Wednesday, August 26, 2020

The Electoral College Map (8/26/20)

Update for August 26.

Day three of the Republican National Convention dawned with the release of a new wave of battleground polls from Change Research, and that had some folks proclaiming "tightening!" in the race for the White House. There are several problems with that premise that violate FHQ's rules for following polls at even a cursory level. Those rules:
1. Never take a poll in isolation. Always try to put it in context.
2. In an effort to contextualize any survey, compare it first to any previous survey in the state by that pollster (if available). If something prior from that firm is available, then USE THAT PREVIOUS POLL FOR COMPARISON FIRST. [Those calling "tightening!" based on the Change Research polls violated this rule.]
3. If no previous poll is available from that pollster in the state in question, then dig a bit deeper than just comparing margins across different polls. It helps to look both at the shares each candidate has in the surveys around the same time and at the share of undecideds. Doing that at least gives one a more robust comparison, one that can help better identify possible outliers. 
Yes, all of that takes a bit more time (and there are certainly other steps one could add to that list), but it ultimately yields something better than a shallow, knee-jerk reaction to any poll release. Speaking of poll releases...

Polling Quick Hits:
(Biden 53, Trump 39)
It is still tough to imagine that Virginia has pushed as far into the Democratic coalition of states over the last (at least) three cycles. What had been a relatively comfortably red state before then has seemingly become reliably blue; blue enough that the commonwealth has not been polled at nearly the rate it has been in recent cycles. And what is more, the latest survey of Virginia from Roanoke College nudges the average margin there up to push the Old Dominion off the Watch List below. Virginia is a Strong Biden state and no longer (immediately) threatening to shift into Lean territory. [And incidentally, since the last Roanoke survey of the commonwealth, Biden has gained a couple of points while Trump has stayed steady at 39 percent.]

(Trump 46, Biden 45)
Over in the Badger state, Trafalgar Group was back in the field during the Democratic convention and into the weekend following it. Some decry the methods out of Trafalgar, but one can take this poll for what it demonstrates relative to the firm's last survey of Wisconsin: no movement. Yes, adjust for the perceived bias if one must, but since the firm was last in the field in the state, Trump has maintained a one point lead with the president at the height of his range of results there over the summer. Meanwhile, Biden has not been as low as 45 percent since a Redfield poll last month that had him up by ten points. If the dual storylines out of this one are 1) outlier and 2) shows no movement, then it is not really the best of news for the incumbent.

Change Research
(late August battleground poll wave)
Looking at the latest wave from Change Research, the take home lesson continues to be that the race is mostly stable. At the end of the day, the candidate's shares of support moved one and maybe two points in five of the six states since the early August series of polls. Again, stability rather than movement is the message. The exception is Arizona where Biden maintained his slim lead and both candidates saw increased support (beyond the one or two point shifts elsewhere) at the expense of undecided/other. Sure, this set of surveys was in the field the weekend after the Democratic convention, and they are not showing much of a bounce for Biden, but they are not exactly showing any tightening either. Minus the late June and early July waves from the Change Research dataset, this looks like a pretty consistent Biden advantage where it matters most. That can still change, of course, but time is ticking off the clock.

Arizona: Biden +2 (Biden +4, Trump +3 since early August wave)
Florida: Biden +3 (Biden -1, Trump +2)
Michigan: Biden +6 (Biden +2, Trump +1)
North Carolina: Biden +1 (Biden +1, Trump -1)
Pennsylvania: Biden +3 (Biden +1, Trump +2)
Wisconsin: Biden +5 (Biden +2, Trump +1)

NOTE: A description of the methodology behind the graduated weighted average of 2020 state-level polling that FHQ uses for these projections can be found here.

The Electoral College Spectrum1
NE CD2-1
(273 | 286)
(302 | 265)
(308 | 236)
(319 | 230)
NE CD1-1
(334 | 219)
ME CD2-1
(353 | 204)
ME CD1-1
NE CD3-1
1 Follow the link for a detailed explanation on how to read the Electoral College Spectrum.

