Welcome to our Election Update for Tuesday, Oct. 23!
It feels like every time you hit “refresh,” Democrats’ prospects in the House tick up. The Classic version of our forecast currently gives the party a 6 in 7 chance (86 percent) to take control.1 In the Senate, things look less promising for Democrats: Republicans have a 4 in 5 chance (81 percent) of holding on. But remember, there remains plenty of uncertainty in our forecast, so you should brace for a variety of outcomes. In the Senate, our 80-percent confidence interval spans everything from Democrats picking up two seats to Republicans picking up three. And in the House, there’s an 80 percent chance that Democratic gains will be somewhere between 20 and 61 seats — that’s a big range!
What fuels this wide range of outcomes? A big part of it is lack of data. Even at this late stage in the election cycle, many states and districts — even potentially decisive ones — have seen little to no polling, which forces our model to lean on other factors like an area’s fundamentals, possibly missing sentiments stirring on the ground. Of course, we don’t need polls everywhere; many states and districts are safe for one party or the other. But if we just look at places that our model considers to be competitive, some get all the polling love, while several others are woefully under-surveyed.
Take the Senate, for example. As of Monday afternoon,2 we had collected 52 polls of the U.S. Senate race in Florida between incumbent Democratic Sen. Bill Nelson and Republican Gov. Rick Scott. But in North Dakota — whose Senate seat is just as competitive as Florida’s — we had only eight polls of the matchup between incumbent Democratic Sen. Heidi Heitkamp and Republican Rep. Kevin Cramer. If we really want to understand the scale of polling disparities between the two races (and other competitive races in the Senate), we can be even more sophisticated and compare them by looking at the poll weights used by our model. It works a little something like this: Based on its recency, sample size, quality of the pollster, etc., each poll used in our forecast is assigned a weight — a coefficient where 1.00 is average. If you sum up all the poll weights for a race, you get an overall number that tells you just how robust the polling is. Here are the poll weights for this year’s competitive3 Senate races.
Some Senate races are poll hogs
Likely, lean and toss-up Senate races by amount of polling according to FiveThirtyEight’s aggregate poll weights, as of Oct. 22
State | 538 Rating | Aggregate Poll Weight | |
---|---|---|---|
Mississippi* | Likely R | 1.39 | |
North Dakota | Lean R | 2.39 | |
Montana | Likely D | 4.06 | |
Minnesota* | Likely D | 4.70 | |
West Virginia | Likely D | 5.64 | |
Indiana | Likely D | 7.34 | |
Missouri | Lean D | 7.35 | |
Nevada | Toss-up | 7.58 | |
New Jersey | Likely D | 8.07 | |
Tennessee | Lean R | 9.45 | |
Arizona | Lean D | 9.83 | |
Texas | Likely R | 11.44 | |
Florida | Lean D | 13.13 |
As you can see, we really don’t need this much data from Florida, where the aggregate poll weight is 13.13. And you can’t shake a stick on Twitter without hitting someone complaining about how overpolled Texas is, especially given that the campaign there is not particularly suspenseful (Republican Sen. Ted Cruz has a 4 in 5 chance of victory). Instead, pollsters should try polling North Dakota (an aggregate poll weight of 2.39) or Montana (4.06). Really, we’re in the most desperate need of polling in the special election in Mississippi, but that’s kind of a special case: It’s possible that no candidate will get a majority of the vote there on Nov. 6, which will send the election into overtime.4
Turning our attention to the House of Representatives, we’ve collected 432 polls covering 142 districts. That’s pretty good at first glance thanks to efforts like the New York Times Upshot and Siena College’s live polling project. But four of the closest districts — those that are “toss-ups” or “lean” toward one party — have seen no polling at all.
