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

House outlook for 2013 (Take 1)

August 30th, 2012, 12:00pm by Sam Wang

(Welcome, DailyKos and FiveThirtyEight readers.)

A long-term outlook based on conditions through mid-August (pre-Ryan/Akin/convention) shows a possible loss of control of the House by Republicans (Democratic takeover, 69% probability). The 2013 seat margin is headed to being narrower in either direction than the current Congress. An election based on today’s conditions would give Democrats 50-51 seats in the Senate and ~224 seats in the House — a knife’s edge in both chambers. This picture may change.

Caveat #1: The quality of House district data is unlikely to be as good as it was for the last two cycles because polling is down. But perhaps our data source (HuffPost/ or another aggregator will know better. This is a situation where a more complex approach for filling in missing data might help.

My first estimate will therefore be based on the national popular Congressional preference. A simple inspection of election returns for 2000-2010 shows the well-known fact that the D-minus-R House popular vote share is strongly related to the corresponding margin of House seats:
National House seat margin vs. vote margin, 2000-2010 This graph shows that each 1.0% of popular-vote margin translates to a 6.0-seat advantage. Also, the intercept is nearly at the origin, i.e. no structural advantage for either side. A nearly-tied popular vote would translate to a nearly-tied House. So: 6.0 seats per percentage point.

However, note that individual data points deviate from the fit line by 7 to 17 seats in either direction. That is the first source of uncertainty, equivalent to 1-3 percentage points in national vote.

Next, let us use the generic Congressional ballot to estimate national popular vote. I will apply the Wisdom-of-Pollster-Crowds principle that has served so well in Presidential and Senate races, and assume that aggregating polls will approximate the actual popular preference.

Request to readers: Some have claimed that the generic-preference ballot does not predict seat outcomes. If you have recent quantitative data on this subject that is appropriate (i.e. that spans multiple pollsters, and surveys the last few Congressional elections), please add it in the comments section.

Here are this year’s polls. The average is a Democratic margin of 2.0%.

House generic preference Jan-Aug 2012 phone only

I have left out robopollers because uses averages, which does not allow me to easily use the median to address extreme pollster bias. The Internet polls mainly added noise, so they are left out too for clarity.

Update: a commenter took issue with the inclusion/exclusion of pollsters. My general approach, as longtime readers know, is to accept all polls. The above procedure was done strictly for display purposes, given’s software capabilities. Sadly, they do not use medians, which addresses the outlier problem. The median of all available generic Congressional preference polls since 8/15 is Democrats +2.0 +/- 2.0 % (median +/-SEM, n=7) (SEM corrected as per correspondence with Alan Cobo-Lewis, link to come).

Considering the possibility of movement in opinion between now and November and the fact that seat outcomes don’t precisely reflect popular vote, the total 1-sigma uncertainty is about +/-4%. This leads to a prediction of R+2% to D+6% (68% CI). This corresponds to anywhere from a 12-vote Republican majority to a 36-seat Democratic majority. The probability of retained Republican control is 31%, a knife edge situation.

All of the analysis above comes from data taken before the addition of Paul Ryan to the national ticket. Since then, the two most recent polls show the Democratic lead expanding to 4-8%. If that held, of course it would change the outlook. I’ll have more on this as it develops.

Tags: 2012 Election · House

19 Comments so far ↓

  • chuck

    American Research Group did an accuracy assessment from 2010 and rated the pollsters:

    1. Ipsos- internet
    2. Survey USA – website says internet
    3. Rcp
    6. Pollster
    7.PPP robo
    10 CNN
    11. Hill
    12.Rasmussen robo

    There seems to me to be an enormous unwillingness to listen to Internet/Robo Pollsters in the USA. You Gov, Ipsos etc. poll in other countries and have good records. Ditto Angus Reid who had the 2nd best prediction in the Wisconsin recall. How many more elections will it take before yourselves, TPM, Pollster etc. will consider using online? There are not enough media oulets left to fund pollsters and those non American media outfits still standing like Reuters and the Economist will fund on line polls since they are cheaper and appear to be more accurate.

    • Sam Wang

      Chuck – Read the post again. It was for display purposes, and also because Pollster does not calculate medians, which I always use. (Medians are so powerful for this application! Everyone should use them.) Your objection does not change anything.