2 The numbers in the parentheses refer to the number of electoral votes a candidate would have if he or she won all the states ranked prior to that state. If, for example, Trump won all the states up to and including Pennsylvania (Biden's toss up states plus the Pennsylvania), he would have 286 electoral votes. Trump's numbers are only totaled through the states he would need in order to get to 270. In those cases, Biden's number is on the left and Trumps's is on the right in bold italics.

To keep the figure to 50 cells, Washington, DC and its three electoral votes are included in the beginning total on the Democratic side of the spectrum. The District has historically been the most Democratic state in the Electoral College.

3 Pennsylvania
 is the state where Biden crosses the 270 electoral vote threshold to win the presidential election, the tipping point state. The tipping point cell is shaded in yellow to denote that and the font color is adjusted to attempt to reflect the category in which the state is.

The thing about the Change Research battleground polls is that this wave is consistent with the averages at FHQ. No, the margins are different, but the order of the six states matches the order on the Electoral College Spectrum above. And needless to say, that match did little to disrupt the order as it appeared yesterday. Even with that outlier in Wisconsin from Trafalgar added, the Badger state maintained its spot in the order but did slip back onto the Watch List below. Wisconsin's margin is now within a fraction of a point of the Lean/Toss up line among Biden's current coalition of states. But while Wisconsin moved back on the List, Virginia came off it (and switched places on the Spectrum with Maine on top of it).

There remain 12 states and districts to watch along with underpolled Nevada. It is that group that most likely could see some change in classification here at FHQ upon the introduction of new polling data.

There were no new polls from Nevada today.

Days since the last Nevada poll was in the field: 118.

NOTE: Distinctions are made between states based on how much they favor one candidate or another. States with a margin greater than 10 percent between Biden and Trump are "Strong" states. Those with a margin of 5 to 10 percent "Lean" toward one of the two (presumptive) nominees. Finally, states with a spread in the graduated weighted averages of both the candidates' shares of polling support less than 5 percent are "Toss Up" states. The darker a state is shaded in any of the figures here, the more strongly it is aligned with one of the candidates. Not all states along or near the boundaries between categories are close to pushing over into a neighboring group. Those most likely to switch -- those within a percentage point of the various lines of demarcation -- are included on the Watch List below.

The Watch List1
Potential Switch
from Toss Up Biden
to Lean Biden
from Toss Up Trump
to Toss Up Biden
from Toss Up Trump
to Toss Up Biden
from Strong Biden
to Lean Biden
Maine CD2
from Toss Up Biden
to Toss Up Trump
from Strong Trump
to Lean Trump
from Toss Up Trump
to Lean Trump
Nebraska CD2
from Lean Biden
to Toss Up Biden
from Toss Up Biden
to Toss Up Trump
from Lean Biden
to Toss Up Biden
South Carolina
from Lean Trump
to Toss Up Trump
from Lean Biden
to Toss Up Biden
1 Graduated weighted average margin within a fraction of a point of changing categories.

Methodological Note: In past years, FHQ has tried some different ways of dealing with states with no polls or just one poll in the early rounds of these projections. It does help that the least polled states are often the least competitive. The only shortcoming is that those states may be a little off in the order in the Spectrum. In earlier cycles, a simple average of the state's three previous cycles has been used. But in 2016, FHQ strayed from that and constructed an average swing from 2012 to 2016 that was applied to states. That method, however, did little to prevent anomalies like the Kansas poll that had Clinton ahead from biasing the averages. In 2016, the early average swing in the aggregate was  too small to make much difference anyway. For 2020, FHQ has utilized an average swing among states that were around a little polled state in the rank ordering on election day in 2016. If there is just one poll in Delaware in 2020, for example, then maybe it is reasonable to account for what the comparatively greater amount of polling tells us about the changes in Connecticut, New Jersey and New Mexico. Or perhaps the polling in Iowa, Mississippi and South Carolina so far tells us a bit about what may be happening in Alaska where no public polling has been released. That will hopefully work a bit better than the overall average that may end up a bit more muted.

Recent posts:
The Electoral College Map (8/25/20)

The Electoral College Map (8/24/20)

One Thing About Convention Bounces

Follow FHQ on TwitterInstagram and Facebook or subscribe by Email.

No comments: