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

Why Your Vote Matters: In the Midwest, will Republicans lose their grip on power?

November 1st, 2018, 10:43am by Sam Wang

(Written in collaboration with Owen Engel ’21.)

Six of the most important gubernatorial elections next week take place in the Midwest: Ohio, Wisconsin, Iowa, Minnesota, Michigan, and Illinois. In October polling, Democrats lead in all six races. In four states (Iowa, Wisconsin, Michigan, and Ohio), Republicans are at risk of losing sole legislative+gubernatorial control of state government, and in three cases they used their power to draw extreme partisan gerrymanders. Conversely, if Republicans in Minnesota win the governorship and a key state Senate race, they will gain total control. A lot is at stake!

A Republican trifecta opens the door for policy that might not be otherwise achievable, such as abortion restrictions, repeal of gun-control regulations, and anti-LGBT legislation. Conversely, a Democratic governor can veto such legislation.

In all states but Iowa and Ohio, the governor has to sign off on the redrawing of district lines after the 2020 Census. In 2010, trifectas led to extreme partisan gerrymanders in Wisconsin, Ohio, and Michigan – three out of eight nationally. Lawsuits can partially address gerrymandering, but are lengthy, uncertain, and leave the offense in place for one or more elections. A faster and more effective way is to district fairly in the first place.

Democrats’ leads are narrow enough that on average a Republican may win one of them. Here are the October polls:

Ohio. In the race to replace Governor John Kasich (R), former chair of the Consumer Financial Protection Bureau Richard Cordray (D) takes on Ohio Attorney General Mike DeWine (R). This one appears super-close: Cordray leads DeWine by 1.5 +/- 1.9% (median +/- estimated SEM, n=4 polls).

Wisconsin. Challenger Tony Evers (D) may unseat once-rising-star and incumbent Scott Walker (R). Walker has been a polarizing figure, especially in labor relations. Wisconsin Assembly is among the most gerrymandered in the nation, and is the subject of the high-profile case of Gill v. Whitford. An Evers win will prevent that gerrymander from recurring. October polls (n=3) show Evers up by 3.0 +/- 2.0 %.

Iowa. Challenger Fred Hubbell (D) leads incumbent Kim Reynolds (R) by 3.5% in just two polls. Reynolds was appointed in May 2016 after former Governor Terry Branstad (R) resigned to become the ambassador to China.

Minnesota. Governor Mark Dayton (D) is not running for re-election; the candidates are Congressman Tim Walz (D) and Hennepin County Commissioner Jeff Johnson (R). Walz leads by 6.0 +/- 5.5% (median +/- estimated SEM, 3 polls in October). In addition to the key state Senate race (SD13), a seat on the Supreme Court pits incumbent Democrat Margaret Chutich against GOP-backed Michelle MacDonald. Embattled DNC Vice-Chair Keith Ellison is lagging in his race for Attorney General.

Michigan. Former state Senate Minority Leader Gretchen Whitmer (D) leads state Attorney General Bill Schuette (R) by 8.0 +/- 2.0% (n=5 polls). This one’s potentially important for multiple reasons: the extreme gerrymander, the Flint water crisis, and Michigan’s crumbling infrastructure, just to name a few issues.

Illinois: Since summer, businessman J.B. Pritzker (D) has led unpopular incumbent Bruce Rauner (R) by 11 to 22 percentage points (n=10 polls).

Tags: 2018 Election

9 Comments so far ↓

  • Sam Wang

    ArcticStones writes:

    Pundits will always argue about the value of Early Voting numbers relative to polls. (Nate Silver, for instance, asks us to ignore the EV statistics, but then again he has a poll-based business brand to protect.)

    Certainly Early Voting statistics must be analyzed carefully and always in context ­ but we do have some very interesting numbers from Iowa:

    In 2014, registered Democrats held a +1.6 % lead in Iowa¹s Early Vote. So far in 2018, however, Democrats had an +8.1 % edge!

    Moreover, as of yesterday 390,521 Iowans had already voted. That¹s more than a third of the state¹s total 2014 vote, including Election Day.

    Time will tell whether more Democrats are voting, or whether more of them are simply voting earlier. Whatever the final outcome, we are seeing impressive turnout figures ­ especially amongst Democrats.

    Which leaves me with a key question about polling numbers. Polling figures often take into account Likely Voter models, which again are based on assumptions about turnout.

    Strangely enough, I have not seen prominent articles about pollsters radically adjusting their models to reflect the extreme turnout that we are seeing.

    For in the 2018 Midterm Elections so far, we are seeing extremely high turnout figures in many states. In several states we are seeing Early Voting that is very close to exceeding the Total Vote for 2014!

    Arizona, Tennessee and Texas are in a league of their own in that regard.

    • LondonYoung

      A prediction about the 2016 election based on early voting in Iowa:

      Prediction was Clinton by 2
      Actual was Trump by 10

      Maine CD-2 was another huge miss, but some other states were actually quite accurate.

      Just a few days left to study this data, but I can’t find anybody (yet) applying 2016 lessons to 2018 when it comes to early voting.

    • Marco Ciocca

      The issue of the Likely Voter Model is what probably misled pollsters in 2016. Think what the House and Senate maps would look like NOW if these Likely Voter models missed a really determined Democratic electorate by 2-3%. The house would become a blowout and the Senate would be in play. Hope? :)

    • Compton Tucker

      Dear Princeton Elections,

      I’m interested in sampling theory and would like to see something on the uncertainty of variables used to predict your election results. I probably missed this but where should I look?

      Thanks in advance and best regards,

      Compton Tucker

  • ArcticStones

    London, I am well aware of Michael McDonald and have been following his running tally of the Early Vote at electproject dot org, plus his Twitter feed.

    I’ve also been watching what Jon Ralston has to say about Nevada, Ryan Anderson about Georgia, and Michael Li about Texas. They are giving the EV numbers context.

    I am well aware of the dangers of making predictions based on the Early Voting numbers. McDonald missed in 2016, as did Dr. Wang based on polls.

    But that wasn’t my question. I am far more interested in whether pollsters have adjusted their Likely Voter models on the basis of the extremely high Early Voting numbers. The turnout so far does not resemble any Midterm Election in recent decades – they are far closer to Presidential Elections.

    So, my question is: Which pollsters have adjusted their Likely Voter models? How have they adjusted them? And how are the polling results of the new model at variance from the old?

    I think those are the pertinent questions.

    • LondonYoung

      Dr. Wang provides all his work so I don’t quite agree with saying he “missed” in 2016 – a few comments about eating a bug are kinda just editorializing around a model that did as advertised.

      The pollsters, however, are a nightmare. I can never find very much info on what each one is actually doing.

    • Jeremiah

      My understanding is that the model linearly reduced the probability of uncertainty down to zero from some time out. I think this was a misinterpretation of what the historical data was telling us, i.e. the uncertainty should have been constant all the way to election day. If these probabilities had been left in Dr Wang’s forecast would have been less certain and maybe more realistic. (There may also have been some correlation that might not have been accounted for – I’m not sure.)

    • Sam Wang

      Thanks – this is correct. I basically set a floor on uncertainty that was too low. I did this in May 2016 and never revisited it. I didn’t want to disrupt the calculation in late October. In retrospect it would have been better to take my lumps for changing the code – it was a single line!

    • LondonYoung

      But this is just calibrating a model.
      Anyone was free to download your code and adjust it.
      They were also free to just challenge that line here on the forum.

Leave a Comment