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

How surprising was the Sanders win in Michigan?

March 9th, 2016, 10:15am by Sam Wang

On both sides, last night’s elections kept both front-runners, Hillary Clinton and Donald Trump, on track for the nomination. Nothing’s changed. They should both get a majority of delegates, with probability over 90%.

The slight lead for Sanders in Michigan doesn’t affect the overall dynamic.

In terms of delegates, it was a near-tie. Clinton has an overall lead in pledged delegates and won big in Mississippi, so a tie is not helpful to him. Based on where things are headed, Democrats should start looking ahead to the November election. The same is true for Republicans.

However, Michigan was a notable polling failure: the pre-election polling median was Clinton +17%, yet Sanders ended up with 2% more votes. Was it because of independent voters? Maybe, but that can’t be the whole story, since other open primaries did not have this problem. See the open black circles in the graph above.

Sanders’s overperformance is even more surprising in the context of the exuberance effect I mentioned in 2008, and which is visible here. The overall trend is for Sanders to outperform polls in states where he is strong, and for Clinton to outperform polls where she is strong. The fit line has a slope of approximately 1.4.

The two big exceptions are Michigan and Minnesota. Minnesota could be excused because it had only one poll, and an old one at that. For Michigan there is no excuse. There were 5 polls. Independent voter turnout was unusually high – about 28 percent. Under-30 turnout was also high. Somehow, pollsters missed this. But why Michigan?

Generally, any persuasive explanation would have to be specific to why this particular set of data had a problem. For example, “Bernie has great get-out-the-vote” is ad hoc.

Update: Philip Bump has a possible explanation: turnout models in this year’s polls were based on 2008. However, 2008 was an anomalous election because Michigan had been disqualified from sending delegates to the national convention. Consequently, turnout in 2008 was extremely low and skewed toward older voters. In contrast, yesterday’s voting was evenly distributed across major age groups. In other words, older voters were overcounted and younger voters were undercounted.

Although this is a significant polling error, it stands in contrast to many polls that did better. In polling, many judgments go into sampling and weighting. Professional pollsters sometimes make wrong judgments, but the error is only visible in retrospect. Failure happens, and it is useful to understand why. Considering that such failures of judgment are inevitable, it is useful to know that even a 20-point lead does not assure a win. For future cases, it might be best to imagine that such a lead comes with, say, a 2% probability of a surprising outcome.

Tags: 2016 Election · President

83 Comments so far ↓

  • InmanRoshi

    I tend to think a big bloc of institutional voters broke a certain way at the last moment after polling had concluded their final surveys. I suspect the UAW, which never publicly endorsed either candidate. Bernie speaks the language of a UAW organizer, and Hilary is the spouse of the President who implemented NAFTA. I’ve heard a couple of journalists on the ground in Michigan claim that many UAW workers were irate and insulted that Hillary tried to portray Sanders as anti- auto-bailout (he voted against the bailout only because there were provisions to help out the auto industry’s financiers).

    • Kalil

      The exit polls I saw had Clinton narrowly winning union voters, though.

    • 538 Refugee

      “I’ve heard a couple of journalists on the ground in Michigan claim that many UAW workers were irate and insulted that Hillary tried to portray Sanders as anti- auto-bailout”

      I know I was really disappointed at Clinton for that one. But, now is the time for her to learn her true weakness’s with the base because we know Trump will hammer hard on other fronts.

    • Matt McIrvin

      To some extent, this is why we have primaries. If Sanders hits her on an issue that Trump is also guaranteed to bring up, that’s important preparation/vetting for the general election.

    • Olav Grinde

      Matt, I couldn’t agree more! Which I think it’s so valuable for Hillary (the probable nominee) to face a real fight from Bernie, rather than a coronation.

      And one more thing: I am proud of the civility, genuine mutual respect and issue focus seen in the Democratic Presidential Debates. What a wonderful contrast to ad hominem attacks, the shouting and the outright lies heard in the Republican Debates!

  • Kevin

    I didn’t watch the last debate, but coverage suggested it was strong for Bernie.

  • JayBoy2k

    There is a decent article on Huffpost. Larger % of young voters than anticipated and better performance in black communities that the polls indicated.

