Split Ticket

As we enter the final stretch of the campaign, Democratic fretting about the election continues to ramp up. Multiple news articles have documented the “vibe shift,” noting that some within the party have already begun to point fingers at those they believe will be responsible for a Harris loss.

These fears are not entirely unfounded, as there is a very good chance that Trump is sitting in the Oval Office come next January, but according to our model, that chance is no higher than it was last week. Our forecast yet again gives Kamala Harris a 53% chance of winning the Electoral College. She remains the narrowest of favorites in Michigan, Wisconsin, Pennsylvania, and Nevada, while Trump maintains a slight edge in North Carolina, Georgia, and Arizona.

This may come as a surprise to readers who watch other forecasts, which have seen more movement towards Trump of late — FiveThirtyEight and Silver Bulletin both give him a 53% chance of winning the election. This has also led to a number of (very fair) questions, asking what the difference is between our models.

There are two ways to answer this. The first, and most obvious, is that there is really not that much of a difference between a 53% chance for Trump and a 53% chance for Harris — the race is a tossup in both situations. But the second, and more wonky answer is actually to be found in our polling aggregates and the differences in our aggregation algorithms.

Our polling aggregates have not seen much movement towards Trump for a while in the battleground states. Since a round of tightening in early October, polls have stayed largely stable at the state level, even while we have seen more tightening in national polls. Secondly, we just haven’t seen much quality polling for a while in state polls, which further compounds the picture of stability — our aggregates are sticky by design and emphasize pollster quality. We’ve not had much high-quality polling at the state level for a while now, and what little we do have doesn’t suggest much movement.

You can read about our methodology and aggregation algorithm here, but it’s less aggressive by design than Nate Silver’s aggregates. Our reasoning for this is that oftentimes, polling movement is simply just noise, and we should have a volume of good polling in order for trends to truly bake in to the model. This is partly why our aggregate hasn’t shifted as much towards Trump in states like Pennsylvania, where we just haven’t had much quality polling of late.

One other key difference between us and other smart aggregators is that we don’t adjust state polling aggregates based on trends seen in national polling. While we see and respect  the logic behind firms doing this, we don’t know if that’s the best thing to do this time — polling suggests that the national political environment may be diverging based on the closeness of the states, with Harris holding up well in battlegrounds even as Trump gains nationally.

All of this means that our polling aggregates in battleground states are a touch better for Kamala Harris than the numbers shown by other sites. This actually explains most of the differences in our model’s output. Below, you’ll find a table showing the model’s probabilistic output under the polling aggregates used by each site — the way to read the table is “what would Split Ticket’s model show if it used this specific polling average?”.

You can see how the model would change in response to the different pictures painted by each polling aggregator. Were we to use FiveThirtyEight’s method, we’d find Trump a 52% favorite (which is remarkably similar to their current model’s output).

Again, we want to stress that this type of variance is normal. Fractional movement drives engagement, and it sparks countless columns from journalists desperate to write stories and meet deadlines. But the difference between the inference to be made at D+0.3 and the one to be made at R+0.3 is actually negligible. Polls (and election models) are simply not built to distinguish between two candidates separated by less than a percentage point. The responsible (albeit unsatisfying) takeaway is that it’s a tossup, and that’s all we feel comfortable saying.

Finally, a note on early voting analysis, the notorious tea leaves that are read in the weeks leading up to the election. Relative to 2020, it appears that Republicans are turning out in greater numbers in early voting (both in-person and by mail). This has been cause for Republican jubilance and Democratic anxiety. However, with the possible exception of Nevada, early voting tea leaves are just that: tea leaves. Although claims that early voting numbers are good or bad for one candidate go viral on Twitter, there is no evidence that they hold any predictive value at all.

So why is this the case? Wouldn’t looking at votes that are actually cast be better than relying on fickle polls? The answer is no, for several reasons. First, there is intense variation from state to state on a number of factors. Whether or not voters register by party (or need to register at all), whether elections conduct all-mail elections (or even have mail ballot as an option for most voters), and when states offer early voting are all factors that are substantially different across states. This has an impact on how voters choose to cast their ballot, so any nationwide analysis will fail to capture these massive differences.

Second, beware the ghosts of party registration. While it is true that most voters who are registered Democrats will cast a vote for the Democratic nominee (and vice versa for Republicans), party registration is a lagging indicator. This means that voters will often choose their party registration early in their life, but do not change it later on, even if their political persuasions have shifted. 

