Figures: 0,1,2,3,4,5,6

• Data, Terminology, and Methods
• Introduction
• Rough Timeline and Rundown of Election
• Deep, Deep Systemic Issues with the Exit Polls - but How we Can Still Use Them
• Historical Trends
• Side Remarks: No, the Muslim/Arab/Third Party Vote Didn’t lose Harris a Single Swing State
• Side Remarks: The Elusive "White No Degree"
• Side Remarks: Marital Status, Singles, and Incels?
• Side Remarks: Assessing Harris, as a candidate on TV, policy aside - but also with policy
• Stop with the "Garbage In, Garbage Out": Considering Turnout Patterns, and not just group vote ratios, shows a drop in POC and Working Class Vote, Rise in White and Wealthy Vote
• Side Remarks: "Can we compare 2024 to 2020? What about COVID? Doesn’t that make it a super exceptional election?"
• An Election for the Rich
• An Election for White People, and Especially White Women
• Side Remarks: The Asian Curveball
• Side Remarks: How Did Swing States Do?
• Verifying the Prediction?
• Closing Remarks
• Data Analysis: Working Class as a Whole (Tables: I, II)
• Data Analysis: 2016, 20, 24 Race and 2016 Race-Gender Results (Tables: III, IV)
• Data Analysis: Latino (Tables: V, VI)
• Data Analysis: Asian (Null, not enough data)
• Data Analysis: White (Tables: VII, VIII)
• Data Analysis: Black (Tables: IX, X)
• BONUS SECTION: The Youth Vote (18-29) Won’t go Trump, but They’ll sit out the Election (Tables: XI, XII)
• Latino Regional Analysis (where did Trump’s new Latino Male vote come from?)
• Gender and {Gender} × {Income}
• Other Observations (very rough, still working on)

(Back to Table of Contents)

This analysis is based on exit polls from the National Election Pool (NEP) and Edison Research (ER) (referred to as NEP+ER throughout), provided by CNN (2016,2020,2024) (NEP is a collection of news media organizations (ABC, CBS, NBC, and CNN) that work together for exit polling). These tell us the percent of the sample (PS) from some group 𝑖 and the vote ratios of some group 𝑖. As already mentioned, these sample 15k-25k people, which means the groups analyzed here ({White, Black, Latino} × ({male/female} or {<$50k, $50k-$99.999k, $100k+}) each have good sample sizes, although some push it at the lower end (ideally sample size should be order of 1000 participants; so a sample of 100 would be suspect). But these results likely aren’t 100% accurate regarding actual turnout (ie an exit poll may suggest 13% of voters were black, when in reality it may be 14%). Nonetheless, the large sample size lends confidence to these exit polls (I imagine CNN is aware of this).

I’ve also started adding in some results from the Fox News+NORC+AP (referred to as Fox+NORC+AP throughout) exit polls, as provided by Fox News (2020, 2024). They seem to be the only exit poll which isn’t the NEP+ER exit poll. Their exit poll, according to their methodology statement, is conducted by NORC at University of Chicago for both Fox News and AP (so it seems the statement that the NEP+ER exit poll is the "only national exit poll since 2004" is simply wrong) (and also, that means Fox News exit poll is the same as the data used by AP). This exit poll gives similar income-group trends (although while also showing $50k-$99.999k having increased no-votes, they show a more substantial Trump gain here than the NEP+ER exit poll), and predicts the popular vote outcome well (and loss/gains of votes) (see Table II.3b), but their race-based results seem distorted (they indicate that among white+black+Latino, Harris loses only 2.8m votes, Trump gains 3.2m (see Table III.2b); where are Harris’s other 4-5m missing votes? In the "other" races of America? That seems dubious). The calculated turnouts for black and Latino voters from their sampling hints at the reason: 2020 turnout was computed as 54.2% and 44.0% respectively, and for 2024, 45.2% and 43.1% respectively (compared to NEP+ER: 2020: 64.0% and 63.6%, 2024: 49.7% and 51.7%). White voters were represented in 2020 (74/67) = 110.4% more, and in 2024, (75/71) = 105.6% more than the NEP+ER exit poll. That doesn’t make it necessarily wrong (that depends on what the actual proportion of voters was), but it’s worth pointing out the differences of the polls.

