Princeton Election Consortium

A first draft of electoral history. Since 2004

The Nevada bonus is back

November 6th, 2016, 10:17am by Sam Wang


Two states are hard to poll accurately, probably because they have high rates of migration: Alaska and Nevada. In addition, Nevada has a high Hispanic population, which votes heavily Democratic. Based on early voting, it looks like 2016 will be a repeat of 2010 and 2012, in which Democrats outperformed Nevada polls by 10 and 3 percentage points, respectively.

In 2010, poll medians missed the victory of Senator Harry Reid (D) over Sharron Angle (R) by nearly 10 percentage points. In 2012, President Obama led Mitt Romney  in early voting by 7.6% and ended up winning the state by 6.7%, more than his polling lead of 4%.

In 2016, as the Votemaster says, early voting is favoring Hillary Clinton nationwide. And in Nevada, where early voting ended on Friday, Clinton has outperformed President Obama. If her final total matches the 2012 result, it would be a 6-point performance over her polling median, Clinton +1%. In short, the Nevada bonus will be just about halfway between 2010 and 2016.

In the crucial Senate race there, the current median is Cortez Masto (D) +3.0%. It appears highly likely that this seat will go Democratic. That would bring the Senate Democratic/Independent total to a minimum of 47 seats (for easy reference, the RCP counter is here). In Pennsylvania, add McGinty, who is at +3.0%, to get 48 seats. To retain control, Republicans will need to hold Democrats to winning only one of the following five races: Indiana, Missouri, New Hampshire, North Carolina, and Wisconsin. That is an uphill climb for them.

Tags: 2016 Election · Senate

107 Comments so far ↓

  • Olav Grinde

    Based on what we know about the unexpectedly massive Hispanic early voting in Nevada, and in Florida, it appears that Mexican-American and other Hispanic voters are building a wall against Trump.

    Oh, the irony!

  • cvermar

    So I am the new group on Facebook “Pantsuit Nation”: http://mashable.com/2016/11/06/pantsuit-nation-secret-facebook-group-clinton-supporters/

    It has quickly grown to over 1.4 million members. Remarkable stories shared from people (mostly, but not all, women) all over the country. The enthusiasm for Hillary was there, all along, but no one wanted to say it out loud. Anyone else in the group?

  • Mike Rappeport

    I may have missed it but having just read all the comments, I was surprised nobody said what seems to me the final nail in Mr. Trump’s coffin. Assume that Jon Ralston is right (which I think almost certainly true), and the Hispanic (early) vote is showing that Nevada is really significantly more Clinton than the conventional interpretation of the polls indicate. Then unless you think the Spanish states are uncorrelated (and I can’t believe that) then surely what is true of Nevada is likely to be true for all the Hispanic states which in this context means Florida and Arizona. And if so this race is over, and indeed Sam has probably been right all along (which knowing Sam I have always believed).

    • Ileana

      A caveat about the Florida vote. Cubans have tended to be more conservative than other Hispanic voters and traditionally GOP voters. While the Cuban percentage of the Hispanic population in Florida is at its lowest point – just under 30% of all Hispanics are of Cuban origin – it’s still the case that Florida’s Hispanic population cannot be considered identical to that of other states. According to Pew, 78% of Nevada Hispanics are of Mexican origin; in Florida, only 14% are. So take comparisons of Florida with western states with a grain of salt.

  • Greg Gross

    Sam – Any insights on why the regression to the mean is seemingly stalled? For more than three days, the MM has hovered around 2.6, with an occasional tick upward, only to sink back to 2.6.

  • Eric Walker

    I am confused by the projected EV total of 314. I went to the CSV spreadsheet file and noted the estimated win probabilities there, but when I add all states above 50% (even taking one EV off Maine), I end up with 322.

    Am I somehow comparing apples and oranges?

    • Sam Wang

      Yes, you are calculating the mode. The banner lists the median, which is the midpoint of all possibilities. Today the mode is greater than the median because of several states that favor Hillary Clinton, but just barely.

