PEC, Election 2024, and more – we’re back
Hello, PEC fans. The Princeton Election Consortium is back! The data feed is live, and there are new projects. Today, just a brief note.
The data feed is live. You can get it by clicking in the banner above or in the right sidebar. As before, the feeds are clean statistical snapshots of state-level polls for President, U.S. Senate, and U.S. House. This has the advantage of (a) rapid response to new surveys, (b) identification of key states, and (c) reporting in terms of virtual margin (“meta-margin”).
As in 2020, we’re not emphasizing win probability. People get carried away with probabilities. Considering that the overall 2016 polling error was about 2 points, I prefer to think of this year’s race as close. The meta-margin reveals that the race is basically within polling error. Of course this can be turned into a win probability (and you can dig that out of our GitHub if you really want), but it seems better to emphasize the closeness.
For example, here’s the meta-margin for the Presidential race:
As you can see, the entry of Kamala Harris has brought this year’s Electoral College snapshot to a place it never went. In that respect, her entry was indeed a game-changer. At the moment, the key states with highest voter power are North Carolina and Georgia.
Harris’s entry has also brought the race closer to the Senate and House snapshot trackers.
In our modern age of partisan entrenchment, the Presidential and House trackers would be expected to stay close to one another. The fact that they didn’t before Harris’s entry is an indication of Biden’s weakness as a candidate. Harris is probably winning back a portion of three groups: double-haters (people who disliked Biden and Trump), Kennedy supporters, and undecideds. These are, of course, overlapping groups. In coming weeks we will see how far she gets.
The Senate tracker shows a knife-edge race for control. Democrats and their independent allies are ahead in enough states to add up to 49 seats after the election. After that, the critical races are currently in Montana and Florida. (Note that I am tracking down a possible glitch in the Senate tracker. More on that in coming days, I hope.)
Posting here will be a bit lower-frequency than in past years. I have other projects you can read about over at Fixing Bugs in Democracy, whose title says it all. Also, check out VoteMaximizer.org, which arose from work ideas developed at PEC over the years.
Welcome back!
Thank you Dr. Wang!
I use your site to guide my donations: what’s the most effective way to spend my money.
Thank you! Scroll down for the ActBlue/WinRed. Also, see the new resource votemaximizer.org, which is rich in information.
I am personally very attentive to Senate races these days.
Thanks for getting up and running. Where can I find the polls you use to create your “status of the race” data and graphics. I used to know where it was but I’ve forgotten since 2020. 🙂
Pleased to see you, and everyone else!
Data: https://projects.fivethirtyeight.com/polls/
Code: https://github.com/Princeton-Election-Consortium/data-backend/tree/master
Good to see you back. I’ll check in more often now. 🙂
Excellent, glad you’re doing it again!
Good to have your site back Sam, I look forward to that jittery line moving upwards. Can you remind us of what quality cuts you put on the polls you aggregate? And where are you pulling your polls from? I recall it was Huffington Post at one time, is it still?
Hi, glad to see you!
The sieve for President: (a) all states, with 2020 margins where there is no data; (b) from FiveThirtyEight take all polls no matter what, (c) where one organization releases multiple versions on the same day, pick likely-voter over registered-voter surveys, and two-candidate (i.e. Harris v. Trump) over other (i.e. including Kennedy Jr.). Then (d) calculate the median, estimated SD, convert to SEM; (e) convert to probability using a t-distribution, (f) fast-calculate the whole distribution using conv() in MATLAB.
For Senate, it’s the same except the time window is slightly longer but I have to look that up. We are tinkering to try to get faster response times, which of course everyone wants, including me.
If you have the patience, poke around the codebase: https://github.com/Princeton-Election-Consortium/data-backend/tree/master
Excellent. So glad to see PEC’s prognoses are back.
Greetings!
Welcome back, Sam!
Welcome back, Dr. Sam! You’ve been missed and your return is…quite timely. Thank you!
Happy to see you’re back!
Question: Is there an easy way for readers to get at the data behind the presidential race EV histogram?
