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

Midterm traffic – thank you

September 27th, 2014, 9:36am by Sam Wang

Some of you are asking if the PEC polls-only prediction is wrong. I say no: the prediction is OK…so far. 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. However, we’re only a touch below the range I have stated: R+0.2% to D+2.5%. We’ll see very soon. Also, next week, as planned from the start, the PEC model will start incorporating an element of “random-walk-from-the-present.” At that point, any inaccuracy will tend to fix itself – like the other predictive sites.

In the meantime…thank you for reading, both here and at The New Yorker. The Princeton Election Consortium is getting excellent midterm traffic! We’re not far below the level for August-September 2012, which is pretty good considering it’s not a Presidential year.
As you might guess, last week’s dust-up created more traffic…though not that much direct linkback from our friends at the ESPN subsidiary.

In the meantime: watch the banner above, ignore isolated polls, and use the right sidebar…including ActBlue (Dems) and the NRSC (GOPers)!

Tags: 2014 Election · Senate

37 Comments so far ↓

  • Bert

    I’ll say this again: Mississippi could be a surprise this November if the Tea Party doesn’t fall in line and vote for Cochran. I don’t expect Childers to win. But look at the current poll: Cochran at 46, Childers at 31, and undecided at 15. Imagine if that undecided were to vote Childers just to spite the GOP establishment? It’s a long shot, but it might be an outlier to watch for. The anger at Cochran among conservatives in MS is still palpable.

  • Insidious Pall

    Like most other political junkies and neophytes, I tend to troll around the various poll aggregation sites. PEC is easily the most fun. I check the Meta-margin daily, and the discussions and presentations for the truly hardcore are nonpareil.

  • RB

    The tea party would not vote for a Democrat just like a Green would never vote Republican. Some may sit out, but there is no shot at MS flipping this year(especially with a Generic R+4 ballot nationally). If I was a Democrat I would be far more concerned with NH and even MI before MS.

    • Bert

      I’m not suggesting Dems divert funds away from NH to MS. I’m just saying keep on eye on MS because it’s interesting. McDaniel still hasn’t endorsed Cochran and numerous commenters on his FB page say they are voting Childers just to spite Cochran. We’ll see if that’s just empty talk on election day.

  • Joseph

    I just summed up the “Powers” in your The Power Of Your Vote panel, and came up with 209. Assume that is dollars, and that you want to donate about that amount. What about donating, roughly, in percentage of the “Power”. Thus, you’d donate $100 (or $100,000) to AK, $33.50 to AR, $24.60 to IA, etcetera.

    You could call it the PEC Power Percenter or something….

  • A New Jersey Farmer

    The undecided vote in the states that hold the balance will be key. I find it interesting that with all of the media saturation people could be undecided this late in the race. Only Cassidy in LA is close to approaching 50%.

  • MarkS

    I’m so confused. The snapshot makes a big move, but the election-day prediction doesn’t move at all. Is that because it’s rounded? To the nearest 10%? I’m all for not giving overly-precise numbers, but an explicit error-bar quote would be helpful.

    And next week you’re changing your extrapolation method? What’s up with that?? It really would be much better to stick with one method per election, and only tweak it in the off-season.

    • Joseph

      My guess: A “big move” in the meta-margin is still a snapshot in time. “Election Day Probability” is looking at the trend over many such snapshots. As regards “random walk from the present”, I suggest we wait and see what that means before questioning it.

    • Sam Wang

      No, you are not understanding. It was the same method all along: long-term plus short-term, once the short-term becomes predictive. Go read the older explanation.

  • RB

    Mark I am no computer expert, just like politics-albeit from the right-but I think Sam is using polls that go back into the summer which are RV polls. He said he will put more weighting on recent polls next week. But the movement is clear on RCP. Go back and look at the generic ballet question in Aug using RVs and look at Sept using LVs. I suspect once the weight is added the long term predictions will point to a narrow GOP win. But look at the big picture. It appears from where I sit KY-MT-SD-WV are gone. AR-LA-GA-AK appear to be almost locked down. Watch for the movement in KS in a week or two for Roberts. Voting for an I might sound good initially, but if the KS voters get the memo that Orman will caucus with Reid that race will change drastically(the Roberts advertisements are flooding the system as we speak). CO-IA look like dead heats, but have trended R of late(that could change in IA when the Selzer poll is released today). Scott Brown is making Shaheen work in NH.

