Proofiness - Charles Seife [54]
In the 1960s, the navy was getting quite annoyed that they couldn’t reliably keep track of the material that they stored in warehouses. Any catalog of material in a warehouse had huge errors, even after the warehouse staff went around physically inventorying everything in the building. Fed up, they asked an industrial engineer to find out what was going on. The engineer was surprised to discover that when he had warehouse workers count and then recount the inventory, the two counts didn’t match up—there was a 7 percent difference between the two inventories. The very act of counting was hugely error-ridden. As one inventory specialist told me, “Humans counting things, I don’t care what the things are, they’re going to be off.”
So long as we count votes, we make errors. And so long as we make errors, we can’t be perfectly precise; there’s inherent imprecision in our determination of who won a vote. In a blowout election, where the margin between the two candidates is much greater than the errors, the errors can’t affect the outcome of the election. But when the election is as close as Minnesota’s, the vote might be rendered meaningless—the truth of the real winner of the election will be obscured—if the errors are large enough.
This raises an obvious question: how large are the errors in counting votes? It’s hard to say for sure, because there are so many ways things can go wrong. There can be problems with the ballots—they might be read incorrectly by voters or by counters. The ballots might be mishandled and either left out of the tally or double-counted. The ballots might be tallied properly, but people might make stupid mistakes when recording the numbers or adding them together. And then there’s the unavoidable error of occasionally losing count—skipping or double-counting a ballot or pile of ballots.
Minnesota’s errors were probably quite small compared to those in other modern elections. The ballots all around the state were of roughly the same design, sheets with ovals to be bubbled in, meant to be fed through an optical scanner. These are considered to be the best kind of paper ballot, as they’re relatively easy to fill out and easy to interpret. (There are none of the hanging-chad problems that bedevil precincts that use punch cards, for example.) But there were definitely lots and lots of other errors. Hundreds upon hundreds.
It’s surprisingly easy to find missing votes if you have the right data. The Minnesota election was commendably transparent. Throughout the count and the recount, the state government shared all of its data with the public and the press. Every day, you could look at the secretary of state’s website and download the total votes tallied (and retallied) in each precinct. Shortly after election day, you could also see the voter turnout—how many people showed up at the voting booths on election day and how many cast absentee ballots of various kinds. In theory, these two sets of data should match exactly: in each precinct, for every voter who casts a ballot, you should have a ballot that is counted. But in nearly 25 percent of Minnesota precincts, the two numbers didn’t match—there were either more voters than ballots or more ballots than voters. This meant that one of the numbers was wrong.
In some cases, the voter turnout numbers were clearly at fault. Ramsey County’s voter turnout lists seemed to have been lovingly maintained by a pack of wild raccoons. One precinct, which had roughly 1,000 voters in the district, somehow claimed to have a voter turnout of 25,000—mostly military overseas ballots. (This error was corrected after several weeks, but it was pretty clear, for various reasons, that none of Ramsey County’s original turnout numbers could be trusted.) In other cases, though, the turnout numbers seemed to be solid. This was where it was obvious that people