  • Dave Andersen

    With due respect to Andy (votemaster) – his analysis of the early voting cheats a bit: (1) It claims that higher turnout of women in early voting may presage a higher turnout of women on election day; but (2) a higher turnout of white voters in early voting presages a higher turnout of *non*-white voters on election day. You can’t have it both ways.

    • blaneyboy

      Wouldn’t that depend on what “higher” means? “Higher” than last year, with a larger population this year? Or higher than last year, with exactly the same size population? If more women are voting early in 2016 than voted early in 2012, I would think, on the assumption that more women are registered to vote in 2016, that it’s likely that more women will also turn out on election day. There will only be fewer on election day if the number of women voters is fixed.

  • Rachel Findley

    Question:
    Any estimate on the likely time lag in voters’ decisions (to vote or not to vote, or for whom) based on the FBI saying there is nothing incriminating to Clinton in their latest computer check?
    I think the announcement is too late to show up in the polls. And the grotesque imagery lingers in people’s minds.
    But do we have any idea how quickly people’s voting behavior might be affected?

    • BrianTH

      Well, I know sometimes in polls there is basically an immediate effect, and in fact there was on the front end of this issue when Comey issued his first letter (which we know thanks to those who released daily samples). However, sometimes the effects of an event build over days to a peak, but that may in part reflect additional secondary events like media coverage of the primary event.

      Long story short, I would guess to the extent the second Comey letter is going to have an effect, at least most of that effect will be available on election day tomorrow.

  • Scott J. Tepper

    Ohio could be a good test of what GOTV is worth. Mrs. Clinton has a strong ground game there. Trump, not so much. Will GOTV be the difference?

  • TeddyVienna

    So Trump, on the second-to-last full day before Election Day, decided to go to Minnesota of all places, where he pitched a “Somalis as scapegoats” argument.

  • Selva

    Weird or what? Now FBI says review of the newly discovered emails found nothing new. Assuming the recent tightening had something to do with James Comey’s letter, is there still time for the opinion to revert back towards the “mean”?

  • RA

    How independent are polling results from state to state? Let’s assume Trump wins Colorado and the polls were off by a large enough margin for that to happen. Given that, what are the odds Trump wins all of Fl, Nv, Nc, and Oh?

    With states not being perfectly independent, the odds are at least somewhat higher that he wins the others given that he wins one, but how much higher?

    • Scott J. Tepper

      Based on what Sam has calculated the odds are less than 1%.

    • Michael Coppola

      CO is currently Clinton +4%. If there is a correlated systematic polling error of 5% in Clinton’s favor, then Trump will win in an electoral landslide. You don’t need any statistical modeling to tell you that.

      OTOH, you can look at the Trump +2% map to see what a less extreme polling miss might look like.

    • BrianTH

      It all depends why. For example, if the median polls systematically missed on Hispanic turnout and/or Hispanic margin, you would get one possible pattern of effects in other states. If they missed on something like female turnout, or Republican women crossover, or so on, you could get different patterns.

    • Ketan

      That’s a fun modeling question! If a particular state has a 3% polling miss, I’d say it’s more likely that not there’s a 2-4% shift across the board. (PEC says 2.6% is enough to make election a tossup.)

      That said, there will be multiple states that have a 3% errors relative to the polls; especially the ones that are not close(ly watched). So, it’s key that you picked CO without any advance knowledge that it would/would not match its polls.

      Folks say NH is the state to watch to detect how well the polls did. Not sure the reasoning.

    • Matt McIrvin

      It’s just because NH is a small Eastern state that gets polled a lot because it’s a battleground state, but also closes and reports returns very early in the evening on Election Day, because it has no very large cities with overworked urban precincts.

      NH is also atypical even for a swing state, in that it’s so small and so white, and has a long history of being pampered by Presidential candidates because of its role in the primary process. That could work against its utility as an indicator, but as long as overall polling swings and misses are heavily correlated across the whole country, it may be OK.

    • BrianTH

      I like Florida this year as a place to look for surprising results (either way).

  • TeddyVienna

    What’s the impact of TODAY’S news from the FBI? I’m guessing it’ll be next to impossible to find any polls that reflect the news, but it may sway a few votes on Tuesday.