  • RB

    DeMoines Register/Selzer—Ernst 44 Braley 38…they are one of the best in the business and should be a clear marker that Ernst is in the lead in IA…Selzer is the ‘gold standard’ of IA… I suspect Braley will release an internal… Treat it for what it will be.

    • Lockdown

      There are major red flags with the DMR poll, though. It has Ernst with an absurd 58%-15% margin among “rural” voters, and Braley losing his own district when the Democratic nominee in IA-01 is doing fine. The latter assertion seems totally ridiculous, and it would take a margin of victory similar to Chuck Grassley’s 2010 win for Ernst to be winning rural voters 3:1.

  • Bruce Sands

    Hello Dr. Wang
    Do the polls reflect the get out the vote effort of the parties? Or is this effect hidden? If effective, how does registering new voters, and getting more registered voters to actually vote impact the results? And do we see this effect in polling?
    Thank you
    Bruce Sands

  • RB

    Lockdown- unskewed polls did the same thing you just did in 2012-the numbers are the numbers. We can ‘t dismiss DMR and Quin cause you don’t like the results. BTW the Farmer comment by Braley may be driving the rural #s

    • Ditto

      What’s the explaination for Braley losing his own district?

      I also question polls this cycle. Not bc of partisanship. But bc they don’t make much sense when you dig into the assumptions of those who bother to provide them. We also seem to have fewer polls this cycle so that adds to my wanting clarifucation about the ones available.

      I forget the state, but it was one where even in 2010 Black turnout was higher than what the likely voter model in polls are predicting for 2014 in the same state.

      Someone asked how can we trust these assumptions, and, then, as now the “don’t unskew the polls” was thrown out.

      I don’t think there is anything wrong with questioning likely voter models, especially if they don’t reflect long term voting patterns. I don’t consider a single election that happened to be a wave (2010) a good basis for making thr same assumption for 4 years later.

      I don’t see the diffetrnce between these assumptions and secret sauce.

      I wish Professor Wang would discuss the polling assumptions. Ultimately, the polls may be righy but I don’t think its a bad idea to question them.

    • Ditto

      Here’s cook discussing likely voter models

      In 2012 many did not predict the 4 point spread between Obama and Romney


    • Ditto

      I am so fascinated by the subject and how pollster likely voter models always seem questionable that I looked up 2012 at about this time, and arguably the best year for the Democrats, and sure enough major likely voter models had Romney ahead of President Obama. In my Google search I found this;

      Sure Daily Kos is a Democratic site, but that doesn’t make them wrong on the data.

      Can you explain this?

    • RB

      The Q poll and the DMR poll are showing an Ernst lead. PPP is rumored to do the same Tues. From what I am reading many on the left are now distrustful of the polls. In Sam’s twitter we have people bashing the FOX polls without realizing they are considered good pollsters with a D and R pollster conducting their surveys. In CO I read how the Q polls were just wrong and now Gardner has led the last 4 polls. Just saying…

    • RB

      BTW Ditto-Go back to 2012 and look at Unskewed polls-That is exactly what they did-they went through the demographics and party ID and adjusted the polls to what they thought the party ID and demographics should be(usually from previous elections)…as I said the numbers are what they are. From my standpoint, I may not like the current NC numbers, but I do believe what they are showing. Obviously these races are fluid and things could change in states like NC -CO -IA. But if I had to guess today Hagan-Ernst-Gardner wins.

    • Ditto

      If you are simply going to dismiss me without answering the the criticism and questions, there’s no real discussion.

      I asked you a very specific polling question: the likely voter models were off in terms of the Presidentual race 2012 in that they seemed to favor Romney more than the actual out come.

      The unskewing you refer to would have increased the error.

      There is a lot more data for building Presidential likely voter models than for Senate races.

      So why the error?

      You dont answer other than to miss what 2012 tells us

      Presidrnt Obama did not swing 5 to 6 points in the last month of the election. Yet the polling data would have us believe he did by the time hd beat Romney by 4 points.

      If you don’t know the answer to my concerns about likely voter models that is okay. But please stop dismissing them as some kind of partisan rose colored glasses.

      The point is simple: even where democrats are performing well these models find us losing like they did in 2012. Again- why? What is it about the assumptions that causes them to be off?

