For all of the energy and drama—even melodrama—surrounding the 2024 Presidential Election cycle, the focus has largely settled on a handful of “battleground” states.
Even within those, Donald Trump is generally favored to win four of the states—Nevada, Arizona, Georgia, and North Carolina—leaving three states—Michigan, Pennsylvania, and Wisconsin—hanging in the balance. Within all of the battleground states, however, the final AtlasIntel Swing State Poll series gives Donald Trump a slight lead, although not outside the margin of error anywhere.
Because voter turnout is the final determinant of electoral success, I believe it is unwise to use a poll, or even an aggregate of polls such as the RealClearPolitics series, as a basis for forecasting actual results.
The key word in poll analysis should be “if”: If the electorate sampled by the polling organization is accurately reflected in the sample population, then the results will accurately reflect voter sentiments for the election. If the voters turn out in the same proportions of Democrat and Republican (and, by extension, the other demographic subcategories frequently tracked by pollsters), then the poll results will reflect the outcome of the election.
Still, understanding how the polls are assessing voter demographics can offer a rational framework for assessing election returns as they are published, and thus whether the results are matching expectations or not.
I have been leaning heavily on the AtlasIntel poll series for doing these deep dives. As I have mentioned before, I am very impressed by their presentation. Their slide deck format is very useful for peeling back the layers and looking at the data behind the headline poll results.
I also take note that Atlas Intel was the most accurate polling organization in the 2020 Presidential Election Cycle, a point highlighted by polling guru Nate Silver in his 2021 tweet thread dissecting the 2020 election poll results.
Atlas Intel also is rated as “least biased” by Media Bias/Fact Check.
Media Bias/Fact Check is an organization that purports to assess the levels of partisan bias of various media sources.
Founded in 2015, Media Bias/Fact Check (MBFC) is an independent website that has promoted awareness of media bias and misinformation by rating the bias, factual accuracy, and credibility of media sources, large and small. Media Bias/Fact Check relies on human evaluators to determine the bias of media sources and the level of overall factual reporting through a combination of objective measures and subjective analysis using our stated methodology.
Such evidence and testimonial does not prove that AtlasIntel is affirmatively the most accurate polling organization, nor is it affirmation that it will have the most accurate polls again in 2024. Such data points do, however, lend credence to my own assessment that an organization which puts significant effort into preparing its poll results is likely to also put significant effort into generating its poll data.
For these reasons, and for the accessibility of their poll reports, These are the polls I use when doing a deep dive on polling.
We will begin with the obvious: AtlasIntel shows Donald Trump having a small lead in both the national polling as well as in the battleground states.
We should note also that this poll is not outside the margin of error. Statistically, it shows Donald Trump and Kamala Harris as tied.
What is outside the margin of error are some of the more intriguing crosstabs.
Nationally, Donald Trump is presumably pulling about 4.4% of Democrats, while Kamala Harris is pulling 7.2% of Republicans.
Among independent voters Donald Trump leads 48.9% to 43.2%. Depending on the relative proportions of Democrats, Republicans, and Independents who turn out nationwide, Trump’s advantage among Independents overcomes the slightly larger crossover advantage Kamala Harris is enjoying by pulling 2.8pp more Republican votes than Donald Trump is pulling Democrat votes.
Donald Trump enjoys another advantage nationwide in that he’s pulling more of Joe Biden’s 2020 voter coalition (6.9%) than Kamala Harris is pulling of Donald Trump’s 2020 voter coalition (3.7%).
Partisan defections appear to favor Trump nationally in another way as well. 9.8% of Democrats disapprove of how Joe Biden has performed nationally, while only 5.2% of Republicans approve of how Joe Biden has performed nationally.
Looking at the 2020 voter coalitions, Trump’s advantage is even greater. 15.4% of voters who supported Joe Biden in 2020 disapprove of his job as President, while only 2.5% of voters who supported Donald Trump in 2020 approve of Biden’s job as President.
Does the disparity in the partisan breakdown of Joe Biden’s job approval numbers outweigh the partisan breakdown of how people intend to vote this election cycle? We certainly cannot rule that out.
How well does the partisan breakdown translate into the battleground state results?
For the sake of length I am confining my analysis to Michigan, Pennsylvania and Wisconsin. At present there is broad consensus that Donald Trump is highly likely to win in the other four battleground states, and the closest of the battleground races are these three states. We will find out after election day if my take on that was right or wrong.