  • 538 Refugee

    Given the large number of ballots already cast it is conceivable that the GOTV is already being picked up by pollsters and factored in. While many of us hope our favorites have hidden support, let’s wait until after the election when we can quantify it if need be.

  • Ruth Rothschild

    Looks like this article in the Huffington Post could help Hillary down-ballot and maybe among voters still sitting on the fence. As most of us at this site knew, much ado about nothing with last week’s Comeygate and a lot of unnecessary drama created by Comey raising red flags before he had all of the info. I suspect that this isn’t going to help his career too much or peoples’ faith in the FBI to be impartial:

    http://www.huffingtonpost.com/entry/james-comey-congress-emails_us_581f8ff5e4b0aac62485196a

    • Tom Benjamin

      This hurts Hillary. I think this letter is part of the same ratfucking. Look at what is happening. CNN is all over it of course. Clinton and her surrogates want to talk about her closing arguments, not James Comey.

      Trump isn’t changing his stump speech and Trump surrogates are happy to say Comey is wrong, Hillary put national security at risk, lied and yada, yada, yada. It is all bullshit of course, but it is bullshit the Republicans want dominating the news coverage. Tomorrow will all be about emails again.

      Comey hurt Clinton when he editorialized rather than just announce the case was closed without charges. He hurt her when he released the FBI files. He hurt her when he sent the letter two weeks ago and he hurt her today.

  • Greg

    Hey folks, I agree with Sam that Hillary is likely to win, though I would put her odds in the 90-95% range. It’s all going to come down to NV, FL, NC, and NH. Hillary probably needs to win one of the four to clinch the presidency.

    Get involved! You can go to https://www.hillaryclinton.com/calls/phonebank/ and start calling voters in these states. Did I mention there are competitive Senate races in each? Don’t forget to put in a plug for the Senate candidates!

    And Sam, thanks for raising awareness of Congressional races! I hope Democrats are paying this much attention to you in 2018!

  • M. Leo Cooper

    The conjectured existence of a faction within the FBI actively working to elect Trump reminds me of what?
    Yes, the “Business Plot” to overthrow FDR in 1933.
    https://en.wikipedia.org/wiki/Business_Plot
    As Yogi Berra allegedly said, it’s deja vu all over again.

  • AAF

    Have you calculated the actual meta-margin, based on the voting results, in each of the races you have analyzed, and compared it to your final meta-margin projection? I can’t find anything about that in your archives.

    • Eli

      I asked this and commenter Josh gave these 2012 results:
      “Final national polling median ’12: Obama +1
      Final MM ’12: Obama +2.78
      Final actual MoV: Obama +3.9″

      (My concern about this election is whether we might have more uncertainty than usual about poll/vote mismatch, such as in the polls’ likely-voter modeling. This election deviates from the priors you’d set from previous elections — I’m betting that educated Republican voters, who usually vote at a very high rate, will show lower this time — and I’d love to know how that’s being dealt with.)

    • Eli

      And now the article titled “Is 99% a reasonable probability?” goes into this question in detail.

  • BritJax

    So just for fun, I looked at all of 538′s final projections for statewide races (President and Senate) during the 2012 and 2014 cycles, where the leader’s predicted probability of victory was below 90%, according to their model. Here’s what I found:

    1. The mean predicted probability of victory for the leading candidate in those races, according to the 538 model, was 75%.

    2. 89% of those candidates actually went on to win the election in that state.

    So, if you add in Nate’s spectacular failure to predict the Republican primary, on top of a model that was already improperly calibrated, it’s pretty easy to see why they went so far overboard with the uncertainly and volatility during this cycle.

    • TheGhostofSimonBolivar

      The 538 is now a wholly owned subsidiary of ESPN. Read this from Christopher Bates in their Featured Stories section over at Electoral Vote. Explains a lot of what we are seeing. And why.

      http://www.electoral-vote.com/evp2016/Feature_stories/538.html

    • Arun

      Thank you!! This was the exact issue I have with the 538 model. Instead of realizing that probabilistic models are not absolute, he chose to model the unmodelable and added arbitrary parameters like state elasticity, trend line corrections, and complex interactions. I made a comment on the 538 board to this effect and got a lesson in elementary probability from someone there. The 538 model as it stands today has little going for it other than Nate’s reputation.