In essence, this means we cannot effectively tell what percentage of registered Democrats or registered Republicans will end up voting for Harris or Trump, respectively, even if we know the number is very high. There is a meaningful difference between an electorate where registered Democrats break 95% for Harris and one where they break 90% for her, and we generally just cannot tell which scenario we’ll get ahead of time.

The same can be said for registered independents or unaffiliated voters, who are even more of a black box. Moreover, being a registered independent or unaffiliated voter is not the same as being a moderate, persuadable swing voter — many registered independents have backed a single preferred party for years. Independents compose an increasing amount of the electorate every cycle, and it is impossible to tell how they will break without making some major, often incorrect, assumptions.

In fact, the way independents will break generally depends on the state, and may even depend on their registration policies; for instance, Nevada has automatic voter registration for all eligible voters, which means a lot of young voters get registered at the DMV; if they don’t opt to register for a party, they get marked as independent, which means this pool is disproportionately young and of even lower propensity than normal.

These problems are compounded by the fact that there are demographic differences in how voters choose party registration, especially when it comes to age. Older voters are more likely to have registered as a Democrat or a Republican, but younger voters are overwhelmingly more likely to register as an independent or unaffiliated voter. Again, this does not mean that these young voters are more moderate; many are Democrats in all but name.

Consequently, if we were to be given a batch of early votes that are 40% Democratic, 35% Republican, and 25% Independent, it is impossible to give any particularly exact estimate for how this batch will actually vote. This is important, because politics is a game of inches. We can probably guess that this batch of votes will not break 80–20% for Harris, but we cannot say if it will vote 60–40% Harris, 55–45% Harris or 51–49% Harris. In a swing state, that could make or break the entire election. 

Additionally, there is no particularly good basis for comparison in 2024. Since 2020, a year defined by the pandemic’s effect on voting, many states have completely overhauled their voting systems. That year, Democrats overwhelmingly chose to vote by mail if they could, while Republicans broke for in-person options amidst Trump’s conspiratorial stance toward mail voting. This caused a massive divergence in vote method; Biden racked up massive margins with mail votes, while Trump trounced him in Election Day votes.

Since then, however, there has been evidence of reversion to the status quo. Many Democrats, now no longer feeling pandemic pressure to vote by mail, have returned to voting in person. Similarly, Republicans have made a stronger effort to encourage their voters to vote early or by mail, which may already be bearing fruit.

These shifts, however, will not ultimately impact the final results. It is best not to think of voters as “mail voters” or “in-person voters” in separate pools. Instead, “Harris voters” and “Trump voters” will choose mail or in-person voting as their preferred method, but it does not matter which one they opt for; a vote cast by any means is still a vote. Take the below example from the past two elections in Maryland.

In both elections, 67% of Marylanders chose the Democratic candidate, while 33% chose the Republican one. Under the surface, however, there were large changes in vote method choice. In 2020, 66% of Joe Biden’s voters chose to vote by mail, while just 9% voted on Election Day. In 2022, only 37% of Wes Moore’s voters cast a ballot via mail, while 44% did so on Election Day. This had a major impact on what the vote method composition looked like (Election Day was just 14% of 2020, while it was 52% of 2022), but the final result was the same. Voters simply migrated from one method to another.

Basically, even if we are armed with party registration figures (and some states do not even give us that), early voting data tells us little about how a state will ultimately vote. We do not know confidently what Election Day turnout will look like, or how Harris or Trump will perform in any kind of vote method. And comparisons to 2020 are often deeply flawed, as voters often migrate between methods. In fact, outside of highly-specific scenarios like runoff elections, early voting has often misled observers into prematurely and incorrectly forecasting landslides and waves.

So if you find that a state is suddenly showing massive turnout disparities in a presidential election, we would caution against using this as evidence of a wave. Instead, it’s just more evidence that partisan voting habits are changing, but almost all of those voters are still more likely than not to show up eventually in some way.

The one and only reliable way to know how a state will vote? Watch as it posts returns on election night. 

Senate

Our Senate update should come as no surprise, as Republicans remain significant favorites to flip the chamber. With 10 days until the election, it does not appear that Republicans are poised to break through in any of the major swing states that Democrats already hold. If they do, it is likely to happen only if Trump carries the state and drags them across the finish line.

House

The race for the House remains just as close as the one for the presidency; Democrats remain marginally favored to flip the chamber, but at 52%, this race is tighter than ever before. The generic ballot has seen substantial tightening of late, especially as national polls have tightened, and this has resulted in the race for control becoming increasingly close.

Like our model says, we’d probably pick the Democrats if we had to; they retain a commanding edge in money and broadly have a fairly good slate of challengers. But control for the chamber will very possibly just come down to the party that overperforms the polling on election night.