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


September 7th, 2016, 11:00am by Sam Wang

Nervous about recent changes in polls? As usual, don’t pay attention to single surveys. However, it is true that the Presidential race has narrowed by a few percentage points; the median of national polls taken over the last week is Clinton +2%. PEC’s state poll-based analysis will probably continue to move toward Trump for at least a week as it catches up with national surveys.

There is a great way for you to redirect your tension – to downticket races. Current polls favor the Democrats to not only win the Presidency, but also potentially to take control of the Senate – with an outside chance of taking the House as well. These outcomes have major consequences for legislative and political action in 2017-2018. Under these conditions, where should you put your energy?

The first answer is: close Senate races. PEC calculates the influence that an individual voter can have (see The Power Of One Vote on the right), based on close polling margins and highlighting states of smaller population. In the top states, shoe-leather get-out-the-vote makes the largest impact. The closest races can also be found by clicking the ActBlue site at left (the Republican site, the NRSC, lacks this capability).

The second answer is: the House of Representatives. Republicans are highly likely to lose seats. But will they retain their majority? If they do, there is preliminary talk on the right of impeaching President Clinton in advance. Such a remarkable act of polarization would probably slow down legislative progress, to put it mildly.

I have previously estimated that to have a chance at taking the majority, Democrats would need to win the national popular vote by 6-8 percentage points. This is hard but not impossible. It will come down to close Congressional races such as New Jersey’s 5th Congressional District, which trends about 4 percentage points Republican compared with the rest of the nation. It would be cool to have a web application where one types in a ZIP code, then gets back information about the nearest close Congressional race as defined by the Cook Political Report. Anyone?

Tags: 2016 Election · House · Senate

49 Comments so far ↓

  • Robin Colgrove

    As a citizen of Massachusetts, I deeply regret _not_ going up to New Hampshire to do door to door Canvassing/GOTV in 2000, when it was clear in advance that the election would be close (I had no idea it would be decisive!). In 2016, I hope to make up for this and work to support Maggie Hassan, since Senate balance may tip by a single seat one way or the other. I have also been trying to convince my, um, comrades on the academic Left that -as Sam writes- shoe leather is the most potent use of our resources. I have read that poli sci data back this up, that door-to-door canvassing yields larger effects per unit of labor than phone, mail, or e-mail, but I would like to be able to cite specific references from peer-reviewed literature, to be most convincing to my tribe (data geeks). As a biologist, I don’t even know where to start to find high-quality research on this topic. Are there good published references I can cite here to encourage people to get up from their keyboards and ring some doorbells?

  • Emigre

    Your “tribe (data geeks)” might really enjoy this one:
    Political Analysis (2010) 18:297-315 doi:10.1093/pan/mpq008 Baseline, Placebo, and Treatment: Efficient Estimation for Three-Group Experiments.

    This article shows the potential advantages of a three-group design (baseline, placebo, and treatment). We present a maximum likelihood estimator of the treatment effect for this three-group design and illustrate its use with a field experiment that gauges the effect of prerecorded phone calls on voter turnout.

  • Charles S

    I’ve been frustrated by the lack of an application that allows the user to find the nearest contested House race too! has a map of contested House races, but it isn’t zoomable.

    I did create an ActBlue contribution page for all the contested House races currently held by a Republican:

  • Anthony

    To answer your question Sam, yes I am nervous. Hopefully the upcoming debates are a complete disaster for Trump. Even Trump at a 7% chance of winning causes me to lose some sleep at night. If the polls continue to tighten there may be some serious health consequences for me.

  • Sophia

    Sam, I would really love for you to talk about the brain and how it works in regards to politics. I would guess that the undecided young male age 18-26 would be more likely to go risky and vote Trump than the undecided mom age 27-45 regardless of race. Also, it seems that Trump speaks directly to those who are in the fear part of the brain that lives in fight, flight or freeze. Being poor will put you there. Help me here. You are the the expert. Also, how many undecideds are male or female and do we have an idea of the age range? Shouldn’t we know more about the undecided voter since we are entrenched in our polarization?

