Proofiness - Charles Seife [53]
There were, of course, errors that favored Coleman rather than Franken. During the post-election audit, the hand recount in random precincts required by Minnesota law upped Coleman’s lead to 215 votes from 206. And during the painstaking hand recount that followed, many more errors were corrected, some of which were to Coleman’s benefit. These errors weren’t hard to find—after all, hundreds upon hundreds of votes had vanished.
Voting has a lot in common with polling. They have a similar purpose; both voting and polling attempt to figure out what the population is thinking about a subject. Also, like polling, voting is subject to error, though the errors are of a different sort. Voting, by its very nature, avoids many of the problems that plague polls.
Polling tries to get at an underlying truth, but that truth is obscured by errors, both statistical and systematic. When pollsters ask a sample of a few hundred or few thousand subjects a question, there’s the unavoidable statistical error that is introduced by the leap-of-faith assumption that the sample represents the beliefs of the entire population. The laws of randomness ensure that even under the best of circumstances there’s an underlying error that obscures the truth. And circumstances are usually not that great. Pollsters have great difficulty getting a sample that is truly representative of the whole population. Often this causes a systematic error—a bias—that obscures the truth even more. And of course, if a poll is badly worded or improperly conducted, it can cause systematic errors that can make your poll all but meaningless.
Voting avoids these problems almost entirely. In a vote, the truth—the victor of an election—is determined by the will of the population that comes to the voting booths and casts a vote. Unlike a poll, which queries a sample of a population, a vote uses the results from the entire universe of voters who cast a vote. The people whose opinions you are trying to figure out are, by definition, exactly the same ones whose votes you are counting. There is no leap of faith that comes with assuming that a sample truly represents the opinions of the whole population, so there’s no statistical error at all. Similarly, you don’t have to worry about pulling a Literary Digest—since you’re looking at the entire universe of votes cast rather than a sample, you don’t have any systematic error due to a nonrepresentative sample. The votes that are counted and certified are, by definition, perfectly representative of voters whose opinions need to be considered.50 Finally, the issues put to a vote are (generally) so black and white that there isn’t (generally) a systematic error caused by the phrasing or presentation of questions. In theory, vote tallies should be fairly “pure” numbers, close to the realm of absolute truth.
Because of this, voting is pretty much immune from the big errors that obscure the truth in polls.51 The huge, multiple-percentage-point errors that make most polls meaningless simply aren’t germane to elections. But this doesn’t mean that elections are error-free. Far from it. It’s just that the errors are more subtle. They’re smaller—a fraction of a percentage point of the total votes cast—but they’re there.
This shouldn’t come as a surprise to anyone reading this book. There’s no such thing as a completely pure number, no such thing as a measurement that’s always