      RCP, 538, and Pollster are aggregators, not pollsters. So am I, and I have been doing it since 2004. However, thank you for your input.

  • Chris R

    Sam, first time writer, long time reader:

    Do you factor in something like the Democracy Corps poll of swing districts (I don’t think it conflicts — in fact, they’re coming to a similar conclusion, or were as of a month ago) in the generic preference #s or is it apples/oranges?

    • Sam Wang

      Chris R – thank you for writing. Everything is in the post. I try to avoid “factoring in” anything, unless it is statistically simple to handle and has multiple sources to allow aggregation. The methods are as close to transparent as possible, i.e. you could replicate most of the work in 10 minutes yourself.

      In this case the generic Congressional preference poll is a simple quantity measured by multiple organizations. Swing district data would be great, and may be available from commercial aggregators in the coming weeks. We rely on Mark Blumenthal and Co. at Pollster for our data feed, and I read RCP often.

  • Sam Wang

    BigAngryBubba – Glad to see you back, Representative Akin.

    I agree, cysts should not hold office. Sorry, I mean hole office.

  • LondonYoung

    It is strange that the “seats vs votes” graph passes through the origin given the R advantage in Cook PVI. In short, city districts are nearly “pure D” zones, while the ‘burbs are just “mostly R” – thus at 50/50 vote the R’s should control the house 239 to 196.
    I can’t imagine the decennial gerrymander did anything but increase this effect.
    But the data is the data, and I see nothing wrong with your approach.

    • Sam Wang

      LondonYoung – I agree that it is notable. But there’s the graph. The Wikipedia link goes to data.

      It seems likely that gerrymandering is good for protecting individuals, but hard to get to work for all conditions of electoral tides rising and falling.

      I am most interested in the reasons for the deviations. These may arise from reasons like what you cite. Note many candidates run unopposed. Large-N statistics are always good for linearization, wot?

  • Pat

    This could be an interesting source basically arguing that a tie in the popular vote does not equal to a split house, but would tend to favor the incument majority.

  • LondonYoung

    Sam – Don’t even question linearization aloud or you may destroy all of what little math exists in the social sciences … And never ever ask a social scientist “what are the assumptions underlying the validity of the standard formula for a linear regression”.

  • Olav Grinde

    So, uhm, what are those underlying assumptions?

    • Sam Wang

      It’s all in the post. Median and uncertainty fed into t-distribution for probability, same as EV analysis for single states. Uncertainty as described in post. 6.0 seats per percentage point margin. Recent polls or all 2012 polls give same answer.

      That’s it.

      The hardest parameter is how to account for asymmetry in the distribution of the amount/direction of movement between now and November. I am concerned about this parameter. For this reason the seat margin is a rather hard thing to estimate.

  • LondonYoung

    Ebenezer – one has to love that graphic in the very next section “Interpretation”.

    In the case of the house, we know that at 100% in either direction the favored party has exactly a 435 vote advantage. So we know for sure that a 6 vote per point model must break down somewhere!

    But with Intrade at 90% for R’s to hold house, and Sam’s model looking quite reasonable, I would save all my “Obama to win” money and put in on “Dem’s to take house”.

    • Sam Wang

      I am thinking about such things. My advice is to wait a little. In a few weeks we’ll have clearer trends. InTrade prices won’t move that much. Also we must design an arb…

    • Sam Wang

      LondonYoung, just looked at InTrade. I see some Pres/Congress bets that are appealing as well. I’ll have to chew this one over.

  • San Fran Sam


    To what extent are the voter suppression, er, ID laws going to affect the outcome of either the Presidential and House races?

  • LondonYoung

    Each of the intrade contracts is just ten bucks notional, so one would hope that some grad students could both put a little money in their pockets while at the same time improving the market observable probabilities for the rest of us.

  • jonah gelbach


    Great site and all that.

    Maybe someone has mentioned it, but not that I saw. Aren’t years ending in 2 likely to differ structurally from other years? To wit, don’t we have to worry about redistricting here? Don’t know if the data exist before 2002, but it seems it would be useful to find a way to adjust for party control of state redistricting processes. Perhaps you could do that at the national level by just counting the number of house seats subject to GOP control over redistricting. Then you chew up only one df…..

  • Pete

    Where do I find the listing of the current polls you are looking at?

Leave a Comment