  • Omer

    Hi Sam,

    You and Nate Silver have had a lively, if not direct, online debate this election about models. What would be a good metric to use after the election to see whose model was more accurate? I was thinking of of some sort of sum of correctly called states weighted by model confidence, but I haven’t found a formula that I’m really satisfied with. Do you think we could get many of the forecasters to agree to a particular metric?

    • Sam Wang

      A media organization is working on that.

    • Suvro

      I thought that the Brier Score – https://en.wikipedia.org/wiki/Brier_score – is one good metric to compare poll prediction models.

      I am not so sure whether all forecasters (and more so the mainstream media) will use any such metric.

      However, I was glad to note that in today’s show Fareed Zakaria had a good discussion between Nate Cohn (Upshot) and Harry Enten (538). Zakaria brought up PEC when discussing models.

    • Sam Wang

      A birdie tells me that at least one media organization will do exactly that after the dust clears on tomorrow’s election.

    • Laurence

      The Brier score is a nice metric for evaluating probabilistic forecasts.

      Nice breakdown of this metric applied to predictions in 2012:
      http://rationality.org/2012/11/09/was-nate-silver-the-most-accurate-2012-election-pundit/

    • Roger2

      Nate Silver’s arguments are awfully cogent. He says, “The point is that however you slice the data, you end up with a Clinton lead in the range of 2 to 3 percentage points.” And elsewhere he points out that there are a high number of undecideds.

      I think it’s pretty clear that Wang’s model is wrong somewhere. 99%? If only that were true!

      But no, 99% makes no sense at all. I don’t think there is a magic bullet somewhere in Wang’s spreadsheets that can account for the results entering into margin of error territory.

      Say what you want about Silver, his arguments are clear and it’s hard to fault his model for being cautious.

      99% on the other hand is a travesty.

    • Gelatinous_Cube

      Drew Linzer alluded to Brier scores in a tweet. But would a Brier score be a fair test when one of the points of contention across models is the correlation of polling error across “similar” states?

    • Lorem

      Roger2, I think faulting a model for being overcautious is exactly the right thing to do. I think it’s a travesty that many seem to think that excess caution is better than excess certainty. They are exactly the same type and gravity of error, but for some reason one is deemed perfectly acceptable or somehow even laudable.

      If I wanted blind cautious, I could have nailed “65%” on the wall some time last August, and that would’ve probably done about as well as Silver’s model. Comparable average accuracy and much more stable.

    • Arun

      Roger2,

      One must be careful in interpreting the 99% probability. Assuming the state polls are predictive of the outcome, which was true in 2004, 2008, and 2012, the odds are indeed 99%. However, take a look at the meta margin. Applying a 2.6% offset in Trump’s favor to polls in most states will have minimal impact on win probability. However, if you apply this offset to Florida and NC, this will have a pretty dramatic impact on the win probability. It is up to you to decide if the polling used in the model accurately represents FL and NC to within 2.6% or not.

    • Michael Coppola

      If it’s “pretty clear that Wang’s model is wrong somewhere” then please tell us where. Unlike the 538 model, the PEC model is simple, straightforward, and completely transparent.

      If the polls are wrong, then any poll-based model will be wrong. GIGO. But fudging a model to the point where it doesn’t actually make a projection isn’t an improvement.

    • alurin

      @Roger2: Any scientist will tell you that the more precise the prediction, the more useful the model. A model that predicts everything predicts nothing. Silver has his ass covered no matter what the outcome of the election.

      Of course, we cannot directly observe the true probabilities; to rerun this election a few thousand times in order to obtain the distribution would be a violation of the ethical treatment of human subjects! Still, Sam’s model is making a clear prediction. If Trump wins, there’s a problem with Sam’s model, but it tells us nothing about Nate’s model.

      This difference is probably because Sam is a scientist, and Nate is something else.

