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


July 27th, 2016, 9:29pm by Sam Wang

The HuffPollster feeds are changing structure to offer more options. We are adapting. Tonight, some routine diagnostics…

President: working normally, with a new polling rule from HuffPollster. When a pollster makes it available, HuffPollster has started to use the 3-way matchup (Clinton, Trump, and Libertarian Gary Johnson) to calculate the default result. Pollsters don’t always poll this particular combination. Some are still just polling Clinton vs. Trump. Still others poll Clinton vs. Trump vs. Johnson vs. Stein.

In previous elections, we have accepted HuffPollster’s judgment as to which option to report; our plan is to do that again this year. The reason is that third-party candidates usually fade in the finish. Our approach should give a polling median that lies somewhere between the two-candidate margin (which underestimates the third-party vote) and the polled multi-candidate margin (which overestimates the third-party vote). This is probably about right.

As of today, the only notable consequence of the new rule is that Ohio goes from a tie to Clinton +4%. This could go the other way too – there are just 3 polls there, which leads to some fluctuation. In the coming weeks, polls will become more frequent.

Update: The actual approach is to simply accept the feed that is provided for each state and parse its numbers. The exact choice of polls is at HuffPollster’s discretion. The above text is simply a description of what that feed appears to be today. You can see each state’s list of polls in tabular form by going to their chart and appending ‘.csv’, which gets you a file of the polling data. For obvious reasons, I am motivated to continue using that feed for as long as it is stable.

Senate: working normally. (NH currently updated manually)

House: working normally. This calculation uses HuffPollster data, and uses a last-3-polls or last-21-days rule,including only the most recent survey from any given pollster. Today, only 3 polls contribute to this estimate. Two were taken July 22-24, and show Democrats +2% and Democrats +3%, making the recent change a reflection of the post-convention bump for the Republicans.

This is probably a good time to say that I’m always looking for local talent! I can use help with development and maintenance of the calculations. Report apparent bugs in comments, please…

Tags: Site News

41 Comments so far ↓

  • Marlene Snyder

    Sam–I think I speak for all of us in saying that we appreciate your diligence. Thanks!

  • Michael Hahn

    Thanks for all of your efforts to maintain this site.

    Pity, though about the feeds!! I was so hoping that the change in Ohio was real!

    • Sam Wang

      It is a real change, in the sense that the 3-way matchup does give Clinton +4%, as opposed to the 2-way matchup, which gave a tie. As the election approaches, we will have much more data in this rather important state, and it should fluctuate less.

    • Michael Hahn

      AHH!! That makes a lot of sense! Thanks for the extra clarification. Interesting that the three way has such an impact!! I wonder if that will hold as the election nears.

  • W.Dow Rieder

    Will comment if I see something–and I’d like to add my thanks. This site is a real antidote to the wild swings of trying to following political news these days.

    • Xecky Gilchrist

      Indeed – I’ve been promoting it as such here and there about.

      Thank you so much for this clearheaded service, Prof. Wang.

  • Amitabh Lath

    I fear we will never be able to distinguish the post convention bump for Clinton from the Trump-Putin-hacking brouhaha consequences. Couldn’t he have waited a couple of weeks? Doesn’t he realize there are people trying to make quantitative empirical evaluations?

    • Lorem

      I fear that trying to get the consequences of the two entangled was precisely his objective (the villain!)

  • David Driscoll

    Probably not a bug that affects anything, but the Huffpost Pollster link for Florida in the ‘Margin’ column on the right sidebar (currently reading ‘tied’) is broken. It points to, but it looks like the correct link is

    All the rest of the links have the former format and work, so I don’t know why Florida is different.

  • Richard Vance

    PEC is my refuge from the hubris.
    Thank you Sam and all the elves.

  • Rudy Shankar

    Your site is a welcome shelter in a storm of statistical hail. As you have cited maybe it does not capture eyeballs ..sadly the only metric used to measure a site’s value..but at the same time it does not suffer from the swings due to uncertainties baked into traditional poll analysis.

    Great job!

  • Indomitable Ted

    Interesting bit on the influence of a three way match result in Ohio. I wonder if that can be attributed to Kasich holdouts that still haven’t forgiven Donald Trump. I don’t know how one would go about testing that, but it’s a curious thought.

  • David

    Hi Sam (and the PEC community). Two things.

    First, your site for Florida Clinton-Trump polls from the Huffington post seems to be broken. It is just a “google away”, but you might want to fix it, in your infinite free-time.

    Second, but more important, I’d like to generally thank you for everything you’ve done on this site. It is a great service to the US.

    (I’d offer more useful comments but numerics and data-analysis are a bit far formal/stringy physics.)

    • Amitabh Lath

      I was introduced to Sam’s blog by a prominent string theorist. I think what draws us to this is the struggle to extract some useful and lasting insights from noisy and sometimes irrelevant data.

      Of course, who is leader of the free world is also very important.

  • A New Jersey Farmer

    Was there an actual Trump bump?

    • Matt McIrvin

      Personally, I think there was one, of a few percentage points nationally. If it’s a typically temporary convention bounce, what we’re seeing now should be close to its actual apex (though any D bounce we see will probably fade too).

    • Jay Sheckley

      Yes. It says here on PEC it came to about 1%.

  • Matt McIrvin

    The reason is that third-party candidates usually fade in the finish. Our approach should give a polling median that lies somewhere between the two-candidate margin (which underestimates the third-party vote) and the polled multi-candidate margin (which overestimates the third-party vote). This is probably about right.

