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

Post-Labor Day movement: for Republicans, a Kavanaugh effect?

October 5th, 2018, 9:24am by Sam Wang

We’re off to a late start this year; more balls in the air, especially gerrymandering, and a busy term.

The House and Senate calculations are running for 2018, though we’re still polishing the display. The calculations are designed to pick up movement quickly. Here’s something notable: since Labor Day, both indicators show distinct movement, about 2 percentage points, toward Republicans.

Now a word about the calculations, and a preview of the Senate snapshot.

The House graph shows a moving median of the generic congressional ballot: last three polls or two weeks, whichever is greater, and a 90% confidence band. That’s a snapshot of current conditions as measured by opinion polls. It captures genuine shifts of D-versus-R opinion, as well as transient changes in voter enthusiasm.

Compare this moving curve with my estimate of where the national House vote needs to be for control to flip, D+6%. Finally, the graph shows an extrapolation based on special elections, which in the past have been predictive of the November outcome.

For most of the summer, Democrats have been favored to retake the House, though not overwhelmingly so. The threshold is high because of a combination of population patterns and partisan gerrymandering. Currently, Democrats need to take 23 seats to win control of the chamber. Without gerrymandering, they would only need to gain around 11 seats.

Since the beginning of September, the House estimator has moved substantially, and now shows control as a perfect tossup. Considering the special-elections result, conditions may well swing back toward the Democrats.

Is this shift a random fluctuation? Probably not, since an independent calculation gives a similar result: the Senate snapshot.

Here is a graph of the average of the entire Senate outcome distribution, from 10 key races (AZ/FL/IN/MO/MT/MD/NJ/NV/TN/TX). Data come from RealClearPolitics and Because this calculation uses polls only, it gives about as high a time resolution as we are likely to get. Again, note the sudden shift toward Republicans starting at the beginning of September.

It’s possible to compare Senate and House movement directly by calculating the Senate meta-margin. Recall that the meta-margin is defined as how much polls would have to shift across the board in order to make D-versus-R control a perfect tossup. Currently, the meta-margin is R+2.8%. On Labor Day, it was R+1.4%. That 1.4-point shift is comparable to the two-point shift in the House calculation.

So far this year, the 2018 rule of thumb has been that an overall shift of 1.7 points in Senate polls leads to a 1-seat shift in the likely number of seats. With Senate control hanging in the balance, small shifts are highly consequential.

However, there’s a really important caveat: midterm polling errors can be large. In 2014, the overall error was close to 5 points. So while Republicans are favored to retain the Senate, the risk to their control is very real, perhaps 30% even based on today’s snapshot.

What’s behind the September shift? One obvious possibility is the Kavanaugh Supreme Court nomination, which has gripped the nation’s attention. Trump’s approval ratings continue to be stable and dismally low, and it is no surprise that intensity on the Republican side has been faint. But with such a highly visible Supreme Court fight, many voters may be reminded of why they supported Republicans in the first place. That would tend to boost GOP voter enthusiasm.

Tags: 2018 Election · House · Senate · Supreme Court

8 Comments so far ↓

  • Chris C.

    Kavanaugh probably did have some effect as it seems to have mobilized the base. I’m skeptical that the bounce will last though. Even in normal times, bounces happen and they tend to be for a finite amount of time. That’s even more true in the Trump age. He will say or do something, or Manafort/Cohen/Mueller will be in the news. Or tax evasion. O r something. Remember when, long ago it seems, that article came out in the NY written by a secret senior staffer in the White House? No one is even talking about that anymore. Or the McCain funeral? Kavanaugh will be confirmed and GOP anger will subside. They won. I doubt this is still much in the news even by later next week.

  • ArcticStones

    I am genuinely puzzled by the notoriously short-term memory of American voters. We see it time and time again.

    That said, I think many American women will remember the message they heard, loud and clear, from Republican senators, including Susan Collins: “We heard Dr Blasey Ford’s testimony, and we don’t care.”

    • Sam Wang

      ArcticStones, I hate to say this but you are stepping on a landmine. However much your sentiments may have a factual basis, it makes readers think I allow that kind of thing. The WordPress software makes it hard for me to police longtime commenters. I can keep out any new riffraff, but you are old riffraff! ;-)

      As far as I can tell, Dr. Ford’s accusations are probably true, but it will be a long time before we ever find out. The whole thing is so polarizing.

      So I ask, just please don’t go there. I’m sorry!

    • LondonYoung

      There are many techniques for analyzing time series data for reversion patterns. If it seems they are there, one might analyze a “shock” and then model the effect as, say, Shock = Shock_LongTerm + Shock_ShortTerm * Exp( – DecayRate * Time ).

      On this forum we have now-and-then discussed mapping today’s data onto probabilistic outcomes on a future election day. I get the feeling more work could be done there.

    • Sam Wang

      Could do. It’s implicitly why I plotted the red “Special elections” line on the House graph.

  • ArcticStones

    I know it is just one poll, but I wonder: How should we read the new CNN (SSRS) poll that has the generic Congressional poll at +13% Democrats?

    Does this indicate that the Red post-Kavanaugh bump may already be fading?

    • Sam Wang

      That’s unusually good for Democrats…I’d have to think about that. Compare it to other surveys.

      If it holds, I’d say that survey covers the whole nation, while the post-Kavanaugh bump seems focused in red states (see the movement in the Senate number, driven by TN and TX polls). Kind of like the 2016 home stretch, where R’s overperformed polls in their home states but not in blue states.

    • ArcticStones

      Perhaps even more striking is this, with reference to a weekly poll by The Economist/YouGov:

      “The gender gap in the generic ballot grew from 35 to 45 pts (!!) among college-educated voters in our polling from last week to this week. Seems to be corresponding with a better overall generic ballot for Democrats. It is not a good long-term strategy to tick off suburban women.”

      — G. Elliott Morriss, The Economist

Leave a Comment