  • Alan

    I have followed presidential elections closely for sixty years. This is the FIRST election where I have seen people stop talking about politics with their family, their friends, their co-workers, their neighbors. On a street whete I saw dozens of Bush or Gore signs in 2000, there is one lonely sign this week.

    On both sides, I sense a mood that says not “the other candidate is less qualified” but says “the other candidate belongs in prison and those who vote for the opposition are traitors.” This is the first election where elected government officials have advocated the execution of the opponent’s candidate.

    In my county (45% Hispanic) early voting was double that of 2012, with people driving five miles and waiting in hour long lines to vote.

    Does any poll measure the level of emotion this election is generating?

  • Donna Gresh

    I’m confused by the histogram for the Senate. The probability given for Democratic control is 80%, but the blue bars add up to something more like 75%. Just the largest red bar is close to 20% all by itself.

  • Not gonna be year 109

    Not to start trying to “unskew” the polls, but I think there are 3 good solid reasons that I have been making for the last couple weeks for why the final vote is going to be much more in favor for Clinton than the final polling average will show across the entire nation.

    1) As Dr. Wang mentions above, states like Nevada (along with Arizona, Colorado, and Florida) that have higher numbers of hispanic voters are going to see the biggest differences due to the fact that hispanic voters are historically underrepresented in polls and usually are adjusted using historical data–hispanic turnout/registration/early voting is showing that it is significantly higher this year.

    2) Decline in landline use. This one kind of shocks me that I haven’t seen more sites like this discuss it. Maybe I’m missing something and the pollsters are doing a better job of accounting for it, but I think this was a big part of the discrepancy for the polls appearing to be close for Obama in 2012 and the actual vote being +4 Obama. Meanwhile, in 2008 the final polls were almost dead on. If we look at the trend of landline use since 2008, we see that it has steadily decreased by about 4-5% every year since 2008 (was at >80%) and now is heading south of 40%. Landline holdouts are predominantly older, rural, and/or uneducated (all Trump stronghold demographics). Even if some of the polls are using a mix of cell phone and landline or all cell phone–there still is the fact that cell phone users are much more more likely to be on a “do not call” list or will simply not answer if they don’t recognize the #. Regardless, I genuinely think that pollsters will need to do some serious examination into their methodology after this election, and that will play a large role.

    3) GOTV–we’ve discussed this one ad nauseam so no need to go into why this will impact for a more Clinton leaning final tally versus the polls.

    I was going to mention the fact that early voting has increased and a large percentage of people voted early pre-Comey when the polls were much more favorable to Clinton, but I have read some analysts talking about that early voting is included in the polls, but I am not seeing that reflected in the early voting results versus the current polls–maybe there are far more registered dems voting Trump than there are registered republicans voting for Clinton. Anecdotally, I don’t know of any dems voting Trump, but I know plenty of R’s voting Hillary or not voting at all.

    TLDR: for the 3 (4) reasons above, I predict the final polling average is going to be 5% – 7% below Clinton’s actual final tally.

  • TheGhostofSimonBolivar

    Unrelated. But it looks like Tim Kaine believes the FBI has been compromised politically.

    http://fusion.net/story/366582/tim-kaine-fbi-support-trump/

    The key phrase here is “actively working to help the Trump Campaign”. Tim said “actively working” twice. To put added emphasis on it. If he believes that, its a safe bet HRC believes that as well.

    When this election is over hard to imagine that some very tough questions will not start being asked of the FBI. Be interesting to see where it all goes.

    • fred flint

      There appear to be violations of the Hatch Act. Not sure if that can be applied to anyone in the FBI or just appointed people like Comey.

      I would expect things to move quickly once the election is over. Can’t have any appearance that President Hillary is being vindictive.

    • Jeff

      Despite being 3 years into a 10 yr term I would be shocked if Pres. Obama does not ask Comey for his resignation after Hillary wins. He will clean house for Clinton.

    • Phoenix Woman

      Tim is right. The NY office especially is filled with Giuliani’s hacks.

  • Joeff

    Site wouldn’t load this morning. Hackery or heavy traffic?

