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

Senate Polls Meandering Toward Presidential Race

October 14th, 2016, 9:27am by Sam Wang

The last significant movement in the Presidential race occurred after the first debate. Here are national polls, unraveled day-by-day:

I calculated this for each date by taking the median of all polls that were surveying voters on each particular day. Then I smoothed it with a 3-day rolling median. It’s not perfect, but it does reveal that after the first debate Hillary Clinton gained 4 percentage points on Donald Trump. The smaller bumps are probably noise.

Oddly, the Access Hollywood video and ensuing sexual-assault scandal for Trump have not had a consequential effect yet on the Presidential race. To my thinking, the more important question is whether it will enhance the Democrats’ ability to tie Trump to downticket races.

There’s a story going around that Senate races are flat or trending Republican, even as the Presidential race is moving toward Clinton. Although it is true that several races have recently become competitive, the overall picture shows a fair amount of similarity. See the PEC aggregated-polling snapshots:

History of Popular Meta-Margin for Clinton
The similarity in these two graphs is remarkable. This was also the case in 2012:

However, there is one reason that a casual observer might form a different impression. Look at the graphs closely, and you will see periods when the Senate aggregate has lagged behind the Presidential Meta-Margin by 1-2 weeks. In other words, the Senate aggregate can give us a time-machine look at late September – around the time of the first Presidential debate. I suspect that Senate polls will move slightly toward Democrats in the closing weeks. The caveats are in Indiana and Wisconsin, where Democrats Evan Bayh and Russ Feingold are struggling more than one would have expected a month ago. Those races bear watching.

Tags: 2016 Election · President · Senate

50 Comments so far ↓

  • Ken L

    Rather than lagged I would characterize the Senate time series as slightly lower frequency or a little smoother. I notice the number of contributing polls is lower (roughly 1/3) or am I missing something?

  • Amitabh Lath

    If these are based on the same polls by the same pollsters, what mechanism could cause a time shift between the two?

    Have you tried a Kolmogorov–Smirnov test with the Senate and Meta margin distributions?

    • Tony Asdourian

      Actually, there is no better party drink than a Kolmogorov-Smirnov martini.

    • Joel

      My assumption, without digging deep into it:

      State polls take longer to collect. You have to collect a similar sample size from a smaller overall population? It seems like the population is large enough in both cases that it shouldn’t matter all that much, but maybe it does.

    • Amitabh Lath

      Joel, the meta margin is also based on state polls so I am surprised the Senate question has a lag.

    • Matt McIrvin

      The problem with the Kolmogorov-Smirnov martini is that how good it’s going to be is algorithmically uncomputable.

    • Josh

      I think there’s a very simple answer to this question: more pollsters are polling for the presidential race than for down-ballot races.

      If you only survey for senate races once a week, as opposed to several times a week for battleground states like PA or FL or NH, your snapshots of those senate races are going to take into account data that is, in some cases, approaching a month old.

      It’s not even hard to see for yourself–just look at the HuffPollster feeds for the better-surveyed senate races like those in WI, NV, NC etc. Some of the data Sam is working with is from mid to late September. Working with data that old can’t possibly give an accurate current snapshot of what these races look like.

    • Slartibartfast


      The secret to a good Kolmogorov-Smirnov martini is in the mixing technique. If the mixing is too strong you introduce chaos which destroys the topological structure of the ergotic orbits.

      In addition to the Senate meta-margin being noisier due to less (and lagging) data, I would think that the gaps between when polls are in the field would mean that it’s harder to sort out the impacts of individual events as well, particularly in this kind of charged partisan environment.

  • Michael

    Reports are that Trump’s strategy now is to drive down Democratic turnout by driving up Clinton’s negatives so much as to dispirit Democratic voters both for her, and for down ballot races. What assumptions are made about turnout in forecasting the outcome of an election? Is it plausible that a dispirited electorate would stay home sufficiently to have an outcome different than your prediction?