    • Olav Grinde

      The Sanders Campaign has long claimed that Bernie has been struggling with the handicap of being unknown, and that he would start getting many more minority votes as his message became more known.

      As JayBoy2k points out, I understand that Bernie got a far higher portion of the black vote in Michigan than he has in other states.

      I know that one state does not make a pattern. But I would be curious to know whether we might be seeing some decisive Sanders-momentum amongst minority voters.

      Sam, I am not suggesting that this changes the probable outcome – a Hillary nomination – but I would be very interested to hear your thoughts on this possibility.

    • Matt McIrvin

      He got hammered among black voters in Mississippi on the same day, though. I suspect it’s not so much momentum as a North/South thing: northern African-Americans have never been as unanimously in support of Hillary Clinton as southern ones, though she still gets a majority.

      Also, Sanders carried Arab-Americans, an important demographic in Michigan. I think I agree with the people who regard this as having to do with Clinton’s perceived foreign-policy hawkishness and connections to megadonor/Israel hawk Haim Saban.

      Neither of these groups are likely to switch to Donald Trump, a thing that’s worth thinking about when judging how much general-election weakness this demonstrates for Clinton.

    • Amitabh Lath

      The point should not be who voted for whom and for what reason, but why the polls missed it so egregiously. Did they not call up any Arab-Americans? Did they just cut and paste African American response from other states?

    • JayBoy2k

      This is the Monmouth pollster on their biggest factor — they had Clinton at 13% up
      Quote “[T]he biggest difference seems to be among white voters and gender,” he said in an email. “We had 49/48 Clinton among white voters compared to 42/57 in the exit polls. We had women at 59/36 versus 53/46 exit poll. We had men at 48/49 versus 44/54. So, it’s a little bit of everything.”

      That’s a big difference from +1 to -15.. Who were they asking in the poll? Significant that women were more negative than men, poll versus actuals?

  • Matt McIrvin

    This was Clinton’s first major test of the cycle in the Rust Belt; if you wanted to construct a narrative you could say it forebodes general-election trouble for Clinton vs. Trump in Michigan, Pennsylvania and Ohio, since railing against international trade is a thing for him just as it is for Sanders.

    (And Clinton getting the Obama 2012 map minus MI, PA, OH and CO is a 269-269 tie and a decision in the House, which is a great hook for an election story!)

    • Todd S. Horowitz

      Is there any evidence that primary results predict general election results?

  • Petey

    “Somehow, pollsters missed this. But why Michigan?”

    Well, gotta resort to anecdotal intuition, rather than data, but…

    I thought Sanders had an outside shot in Michigan despite the crazy poll deficit because I remember the ’88 campaign where Jesse Jackson won a surprise victory running on a social democrat agenda.

    So, I thought Michigan would be receptive to Sanders. Don’t get me wrong: I was still pretty surprised. But I did have some ideas ahead of time it might be close.

    Back to data: zero idea why the polling was such a failure specifically there.

    (FWIW, I’m old enough to remember what is widely being cited today as the only comparable precedent for a polling miss this big: the ’84 NH Dem primary. And for reasons I can no longer explain in detail, I was 100% sure that Hart was going to win that 48 hours out, despite knowing the polls.)

    • InmanRoshi

      Good thoughts.

      Some past Michigan Primary winners…

      George Wallace
      George Bush over Ronald Reagan
      Jesse Jackson
      Pat Robertson
      John McCain over GW Bush.

    • Petey


      Yup. Michigan since ’68 has had an odd outlier vibe that is strongly populist, which expresses itself in either nativist or social democratic form.

      And to add one more to your list:

      IIRC, Pat Buchanan put up a very strong challenge there in either ’92 or ’96, until the establishment candidate blanketed the airwaves with a late ad noting that Buchanan drove a German Mercedes, which totally killed him in that state.

    • Olav Grinde

      Petey, these days I don’t suppose it would have quite the same effect to point out that a candidate is using a Chinese-produced iPhone? Wearing garments made in Bangladesh or Italy, or writing books that might be printed abroad? ;)

  • Petey

    Phillip Bump at the WaPo has a VERY plausible rationale for why the polls were so wrong….