    • Greg Gross

      I was wondering that, too. C’mon Sam, do tell.

    • Sam Wang

      Traffic seems similar to the preceding two days, which were okay. Investigating…

    • Greg Gross

      Uh oh: Perhaps Putin does not like Sam….

    • JMM

      WaPo site wasn’t resolving by DNS for me Friday evening. In this day and age there a any number of groups, associated with nation states or not, that could DDOS a significant number of sites like PEC, 538, or WaPo.

      As much as I value the analytics, the key is still turning out the vote, whatever the web says or can’t say.

    • 538 Refugee

      The error looked similar to something I’ve seen when the database connection failed for some reason. If your server has restart scripts for services I’d look at those log entries.

  • SH

    As someone without much of a statistics background, I am unable to follow much of the nitty-gritty analysis that is put forth by you, Prof. Wang, and by other intelligent folks in the comments, but I am continually drawn here, regardless, simply for the non-sensationalized take this website provides.

    I came for the prediction, but stayed for the content, and I thank you for being a level head. You and Beyonce make a beautiful couple.

  • Robert P Wolff

    Sam, the conventional wisdom is that a great ground game can add 2+ points to the score. But your predictions are based entirely on polls without adjusting for ground game. Does that mean that you think the conventional wisdom is wrong?

    • Sam Wang

      No, it means I don’t have a good way of quantifying that without introducing possible bias. That’s what the Clinton +2% tool in the right sidebar is for.

      With a more sophisticated organization, PEC could make that more interactive. However, we work on a shoestring around here.

    • anonymous

      Can we donate to help out? Even though you are a private uni.

    • JB

      I’d think the +2 refers to a “great ground game” versus an ordinary ground game. What does the conventional wisdom say about a GREAT ground game versus an utterly incompetent or non-existent one? +3? +4? Maybe as many as +5?

    • pechmerle

      Don’t think that article is much help to us in estimating effect of GOTV in a huge national election. The study focused on certain local elections in Nov. 2001 [sic] — in other words, a non-presidential year to start with, and a low likelihood of voting. I don’t doubt for a moment that in that environment GOTV can affect outcomes; sometimes you are talking about a difference of a few hundred votes changing an outcome. But I doubt that their result is per se transferable to the considerably different context of a national election, with its heavy news saturation, extensive political advertising, etc.

    • Broadcast James

      I’d donate to a PEC Scholarship Fund. Perhaps for study in political science.

    • Nathan Duke

      I saw someone mention the Michigan/Robby Mook thing. I could, of course, be wrong, but my guess is that it’s a head fake. She’s led in Michigan the entire time. And, you’ll notice, Trump is going back there. This would possibly keep him away from Florida and Ohio where she’s still spending time. It’s possibly similar to 2008 when Obama pretended they were investing in Arizona at the end of the race, thereby forcing McCain to go back there. Just my guess.

    • ravilyn sanders

      How much would it cost to fund one or two grad students for you dedicated to polling, analysis, etc? Would it be possible to link together grad students and undergrad volunteers from universities from all 50 states to cobble together a polling outfit, with open source results? Would there be interest in setting up a GoFundMe project to raise funds for Sam?

    • Craigo

      I’ve worked in field operations for multiple campaigns and firmly believe in its effectiveness, but I don’t know if there’s a good way to measure its contribution prior to voting.

    • Ketan

      Let’s say one side has a much better (like huge) ground game i.e. they have spent all of sept-oct talking and cajoling. Imagine they stop trying on Nov 1.

      Polls of Likely Voters on Nov 1 would include those newly excited voters, and those polls would accurately include the ground game impact.

      I’ll defer to Craigo (or other campaigner experience) but I think that GOTV takes time, and you don’t get most of your impact from just the last couple days of the campaigning.

      So the polls (and maybe the trend) should include the effect of ground game. (Reasonable?)

    • Ken McConnell

      Given the 2012 election had approximately equal ground games (maybe slight bias one way or the other) to the Dems in this election, and this election has a major party with an abysmal ground game, might that give some insight for future elections?