Thanks!
https://github.com/Princeton-Election-Consortium/data-backend/tree/master/matlab/outputs
Hi Sam, Do you have a histogram for House seats, similar to the one for Senate?
Not at present. That would involve estimating a conversion between the generic ballot and seats, among other things. Good thought, stay tuned.
Not sure where to comment, but might want to look at the coloring on the senate slide — since the presidential election is a toss-up/favoring republican, you might want to draw the line of control in the center of the 50/50 senate split to reflect this uncertainty.
the senate meta margin for democrats is objectively NOT sitting at 50 seats today.
I agree that the right way to color that is to split the 50-seat bar. There is a MATLAB version sitting backstage at the Github that shows that. We can fix it, but in the meantime your critical sense is serving you well!
Do I understand correctly: that as of 10 August, your data points to a Trump victory, with Kamala Harris attaining only 251 Electoral Votes?
I have a hard time seeing the same. Does this take into account the recent polling in swing states?
Hi, and glad to see you back!
As in previous years, the snapshot and prediction only use state polls. The current time window is 2 weeks or 3 polls, whichever is more. Every swing state has data. The current update times are 10am, 4pm, 10pm. We’re about to increase that to once every hour.
I have to revive the HTML table of states. In the meantime, see our data source, FiveThirtyEight, which has median margins of: Pennsylvania (Trump +1%, 9 polls), Arizona (Harris +1%, 7 polls), Georgia (tie, 5 polls), Wisconsin (Harris +2%, 6 polls). Note that the margin for Pennsylvania is in the same direction even if the time window is shortened to 1 week. So, I think a meta-margin toward Trump is correct.
Give it a little time?
Thank you! Encouraging to hear this is most likely a time lag.
Ah, my last comment was meant as a grateful reply to your response, but it didn’t post under it.
the recent david wolfson nevada poll doesnt seem to be in 538. i wonder if their data is stale. great yo see you back dr. wang
On casual reading the survey seems OK, including methodological details. Maybe they missed it? Write them at polls@fivethirtyeight.com.
Great to see you’re cooking again!
Sam, I have a question about your MoneyBall Politics – specifically about the Montana Senate race. At the outset, with less than a million voters, Montana might seem a good place for me to send money to Jon Tester’s campaign.
However, if I recall correctly, an astonishing 200 million dollars is already being invested in this race, which is an astonishing amount of money per voter.
Does your Moneyball calculation take into the huge amount of money being invested? For surely this will eclipse anything I might donate and thus greatly reduce its actual impact.
That’s a good point. It’s not accounted for. Right now the formula, detailed over at my Substack (which everyone should subscribe to), divides by state population as the normalization factor. One could instead divide by total funding.
Or…move to Montana for a few months!
Total ad spending for the Montana Senate race per 7 August was $ 205.1 million.
https://www.axios.com/2024/08/07/senate-campaign-2024-election-ad-spending-republicans-democrats
Good to have you back, Dr Wang – one of my go-to sites for following the shifts in US elections from this side of the Atlantic.
Good morning, Sam! Is the recent Cook Political Swing State Project poll included in PEC’s calculations? Or is this poll “outside” the database? (Carried out in cooperation with BSG and GS Strategy Group)
https://nitter.poast.org/CookPolitical/status/1823670209581817941#m
(NB. Nitter is a way to access Twitter/Xitter posts.)
so the ev is now showing a narrow d win. which poll pushed it over? was it fla?
Does anyone know where I can find reasonably accurate, updated data on the following:
– Number of voters purged from voter rolls, per state
– Number of newly registered / re-registered voters, per state
I have searched and asked numerous people, but have so far been unable to find these two simple overviews.
(Very partial data occasionally appears in news articles. However, these articles fail to refer to any updated sources. Usually they don’t refer to sources at all.)
Arctic Stones try Michael Mcdonald at u of florida. he seems to have a lot of data and seems to know where everything is.
Paras, many thanks! I have just email my question to Prof. McDonald.
The NYT and Chris Cuomo showed registration data recently so its available. ive also seen a chart with new residents, and move outs by affiliation. So the data should be available somewhere.