  • Olav Grinde

    Credit should be given where credit is due: Hillary has won the Southern states, hands down.

    However, as I understand it, Bernie has won an impressive 9 / 12 states outside the deep south: New Hampshire, Colorado, Minnesota, Oklahoma, Vermont, Kansas, Nebraska, Maine and Michigan. (Losing Iowa, Nevada and Massachusetts.)

    Sanders he has won 56 % of the pledged delegates in Northern and Western states, while garnering a far-more-meagre 30 % share in the former Confederacy.

    If my counting is correct, the only Southern states yet to be decided in the south are Florida and North Carolina.

    What happens if Bernie maintains his 30 % share in the South, while increasing his delegate share elsewhere from 56 to 60 %? You might think that a tall order, but a good friend tells that Bernie would then go to the convention with a 2097 to 1955 lead.

    With superdelegates deciding the balance…

    • Mark F.

      Current polling shows Sanders down by huge amounts in Ohio, IL, NC and FL. Do you expect him to win any of these states or even approach 50% of the popular vote in any? If he has a good night and fights Clinton to a draw on March 15, she maintains her delegate lead and then he needs to start winning states with 60-65% of the vote or more. Unlikely.

    • Froggy

      Most of the delegates still up for grabs are in places that look more like Michigan (where Sanders got 53% of the delegates), than in the places were Sanders has gotten 60% or more of the delegates (NH, CO, VT, NE, KS, ME, MN). Except for Colorado, not a lot of minorities in these states, and among them only Colorado and Minnesota have large cities.

      Next week we have three non-southern states with large cities and significant minority populations (OH, IL, and MO), with many more such states in the wings (CA, NY, NJ, PA). Your math works, but for Sanders to even continue getting 56% of the delegates in non-southern states seems like a tall order.

    • InmanRoshi

      Sure, if you could just draw a hard line around the “Deep South” as a homogenous clump on the map, but I’m not sure it works that way. Why would we expect Maryland to be so much significantly worse for Hillary than bordering Virginia was? It has the 4th largest % of African Americans of any state in the country.

      You can classify Texas as the “Deep South” because it was a former Confederate State, but it could be equally classified as part of the Southwest with huge concentrations of hispanic voters who Hillary did quite well with on March 2nd. Bernie isn’t merely going to have to hold serve or pull out slight wins in New Mexico or Arizona, he’s going to need to win by huge margins to cut into Hillary’s delegate lead.

    • RDT

      Really replying to Mark, below… My guess is that Ohio will be not all that different from MI (which, when you get past the surprise factor was nearly as much of a “tie” as MA and IA), come closer than the polling in IL, and that Hillary will win FL solidly and NC fairly solidly.

      I’m guessing that there’s a combination of anger and strategic Trump/Kasich voting in open primaries that is making the upper-midwest states hard to poll, so that Hillary’s lead in OH may not be more solid that her lead in MI. But I would be very surprised if the same applied to FL.

    • Matt McIrvin

      Do you expect him to win any of these states or even approach 50% of the popular vote in any?

      The surprise in Michigan certainly suggests that anything is possible. The question is whether the problem with the polls there was a one-off, something to do with Michigan specifically (like the use of the messed-up ’08 primary as a turnout model), or some indication of wider failure in modeling the Rust Belt, or something having to do with a dramatic change in the nature of the race nationwide (which I imagine Sanders supporters are hoping).

    • Matt McIrvin

      There are two polls in Illinois now that show the D primary race nearly even, and some overlap the time period of other pools showing Clinton up 65-25. That makes no sense; somebody is just blatantly wrong, unless at some point Bernie gained 25 points literally overnight.

  • Sam Wang

    If you want to talk about FiveThirtyEight, do it over there please!

  • Amitabh Lath

    The “polling working fine in other contests” could well be due to lots of the same mistakes which either didn’t amount to much or canceled each other out in the other states.

    My hypothesis would be demographics and response rates. Perhaps the Dem contingent in previous states has been somewhat homogeneous and Michigan presented a specific challenge: lots of hard to sample college students, an Arab ethnic population that may have voted as a block (and had a low sample rate), an African American population that showed an age effect that was not calibrated for, etc. etc.