    • Mikey

      Hi Sophia, I think the Reuters Polling website is great, you can really drill sown into the demogs you are interested in- apologies if you already know about it. Anyway, I think the premise about question re:mums not supporting Trump is not borne out by the polling data. Hillary currently has a 19% point lead over Don overall among childless 30-49 year old women of all races, but has only 2% points among moms in that age group. If we control for race, as you suggest, Donald leads by 12% points among young white moms, Hill leads by 13% among the childless 30-49 year old white women.,RESP_AGE:-9|-3,Z8A:2

  • Terry

    Needed this today.

  • Canadian Visitor

    Dr. Wang: Is there a reliable statistical model to forecast turnout in US presidential elections? To ignorant and naive foreign observers, the fall-off from ‘registered voters’ to ‘likely voters’ to ‘actual voters’ looks like the principal risk for candidates who lead steadily in polls. Any light you or your learned community of commenters can shed on this aspect would be greatly appreciated. Thank you.

    • Sam Wang

      In Presidential election years, pollster models do fine on average. In midterm years when turnout is low, not.

    • KR

      When (if) will you look at House expectations with priors?

    • Sam Wang

      I probably won’t, mainly because the relationship between votes and seats is tenuous. I think a good approach will be (a) tracking the generic Congressional, and maybe adding random drift to it; and (b) adding up race-by-race totals based on sites like the Cook Political Report.

    • Kevin O'Connell

      That is particular evident in this week’s polls. For instance, the CNN poll showing Trump +2 in Florida among likely voters show Clinton +2 among registered voters. Ipsos/Reuters is more dramatic, a national tie in a four way race in the likely voter screen becomes a 41-35% Clinton lead among registered voters. WaPo was similar, though Clinton led in both, +5 among likely voters, +10 among registered voters.

      This result is particularly weird this year, given Clinton’s growing support among college educated voters, as college educated are VERY likely to vote.

      Given Clinton’s growing strength among the college educated, any thoughts as to why the likely voter/registered voter numbers aren’t closer?

  • Sharon Machlis

    I coded a proof-of-concept demo app — enter an address and you can see competitive races within up to 100 miles.

    • Olav Grinde

      That’s awesome! Worked like a dream.

      Perhaps in a later version there can be an option for distances > 100 miles?

    • Sam Wang

      Wow, that is great.

      Do you know if that platform can conveniently add other information: labels on the map, text, and links?

    • Amitabh Lath

      Impressive and fast!

      I just used it to find NJ-05, held by Scott Garrett. I remember going down to DC with a Princeton physics guy at the time of the higgs discovery, and Garrett was the only NJ rep who would not meet with us, because we weren’t from his district. I’m not surprised that he is on the bubble.

    • Sharon Machlis

      Thank you, Dr. Wang. It looks like labels can be added with markers, such as this
      One would first need a list of the Congressional Districts with latitude/longitude of a central area.

      Links and brief district info for each district are available now by clicking on a district, although that may not be initially obvious to users. More info can easily be added to those pop-up windows, I kept it very brief for now since I didn’t want to take & use too much information from the source.

      And thanks, Olav Grinde, distances > 100 miles could be added as well.

    • Sam Wang

      Let me get right on that with latitude/longitude…

      Information from the source: you mean not copy material that was created by someone else? If so, perhaps Wikipedia/Ballotpedia resolves that. I was just thinking a few convenient links so that people could click through without too much effort.

      Distances: maybe 200 and 500 miles, the last being mainly to allow the user to generate a convenient map of, say, competitive districts in the Northeast.

      Also…I just emailed you! -Sam

    • Sam Wang

      …here we go. Courtesy of PEC reader Lyndon Estes, Congressional district centroids. They are listed by state number, not state name. Here is a list of state codes.

      What about this 3-line format when a CD is clicked?
      District: NJ-05 Wikipedia Ballotpedia
      Democratic candidate: Josh Gottheimer
      Republican candidate: Scott Garrett (incumbent)”

      …or if that is too complicated, maybe just the first line.