Insofar as there are errors in the data (and this becomes a greater issue with the inclusion of other data-sources; ie the way the CNN exit poll determines how a person has income <$50k and is black, and the way the US Census Office does, may differ), trends based on the exit polls for these elections will likely share the same issues (ie if a group is under-represented in each), and so the trends should be more reliable. For example, if analysis the 2020 and 2024 polls show group X turned out at 65% and 55%, but in reality they turned out at 70% and 61%, the actual turnout values may be wrong, but the downward trend is still observed. Hence, I stick with the CNN exit polls for 2016 and 2020, even though more accurate data may be available, in aim of shedding a more clear light of trends in 2024.

As discussed here, there are deep deep systematic issues with these exit polls. Still, from what I can tell, they are useful for discerning trends.

National Eligible Voter Population and General Turnout: precisely computing election turnout is surprisingly tricky. For example, while one must be 18 or over to vote, not all 18+ can vote (ie felons in many (all?) states, non-citizens). For much of the analysis thus far, I’ve been using results from Wikipedia and other sources. Thus, I’ve taken turnout from Wikipedia for 2016 (60.1%) and 2020 (66.6%). I had originally found that the eligible voters for 2024 are around 244m (which seems to hold), and that turnout would be at ~63%. And for 2016 and 2020, I’ve taken the eligible voter population from computing EV from these values, based on total-votes/turnout = total eligible voters. Later I found a total of 246.05m from Pew (2024, P24), so some results are slightly out of tune. Even so, upon digging, I’ve found this problem isn’t as simple as it seems.

For example, the UC Santa Barbara "The American Presidency Project" (last updated 2024) (APP24) offers 230.9m eligible voters in 2020, and 242.69m in 2024. But if we use the 242.69m number for 2020, we get a turnout of 65.3%, not 66.6%. The 66.6% number comes from the "US Elections Project" in 2020 (UEP20), when they estimated eligible voters was at 239.9m. In fact, APP24 and UEP20 computations of eligible voters were apparently done by the same person, Dr. Michale McDonald, but the new estimate (242.69m) is due to new census data. So it appears the actual turnout was at 65.3%.

For 2024, such issues persist. For example, the P24 article above suggests 246.05m; with results we have now, we can more precisely say this gives a turnout of 63.3%. Meanwhile, the Election Lab at University of Florida (2024) (ELUF24) computes 244.667m (with a turnout of 63.68%).

Overall, I will work on updating (updated income tables, race tables for 2020 and 2024), but will update in the future, using the 2016 eligible voters of 230.932m from the APP24, which with total votes of 136669276 votes, gives an actual turnout of 59.2%; for 2020: the APP24 turnout of 65.3% and eligible voters of 242.69m; for 2024: I will use the ELUF24 results, with a turnout of 63.68% and eligible voters of 244.667m.

As is, these shouldn’t affect results too much, but will make me feel better. Some trends comparing 2024 and 2016 may be different however, as the actual 2016 turnout is 0.9% lower than the oft-cited 60.1%, and the turnout for 2024 I will use now is 0.68% higher than I was using prior. Likely 2016 values are ~1.5% too high (ie 50 → 49.3), 2020 values are ~2% too high (ie 50 → 49.0), and 2024 values ~1% too low (ie 50 → 50.6)).

Information about the eligible voter population for Black and Latino voters is obtained from Pew (here is a link to their data on Latino (Hispanic) voters; you can find the other reports from there). They also report eligible populations for Asians, although their sample size is small in the CNN exit polls, so I didn’t carry out a race-income and race-gender analysis for Asians. This data includes total population for a racial group, percent of the total eligible voter population which is of a racial group, and the eligible voter gender ratio within a racial group (male:female ratio, US - 49:51, Black - 47:53, Latino - 49:51, Asian - 47:53). To find white eligible voter population (white_EVP), I took the difference of the total US eligible voter population (US_EVP) and the black, Asian, and Hispanic summed eligible voter populations (BAH_EVP):

While this erroneously rolls into "white" several minorities who are not white, the predominance of whites among the remainder means the US_EVP - BAH_EVP value will give a good idea of the white_EVP, despite these errors. If you have a direct source for the eligible white voter population in 2016, 2020, and/or 2024, please email me at glaznaruost@gmail.com. For the white eligible voter gender ratio, I used the US collective ratio, 49:51.

Note that the eligible voter gender ratio already accounts for gender imbalances within a group (ie if, for some reason, a group 𝑖 was 90% male, and 10% female, then the expected ratio would be 90:10).

