Princeton Election Consortium

A first draft of electoral history. Since 2004

And Then There Were Four (or Five?)

September 23rd, 2014, 1:06pm by Sam Wang


In the last few days the Meta-Margin’s been bouncing around. It’s a snapshot of current conditions, and will probably keep moving up and down. That bounciness – so vexing to some of you! – is part of how I am sampling day-to-day variation. This is the principal input to our predictive model for the November election.

For a more long-term look, see my new piece in The New Yorker. It details how the battle for Senate control has begun to focus strongly on four states (Arkansas, Colorado, Iowa, Louisiana) and maybe a fifth (Alaska) – and what activists are likely to do to sway the outcome in those races. Those activists might also be interested in the NRSC (Republican) and ActBlue (Democratic) links at the left.

Update: Back in July, I wrote a piece here at PEC with a very similar title. I am interested to see that three of the four states have not changed, though Kentucky (then) has been replaced with Arkansas (now).

Tags: 2014 Election · House · Senate

26 Comments so far ↓

  • Stuart Levine

    Sam–I just noticed that in one of the “banners” at the top of the page, you now omit the likelihood of “Probability of 50 or more Democratic+Independent seats if election were held today” and only have the probability as of election day. While to a casual reader who doesn’t understand the concept of probability (e.g., most newspaper reporters), this avoids some confusion. However, I was wondering what factors went into your decision to omit this probability point.

  • Amitabh Lath

    As for unskewing, you are correct not to. It is not an aggregator’s job to correct for pollster errors. The mechanism causing this bias needs to be understood by the pollsters themselves, and corrected for properly. I know Gallup had initiated some such effort internally after their 2012 debacle, but there must be venues like conferences and workshops where analyses like these can be presented.

  • Amitabh Lath

    Thanks! I didn’t expect to create homework for you.

    This table is simply amazing. I would have assumed that any systematic bias would get rubbed out as pollsters adjusted to their mistakes from previous cycles. A persistent effect like this says that pollsters lean Republican because there is no penalty to pay.

    • bks

      There’s no penalty to pay for the margin. It’s only when you get the winning candidate wrong that your organization gets pilloried — take Dean Chambers of UnSkewed polls, please! –bks

    • Craigo

      I think there’s a red queen effect in play – pollsters have to run as fast as they can just to keep up with the changing electorate. Or alternatively, they’re always fighting the last war.

  • Amitabh Lath

    Sam, have you considered that polling systematic effects may not be symmetric? In other words, most of the mistakes pollsters might make in sampling, screening, weighting etc. would lead to underestimations of Democratic-leaning populations.

    I ask not only because outfits that get it wrong (Gallup, Rasmussen) tend to err on the Republican side, but also current outliers (like the recent Quinnipiac polls) spring up predominantly R as well.

    • 538 Refugee

      I did some reading on PPP. They tend to be pretty accurate, but when they miss, they also do so in the Republican’s favor.

    • Amitabh Lath

      Yeah, asymmetric biases can crop up anywhere. My thesis experiment was one such. All possible mistakes went towards making my measured value smaller, none made it larger. I can imagine a similar problem with estimating likelihood of Democratic voter turnout.

    • Sam Wang

      This indeed did happen in the last four elections. Democratic outperformance relative to polls, using mostly* median-based methods, comes out as follows.

      Election Outperformance (mean±SEM)
      2004 Pres. (nat’l) R by 1.4%
      2006 Senate D by 2.3±0.7%*
      2008 Pres. (nat’l) D by 0.2%
      2008 Pres. (state) D by 2.4±0.9%
      2008 Senate R by 0.5±1.1%
      2010 Senate D by 2.6±1.2%
      2012 Pres. (nat’l) D by 2.4%
      2012 Pres. (state) D by 2.9%
      2012 Senate D by 2.3±0.7%
      Average D by 1.5±0.5% (SD 1.6%)

      The Presidential state-level findings come from comparing the final Meta-Margin with the margin of victory in the tipping-point state, i.e. the state that pushed the winning candidate just over the electoral-vote threshold for victory.

      The last item, 2012 Senate, is something I missed before, I think because I was looking for missed calls. Anyway, it’s a fairly impressive string of similar biases. You intrigue me, old bean.

      I still can’t unskew anything. No way. One big reason is that it isn’t consistent over time. In addition to the above data, I am pretty sure it was less than 1% in either direction in President-2004 at the state level. However, I could still revive the D+2% and R+2% bias calculations for the sidebar.