    Pollsters are not going to do any sort of public post-mortem where we could look at raw data and maybe learn something. They will just put this behind them, and pat themselves on the back if they call Florida close enough.

    • Matt McIrvin

      I like Philip Bump’s hypothesis as cited by Petey below. The 2008 Michigan primary was anomalous: they scheduled it earlier than party rules allowed, Michigan had its delegates stripped from the convention, and Obama actually withdrew his name from the ballot (he wasn’t allowed to do this in Florida, which had the same problem).

      So the Michigan primary didn’t matter, and Obama’s name wasn’t even on the ballot. All that suppressed and warped turnout so severely that 2008 would have been a very bad model for 2016 turnout. If pollsters were relying on 2008 data for weighting purposes, you’d expect bad results.

    • Matt McIrvin

      …also, in the two presidential election cycles before that, Michigan Democrats were using a caucus.

      So there was absolutely no even approximately recent model for turnout in a competitive Michigan Democratic primary.

  • Scott

    There is an article at Vox attributing the miss to underestimating the percentage of young voters. I’m not sure about the overlap with the Post’s hypothesis.

    One reason that some people thought Sanders would under-perform is that a number of schools have spring break this week and thought that students would be in Florida instead of at school or home.

  • bks

    Stop freaking out. This is what real data look like. Now if Ohio and one other state show a problem with the models, then search for the problem. Otherwise remember that the roulette wheel comes up black six times in a row a couple times a day.

    • Amitabh Lath

      Good point, fluctuations happen. The average before the Michigan primary was 57%/39% Clinton/Sanders.

      Statistical uncertainties are small in these kinds of aggregations, but let’s assume an overall 5% systematic uncertainty (I totally made that up) that’s a 3.6 sigma effect.

      A two tailed p-value = 3.2 e-4. Really small. But that’s a local p-value. There are 50 states (not all are polled but never mind) and two parties and then the general election polls, congress and senate polls… a global p-value might indicate a one in a thousand chance of a fail of this magnitude happening somewhere in an election season.

      But any more such fails and we become suspicious that the pollsters do not know what they are doing.

  • Latichever

    I’m not sure of how the optics will look if the headline is Bernie winning, and the subtitle is Hillary expanding her delegate lead.

    That said, I’m excited that my vote in Connecticut’s April 26 primary may have some value for the first time since amoebas roamed the earth–unless NY ruins it for us on April 19.

    • Jack Tenold

      If Clinton is expanding her delegate lead, she is winning. A hard lesson that she learned in 2008.

    • Some Body

      When did CT vote in 2008?
      PS: Amoebas still roam the Earth today, FWIW.

    • Froggy

      The Connecticut primary was on Super Tuesday in 2008, February 5, 2008. (I think that was before amoebas, though.)

  • Alan Cobo-Lewis

    Since Florida was also sanctioned by the DNC in 2008 for an early primary, does the Michigan explanation generate a testable hypothesis about the relationship between polls and outcomes in Florida on 3/15/2016? If so, do you have the data needed to make a prediction?

    • Sam Wang

      It could be done. Thoughts: I think Florida has more information on primary electorates, so the error might be smaller. There are many more polls, and one would have to drill into the demographics to see if there was anything weird about pollster assumptions – seems like maybe not since there are so many polls.

      Whatever happens, there probably won’t be a big headline. Clinton leads about about 30 points in Florida, 10 points more than Michigan. So a Michigan-sized error won’t lead to a Sanders “win.”

      Finally, I wrote “win” because delegates are allocated not winner-take-all, but proportionally by vote. This rule gives power to voters everywhere. I actually think that the headlines about Michigan are totally misplaced, on the grounds that the effect of this error is not the state win, but the quantitative effect of the number of votes provided by 20% of Michigan Democratic primary voters. That turns out to be about 1% of all Democratic votes cast so far.

    • Matt McIrvin

      Florida 2008 was also less anomalous in that Obama was not allowed to withdraw his name from the ballot.

  • Sam Wang

    This is not that big a deal.

    Amit, polls are not pure sampling problems involving a simple lookup in a t-table. Many judgments go into sampling and weighting. This is a field populated by professionals. Sometimes those professionals are unable to identify every single systematic error. Once in a while, a failure happens. It is useful to understand why.