    • Dave James

      Just because people tend to use apps more if they can be lazy, I suggest the option of zip code only.

    • TJ Baker

      This is a fantastic tool, Sharon! Well done!

    • Sharon Machlis

      TJ – Zip codes to be added

  • truedson

    Not so much nervous, but similar to what Kevin Drum wrote today….

  • TJ Baker

    I’m not nervous as much as I continue to be seriously baffled that a man such as Trump can be so competitive and a real potential to be POTUS.
    I get that we live in highly polarized times, but good grief!!

    • Anthony

      This just finally proves that no matter who the republicans nominate they will at minimum get 43% of the vote. That is the absolute floor.

    • DebbieR

      I’m working overseas at the moment, it’s not fun! People are at me everyday to talk them off the ‘Trump ledge’.
      There is an overwhelming sense of bewilderment and palpable fear at what is happening. One thing is consistent, confusion at how on earth, given the level of ignorance and infancy demonstrated by Trump, he is getting the level of support he is getting.
      It is an embarrassment.
      People are losing their minds over this, that I can tell you!

    • Mikey

      I think Trump would be crushed in the polls if his opponent was the Barack Obama of 2008, or the John Kerry of 2004, or Al Gore of 2000. Don is only in with a chance because the Democratic Party managed to pick a candidate who could rival him in unlikability. I know a lot of commentators here are enthusiastic supporters of Hillary and will not enjoy or agree with that opinion, but the reality is most other people do not like her and think she is dishonest and corrupt- and the more people see her, the more they dislike her. I’m not sure why Hillary’s Favourablity rating aren’t showing up on the feed to the left, but if you click on the Huffpost link you will see she is strongly disliked.

    • Sam Wang

      No, you are probably wrong. Any Democratic nominee would likely be demonized the same way. That’s the polarization I have been writing about. Maybe you don’t remember what was said about any of those people.

  • 538 Refugee

    This might help with the nerves some. ;)

    “Nearly a dozen GOP strategists said in interviews this week that they are taking steps to buttress Republicans in strongholds where private polling shows Trump — his recent uptick in public surveys notwithstanding — is threatening to drag down candidates running beneath him.”

    • Veronica

      I read that this morning. Yeah, I’ve known that Hillary has a very good ground game: 50-state strategy! That’s really smart on her part. Not only does she increase her odds of winning, she also sends the message to all of us that we count, even those of us in red states. That’s reassuring. And I did know that her poll numbers were expected to go down some. She’s still in the lead. That’s good, but still, I’m a little worried, pretty much like most other people who want her elected.

  • Josh

    ” As much has the Clintons are hated, it appears that the incumbent is also moving the needle.”

    ” Any deviation or variation from the President’s approval rating, is variation on account of the Sec. recent media driven scandals.”

    Maybe I’m reading these statements wrong, but don’t they directly contradict one another? The first implies that Hillary’s success has been dependent on Obama, while the second argues that the changes in Obama’s favorability are a result of how people see Hillary. And isn’t it also possible that the close tracking of Obama’s favorability and Hillary’s lead is simply correlation? After all, there’s a tremendous overlap on the Venn diagram of Obama supporters and Hillary supporters…

  • Michael

    Is there reliable data on the effect of debates on polling and voter preferences? Emphasis is always put on the first debates, and in 2012 Obama’s performance there was lackluster, at best. But, he was already president. Is there greater effect in an “open” election?

    • Gopa

      Yes, after the first debate, the gray region or the lower confidence bound, dropped well below the 270 Redline. Andrew Sullivan had a near meltdown after the first debate. It pulled back up after the second and third debates. In the present scenario the lower bound is way above the Redline. Let us hope it stays that way.

  • Josh

    @Gopa One of the problems I have with your hypothesis is that the Meta-Margin lags national polling, sometimes by several weeks, so if discrepancies in the MM and Obama’s favorables really were solely the result of Hillary’s perceived negatives, that would mean people weren’t discounting Hillary until weeks after those “scandals” broke.