      You don’t need to unskew anything to answer my concerns about bias. For example, I would be willing to buy your response about the rural numbers if the same data wasn’t saying soneghubg really strange about his own district

      I want you to explainn without dismissing the question.

      By the way, what I think is happening is that the voters don’t much like either part ly this election cycle

      Not exactly favorable to the Democrats if it is

      Whereas the pollsters seem to be arguing through their models that this is more of a wave year than 2010 (which is what one would have to assume in claiming blacks will vote less than 2010 in /014).

      I just want an understanding of the assumptions being made rather than to have you erroneously try to read my mind.

    • Sam Wang

      Play nice! I think you’re both saying something interesting.

    • RB

      A few points-There actually was a swing in the last election. When Romney won the first debate there was real movement. When he was unable to repeat it in 2nd or 3rd debate the movement slowly reverted back to pre first debate. Also, with Sandy hitting so late in the cycle it is very difficult to measure the effect that it had in 2012(I suspect it helped Obama)—with regard to being down in his district—In my home state of NJ Christie won with 60% of the vote, but has never led Clinton in a Garden State poll(even before the bridge)—It happens. Braley has been a poor candidate-He had the ‘farmer’ comment and his petty lawsuit with the chickens have absolutely taken their toll, especially, I suspect, among rural voters.

  • Bill Mason

    I check your meta-margin chart several times a day. I also download daily the first 18 lines of your file Senate.polls.median.txt to check the individual state races.

    I have a couple of questions. In your paper “Origins of Presidential poll aggregation” you say that for this paper you redid your calculations using the t-distribution to replace the normal distribution.

    But it looks like you have reverted back to the normal distribution for the 2014 calculations. Specifically, in line 11 of senate_median.m you use matlab’s “erf” function, which I think is the normal distribution, to calculate the Democrat win probability. That win probability is used to calculate the seat distribution probabilities in lines 22-26.
    Why do you use the t-distribution in your paper but the normal distribution for real time calculations?

    Also in the file Senate_November_prediction.m you initially use a “black swan factor” of 1 for the degrees of freedom in “tcdf”, which I think is the t-distribution function. But for the last 35 days before the election you switch the blackswan factor to 3. Why the switch from 1 to 3 degrees of freedom?

    • Sam Wang

      Thanks. I appreciate the close reading.

      I agree the use of erf would be better as tcdf, and I will fix that. However, for a snapshot it might not matter that much since what we are doing there is gathering daily fluctuations. The big picture, i.e. where the snapshots go in time, isn’t affected much.

      For the November prediction, my thought there was that on shorter time scales, black-swan events should become less likely because there is less time for them to happen.

      There’s actually a much bigger issue: the assumption of no overall trend over time from June until now. If you look at the Meta-Margin over time, there is a real question developing at the moment as to whether this is true. I think this is the big issue.

  • Alan Koczela

    Dr. Wang,

    First, I would like to thank you for doing this work. I know you only receive “psychic” income from it. Like many, I find it very useful and fun. Your work demonstrates that polls, no matter their source, are extremely useful in predicting elections. However, I find them to be very “rough” predictors given the extreme swings in your meta-margin.

    Second, please ignore the backbiters that inevitably appear when the meta-margin flips or goes outside your predicted range. As you know, science is best described as a process striving toward the truth through the path of constant failure. If your earlier predictions do fail, I know you will make the necessary adjustments to improve the model in the future. I commend you for your selfless work.

    Please do what’s best for your model, stick with what it tells you and ignore the gutter-snipes.


  • A New Jersey Farmer

    I’ve read the discussion and looked at the polls. No matter how you slice it, the trend is towards the Republicans. As more polls begin to show that, the news will take on a life of its own. If the Democrats don’t effectively fight back over the next 10 days this could get uglier.

  • Amitabh Lath

    Multiple states (AK, IA, CO and others) all moved towards R within a day or two of each other. If the Senate races are supposed to be (mostly) independent of each other, then I find this puzzling.

    I suppose the correlation could be due to a multi-state ad blitz by some outside PACs that caused Republican-leaning undecideds to finally break for their candidates. Or it could be a couple of polling firms with conservative-leaning Likely Voter screens releasing multi-state polls on the same day.

    • The Live Toad

      538 certainly believes that state swings are correlated. This may be evidence that they are right.