    For example, the Nevada Senate race between Harry Reid and Sharron Angle, where Angle led every homestretch poll but Reid won, made me think about the high mobility of voters in and out of that state. It made me generally cautious about that state and Alaska, which has similar characteristics.

    In the case of Michigan, I think Philip Bump might be on to something with the 2008 turnout hypothesis. Considering that such failures of judgment are inevitable, it might be good to assign a minimum probability based on how often they happen. Maybe 2%, so that Clinton’s win probability was actually 98%. That level of belief seems about right to me.

    • Amitabh Lath

      I’m a little wary of post-hoc assignments of uncertainty, (they tend to come up exclusively when things go kaput) but you (and Bump) make a compelling case about these specific states. So rather than a chance of a few in 10k we go to few in a hundred. That’s a significant loss of certainty in your measurement. It might be good to get an understanding of various states and their peculiarities and overall systematic.

      PS: Fields populated by professionals brought us cold fusion.

    • Some Body

      Reminds me of the disagreement you had with 538 back in 2012 about Obama’s probability of winning the general election, where Silver was factoring in the chance of systematic polling error, and you were reluctant to. A bit of a reversal in roles now.
      Anyway, with ~20 pts margins at stake, I’d put the chance of systematic error being large enough to wrongly predict the winner considerably below 2%. The one precedent in US primary polling was back in 1984, so far more than 50 contests per miss of this magnitude. Actually, the >99% probability 538 had down wasn’t wrong (as stated; the exact number they rounded that way may have been overconfident, or underconfident).

    • Matt McIrvin

      I think the probability of a major miss is much higher in a state primary than in a general presidential election: turnout is much lower and more variable in primaries to begin with, so large sampling mistakes based on a bad turnout model are a bigger danger.

      And this time around, for the specific case of Michigan, there was actually no good prior election to serve as a model, and the most recent one was actively poor.

  • Mark F.

    Looks like Kasich may win Ohio next week, and Rubio will probably lose Florida How will that impact Trump and Cruz?

    • 538 Refugee

      Interesting. In the places where the outside money is finally attacking Trump hard he is going down in the latest polls. ‘Yuuge’ nosedive in Ohio and a more modest one in Florida that has Rubio up to 9 points down. Nationally Trump is on a long steady incline though up to 40%. If Trump is the eventual Republican nominee the blueprint is already in place for the Democrats.

    • whatever next

      Good for Cruz – he’s more likely to get a larger proportion of Rubio voters than he would have done from Kasich ones if it had been Kasich dropping out.

      Or, specifically, 2nd preferences show him likely to pick up more Rubio supporters than Trump.

    • Sam Wang

      Good for Trump. Keeps Kasich in, field remains divided.

    • Matt McIrvin

      Nationally, Trump’s overall support level doesn’t seem to be dropping in the latest polls. It’ll be interesting to see what happens after yesterday’s escalation of ugliness, Trump’s rally cancellations and claims that he is being victimized by protesters.

      The main trend seems to be that, in the role of great establishment hope, Kasich is gaining and Rubio is dropping.

  • truedson

    Surprising that no one else is picking up the Bump explanation. Not TPM, Krugman, who just posted on Michigan etc… I guess we will find out next Tuesday.

    • Matt McIrvin

      Usually they just use the Michigan surprise as a lead-in to talking about why Sanders appeals to some people, as opposed to specifically talking about why the MI polls were wrong about it.

  • bks

    The Bermuda triangle seems to have claimed the Virgin Islands’ GOP delegates.

    • Mark F.

      All 9 elected delegates from the Virgin Islands are uncommitted.

    • MAT

      That makes 117 unbound GOP delegates that I’ve identified so far: Penn, ND, WY & the Virgin Islands. It’s also clear operatives are starting to make moves to grab those positions. Pass the popcorn, this is going to get interesting…..

    • Sam Wang

      See Taniel’s spreadsheet for a few dozen more unbound delegates.

    • Petey

      “That makes 117 unbound GOP delegates that I’ve identified so far: Penn, ND, WY & the Virgin Islands.”

      Are you including Colorado?

