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

Is the Senate Kavanaugh bounce partially ending?

October 24th, 2018, 3:29pm by Sam Wang

People often ask if polls move opinion. Not that many people pay direct attention to the numbers. However, polls do set the tone for what journalists and pundits write…with some delay. Combined with the time it takes to conduct and release a poll, this means that news articles can be a lagging indicator of the state of play.

Currently, political writers have noted the turn of Senate races against Democrats. This comes from several key races in Republican states – Tennessee, Texas, and North Dakota – moving toward Republicans at the same time that the Kavanaugh confirmation hearing came to a crescendo. Few of them have noticed that opinion seems to be swinging back:

Recent poll medians currently show the leaders in those states as Blackburn (R-TN) by 3%, Cruz (R-TX) by 8%, and Cramer (R-ND) by 12%. So in red states, post-Kavanaugh sentiment is partially holding. It’s unlikely that Democrats will win any of them, though I am keeping an eye on Tennessee.

Leaving those three races aside, Democrats have a narrow path to 50 seats – and split control. This would involve winning Arizona, Florida, Indiana, Missouri, and Nevada, all of which are within two percentage points. Home-stretch polling error in midterm elections has been as high as 4 percentage points, so it’s possible. However, if the error goes in the other direction, Democrats could go as low as 45 seats. The median result is currently 47-48 Democratic seats and 52-53 Republican seats.

Why, in the face of high disapproval for Trump and pro-Democratic national sentiment (see House data), do Democrats still face an uphill battle in the Senate? Most of you already know the reason: lots of pro-Trump states have Senate races this year. But this year’s outcome also sets the stage for 2020, when Democrats have a far more favorable playing field. Even if the Senate does not flip for 2019, control in 2021 will be highly consequential for whoever is president – and for all of us.

Tags: 2018 Election · Senate

6 Comments so far ↓

  • Matthew McIrvin

    But your House Meta-Margin just plunged again. It seems really, really noisy.

  • LondonYoung

    Things are way more partisan now than they were in 2001, and I suspect the GOP would just keep control if there was a 50 50 split – but that would be decided by the likes of Murkowski and Collins.

    But with the house in the hands of the dems, this really only affects confirming judges. So maybe the GOP would keep control of judiciary and agree to split the rest.

  • ArcticStones

    I have a related set of questions about polling:

    In state after state, we are seeing extremely high Early Vote turnout – in some cases even exceeding that of the 2016 Presidential Election…

    To what extent have pollsters adjusted their Likely Voter models to reflect the extremely high Early Vote? Have any? And if so, which?

    If so, how has this impacted the Meta-Margins?

    If not, and if also the total vote is historically high, what sort of discrepancies between predictions and actual results are we likely to see?

    (Perhaps I have missed them, but I have not seen articles where pollsters talk loudly about having to adjust their Likely Voter models based on extremely high turnout.)

  • Michael K

    Can you clarify something about the Senate Meta margin?

    As I type this, it is showing R+5.0%. However, if I’m reading the ‘Close Senate Races’ table correctly, it looks like the ‘tipping point’ race is Tennessee, which is currently R+3.0%. So if all the local race poll averages are off by exactly 3.0% (too optimistic for Rs), then I believe Democrats win 50 seats (the 43 likely D seats plus NJ, FL, MT, AZ, IND, MO, NV) and the potential 51st seat (TN) would be on knife’s edge.

    I assume that the difference between the 5.0% and the 3.0% comes from Democrats facing more local-polling-error risk because they need to win more of the competitive races. Whereas Republicans have many places where one case of local-polling-error towards Ds could bail them out even if the systemic-polling-error is 3.0% towards Rs.

    But this involves making some assumptions about the magnitude and likelihood of local-polling-errors versus various the magnitude and likelihoods of various systemic-polling-errors, doesn’t it? I gather from the Meta Margin FAQ that you are using “the amount of variation observed in similar past [Senate?] races.”(?) Can you highlights what those historical variation numbers are (local and systemic), and to what extent local-polling-error and systemic-polling-error are related (e.g. does higher systemic-polling-error historically translate to higher local-polling-error)? Thanks.

    • Sam Wang

      Generally I think you have the right idea. I have to inspect the data to give you an exact answer, but a brief take:

      If polling margins were all off by 3 points, then the R+3% race will still only be a tossup, which would be about 50.5 seats on average (only counting the tossup, and counting all the other probabilities as certain). That’s midway between 50 and 51, so you are right, that would suggest a Meta-Margin of 3%.

      One explanation for the gap is that there are a bunch of super-close races that are more D-leaning than Tennessee’s R+3%. Each of them has a downside risk for Democrats, even if polls are off by 3%. But I’ll have to drill into that.

      Generally we calculate probabilities using estimated SEM, then turn that into a probability using a long-tailed distribution. We further assume there is always some minimum uncertainty, even if the SEM is super-small.

Leave a Comment