Over the past few weeks, there's been some disagreement among the election forecasters seeking to predict this year's Senate races. In particular, Sam Wang of the Princeton Election Consortium continued to give Democrats strong odds of holding onto the chamber, while Nate Silver and most other forecasters saw the GOP as holding an edge.
But in recent days, as several terrible new polls for Democrats have come in, Wang's daily snapshots (which are different, for reasons explained below, from his overall election day probability) have come closer to agreeing with Silver's forecast.
Check out Vox's election forecasting hub for the latest numbers, but here's a rundown of why things have changed:
Polls have turned against Democrats in three crucial races
In three key races — Colorado, Iowa, and Alaska — Democrats had seemed to be narrowly favored in mid-September. Yet several new polls in each state suggest that things may have changed — and that Democratic candidates may now be underdogs in all three states:
- In Colorado, Rep. Cory Gardner (R) has led the most recent four polls, and posted margins in the high single digits in two of them. Polling in Colorado overestimated how well Republicans would do in the 2010 Senate race and 2012 presidential race, but considering that nearly all polls before this showed Senator Mark Udall (D) ahead, this isn't good news for Democrats.
- In Iowa, the most recent four polls have shown either a tie or a 6 point lead for Republican Joni Ernst. And the most recent poll showing Ernst up is from the highly respected Selzer & Company, which FiveThirtyEight evaluated as "among the best in the country."
- In Alaska, a state where we've gotten little polling, Republican Dan Sullivan has led the most recent three polls. Even more notably, since July, Senator Mark Begich (D) has only ever led in polls sponsored by Democratic firms. Silver also notes that polling has frequently overestimated how well Democrats will do in Alaska.
The forecasters have moved in response
In early September, there was an evident split among the six main Senate forecasts. Those that used only polls of individual races — HuffPost Pollster, Daily Kos, and Princeton Election Consortium — showed a small Democratic advantage. But those that also incorporated factors called "fundamentals" — FiveThirtyEight, the Upshot, and the Washington Post — gave the GOP a lead. (These "fundamentals" can include both state-level factors like candidate fundraising and how the state voted in the 2012 presidential race, as well as national factors like the generic ballot.)
Now, though, that split is gone — two of the polls-only models also give the GOP a lead:
Two things have been going on here — the modelers that use fundamentals have been weighting them less as the election gets closer, and new and favorable polls for Republicans have come in. But overall, the forecasting consensus is now quite close to where FiveThirtyEight and the Upshot were at the beginning of this month.
What's happened with Sam Wang's model at Princeton Election Consortium is a bit more confusing. He takes a daily snapshot — and his snapshot, which has all the most up-to-date polling, now predicts the GOP will retake the chamber. However, his overall election day prediction of party control is based on combining all of those daily snapshots since June. This methodological choice makes Wang's election day prediction slower to respond to genuine movement in the polls, if it occurs.
Wang wrote on Saturday, "Recall the central assumption: the Meta-Margin range from June through now is representative of the future. Obviously, this week it is on my mind that the assumption won't hold up." Indeed, his newest numbers give the GOP a larger advantage than at any time so far. As of Monday morning Wang had pulled his favorable election day probability for Democrats from his site, and replaced it with the word "Recalculating."
Update: In a new post Wang writes, "To state the obvious, some recalculation is necessary. That's the bad news, from a modeling standpoint." He writes that he's now considering looking only at snapshots since September rather than since June.