As we wrote last week, election forecasts have been looking better and better for Republicans lately. This is mainly because GOP candidates have taken consistent leads in the purple states of Iowa and Colorado, while Democratic incumbents in Alaska, Arkansas, and Louisiana have fallen even further behind.
With the election 15 days away and the GOP's advantage looking clear, forecasters have recently been tackling an interesting question — how often has Senate polling in previous years been wrong?
- Nate Silver examined polls of Senate races from 1990 to 2012 in the final 21 days of the campaign. He found that consistent bias of the polls of a few percentage points in one partisan direction or another was common. But, he argued, there's little consistency year to year that would help indicate which direction this year's bias might be in.
- Meanwhile, Sam Wang pointed out that, in 2010 and 2012, Senate Democrats outperformed the polls in close races by a couple percentage points.
- Back in July, Josh Katz of the Upshot analyzed all Senate polling from 2004 to 2012. He found that candidates leading by less than four points in polls two weeks before the election only ended up winning 67 percent of the time. In the final week of the campaign, accuracy rises to 81 percent. (Right now, FiveThirtyEight estimates that there are six Senate races where the leader is ahead by four points or less.)