In polling Michigan, AtlasIntel constructed a moderately-sized sample population of around 1,200 likely voters, resulting in ±3pp Margin of Error (MOE).
The population’s political breakdown is fairly even at 32.5% Democrat and 32.4% Republican. If one party or the other is over-sampled the poll size alone does not clearly reflect that, making the poll results truly “middle of the road.”
That being said, Donald Trump is favored to win—although the margin of victory is within the margin of error.
Is Jill Stein playing a “spoiler” role here? Quite possibly. At 1.7%, she could be pulling just enough votes away from Kamala Harris to give Donald Trump the lead.
That spoiler effect is likely magnified by Donald Trump’s greater apparent success at party defections, with 6.2% of Democrats likely to vote for Trump while only 3.2% of Republicans in Michigan are breaking for Kamala Harris.
Donald Trump’s edge in partisan defections and crossover increases when we look at the 2020 voter coalitions. 7.3% of Joe Biden’s voter coalition is prepared to turn out for Trump this election cycle, while only 4.5% of Donald Trump’s 2020 voter coalition is prepared to turn out for Kamala Harris.
Partisan defections loom even larger when we look at Joe Biden’s approval numbers. 9.3% of Democrats disapprove of the job Joe Biden has done as President.
Looking at the 2020 voter coalitions, the defection disparity is even greater: 15% of Joe Biden’s 2020 voter coalition is now turning out for Donald Trump, while only 1.4% Republicans approve of the job Joe Biden has done as President.
If the Michigan votes, when tallied, break down along party lines, these percentages of defections in both directions would most likely result in a Trump victory in Michigan.
In Pennsylvania, AtlasIntel was able to assemble a much larger sample population of 2,049 likely voters.
The much larger sample population allows for a smaller ±2pp MOE. We should be able to place greater confidence that the poll data matches the demographic realities of Pennsylvania.
We should note, however, that Pennsylvania was sampled with a higher percentage of Republicans (42.9%) than Democrats (39.3%). Is this an oversampling bias that skews the results? That would be part of the “ifs” pointed out above. We will know after election day if that was the case.
Pennsylvania is going more strongly for Trump than Michigan, according to these poll numbers.
With a 1.8pp lead, Donald Trump is just within the MOE. Statistically, we would still call this a tie race.
What is perhaps more significant, however, is that 2% of poll respondents were still undecided as of November 2. If that entire 2% goes to Kamala Harris, she wins Pennsylvania.
The existence of the undecideds means we are not at present seeing a strong “spoiler” effect, as Jill Stein is pulling only 0.9% of voters in Pennsylvania. Even if all of her support were to wind up going to Kamala Harris, that would not change the outcome of the race.
In fact, if anything, the undecided voters are defections from Jill Stein, who a month ago was polling much stronger and presenting a significant “spoiler” effect in the state.
Amplifying the significance of that undecided cohort is the greater level of partisan defections in both directions, with 8.3% of Pennsylvania Democrats turning out for Donald Trump and 8.9% of Pennsylvania Republicans turning out for Kamala Harris.
However, to make sure we realize these polling waters are still quite muddy, 4.7% of Joe Biden’s voter coalition is turning out for Donald Trump, while 2.7% of Trump’s 2020 coalition is turning out for Kamala Harris.
At the same time, a much larger percentage of Democrats in Pennsylvania are unhappy with Joe Biden than in Michigan. At 12.9%, that is more than 3pp greater than in Michigan, a differential that is outside the MOE in both states.
Curiously, “only” 10.8% of Joe Biden voters from 2020 have soured on him in Pennsylvania, as opposed to 15% in Michigan.
Is this a signal that that partisan defections will play less of a role in Pennsylvania? We cannot discount that possibility.
Wisconsin is once again the state with the smallest sample population at 728 respondents, and thus the highest MOE.
While Wisconsin has a higher percentage of Independent voters than Pennsylvania, we should note that the Republicans have the same 3.6pp edge in representation within the sample population as in Pennsylvania. With the smaller sample size and greater MOE, the potential for oversampling error is thus magnified.
With those caveats in mind, AtlasIntel is showing a slightly tighter race in Wisconsin than in Michigan, with Trump leading by a slim 1.3pp.
Once again, we are seeing that undecideds are in a position to determine the outcome. At 1.5% if undecided voters go all one way, towards Harris or towards Trump, within this polling universe that will determine the outcome of the election.
