The Plural of Bad Polls Isn't Data Either

We've heard the old saw over and over this election season: "the plural of anecdote isn't data." That's true. You can easily find stories of evangelicals for Obama who've always voted Republican before alongside PUMA blogs from those who've always been Democrats. And there's no way to extrapolate from that what's really going on. Long lines here or there (we had a huge line in a very red district of a very red state with very few interesting races this morning) don't tell you who will win. But here's a news flash: neither do the polls.

We know the polls are wrong. Period. Full stop.

We know it because they disagree with each other far beyond their margins of error. They can't all be right. In fact, they're not even mostly right. It's true that they all agree one candidate will win...but is that data?

My answer: NO! If Obama wins, it will not validate the bad polling that's been done this election cycle. The pollster who comes closest to the final number will claim validation, but won't legitimately have it. Correlation is not causation. Throwing darts at the wall blindfolded and hitting the bullseye without even knowing where the dartboard is does not demonstrate skill. It's dumb luck.

Yet because all of these bad polls agree on the outcome, we've accepted them as accurate. "Just average them together" we're told and you'll have the number. Well, no. A poll that's off ten points (and we know that some of them are) could just as easily be a McCain squeaker as a comfortable Obama win. We're buying a pig in a poke with these polls. They're all--every one--based on assumptions that this election will be different from all that have come before it. What if those assumptions are the underlying source of the error.

What if the Shy Torys and the PUMAs do vote instead of staying home discouraged? What if the bad polls have a fatal flaw at their heart? What if McCain wins?