    There’s a danger here, isn’t there? Since Johnson is right of center and Stein is far-left, including Johnson but not Stein should record a third-party vote that is slightly biased against Trump. I’d expect that would exaggerate Clinton’s lead, if only by a small amount.

    • Sam Wang

      There is that risk, but it is unclear what to do. There could also be a move toward Johnson. Note all the conservatives who have turned against Trump.

      I do have a preference to present only one estimator. In that context, a choice has to be made.

      One answer is to increase the uncertainty in projecting to November. My current preference is to trust the wisdom of crowds of pollsters. I believe what HuffPollster does is offer all measurements from each pollster, in which case the answer is to take the most refined measurement that each one offers. Which is approximately what my algorithm does now.

      Full disclosure: I have to drill into it to make 100% sure of that; changes in the feed put me on edge a bit.

    • Matt McIrvin

      I also suspect Stein’s support will fade faster than Johnson’s because she’s more obviously deranged. But that is pure gut feeling.

    • bks

      I think you’re right to be on edge about HuffPost changes. No one is going to get a bonus at HuffPost for continuing to post the polls accurately. The only thing that the honchos will reward is the thin, thin candy shell that attracts eyeballs.

    • Sam Wang

      No, that’s not true! They are committed to accuracy in the original data points. I have full confidence in them, which is why I use them.

      However, their main priority is getting their own charts and feeds into a form that they like, for immediate public use. This might lead to instability in exactly what data is available in each chart’s associated feed. By “instability” I mean “what our Python code extracts.” It is written to work with the current structure of the feed, and so I am sensitive to the possibility of surprises. That’s why I mentioned the .CSV files; I think there are readers here who might look at those and give a shout if PEC goes off the rails.

      Generally, I am under the impression that their state-level feeds will be more stable, simply because there are fewer polls per state.

    • bks

      Trust but verify.

    • Mark F.

      Johnson is running to the left of Clinton and Trump on civil liberties and foreign policy (less interventionist, more anti-war, against stuff like the Patriot Act), and to the right of both on economics (cut taxes and government spending). Add in being pro-choice and pro-gay, and it’s an interesting combo.

    • Jay Sheckley

      Study Huffpo’s pie charts of nonbinary polls, and you’ll find far heftier support for Johnson than for Stein. So the possible distortion is less than I’d assumed. It’s merely anecdotal, but some Stein fans told me they won’t vote 3rd party in a swing state. Also among those I’ve spoken to, Steinheads and those who “Feel the Johnson” find binary polls highly offensive. Some would either not respond to a binary poll, or would poll in a way we’d find random. Sam’s inclusion of nonbinary data should reveal truths. My worry is pro-Trump distortion from polls which ignore Johnson, especially in states where the race is close.

  • Sean

    In an especially toxic environment this year this site is refreshing in that it’s all about the numbers. This type of analysis isn’t what the mainstream likes. They like to hear that their favorite candidate just swung +10% overnight. Of course they can always go to fox “news” to get that fix.

  • Adam

    I saw Nate came out with his explanation of various model differences. He seems very confident that national polling is also reflected in every state.

    In elections like this, in an increasingly polarized country, I don’t know if that is so true. Trump has, to date, been doing so much worse in areas like Kansas, Mississippi, etc. than the typical GOP candidate, that I do wonder if some of his gains and bounce came from the red staters that weren’t so pro-Clinton as they were not sold on Trump. And such, they are in states that were red, anyhow–it’s just a matter of HOW red they were and are.

    Conversely, this election will be won primarily in OH, FL and PA, with lesser purple states of NC, IA, NV and (if I am completely honest, I don’t think it’s a big play with Kaine, now), VA. Are those purple states moving like Texas, Kansas and Nebraska? I don’t know if you can say that with the certainty that Nate claims.

    • Charles

      I’m inclined to agree with you. Nate also made the point that PA polling showed a significant Clinton lead, but it hadn’t been polled in a couple weeks, so his model had the margin in PA matching Ohio, where it’s narrowed to only a couple points. But the polls released today show Clinton at a nine point advantage in PA, so that prediction seems to be flawed.

    • deb

      why does it have PA only up D 1% on the right/ When you click on its 4%

  • The Ohioan

    I was looking at the Pollster’s 3-way presidential polls, and it’s troubling that Johnson seems to be taking more votes from democrats and independents than from Republicans. I would have expected this from Stein, not from Johnson. It looks like Gore facing Nader, times 2.

  • A New Jersey Farmer

    Interesting that Rasmussen has a poll out showing Hillary +1. Their previous polls were seeming outliers, having Trump up consistently. Negative bounce?

  • C Hart

    Is the API changing? I had just finished writing some python code for my own model :(

  • Christopher Brandow

    I think that you meant 0.4% for OH

    • MarkS

      No, it’s 4%. The last 3 polls in Ohio, using the percentages when Johnson’s name is included, are Clinton -3%, +4%, +4%. Median is +4%.

  • MarkS

    “The last 3 polls are used, or all polls with an end date within 7 days of the most recent poll are used, whichever is greater.”

    Does “within 7 days of the most recent poll” mean “within 7 days of the most recent poll’s end date”, or “start date”, or “median date”?

  • Scott

    If Rassmussen has Hillary +1, you can bet she’s probably really +4 or 5. Rassmussen is famous for putting a thumb on the scale for the Republican.

  • Michael

    To what degree does your model use “trends” in the polling? Also, to what degree is that factor a good one to use? Thanks.

Leave a Comment