One point we should note about the AtlasIntel poll: While Robert F. Kennedy Jr., is not appearing on the poll results, he is still on the ballot in Wisconsin. Arguably, he might represent that 0.5% of “other” support. If he does and if those voters all break for Donald Trump, he could negate the power of the undecideds to shift the electoral outcome in the state.
This is why predictions based on polls are inherently unwise. There are simply too many variables such as these which can shift the final outcomes dramatically.
At I noted in my previous analysis on the state, Wisconsin is considerably more partisan than Pennsylvania, with only 1.9% of Democrats breaking ranks to declare for Donald Trump.
Donald Trump has actually improved his situation in Wisconsin, as the number of Republicans breaking rank to side with Harris has improved from 6.6% to 4.4%. Trump has also clawed back a portion of the defectors from his 2020 voter coalition in Wisconsin, as only 1.8% of Trump 2020 voters are turning out now for Harris.
By the same token, Harris has clawed back some of Joe Biden’s 2020 support, with defections from that coalition now down to 3.2%, although her success in this regard has been less than Donald Trump’s.
Wisconsin shows that it is the most reliably Democrat of all three battleground states in the north. Only 6.7% of Wisconsin Democrats believe Joe Biden has done a bad job, the lowest among the three states under consideration here.
Wisconsin is more like Michigan in that despite a lower defection rate when it comes to Biden’s job approval, a far larger percentage of Biden’s 2020 voter coalition has soured on him, at 14.7%. Will defections from the 2020 vote coalition prove to be the significant variable that gives Donald Trump the win in Wisconsin?
Do we dare predict from these poll results that Donald Trump is going to win all three of these states? Some will, I will not.
However, what I will note is that these polls put Donald Trump in the lead in all three states.
This is important because if we start building an Electoral College scenario from the 2020 results, and shift Nevada, Arizona, and Georgia into Donald Trump’s column while leaving Michigan, Pennsylvania, and Wisconsin blue, Donald Trump rises to 268 Electoral College votes.
Thus, if I move any one of these three states into Donald Trump’s column, he wins the Presidency. By the same token, if any other state turns into a surprise for Donald Trump, even holding these three states does not get Kamala Harris the Presidency.
Where might we see such a surprise? There are some tantalizing early vote returns which could suggest Virginia is breaking for Donald Trump.
Similarly, RealClearPolitics recently shifted New Hampshire to “toss up”.
Building from the 2020 results, so long as Donald Trump does win Nevada, Georgia, and Arizona—battleground states which went for Biden in 2020—if he picks up any suprise win in any state, he wins.
If Kamala Harris loses any of these three states, in addition to the southern “battleground” states, she loses. If she picks up Georgia or Arizona, she can afford to lose one of these three “blue wall” states, but no more than that.
AtlasIntel has Kamala Harris behind in all the battlegound states. RealClearPolitics has Kamala barely ahead only in Wisconsin and Michigan.
Kamala Harris will have to overperform in multiple states if she is win the Presidency.
Donald Trump only has to perform within the expectations established by the polls if he is to win the Presidency.
Is Donald Trump a “shoe-in” to win? I would not say that. I would not argue, as some have, that Donald Trump is on a “glide path”. There are still too many unknowns to make such claims with confidence, not the least of which is voter turnout; voter turnout always has the power to break even the best of polls.
Still, you do have to like Donald Trump’s chances.
Peter, how did you know that I was just now literally craving NUMBERS? Never mind all the speculation, guesses, and extrapolation- what are the latest numbers? And like Santa Claus, you answer my wish with a large bagful of the latest hard data of numbers. Thank you, kind sir!
That being said, I think this election is going to involve a bunch of factors that are hard to quantify in the polls. A long-standing male Democrat may - possibly without being conscious of it - ‘feel like a wuss’ if he votes for Harris. A vote for Trump is a ‘manly’ vote. That will be countered by the huge fear factor being ramped up on females that ‘Trump is a Hitler who will take away all of our rights!’ How many votes will be decided last-minute because of psychological factors like these?
You’re right, Peter, the winner of this election cannot be called just yet!
But, the election is over in less than 50 hours! What could possibly happen to affect the outcome in two days time? I’d sure like to hear anyone’s speculation on that!
Nice to see some data digging instead of the piles of posts that are just partisans thinking they’ll predict the outcome based on what they want to happen/their priors/vibes. What’s your take on the early vote so far?