    • David Elk

      An interactive tool for an election in <48 hours is a tall order, but here's a very simple static tool:

      https://davidelk.github.io/mm_map.html

      You can see each meta-margin, the median EV, and a 270towin link, for each meta-margin that has a uniquely colored map.

      On the most recent podcast, Prof. Wang mentioned something to do watching polls on Nov 8: look at small, early-closing states like NH or NC. If there is early indication of systematic polling bias, you can use the table above to see what the map might look like.

    • Sam Wang

      A-plus tool. Thank you yet again, David Elk!

  • DonC

    I commented on this yesterday so I think this is correct. However, I’m assuming you can’t use this information without compromising the methodology of the predictions.

    No doubt many have seen Jim Messina’s Op Ed piece in the NY Times. He told the story of Obama asking him, on the eve of the 2012 election, why he thought his predictions were right and that of many pollster’s were wrong. The answer to this is that campaigns are predicting based on voter files and demographics rather than polling, and predictions are way more accurate. It’s why Amazon knows what you’ll buy before you do! LOL (Also of interest is that the NY Times Upshot has reported that 20% of those who self reported as being “unlikely” voters have voted early — yeah polling is hard).

    Because a data driven campaign can accurately predict voter behavior, when Robby Mook says Clinton is ahead by 170K votes in Florida rather than behind by 20K as Obama was in 2012, that’s highly likely to be accurate. Far more accurate than a poll. To some extent “small data” prediction is still a new science, but it is a science which has been honed and proven for many years by businesses.

    If you look at how the Clinton campaign sees things, it lines up well with the PEC electoral map, though Ohio and Arizona are perceived as being toss-ups and closer than Iowa.

    • Clark

      DonC, your small data analysis is a good one. Jim Messina specializes in this and recently talked about it in an interview. Sorry, I don’t have it handy, but you can probably find it online.
      As you most likely know, Jim Messina was the guy who was the data guru for Obama’s two successful presidential campaigns.

    • anonymous

      If that is the case, then we SHOULD worry when Mook said things are too close in MI for comfort?

    • BrianTH

      This sounds plausible in principle, but you still can’t trust selective leaking of internal data.

    • DonC

      First, sorry for the triple posts. The site kept timing out and I didn’t think the post had posted.

      Annonymou — yes we should but I suspect this is more fretting than anything. Win Fl and NC and you can lose MI, PA, OH, IA and NH and still win.

      BrianTH — as is the case of NV, in FL and NC there is corroborating evidence in the early voting. Check out the Upsot which is tracking NC or Steve Schale who is tracking FL.

      Clark — I’ve seen a number of articles over the course of the last year. The Clinton campaign is incredibly data driven. This is an interesting article on how it’s working sans detail of course. https://www.dpreview.com/forums/post/58017853 FYI I don’t think Messina was the data guy. I think he was the Mook of 2012 or Plouffe of 2008. But for sure he understands it.

    • Seth Gordon

      If Clinton were doing WORSE in her internal polls than the public polls lead us to believe, would any insiders in the Clinton team leak this knowledge?

  • DonC

    I commented on this yesterday so I think this is correct. However, I’m assuming you can’t use this information without compromising the methodology of the predictions.

    No doubt many have seen Jim Messina’s Op Ed piece in the NY Times. He told the story of Obama asking him, on the eve of the 2012 election, why he thought his predictions were right and that of many pollster’s were wrong. The answer to this is that campaigns are predicting based on voter files and demographics rather than polling, and predictions are way more accurate. It’s why Amazon knows what you’ll buy before you do! LOL (Also of interest is that the NY Times Upshot has reported that 20% of those who self reported as being “unlikely” voters have voted early — yeah polling is hard).

    Because a data driven campaign can accurately predict voter behavior, when Robby Mook says Clinton is ahead by 170K votes in Florida rather than behind by 20K as Obama was in 2012, that’s highly likely to be accurate. Far more accurate than a poll. To some extent “small data” prediction is still a new science, but it is a science which has been honed and proven for many years by businesses.

    If you look at how the Clinton campaign sees things, it lines up well with the PEC electoral map, though Ohio and Arizona are perceived as toss-ups and being closer than Iowa.

