Crossing
With democracy looking fragile these days, please focus on the substantive stakes of the election, for both this year and for years to come. I know people come here for horserace, but the same statistics can be used to understand where you can be most effective. To learn more, see Vote Maximizer and read my work at Fixing Bugs in Democracy.
And now, horserace.
As longtime readers know, the important quantity to track on this site is not the win probability (which is calculated but hidden over at GitHub), or even the electoral vote count. In my view the most precise measure of the Presidential or Senate contest is the meta-margin, which aggregates all available state polls in terms of percentage points of voter support. For example, the Presidential meta-margin represents how far away Harris or Trump is above threshold for winning the electoral college.
Yesterday between noon and 4:00 PM, the Presidential meta-margin did something interesting: it crossed zero. It’s now at Harris +0.2%, meaning an across-the-board movement of 0.2 percentage points toward Trump would create a perfect toss-up.
Now, don’t get too excited. Precision is different from accuracy. I can have a precision timepiece, but if it is 5 minutes slow, then I am at risk of being exactly 5 minutes late for everything. In the case of the meta-margin, the inaccuracy in 2016 was 2.1 percentage points favoring Hillary Clinton. Trump outperformed on election day, and we know what happened after that. So although state polls now favor Kamala Harris rather than Donald Trump, the election could still very much go either way. Until that meta-margin exceeds two percentage points in either direction, we should consider Presidential polls to be on a knife edge.
However, because the tracker is so precise, you can be relatively certain of the direction of movement. Its fluctuations are generally no more than half a percentage point. We can be confident that surveys have moved at least four points towards Harris. I would also guess that non-response bias, in which a candidate’s supporters are less likely answer a survey if they feel less enthusiastic, is too small to account for this swing.
Is it done moving? From the steepness, I would expect not. The time integration window for the aggregate is currently 2 weeks or 3 polls per state, whichever gives more data, from the FiveThirtyEight API. So even if conditions stop changing, the meta-margin takes at least two weeks to catch up. (If you want a faster-moving snapshot, the most recent poll in each state is featured at electoral-vote.com.)
There is another source of rapidly changing data: national polls. Here is what the median Harris-over-Trump margin looks like, grouping by the last date of each poll. (“-10” means July 21. Don’t judge me.)
This graph is in the same units as the meta-margin. As you can see, the Harris-Trump margin has moved by at least 6 points (from Trump+4% to Harris+2%) since Harris entered the race. Since the meta-margin has only moved by 4 points, we can surmise that it might not be done moving yet. Give it another week.
This kind of a swing is unprecedented in our modern age of polarization. Ever since the mid-1990s, when Newt Gingrich took over Congress, voters have become highly loyal to either party. That entrenchment coincides with the personalization of politics and the nationalization of partisan messaging. The movement since Biden’s dropping out of the race is is consistent with his weakness as a candidate imposing at least a five point penalty to Democrats.
As an alternative, Harris could be an unusually strong candidate. The Presidential data provides only a Galilean frame of reference. However, there is other data stream that suggests that there were something specifically weak about Biden: Senate and House data. If we superimpose Prewidential, Senate metamargin, and the generic congressional preference (which is a good predictor of overall House elections), they look like this:
The sharp convergence of the presidential towards the other two traces suggests that Biden was underperforming. In the coming weeks we should be able to see if all three traces start moving together. If the Presidential tracker keeps moving upward, then Harris might be unusually strong.
OK, we are done with horserace. Now, what can you do about any of this?
As it turns out, this data is quite useful in helping you direct your time and your resources. The analytics here allow calculation of voter power: the ability of one or a few votes to swing the outcome, whether for president, Senate, House, or even state legislatures.
An easy way to use that information is through the ActBlue and WinRed links at the bottom of this site. Another application is Vote Maximizer, a Web app that uses these metrics in the service of maximize your individual power to move politics and strengthen democracy. Vote Maximizer is a new product of the Electoral Innovation Lab, which I lead and which is fully separate from my activities here at Princeton University.
Topics:
What do you make of Frank Luntz’s comment yesterday that, “[Harris is] bringing out people who are not interested in voting for either Trump or Biden, so the entire electoral pool has changed …”
At first blush, I disagree. The differential movement of Presidential sentiment (moved 4-6 points) vs. Senate/House (stable) is not consistent with that claim. Those surveys often ask both questions, so I would imagine differential nonresponse would affect both to a similar extent.
