Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

Post-convention coattails reach Senate races

September 19th, 2012, 2:00pm by Sam Wang

The post-Democratic convention bounce has not affected Presidential race dynamics. Where it may matter is at the Congressional level. The post-Labor-Day bounce has reached downticket races, a fact that has major implications for control of both the Senate and the House. Both chambers appeared to be on a knife edge in August, but for now that is no longer true. If the bounce sticks, the probability of retained Senate control by the Democrats is 88%, the same as the probability of the President’s re-election. But will it last?

On average, each party wins more of the vote at the Congressional level when it wins the presidency than when it loses. To what extent is it driven by explicit messaging by the parties during a campaign?

In August, I said that the addition of Paul Ryan to the Republican Presidential ticket was a game-changer – but not for Romney, who on statistical grounds was unlikely to catch up with President Obama. Ryan is chair of the House Budget Committee, and his nomination allowed the national conversation to center on GOP Congressional budget priorities. Representative Todd Akin’s remark about “legitimate” rape opened the way for social issues to enter the national spotlight.

The Republican convention was light on specifics and yielded no bounce or even a negative bounce. Budget and social issues were embraced by the Democrats at their convention. The result has been a Presidential bounce of about 20 electoral votes, equivalent to 1.5 percentage points in Popular Vote Meta-Margin (the shift in opinion necessary to create an electoral near-tie). This is not particularly notable, since it keeps the race within the 290-330 EV range that we have seen since July.

History of electoral votes for Obama

However, a new wave of polls is notable. It shows that for the Senate, a larger shift has occurred.

First, a caveat. In well-polled races, such as MA-Sen Warren (D) v. Brown (R-inc.), the bounce was sharp and directly associated with the convention, where Warren was a prominent speaker. In other cases it is not possible to say for certain when the shift occurred. Indeed, we still don’t have any post-convention Senate polls for North Dakota or Connecticut, and only one in Montana. But a shift is visible for multiple other candidates, suggesting that some national factor has been the driving force. The most plausible candidate is the Democratic convention.

The Senate picture looked like this just one week ago:

Senate projection 12 September 2012

Now, it looks like this:todate Senate histogram

Projected from today’s standings*, the median predicted November result is now 52D/I to 48R, with a 95% confidence band of 49-54 D/I seats. The probability of retained Democratic control is 88%.

However, there is also the possibility that the current bounce is transient. In this case, polls will drift back toward the Republicans, in which case the probability could be as low as 60% based on last week’s data.

The shift has a weak tendency to be larger in more strongly Democratic states. Democratic candidates have benefited in WI (D-Baldwin vs. R-Thompson, +11% swing), MA (D-Warren vs. R-Brown, +8% swing), and IN (D-Donnelly vs. R-Mourdock, a +4% swing). States showing no real movement are VA (Kaine vs. Allen, +0.5% swing). and MO (D-McCaskill vs. R-Akin, -1% swing, i.e. toward Akin). This is consistent with a recent Pew Center finding of increased Democratic enthusiasm post-convention.

Now that we have evidence of downticket effects of the national race, what should the Presidential candidates do to help their party? In the Senate, the answer is to travel to places where they are running ahead of their side’s candidate. Team Obama should probably have the President (and Michelle Obama, Joe Biden, and Bill Clinton) spend time in MA, VA, and CT. Team Romney should send Romney and Ryan to IN, ND, and WI.

And you, dear reader, should go to ActBlue (D) or Crossroads GPS (R).

The Senate is not the only chamber whose polls have moved. House polls have gone in the same direction. More on that soon.


*Technical note: I calculate the histogram above under the assumption that all polling margins will move together between now and Election Day by an amount diff. Then I let diff vary from -3% to +3%, and sum over all the possibilities. This does not account for state-specific events (Akin’s comment being a prime example), but it is not that far from reality. (Fellow geeks, the MATLAB code is here if you want to play around. To allow for biased drift, set the variable bias to specify drift in one direction; for example, bias=-2 indicates average movement of all margins toward Republicans by 2%.)

HuffPost/ data sources: IN, MA, VA, WI, MO, MT, CT, ND.

Tags: 2012 Election · Senate

21 Comments so far ↓

  • Tapen Sinha

    It is more likely to last for one simple reason. The rank of the undecideds has shrunk considerably. Of course, that is not true for ALL the Senate races. For example, Murphy leads MacMahon in CT by 4 but it is 37 to 33. That can change at the smallest nudge.


  • Olav Grinde

    Indeed, there seems to be considerable disarray on the GOP side. While there appears unity and systematic effort on the Democratic side, we see conservative commentators displaying puzzlement, dismay and even disgust at the performance of the Romney Campaign. More telling, perhaps, are the concerted efforts of other Republican candidates to distance themselves from Romney.