  • DonC

    I commented on this yesterday so I think this is correct. However, I’m assuming you can’t use this information without compromising the methodology of the predictions.

    No doubt many have seen Jim Messina’s Op Ed piece in the NY Times. He told the story of Obama asking him, on the eve of the 2012 election, why he thought his predictions were right and that of many pollster’s were wrong. The answer to this is that campaigns are predicting based on voter files and demographics rather than polling, and predictions are way more accurate. It’s why Amazon knows what you’ll buy before you do! LOL (Also of interest is that the NY Times Upshot has reported that 20% of those who self reported as being “unlikely” voters have voted early — yeah polling is hard).

    Because a data driven campaign can accurately predict voter behavior, when Robby Mook says Clinton is ahead by 170K votes in Florida rather than behind by 20K as Obama was in 2012, that’s highly likely to be accurate. Far more accurate than a poll. To some extent “small data” prediction is still a new science, but it is a science which has been honed and proven for many years by businesses.

    If you look at how the Clinton campaign sees things, it lines up well with the PEC electoral map, though Ohio and Arizona are perceived as toss-ups and being closer than Iowa.

  • Ebenezer Scrooge

    Five uncorrelated tied races would mean a 3% chance of Republicans holding the Senate if Hillary wins. Sam still gives the Republicans a 20% shot, although this is independent of Hillary’s chances. This is the power of correlation!
    (Illinois plus Nevada does not make it 49, since Nevada is a Democratic hold, while Illinois is a gain. But if you think that McGinty is comfortably ahead as well as Cortez Masto and Duckworth, you’re up to 49, and are tossing coins with NH, MO, IN, WI, and NC.)

    • Clark

      The numbers for Feingold seem to be solid. Also, early voting in Wisconsin is also breaking records (which seems to be one of the big stories in this election). It probably was going to be 51 with Bayh, but he seems to be trending down the last week or so. Still, Missouri is razor close. It all depends on turnout in St. Louis and Kansas City. Reminds me of the Jack Danforth, Harriet Woods races decades ago, just less mean.

  • Jungwoo You

    Why is MS showing +3 Trump on the sidebar? Shouldn’t that be much higher? Wondering if I can safely assume >99% confidence…

    • Not gonna be year 109

      I’m guessing that is supposed to be AZ or GA. No way that MS would be listed in the Power of 1 Vote top 15.

    • (((CassandraLeo)))

      I wouldn’t be so sure about that. Polling has been pretty close there at several points, and it’s less populated (3 mil) than GA (10 mil) or AZ (just under 7 mil), so it might be easier to swing the vote there. The South isn’t what it once was. Several of those states will probably be swing states within a few election cycles if current demographic trends continue.

  • anonymous

    If Clinton wins, they only need 50/50 Kaine makes it the majority vote.

    • Dan

      Yes, but I am not at all convinced that all 50 Democratic Senators would stay together for such things as SCOTUS appointments. There is always one or two that don’t stay in line. One huge example: Manchin from WV. There are a host of issues where he has sided with the GOP. And he is pro-life, voted to defund Planned Parenthood, supported balanced budget amendments, and voted against raising the federal debt ceiling.

    • Leading Edge Boomer

      Agree. Managing Democrats is like herding cats. I hope Sen. Schumer and the rest of the future leadership is ready to pry loose a moderate Republican or two, because Manchin will do whatever is necessary to get re-elected in WV.

    • Richard Vance

      Manchin will do whatever is necessary to get re-elected… Yes. Including changing party!!

    • Billikin

      The question is, if the Senate is 50-50 and Kaine is VP, will the Senate amend its rules to prevent filibusters of judicial nominees? If so, then the Democratic leaders can do some horse trading with the Blue Dogs.

  • Richard

    In the last few elections, close Senate races all broke together in one direction.

    • Kalil

      Iirc, it’s usually ‘almost all’ rather than ‘all’ – Ford in Tennessee in ’06 stands out in my memory, for example.

  • smartone

    The best news on a Sunday!

Leave a Comment