Great explanation, succinct analysis. Thank you!
Glad I followed the advice you gave five days ago: “Give it a little time.”
the registration data that arctic is looking for might give a more direct measure of this conjecture. i saw some charts that net registration has turned blue since the candidate change.
Good to see PEC back. Miss the podcast.
Looks like the metamargin/national polls gap is a little smaller this year.
Yes, seemingly about 2 points.
Welcome back!
Sam, you’re a jinx. It crossed 270, you made your post and it’s back to 266 as l write this.
So if the meta margin for Harris soars 20, please, please don’t say the race is sewn up, all the action is down ballot. Especially, don’t promise to do anything if Harris loses. No promises to eat a bug, take a drug, or clean a rug.
I am so sad that you did not absorb the paragraph on accuracy.
That ticker is just the bait. The goal is to get you to get out there and do something.
If the race turns out the wrong way for you, could it be because you spent too much time commenting on blogs? 😉
Touché and it’s great to have you back.
Looking at this data, and considering Trump’s original dominance in the polling, this all reminds me of a poker player with a huge stack, who was coasting to lock up a big night, but then loses a big pot (because he lost his mojo, from his awful pick for a VPand to becoming unsettled at the thought of running against a formidable candidate) and then goes on tilt and never regains his form (i.e.his recent behavior, and tantruming himself into this downward spiral).
Thoughts, Sam?
P.S. Just to be clear, that’s not the Mike Beer who was a postdoc with Saeed at Princeton 2000-2004.
EmptyWheel mentions six things that could significantly impact the race. I’ve added two:
– Big protests at the DNC
– Ceasefire in Israel – or further escalation
– Superseding indictment and/or sentencing for Trump
– Another tumultuous debate
– Further decline in Trump’s mental state
– Far right political violence
– Major false-flag terrorism action
– Last-minute release of AI-generated evidence of a “scandal”
https://www.emptywheel.net/2024/08/17/2024-remains-an-unpredictable-race-six-things-predictable-things-that-could-still-upend-it/
I’m very interested in Sam’s thoughts and those of others here on this.
Any chance you could add Josh Hawley vs Lukas Kunce in Missouri to the Senate races for which you are showing the polling margins?
Thanks, we will keep an eye on that. Not many polls now. The one recent survey shows Hawley up by 9 points.
Nebraska, on the other hand…
Hard not to notice that the meta-margin has moved back towards Trump to the tune of 0.8% (From D +0.2 to R +0.6)
To what do we attribute this? Kamala was benefiting from a post-announcement bump that is now subsiding?
Could be real, could be noise. The tracker here has a certain time delay. To form an opinion on your question, look at the national surveys. Possibly 1 point back toward Trump, one has to squint or calculate.
Thanks, Sam. Yes, I had forgotten about Osborn in Nebraska. Brilliant if you could add both.
Two technical things: Even though I choose “Reply”, that’s not where this comment is placed. Second, for every comment I have to do a new “I’m not a robot” reCAPTCHA. Is it intended to be this strict?
Wondering if the systematic bias of the polls in favor of the Democratic presidential candidate in 2016 (which translated into a 2-point meta-margin bias) has been understood and if it has been corrected in more recent polling.
Vote Maximizer goes a great job of telling us “where” our efforts could make the most impact. But a related problem that I struggle with is “how” to exert my efforts. I’m in a state where my voter power is nil for President and nil for the Senate, so I cannot affect anything by knocking on doors anywhere within reach. Sam, do you have research in hand (not necessarily your own team’s of course) that shows what *methods* of seeking impact in swing states – when you don’t live in one – are most effective?
Money to individual campaigns? Post-carding? Phone-banking? Texting? Do any of the last three there have any measurable impact at all? (I know many people who do them/are doing them and feel better but that’s a very different matter, and potentially a misguided waste of valuable activist energy.) Do you know? Does anyone really know?
How many House seats does the 3.0% meta-lead translate into?
Do I understand correctly that there is now > 90% probability that Democrats will flip the House?
Does a live meta-margin graph exist somewhere? Somehow not able to find it.
On a desktop, on the banner click which chamber/office interests you. Or click on a graph in the sidebar. The President page contains all three superimposed.