    Could there be clearer evidence that the Republican brand is severely damaged, and that it is reducing the changes of Republican candidates to the Senate and House?

    That, at least, is how I read the polls, the pundits and the saturation of media reports.

  • Xtalographer

    Hi Sam,

    Looking at the probability distributions right now, it appears to be bimodal with 2 peaks. Are those peaks basically reflecting the addition or subtraction of Florida?

    Pretty nice to see Obama can win easily even spotting Rmoney 27 EVs.

    • Patrick Draut

      Even better news… Obama can actually spot him these four states, as long as he holds onto Ohio and Iowa, and still win the election. I believe this is why he has spent a decent amount of time in Iowa while on the trail.

      Virginia -13
      Florida -29
      Colorado- 9
      Wisconsin- 10

    • Sam Wang

      You might be just the guy for this other spectral thread.

    • Xtalographer

      Whoops. I missed that last thread… <>… well great minds (particularly biophysicists, though I have no love for NMR) think alike!

    • wheelers cat

      Its not spectral, its fractal.

  • Patrick Draut

    Dr Wang-

    Does your model take into account The Gary Johnson Effect? If so, how? I assume his supporters should pull from the Romney side, but by how much proportionally? Or are a majority of Johnson supporters those that would generally not vote for either Romney or Obama thus ‘staying home’ on election day, not having any meaningful impact on polling or the actual election?



  • Olav Grinde

    One of today’s posts by Nate Silver at addresses a topic recently dealt with here: the accuracy of polls that omit cell phone-only voters.

    The differences described by Mr Silver are striking!

  • Amitabh Lath

    Sam, a couple of quick questions on the error band on the Median EV estimator.

    First, it seems to have gone asymmetric. Longer tail on the low end.

    Second, the size of the band does not seem to be shrinking. It was small in August, but currently it seems to be 60 EV end-to-end.

    Since this technique has hit within +-1 EV a few times, I would expect the error bands to start
    shrinking. Or am I reading things wrong?

    • Sam Wang

      As I have said here so many times, much of that band arises from house effects, i.e. systematic errors. The true uncertainty is much smaller.

      House effects do not get smaller over time, so it will probably stay the same size.

    • Matt McIrvin

      I’d guess that if Obama starts doing so well that it looks like he’s going to carry all the states he did in 2008 (or maybe all of them minus Indiana), which with today’s electors would be 358 or 347 EV, he starts to hit a red-state ceiling that is so hard that the upper tail of even the systematic uncertainty should almost disappear. It’d possibly still be there in Meta-Margin but not in EV. On the other hand, the lower tail might grow in step.

    • Matt McIrvin

      …then again, the situation in Missouri is just soft enough that in that scenario it might look gettable, and the impenetrability of the really red states might just be an illusion caused by lack of polling.

      (If I had to bet, though, I’d still say Obama’s final EV count is somewhere in the 300-330 range.)

  • Peter D

    @Billy: You can get that info for yourself here:

  • Peter D

    @Amitabh As to the size of the distribution, based on this post ( I believe that the MM SD will be fixed at 2.2 until at least 40 days before the election.

  • Peter D

    @Olav I agree! The first strong argument I have seen for a bias term ~=0!

    If this is the *only* difference between the polls, then I would guess that of course the cell-inclusive polls are superior. In that case, what an arb in the markets!

  • xian

    I do think a Bob Dole type or worse collapse could put Missouri and even a few others into play, fwiw.

  • Amitabh Lath

    Thanks Sam, Matt, Peter D. I understand how the error bands could go asymmetric due to the non-uniform quantization of the EV. I had not thought it through.

    And thanks for pointing out the post from Aug 23, I had missed it. The collapse of uncertainty beginning at D-40 days is amazing. It’s like the temperature going down and the molecules finally snapping into place in their preferred orientation.

    Sorry if that was kind of nerdy.

  • Matt McIrvin

    While we were talking, there were some D-leaning polls in Florida, and the distinctive “spectral doubling” from yesterday disappeared.

  • Rick L

    Love the Matlab code. I went through and updated the margins for the most recent three polls in each swing state, added NV and AZ as potential swings (bumping down GOPsafe to 43), and the resulting odds of Dems keeping the Senate was 94.8% with a miniscule, but not 0, bar at 57 Dem seats. Fun stuff!

  • Rick L

    … but of course adding marginally-leaning R states and not also marginally-leaning D ones introduces bias. I reran with all 14 senate races that have a candidate with <10% lead, and the odds of a D Senate are (with latest polls) 82%. Distribution is more Gaussian – not a significant-looking asymmetric tail on the R side.

Leave a Comment