  • we_are_toast

    I believe Trump already has accepted the inevitability of his loss. His goal now is to build an audience for his upcoming Trump TV network. The way to do that is to go full scorched earth on the GOP and the media, and to start a new political party to keep his audience engaged. This should help the Dems.
    We may be witnessing an historic moment.

  • Matt McIrvin

    So it’s similar variation but within an extremely narrow range–which is consistent with the observation of general high polarization and low volatility. Coattail effects just do not seem very large this year.

    • jbaz

      Perhaps a function of the intense dark money spending going on on down ballot races. Most big donors on the Republican side gave up on the Presidential ticket months ago, opting to preserve their control of Congress instead. It has an effect.

  • Ken L

    Are the history of the presidential and Senate meta-margims available as .CSV without having to recompute from polls?

    • Phil C

      I am not aware of a history file for the Senate. For the Presidential history, click on the EV estimator chart in the right sidebar. On the ensuing page, click the “tabular data” link below the EV history chart for the CSV file. The file has the daily history of essentially all the Presidential numbers, including the popular meta-margin (medians, modes, safe, and tossup EVs, one and two sigma EV bands, count of polls, PMM, and the random and Bayesian Clinton win probabilities) The PMM is fourth column counting from last.

  • smartone

    One thing that I don’t understand – HRC was piling money on TV against Trump for months – outspending him considerably and yet it hardly moved the needle.

    Meanwhile big GOP has successfully used big money to really effect races Senate in Ohio and Wisconsin and Indiana.

    • mediaglyphic

      Perhaps we have hit a decreasing marginal rates of return for democrats in the General Election. In these states the Republicans have not.

    • 538 Refugee

      Hillary has been running some negative ads on Trump and some positive ads about herself. The negative side was mainly taken over by a PAC allowing her to focus on the positives. I think that may have helped the debate audience with expectations.

    • Matt McIrvin

      Hillary and The Donald are both exceptionally well-known people. Mere name recognition is not an issue, which probably eliminates the benefit of many forms of campaign advertising.

    • David T.

      I think where you will see the effect of the spending by the Clinton team is in new voter registrations. They already have some very good numbers in Florida.

    • LMB

      Also, Hillary’s ads targetted Trump only till very recently, right? Certainly the Obamas’ and Hillary’s speeches I’ve watched make no mention of the Senate.
      In this highly polarized atmosphere, ads reminding conservatives of Trump’s/Fox talking points, i.e. “we need a GOP majority to keep Crooked Hillary from installing activist pro-abortion Justices, bloating government and otherwise destroying the country” might be very effective, even with those who can’t bear to vote for Trump.

      I don’t actually understand why Hillary, Bernie, etc. are not talking up the need to also vote for Dem Senators and House Reps. Ryan is certainly not similarly silent about his need to keep Congress GOP-controlled.

    • 538 Refugee

      LMB. I’ve read stories that Dems are unhappy with Clinton not pushing for the down ticket vote. Her strategy of picking off moderate Republicans kind of keeps her from sounding partisan though. Check out the Republicans unfavorable vs Democrats on HuffPo.

  • Ruth Rothschild

    I’m posting my questions in this thread because I’m unable to post them in your other article from this morning about doubting polls. Sorry that this isn’t related to this particular post. In that other article, you discussed Johnson and Stein’s effect on polls and a migration of people who support these two back to the two major party candidates. Where does Evan McMullin fit into your equation? While he’s on the ballot in only 11 states, he’s currently doing well enough in Utah that it’s not out of the question that he could win their 6 electoral votes. And, if undecideds in the other 10 states become persuaded by what they see in Utah and decide to go with him and not migrate back to Trump or Clinton, isn’t it conceivable that McMullin, like George Wallace in 1968, could pick up electoral votes in some of the other 10 states? Have you factored McMullin into your calculations? And, would the trend of these McMullin supporters be similar to that for Johnson and Stein supporters (i.e., would they end up migrating back to either Trump or Clinton)? I know that in past elections there have been more than 2 candidates from various minor parties and that they historically haven’t made a big difference in the final outcome. But, with the current polarization in this year’s election, I wonder if having McMullin in the mix, even in only a few states, could end up making the numbers between Trump and Clinton closer than predicted (and maybe too close for comfort to the point that Trump could contest the results). Is this a possibility?

