in a nearby district." A formal review later found hundreds of small anomalies in vote reporting in individual districts. The Times reported election lawyers representing both Barack Obama and Hillary Clinton agreeing that it is "not uncommon" to see such mistakes "made by weary inspectors rushing on election night to transcribe columns of numbers."
Similarly, high-profile recounts in the 2008 Minnesota Senate race, the 2004 Washington governor's race and in the 2000 presidential race in Florida have exposed similar examples of human error in the initial tallies. McDonald said we are "blissfully unaware" of small election-night mistakes in most cases "because you had a definitive winner and the election result is not changed ... It's only on these close races where these warts are exposed."
McDonald said he has found discrepant numbers even in the final certified results. "I'll go back and ask the election officials why the numbers don't add up, and they'll say, 'Oh, we made an error,"" he said. Since correcting such errors requires action by a court, he said, "election administrators will generally just let those errors persist" as long as they do not affect the election outcome.
"If you're a pollster," McDonald added, "you talk about the margin of error. People who delve into election administration often talk about the 'margin of the election.'" If we ignore that potential for error, he warns, we "probably have unrealistic confidence about how human beings actually do large, complex tasks."
[Correction: The original version of this story incorrectly characterized the procedure used to record whether an individual registrant has voted. In Wisconsin, poll books are used by poll workers to record whether individuals have voted, but voters do not sign in.]