A Kooky Kount
The meaninglessness of the Florida recount.

November 14, 2001 4:40 p.m.

ow precise was the major-media recount of 175,010 rejected ballots from last year's presidential race in Florida? Consider some of the factors that played a role in the final product.

The National Opinion Research Center, which conducted the study for the New York Times, Washington Post, Wall Street Journal, and other media organizations, hired 153 "coders" to examine the disqualified ballots. NORC asked the coders if they viewed themselves as Republicans, Democrats, or independents. Thirty percent said Republican, 36 percent said Democrat, and 30 percent said independent. As it turned out, according to NORC, self-identified Republican coders were more likely than Democrats to see a vote for George W. Bush. Self-identified Democratic coders were more likely than Republicans to see a vote for Al Gore.

It also mattered whether a ballot was examined by a man or a woman. "The odds of a coder seeing a mark for Bush were somewhat higher for male coders than for female coders," says Kirk Wolter, the NORC statistician who oversaw the effort. The same was true for Gore. "The odds of seeing any mark at all for Al Gore were also somewhat higher for male coders than female coders," Wolter says.

Aside from those differences, there were simple differences of opinion among the teams of three coders who examined each ballot. Studying the results, the Washington Post wrote, "When researchers from the National Opinion Research Center looked at a ballot, especially a punch card, they did not always see the same thing. For every two ballots on which the researchers unanimously saw a potential vote for Bush or Gore, there was another ballot on which someone saw a vote for Bush or Gore and someone else didn't."

In other words, in one out of every three ballots which contained a discernible marking, the three coders couldn't agree on what it was.

The coders were also hindered, Wolter says, by not being allowed to actually touch the ballots. In the study, the ballots were handled by local elections officials (as they were in the actual recounts last November and December). The coders were just allowed to look at a ballot that someone else was holding. "Our view of dimples was less reliable, perhaps, than a canvassing board's view of dimples," Wolter says.

Take all that together — disagreements on a significant number of ballots, differences traceable to party affiliation and gender, and less-than-ideal study conditions — and you have a recount that is simply not precise enough to come up with exact vote totals in a race as close as the 2000 presidential election.

Using the raw data supplied by NORC, the Washington Post came up with scenarios in which Bush won by 225 votes, 430 votes, and 493 votes. The paper also came up with scenarios in which Gore won by 60 votes, 107 votes, 115 votes, and 171 votes. But the number of ballots about which coders disagreed, or which were affected by some sort of coder bias, was far greater than any of the vote margins in any of the media scenarios.

Which means that Gore partisans who claim the recount proves Al Gore actually won Florida are wrong. It also means that Bush partisans who claim the recount proves George W. Bush actually won Florida are wrong. What the media recount proves is that the election procedure in place in Florida in November 2000 was actually pretty good: Count the votes by machine on election night. Count them by machine again if the margin is close. Declare a winner, with a tightly limited time for the loser to contest the results. When that time is over, certify the winner and close the books. Don't change the rules after the fact — and don't go off on a wild goose chase for a level of certitude that cannot be achieved in a 6,000,000-vote election. Though that's certainly not what they intended, the media recounters proved it can't be done.