    • Matt McIrvin

      Sam’s model uses state polls, so it will reflect them to the degree that McMullin is in them.

      So far, McMullin mostly seems to be a player in Utah, where he’d be taking conservative votes from Trump. Of the other states where he’s on the ballot, the one where I could see him getting a non-negligible number of votes is Idaho, with its large Mormon population. He’s not on the ballot in Nevada or Wyoming, which might have been promising.

      I see a writer at 538 already jumped on the fantasy in which McMullin deadlocks the electoral vote and is chosen as President by a Trump-weary conservative House. I think this is a remote possibility at best. It’d probably drive the Trumpists more violently crazy than a Hillary win would.

    • Ruth Rothschild

      I also highly doubt that McMullin would have the impact to deadlock the race and the electoral votes given that he’s on the ballot in only 11 states and those states put together don’t comprise a large number of electoral votes. . And, you bring up a good point about him taking votes away from Trump in Utah. Of course, given that Utah is his home state, it’s not surprising that he would get a lot of support in that state. But, I’m wondering if, in the next 3 weeks, with all of the craziness with the combined wikileaks stuff and more allegations of sexual assault by Trump that could pull undecided votes away from both Trump and Clinton and make the numbers too close for comfort. Maybe this won’t have enough of an effect to change anything. At this point, with Trump out of control and acting out of desperation, 3rd party candidates probably won’t negatively affect his numbers– his big, uncontrolled mouth and the way he spews BS will likely have more of an effect of turning voters away from him (except for his base support).

  • DeanH

    Polling is cool, but the problem is they are based on similar turnout from the last election. Everything seems to point this is not going to happen because of the disgust towards the candidates. Since Donald doesn’t have a real turnout organization, Rs can expect the worst. Like to know if it is showing up in the polling.

    • jbaz

      An enthusiasm gap has emerged in the polling. Putting us in House Majority flip territory, historically speaking.
      Another indicator is absentee voting. North Carolina provides counts by party affiliation. Already over 400K votes counted and they’re showing a huge dropoff in Republican votes relative to 2012.

    • Chuck

      Actually, no, polls are based on what people tell pollsters. Sounds like you should check out Sam’s last blog entry:

    • Bill

      I agree. Turnout may be very different this time around. By all accounts Hillary has a vastly superior GOTV system. I am encouraged by the NC early voting analysis showing R turnout below previous years. Hope this leads to a Dem senator there and perhaps to a ‘wave’ if that trend manifests nationally.

  • Scott J. Tepper

    Thanks for analysis based on facts and statistics. Once again is concern trolling about the Democrats losing the Senate races in an attempt to increase clicks. It’s refreshing to read a dispassionate report from someone who doesn’t need to add to the fake horse race analyses everybody is still spouting. Only last night did MSNBC grudgingly admit that Mrs. Clinton may now have enough states leaning to her to win at least 270 E.V.s. Up until then Steve Kornacki had her at something around 258 E.V.s.

    • 538 Refugee

      The media has a tough call here so I cut them slack. I’d rather they keep the interest in voting up than not. At the end of the day, higher voter turnout is good for the republic. It is a truer gauge of the people’s will. Those more invested can seek out sites like this.

  • Ed Wittens Cat

    What i just learned from Paul Ryan lol
    If democrats take the senate
    guess who becomes head of the Budget Comittee?
    ;;suspenseful pause;;
    Bernie Sanders!