Lots of close House races are underpolled
Lean and toss-up House races by amount of polling according to FiveThirtyEight’s aggregate poll weights, as of Oct. 22
District▲▼ |
538 Rating▲▼ |
Aggregate Poll Weight▲▼ |
|
---|---|---|---|
CO-3 | Toss-up | 0.00 | |
IN-9 | Lean R | 0.00 | |
NY-2 | Lean R | 0.00 | |
WI-6 | Lean R | 0.00 | |
MI-7 | Toss-up | 0.36 | |
NV-3 | Lean D | 0.52 | |
MN-1 | Toss-up | 0.60 | |
OH-12 | Lean R | 0.74 | |
AZ-8 | Lean R | 0.77 | |
FL-25 | Lean R | 0.79 | |
PA-10 | Lean R | 0.91 | |
WA-5 | Lean R | 0.96 | |
NY-22 | Lean D | 1.10 | |
KS-2 | Toss-up | 1.31 | |
VA-5 | Toss-up | 1.34 | |
IL-14 | Toss-up | 1.43 | |
IL-6 | Toss-up | 1.45 | |
IL-13 | Lean R | 1.58 | |
FL-6 | Lean R | 1.58 | |
NY-24 | Lean R | 1.59 | |
NC-2 | Toss-up | 1.62 | |
NY-19 | Toss-up | 1.64 | |
NM-2 | Toss-up | 1.65 | |
CA-48 | Lean D | 1.74 | |
WA-8 | Toss-up | 1.82 | |
IA-3 | Lean D | 1.85 | |
UT-4 | Toss-up | 1.92 | |
TX-7 | Toss-up | 1.93 | |
CA-39 | Lean D | 1.99 | |
ME-2 | Toss-up | 2.01 | |
NJ-7 | Lean D | 2.05 | |
NY-27 | Lean R | 2.08 | |
IL-12 | Lean R | 2.12 | |
CA-25 | Lean D | 2.16 | |
MI-8 | Toss-up | 2.16 | |
WA-3 | Lean R | 2.17 | |
CA-45 | Lean D | 2.26 | |
MT-1 | Lean R | 2.28 | |
NE-2 | Lean R | 2.29 | |
VA-7 | Toss-up | 2.38 | |
KY-6 | Toss-up | 2.53 | |
FL-26 | Toss-up | 2.62 | |
PA-1 | Toss-up | 2.76 | |
NC-9 | Toss-up | 2.77 | |
TX-32 | Lean R | 2.79 | |
NC-13 | Lean R | 2.85 | |
NJ-3 | Lean D | 3.03 | |
AK-1 | Lean R | 3.12 | |
VA-2 | Lean R | 3.13 | |
FL-15 | Toss-up | 4.83 |
Those four districts are the Colorado 3rd, Indiana 9th, New York 2nd and Wisconsin 6th. The Indiana 9th will be one of the first districts to have its polls close on election night and could be a good bellwether. In our model’s estimation, the Colorado 3rd is a true coin flip between incumbent GOP Rep. Scott Tipton and Democrat Diane Mitsch Bush, a former state representative. Somebody poll these places!
The situation isn’t much better in the toss-up Michigan 7th, which hasn’t been polled since February. Nor in the Nevada 3rd, whose only poll was really more of a subsample of a statewide Nevada poll (it had just 178 respondents). Nevada is a particularly surprising omission, since even qualitative election forecasters agree that that seat is flippable! That’s at least six districts in desperate need of polling — and that’s just among the most competitive seats. Our forecast shows the House battleground as historically broad; dozens of districts that rate as “Likely” Democratic or Republican — races that could determine control of the House — also have little to no polling.
On the other end of the spectrum, pollsters can probably give surveys of the Virginia 2nd District and Alaska At-Large District a rest.5 And the Florida 15th District has an aggregate poll weight of 4.83, meaning we have more and better polling of Lakeland, Florida, than of the Montana U.S. Senate race.
Overall, it appears that pollsters may be behind the curve on which districts are truly competitive given Democrats’ increasingly strong position in the House. For example, some of the highest aggregate poll weights in the entire House are in districts where the outcome isn’t in very much doubt, like the Virginia 10th (4.51), where Democratic state Sen. Jennifer Wexton has an 8 in 9 chance of defeating incumbent Republican Rep. Barbara Comstock. In fact, “Likely Democratic” districts have the highest median poll weight of any competitive category.
Likely Republican races are underpolled
Median aggregate poll weight by FiveThirtyEight rating category
538 Rating | Median Weight | |
---|---|---|
Likely D | 2.04 | |
Lean D | 1.99 | |
Toss-up | 1.82 | |
Lean R | 1.58 | |
Likely R | 0.63 |
As you can see, the median aggregate poll weights gradually move from high to low as you go from more Democratic- to more Republican-favored districts. What that means is we’ve got better polling in districts where Democrats are favored than races that could go either way or Republican. Ideally, we’d see more of a bell curve with toss-up districts having the highest poll weights, but this hasn’t been so. Most striking, however, is the sharp decrease in median poll weights in borderline-competitive Republican districts. That’s the kind of data that would help us foresee a blue wave — and unfortunately, it’s currently our biggest blind spot.