    Frankly I just think you’re overthinking this. Hillary and Obama’s support comes from basically the very same group of people. If there are small, sudden jogs in Hillary’s Meta-Margin that cause it to deviate from Obama’s favorables, it’s probably just noise.

  • C Hart

    I like everything you have to say, except your endorsement of MATLAB. The day I left it for Py was the day my soul was set free.

  • Than

    Can someone explain the pretty big drop in projected EVs for Clinton in the past week, and if it mainly stemmed from the oft spoken Q Poll? Also, why is PEC’s header showing 328 (currently), while 270s is 305? Lastly, and this may be asking for too much editorial, but how is it possible Florida and Ohio could seriously go Trump?!!?

    Thanks for all the great and intense work and insight!

  • Samrat

    I am stealing from the great Rumi, to express my consternation:

    “I have lived on the lip
    of insanity, wanting to know reasons,
    knocking on a door. It opens.
    I’ve been knocking from the inside.”

  • Michael

    How do I ask you, without asking you, if there’s genuine reason for Democrats to freak out?

  • Howard B

    Admittedly, a novice question: might the concept of regression toward the mean in part explain why Clinton’s numbers dropped a bit?
    From an intro statistics class, of course.
    Does it have applicability to polls?

    • Sam Wang

      To my ears, the phrase “regression to the mean” has always been a bit opaque-sounding. But basically yes.

      The voting public today is composed of the same people as one month ago, or two months ago. There is a limit to how much their opinion can change, especially under extremely polarized condiitions. So movement is confined to a little band that corresponds to about 2% of voters changing their minds – or small differences in how likely Clinton and Trump supporters are to respond to a survey.

      If the Meta-Analysis gets away from the average, it is more likely to move back toward the average than away from it. That’s regression to the mean. it is the basis for the Bayesian prior of my predictive model.

      History of Popular Meta-Margin for Clinton

  • bks

    Apparently the black swans are carrying pneumonia this year. Now I’m worried.

  • Kevin King

    Hey, is there a place to look at the history of the 2004, 2008 & 2012 meta margins? I looked around for a few minutes but can’t seem to fin them, if they exist on here.

    Thanks, All!

  • Gopa

    @Josh :Maybe I’m reading these statements wrong, but don’t they directly contradict one another? The first implies that Hillary’s success has been dependent on Obama, while the second argues that the changes in Obama’s favorability are a result of how people see Hillary.”

    No the two are not contradictory, neither are they mutually exclusive. In an sum of variance, you can partition variance on account of different factors. So if the Clinton Meta-Margin is a summation of variance on account of Clinton Legacy and Incumbent Popularity, then my theory is the overall shape of the curve and trajectory matches the POTUS approval rating, while any mismatches between these two lines is attributed to Clinton unpopularity. Then there are numerous other factor are not being counted media bias, systemic erros, random error. Those two circles of the Venn would nearly lie on top of each other, any jitter is on account of the sources of error that I enumerate above. Now if we take a pie-chart instead of a Venn, my take is the biggest slice of the Pie would go to the POTUS rating, the next to the Clinton legacy and the smaller wedges and slivers to other factors. I am turning a blind eye to white racism. If you know how to factor that in, then you hit the bullseye!

  • Gopa

    I find that if you overlay the graphs of the Presidential Approval Rating and Sec. Clinton’s meta-margin, they take the same trajectory and mirror each other. As much has the Clintons are hated, it appears that the incumbent is also moving the needle. Any deviation or variation from the President’s approval rating, is variation on account of the Sec. recent media driven scandals. I think that the POTUS, FLOTUS, and VPOTUS can define the race. They should go to bat for the Sec a little more. Pres Clinton should do more. It is about time the server issue is put to rest. It is imperative that the Clown is called out on his support of Putin which borders on sedition. At any rate this is the only analysis that keeps me sane, even if it makes me nervous. I learned that in the past two elections. The fact I use MATLAB and a neuroscientist redoubles my trust in the methodology and results. Cheers!!

Leave a Comment