All of the major Senate forecasting models, including ours at Election Lab, now rely heavily on averages of public polls.  This raises the question of whether those averages will be correct on Election Day, and whether any misses could affect which party manages to retain control of the Senate.  In particular, there is the question of whether polling misses might mean that the Democrats end up with a slim Senate majority after all.

There are reasons to be skeptical that this will happen.  It’s not just that we can’t easily predict whether the polls will over- or underestimate one party’s vote share, as discussed by Nate Silver and by Mark Blumenthal & Co.  And it’s not just, as Josh Katz and Sean Trende have found, that Senate polls already tend to be pretty accurate at this point in time — especially when candidates have a 3- to 4-point lead, as do Republican candidates in Alaska, Arkansas, Georgia, Kentucky and Louisiana.

The other key point is this: Late movement in Senate polls tends to be in the direction of the underlying fundamentals.  I discussed this movement in the polls in an earlier post, and it’s worth revisiting it now.

The analysis is pretty straightforward.  Estimate a simple model of Senate elections from 1980 to 2012 that relies on only a few factors: economic growth, presidential approval, whether it’s a midterm or presidential year, and how the state voted in the most recent presidential election. Then estimate an out-of-sample forecast for every Senate election between 1992 and 2012. Then compare the polls to that forecast.

Here is the gap between the polls and the forecast for the last 60 days of the campaign:

Graph by John Sides

On average, the polls have not yet fully incorporated even these basic fundamentals by this point.  After today, 18 days before the election, there tends to be further movement toward those fundamentals.  The sharp movement in the last few days of the campaign should be taken with a grain of salt, since the volume of polling drops over the weekend before Election Day, but the overall pattern is clear.

As we noted back in January, a slightly more elaborate fundamentals-based forecast based on elections since 1980 favors the Republicans.  And this is what poses a challenge for the Democrats: Right now they need the polls to move in their favor, but in key states this would entail movement opposite to what the fundamentals would predict and therefore opposite to the trend above.

Of course, it’s not impossible for polls to move in that direction. The graph depicts an average across many Senate races, and not every race may conform to this pattern.  It’s just unlikely that the polls will move enough in that direction for Democrats to benefit on Election Day.