    • Jeremiah

      Allied to this point – we should get rid of the filibuster. I think electoral success should really mean something and that should be implementing the policies you want.

    • Ed Wittens Cat

      good point Jeremiah
      but the Bernie Sanders thing…i worked very hard for Obama in 2012.
      i was not at all inspired to exert the same effort for Clinton, largely because she ganked me and all the other Sandersnistas with (yuk) Tim Kaine.
      but i just called and volunteered to GOTV this weekend.
      this gives me something to work for.

  • Prairie Pundit

    This question may already have been asked and answered. If that’s the case, I apologize in advance! But . . . I’m curious as to why “The Power of One Vote” margin for the Indiana senate race has held steady at Bayh, +14% for the last several weeks, given that various recent polls show the race significantly tightening?

  • Ed Wittens Cat

    Can someone explain to me how Trump plans to avoid another AndersonCooper style question in Debate 3?

    • Avattoir

      You think Fox anchor Chris Wallace is going to challenge Trump remotely like Anderson Cooper did?

    • Kevin King

      I actually think Wallace will. He is a respected journalist, and Ailes is gone from Fox. Witness the fights with Hannity.

    • Ed Wittens Cat

      I have a friend at UNLV that is telling me abt the preparations for Wednesday Debate 3– there are 3 circles of barriers, one of shipping containers stacked 2 high, one of portable fencing, and one of concertina wire (!!!).
      There is an enclosed area called the “Free-Expression Area” to herd the more enthusiastic protestors into. There are emplaced rooftop snipers. UNLV is expecting 30,000 protestors.
      Is this America?

  • Lorem

    I didn’t like that your graphs had different x-axis, so I idly screenshotted them and overlayed them in photoshop (x-axis rescaled and shifted, y-axis untouched). Doesn’t look like one consistently lags the other to me:

    Don’t think I messed things up while overlaying, but it’s possible.

    (red – senate, purple – presidential)

    • Lorem

      *I should have said that y-axis is shifted to make the left ends of the tails line up, but not rescaled

  • pechmerle

    Off-topic, but of interest:
    LA Times comments on the stories about its tracking poll with USC over-weighting one young black Trump supporter:

    • Lorem

      Skimming, it seems they are admitting that the original story was exactly right, despite the headline.
      They removed the high weight that had been criticized (good job!), and this shifted results by 1-2% (as had been originally claimed).

      Conversely, the original story used the single voter as the headline, but the big difference (4%) it found was based on readjusting all the small-group weighting and removing the “last vote” weights, not just from the one voter.

    • MNP


      I’d rather they just stuck to what they did rather than changing in mid-poll.

  • Randal

    “Oddly, the Access Hollywood video and ensuing sexual-assault scandal for Trump have not had a consequential effect yet on the Presidential race.”

    One possibility is that we’ve found Trump’s “I could shoot somebody on 5th Avenue” level for national polling in the general.

  • Larry Guy

    Isn’t it weird that the Iowa prediction markets are showing a 17% Clinton lead in the expected share of the popular vote?

    It’s inconsistent with their giving her only an 85% chance of winning. Arbitrage opportunity?

  • 538 Refugee

    Trump now calling for drug testing before the next debate. Clinton should counter with POST debate testing. I’ve heard hair testing is the most accurate but that could be problematic in this instance.

  • 538 Refugee

    I’m reading more states that could fall like Georgia and Arizona. Time for some Google to the power of N?

  • Lee

    I am having trouble reconstructing this day to day; perhaps it could be computed for me and displayed (please, please). For the United States Senate, which races will I want to pay most attention to on election night given the current polling? That is, which are the central 3 states, such that the party that wins the majority of these 3 is likely to be the controlling party of the entire Senate? I thought that the answer is NC, NH, and PA, but these don’t seem very near the Senate meta-margin at the moment. (Or central 5, etc., if that makes more sense statistically.)

    Thank you!

Leave a Comment