  • JayBoy2k

    Things are moving, but I know neither what is driving the movement nor what this change implies for the future. I like the concept of a “polls only” approach but I am not yet convinced.
    Since we are using many of these same polls since June, they would seem to be as accurate now as they were previously.
    I am intrigued by Sam’s statement that there might be a trend in the historical movement of the Meta-Margin.
    Time to get out the popcorn. This is getting interesting. Great site!!!

  • SFBay

    Got to say, the latest polls and trend are pretty unpleasant right now. The next week will probably bake the cake, one way or another.

  • pgvaidya


    I admire your fortitude. A lesser mind would find this period very disturbing,

    I agree with Alan Koczela. What I have always liked about this site is that it is a site run by a true scientist in a scientific spirit.

    I think that your greatest contribution so far has been your prediction about the possibility of Democratic take over of the house and subsequent painstaking analysis of the gerrymander bias.

    The horse race is fun but using science to real life problem and showing an ability to adjust our thinking after deeper cogitation is what makes the full life of a scientist so much more enjoyable.

  • jd351

    Everyone just take a few steps back and Breathe. Nothing is set in stone, the keys races are still very tight. Undecided are still high in my opinion. The cake is starting to bake somewhat.

    • RB

      JD 351-I think the movement can be looked at on two ways. First switching models from
      RV to LV(this is actually not real movement as opposed to drilling down to the actual 2014 electorate). While all models vary on LV, it was clear the movement made in the Generic Ballet Question was switching from RV to LV(this happened in pretty much all the national poll models)…That being said on a state by state drill down there has been movement in IA-CO-AK as a result of individual candidate mistakes. IA has been out there the longest with Braley’s chicken lawsuit and farmer comments. In CO-Udall stated what he thought the beheaded journalist thought and then has been refusing to debate Gardner on local TV. In AK, Begich ran an ad that was widely panned and backfired on him. Those three states since the summer have had real movement.

    • atothec

      Correct. Post labor day Republican surge is par for the course. The fact that this is happening across the board actually makes me feel better vs. some surging and some failing. To me this says the repub base is consolidating meanwhile pollsters are using 2010 ‘wave’ numbers in weighing likely voters.

      It is not a wave election and according to Huffpo Iowa is on track to set a record for Democrat ballot requests.

      The wild card here is Bannock Street. But since they have been silent all year we won’t know anything until after the election.

      I would not bet against Obama’s ground game which is not being used for the mid terms. These folks are light years ahead of the Republicans in terms of voter outreach technology. Like Google level. In fact I think they have someone from Google consulting?

  • atothec

    I meant which is *now* being used for the midterms.

    • RB

      Atothec-I hear you on the ground game and outreach(which is a concern of mine-from my perspective)—that being said, I would think the polling would reflect some of those voters being outreached. People who already cast ballets and were polled should, I would think be counted as a LV. At some point (probably even now) the polls should be picking up on this hidden vote coming to the polls. So far it hasn’t materialized and a good measure of this is the enthusiasm gap which is favoring the GOP significantly this year(but not as much as 2010-of course in 2010 the GOP was playing on a very blue map in the Senate)

  • Edward G. Talbot

    As a general rule, the only time I am tempted to unskew polls at all is due to explicit partisanship or internals that seem very improbable. But the whole point of poll aggregation is to smooth some of these concerns. So I would conclude things are moving R in IA,CO,AK just like they have been moving D in NC. I wish it weren’t so, but that’s what the data is saying.

    This doesn’t mean that when the polling data has some big conflicts – like say MA-Gov in the past ten days – we shouldn’t do some non-quantitative analysis. But from an objective standpoint, there really aren’t a lot of red flags in the collection of swing state polls to suggest that we need to adjust something. And from a partisan standpoint of what action to take, if Democrats weren’t already in full court press mode in IA,CO,AK, then they were already misreading prior polls.

    I do agree that Alaska is likely to diverge more from the polls than most states, but there’s no real reason to think it will be in the Democratic direction and potentially reasons to think it will go the other way.

    All that said, once we’re a week or two further along and it is largely LV polls being considered, the fact that LV polls have been skewed towards Republicans in the most recent years (think Sam posted on this in recent days) is certainly a worthy topic for discussion of what it means for this cycle.

Leave a Comment