      Last time I ran the numbers, I got that Trump would need 55% of ‘earned delegates’ to get a majority because of the unbounds, but a few more unbounds have been added since…

  • Jim Kehoe

    Is there any modelling on the implications of state-by-state primary results for the general election? I ask because Sanders does better than Clinton, albeit by small margins, in states that the Democrats tend to carry, e.g., Michigan, Minnesota. Conversely, Clinton does better in states that the Democrats have a snowball’s chance of winning.

  • Eri

    Hi Sam,

    Have you ever had a chance yet to see Alan Abramowitz’s prediction at Sabato’s Crystal Ball?

    A two-predictor — just two — model, the racial composition of the Dem primary electorate and a dummy for region. He shows that it explains over 90% of the variance in Hillary’s vote share in the contests through March 8.

    Curious to have your thoughts if any. His mod does not take into account anything else that people have been talking about, e.g., open/close, primary/caucus, etc.

    • Matt McIrvin

      That model is consistent with the most recent Illinois polls, which are dramatically different from earlier polls.

  • MAT


    OK, if I’m reading the rules right in Colorado, there are 3 completely unbound delegates, 13 that can be elected at the State Convention either bound or unbound, depending on if they declare for a candidate prior to election as a delegate, and 21 more delegates selected at the district caucuses that could be bound or unbound. Having been to some district & state delegate selections, it would be political malpractice for the ‘establishment’ not to grab these delegates- this is the type of stuff the Ron Paul guys were great at in 2012. So Colorado brings the potential unbound delegates up to 154, which is 12.45% of the delegates needed for the nomination. This is really worth watching. Any more floating around?

  • 538 Refugee

    Generally a good read detailing Trump laying the groundwork for his run. About halfway down the page is the Romney team’s assessment of the Trump endorsement in the 2012 presidential race. It may come as no surprise that it is just a tad different than what Trump tells.

    article quote:

    In an appeal to Trump’s vanity, the Romney campaign stressed that his endorsement was so vital — with such potential to ripple in the media — that it would be a mistake to dilute the impact with a question-and-answer session.

    “The self-professed genius was just stupid enough to buy our ruse,” said Ryan Williams, a former spokesman for the Romney campaign. While they agreed to hold the event in a Trump hotel, the campaign put up blue curtains around the ballroom when the endorsement took place, so that Romney did not appear to be standing “in a burlesque house or one of Saddam’s palaces,” Williams said. On stage, as the cameras captured the moment, Romney seemed almost bewildered.

  • Olav Grinde

    And here are the results:

    Northern Marianas (D):
    Sanders 2, Hillary 4 (+5 superdelegates)

    District of Columbia (R):
    Rubio 10, Kasich 9

    Wyoming (R):
    Cruz 9, Trump 1, Rubio 1, Uncommitted 1

    Guam (R):
    Cruz 1, Uncommitted 0

  • Amitabh Lath

    Douthat’s 35%: Generally I am in favor of pundits using numbers, but I am confused by this statement in his latest:

    “…if they cannot be mobilized to prevent 35 percent of the Republican primary electorate from imposing a Trump nomination on the party.”

    Where is he getting this 35%? Is that the result of a weighted average of all primaries and caucuses so far?

    Trump’s numbers have been all over the map, hitting close to 50% in MA and single digits in the Wyoming caucuses.

    How would one go about estimating this number properly? Obviously caucuses have a suppression factor built in, closed primaries have a smaller one.

    But the biggest issue is how to assign the non-Trump vote. Obviously some voters (10%? or 30%? ) for fallen candidates will go over to Trump but how to assess that value?

    • Sam Wang

      I just completed a full simulation of primaries. Douthat’s number is almost exactly right – if all elections were held at once. I put the number at 34%. I haven’t calculated the error bar…probably 1-2%.

      However, conditions are about to change. After Tuesday there will be a shakeup, which will probably involve Rubio losing Florida and effectively withdrawing from the race. I do not know what will happen with Kasich and Ohio. According to a PPP poll, Rubio supporters are more likely go to Cruz or Kasich than to Trump. At that point, Trump will need somewhat more support than indicated by my full simulation.