      *done using the RealClearPolitics averages – it was getting to be too much work for a comment.

      …and here is the House data, a lot messier.

      2004 House R by 3.6%
      2006 House R by 4.1%*
      2008 House D by 1.2%
      2010 House D by 2.6%
      2012 House D by 1.3%
    • 538 Refugee

      If you are going to do GOTV it would make sense to push absentee to the folks that are not committed to making it to the poll on election days. Voters egistered to your party but only get out for presidential elections. Absentee seems to have expanded quite a bit from a decade ago. People who work all day may not want to go stand in line. If I were cynical I’d say the stories of long lines we see in some places are simply vote suppression efforts. This might be at least some of the difference you see though. I’m not sure how hard it would be to see if absentee has shifted in a partisan direction.

    • ArcticStones

      Sam, that is an intriguing exchange – and I am fascinated by the historical data you just shared. In my opinion this deserves a full-fledged PEC post!

  • WDR

    Good article, but I found the tables small enough to be difficult to read, with no larger version available.

  • SFBay

    I have several friends in Alaska as it happens. Unfortunately for me they are card carrying members of the Tea Party.

  • Shaun Guil

    In the end I do believe the Democrats will hold both Colorado and Iowa. However, I would like to know what is going on in Georgia? There is not enough polling from that state and it appears to be close.

  • SFBay

    Thanks for the update. I still have irrational hope that the Dems will surprise us and hold on to 52-53 seats. Then again, wishful thinking is not a particularly good prognostic tool.

    • Sam Wang

      I would be pretty surprised by that outcome. Then again, if you personally go turn out a few hundred voters in Alaska…you could help make that wish come true.

    • ArcticStones

      And that, Dr Wang, is precisely the wild card. There will most likely be a highly effective and well-targeted GOTV effort on behalf of Begich –and Democratic candidates in other toss-up states.

      The question I ask myself is: what is the upper limit for the potential of those GOTV efforts?

  • Lewis

    Based upon the last election in 2012, the Democrats did an exceptionally excellent job in the so called toss-up states. I know that the Republicans have rolled out their voter suppression tactics. Ohio and Florida were great achievements. So I am anticipating that the polling models will underestimate the Democrats going to the polls. I look forward to Rove witnessing the “impossible” results again.

    • Matt McIrvin

      2012 was a Presidential election year, which brings an entirely different electorate to the polls. And polling didn’t particularly underestimate the Democratic vote; in fact, it did extremely well at predicting it, if taken seriously.

      In the midterm election of 2010, polling generally underestimated the size of the Republican wave. This year is unlikely to be a giant wave like that one, but the electorate will probably skew older, whiter and more conservative, because that’s the way it generally goes.

    • Sam Wang

      Are you sure those statements are true? See this response from earlier. The Meta-Margin ended 2012 at D+2.5%, yet the borderline states were Pennsylvania and Colorado at D+5.4%.

      In 2010, the last generic Congressional polls showed R+9.4%, while the actual popular vote was R+6.8%. It is true that Republicans gained more seats than expected by most analysts, but one would have to drill into that a bit to understand it.

      In both of these cases, Democrats outperformed polls.

    • RAJ

      Ohio, Florida in 2012 , Virginia in 2013 and Mississippi primary in 2014, Black voter turnout was the difference. Nothing motivates any demographic group more than attempting to take away their right to vote. Black voter turnout in toss-up Senae races will keep Harry Reid as the majority leader.

  • Joseph

    Nice article, Professor!

  • axt113

    PPP released a poll showing Begich down but with a high amount of undecideds, hopefully most of those undecideds will break for Begich

    • mediaglyphic

      Dr. Wang, what do you make of the PPP alaska Poll. 538 has a rant about why the other polls are biased and how their model gives less weight to biased polls. Any thoughts?

    • Sam Wang

      He might have a point, but I don’t think it’s because of partisan polls. I think it has to do with the fact that in any given year, 10% of Alaska’s population is either new arrivals or about to move out of the state. How does one survey that? It reminds me of Nevada in 2010, when polls missed the Reid-over-Angle margin by nine points. If you look at polls that use internet or internet-plus-phone sampling, Sullivan seems to be slightly ahead.

    • mediaglyphic

      why does “The Power of Your Vote” table have Begich with +5% (that seeems to be an old number)? I am sure i am missing something.