On Thursday, Silver’s model was giving Obama a 79 percent chance of winning the election. The luxury of his contraption is its built-in defense against being wrong. He’s not calculating the size of an Obama victory, only the probability that the president will end up on top. If Obama loses, Silver can always say, “Well, that’s okay, my model gave Romney a 21 percent chance of pulling it out.”
When Silver says that 80 to 90 percent of punditry contributes nothing of journalistic value, his modeling is being charitable by about 10 percentage points. On television, the goal is to panel up and talk, with the hope of stirring a disagreement that gets shared like crazy on Facebook. On the Web, the goal is volume, and nothing yields volume quite like political analysis. Politico, the king of volume and analysis, printed this paragraph before the debate between Vice President Biden and Paul Ryan:
“The best case scenario for each? A clear win. The worst case scenario for each? A clear loss. The murkiest scenario for each? A debate that gets fought to a draw, which the press will interpret in different ways but which also likely won’t stop the GOP ticket’s momentum.”
That’s important context. Because no matter your view of polling, no matter how biased you deem the people behind the questions and the modeling, think of the choice: impression-based horse-race coverage by pundits vs. data-based horse-race coverage by statisticians.
Given Silver’s new fame, there’s an 84 percent chance that media outlets will christen at least 10 new poll-modeling gizmos in time for election 2016. “I think I’m taking advantage, in part, of an underserved market,” Silver says. “The fact that I get kind of undeserved attention, both negative and mostly positive, would suggest there’s a market deficiency — that you have a gang of maybe 500 pundits.”
And not enough modelers.
Much has been made of the possibility that pollsters and even modelers will be proved wrong come Nov. 7. Lavrakas says that outcome would be a “major setback” with “dire consequences.”
Silver has a milder take. “I know as a matter of practice that I’m going to have more opportunities if my prediction looks good and fewer if it doesn’t,” he says.
When’s the last time a cable pundit had to face such consequences?
I’m more forgiving on the forecasting models. If the model doesn’t work, tweak it. If that one doesn’t work, tweak it again. But keep the reality-based analysis coming. I’ll take a deeply explained essay on the eccentricities of New Hampshire political forecasting or a disquisition on the house biases of polling organizations over conventional punditry every time.
Erik Wemple reports and writes about the media in his Washington Post opinion blog.
Read more from Outlook, friend us on Facebook, and follow us on Twitter.