  • JayBoy2k

    Chuck Todd on Meet the Press just had a curious statistic . In GOP primary states where the unemployment rate is > 5% and the African American population is >8% , Trump is 9 for 9. There are only 4 such states left. NC, FL, IL on this Tuesday and Connecticut.
    I predict he will not go 13 for 13.

  • truedson

    What is with the Illinois polling today? Clinton goes from 40 ahead to tied??

    • George

      yeah – usually you’d say the outlier is wrong, but after MI (being open primary, as is IL) you have to wonder. But no real movement in OH, which you would think would show at least some of the IL trend. Internet based and an odd methodology. We will have to see if other polls confirm or disconfirm (if we get anymore before Tuesday). Waiting for Yoda Sam to chime in.

    • Sam Wang

      Now that is interesting. I wonder if YouGov and Marist are applying very different weightings from We Ask America and Research America.

    • Matt McIrvin

      More polls today saying the same thing. (One of them has Clinton and Sanders at 31-31 with an enormous “undecided” contingent.)

      Either something huge happened in Illinois in the second week of March (last-minute campaign ad blitz? Sanders perceived as a viable candidate after Michigan?) or some polls have vastly wrong sampling assumptions.

    • Matt McIrvin

      …30-31, that is.

    • truedson

      I live just west of Chicago and this is bewildering in such a drastic change since nothing noteworthy has happened in the last week here.

      It’s the equivalent of Bear fans suddenly becoming Packer fans….I’m stumped.

    • Matt McIrvin

      Hasn’t there been a giant blitz of primary campaign ads? I hear Illinois people talking about them.

    • Matt McIrvin

      Also the race does seem to be narrowing in Ohio now as well, though it’s not as extreme a jump as what happened in Illinois.

      I think Sanders may carry much of the upper Midwest. It probably won’t be enough to get him the nomination since he needs to be winning by gigantic margins to overcome Clinton’s delegate gains from the South. It could prolong the race, though.

    • truedson

      There’s been some Bernie ads but other than sports I don’t watch much. But no real ad blitz. Certainly none that people are talking about. The Chicago Tribune, which didn’t endorse anyone in the Democratic race, hasn’t mentioned anything note worthy. Their own poll in the paper Sunday had Clinton ahead by 40 points. The big news was the Trump melee at UIC Friday. But that would seem to me to work in Clinton’s favor.

      As Sam noted they must have changed their methodology. I still think Hillary will win big here, but it is bewildering.

    • truedson

      Apparently Bernie has had ads linking Clinton to the Mayor who is very unpopular at the moment especially with African-Americans who are normally Clinton supporters. But Emanual’s poll numbers have been down since December so this late swing is confusing.

  • RDT

    It seems to me that the need for weighting must lead to much bigger uncertainties in races where there are big differences in demographic groups. If Clinton and Sanders were equally popular in all age groups, getting the under-30 turnout correct wouldn’t be so crucial to a correct result. Do pollsters take any of this into account when they report uncertainty, or is that just based on the number of responses in their sample?

  • A New Jersey Farmer

    New CBS and NBC News polls show some significant movement towards Sanders in Ohio and Illinois. North Carolina and Florida look strong for Hillary. Interesting to see what the polls say tomorrow. if there is more movement to Bernie in the Midwest, that would at least complicate Hillary’s electability message.

    • DonC

      Not following this. The primary is about winning delegates, not states or the narrative. If it were about states then Hilary would have won in 2008. Winning a large state 51%-49% gives you fewer delegates than winning a smaller state by a large margin. Hence last week Clinton increased her lead while losing Michigan.

      Also hard to see how it affects electability in the general, My assumption is that voters preferring Sanders to Clinton would not vote for Trump but voters preferring Kasich to Trump might vote for Clinton. Even Kasich is having a hard time voting for Trump rather than Clinton!

    • Ryan


      I’m really not so sure that it’s a good assumption that Sanders supporters move toward Clinton. It’s been my biggest fear all along that Sanders and Trump are drawing partially from the same well of low income economic angst.

      Said another way, Sanders and Trump have more in common in terms of economic messaging than one might otherwise think, even if their presidencies would unfold vastly differently.

  • Mark F.

    It’s all about delegates, folks, not winning states. Hard to see Sanders catching up Tuesday even with another surprise win or two.

Leave a Comment