To find income distributions for the whole US, or a racial group, in a given year, I used data from the US Census Office (the file I used is here (clicking will download a PDF file)). The income brackets analyzed are either {<$50k, $50k+}, or {<$50k, $50k-$99.999k, $100k+}, since these are the brackets broadly used by the CNN exit pollsters. I use the computed income bracket ratios as the percent of the eligible voter population, although the actual eligibity values may be slightly different (since this is household data, certain income brackets may have factors affecting their eligibility more than others, etc). For 2016 and 2020 income bracket ratios, I use incomes from those years. For 2024, I use the reported income ratios for 2023, since that is the most recent year available from the Census Office data.

Household vs Income: It’s unclear to me if the exit polls asked for household or individual income. (Edit: It appears its household income, or at least, "family income", as reported by NBC). I tried making a crude adjustment to income distributions to account for the fact that households often have more than one earner (by dividing the median household income and median individual income for a group, and adjusting the income distribution accordingly), and the results are quite a bit more unrealistic than just using the household distributions, but not as bad as I originally thought (was a code error). Thus, I’ve stuck with just the household income distributions, as given by the Census Office, although I’ve presented the individual-shifted results for the whole population and white, black, and Latino.

Interestingly, because group turnout is computed as (PS/PP)*Tᵢ = (PS/(gT/wT)) = (PS*wT)/gT, where wT is the whole total, and gT is the group total, this means that total votes = ((PS*wT)/gT)*Tᵢ*PP*gT = (PS*wT)*Tᵢ*PP, where PP is the party percent vote reported in the exit polls. This means that vote totals for the parties add up the same for any distribution (with some variation for no-vote). However, actual percent votes - ((PS*wT)/gT)*Tᵢ*PP - does change. Qualitiatively however (ie in terms of sign), even these results don’t change. Perhaps the only noteworthy change is, for the whole nation, household income gives Trump +2.9 percent points with <$50k from 2020→2024, whereas the individual-shifted income distribution gives Trump only a +0.9 percent point gain here.

Class terminology: Throughout, I make references to "working", "middle", and "upper" class. I have in mind the standard Marxian definition of "working class" (someone who sells their labor-time for a wage), but this gets mixed with the non-Marxian categorizations of "middle" and "upper", as they are a big part of American political lingo; in reality, "middle class" - a loose term that refers to someone who has savings (and can probably comfortably retire in their 60s), owns a home (or at least, is on track to pay it off), aspires to some degree of professional/educational virtue, etc - includes many working class people, and also includes many bourgeoisie. "Upper class" almost entirely consists of bourgeoisie, although one may argue that wealthy professionals, such as professional sports players, are working class (until they begin investing their earnings, at least). The income brackets don’t necessarily correspond 100% to any of these. For example, <$50k is indicative of working class trends, but likely a large portion (maybe even majority) of $50k-$99.999k are also working class. We may comfortably assume, however, that middle class and upper class predominate in the $100k+ group.

Turnout computation for a given group. For a given group 𝑖’s turnout (Tᵢ), I took the percent sampled (PS) from a group in the CNN exit poll (ie Latinos were 13% of those sampled in 2020), and divided it by the percent in the eligible voter population (PP) of that group (ie Latinos were 13.6% of eligible voters in 2020), and multiplied by general turnout (GT) (ie 2020 turnout was 66.6%):

For example, for Latino voters in 2020, this computes as T_{Latino} = (13/13.6)*66.6% = 63.7%. For 2024, I’ve estimated general turnout at around 63% (based on how many votes remained to be counted from AP reporting), but this may need to be updated in the future. When looking at groups within a group (ie gender within the black population, income brackets within the white population), the percent of eligible voter population (PP) is with reference to that group, not the national whole (ie the income bracket ratios of Latino voters should add up to 100%).

"Actual" vote ratios and totals. Given a group 𝑖’s turnout Tᵢ, I compute how many 𝑖 eligible voters voted for Democrat (D/Dem), Republican (R/GOP), and no-Vote (NoVote) as follows. First, find the Dem:GOP voter ratio for that group from the exit polls (ie, following the above computation for Latinos, Latinos voted 65:32 in 2020, based on the exit poll); keep in mind that these ratios are, by definition, the results of people that voted, and excludes those that didn’t. To figure out "actual" vote breakdowns, I multiply those values by Tᵢ (for Latinos in 2020, T_{Latino} computed above: 63.7%; so, Dem: 0.65*0.637 = 41.4%, GOP: 0.32*0.637 = 20.4%), and find the percent NoVote based on 100% - Tᵢ (ie Latino NoVote: 100% - 63.7% = 36.3%), which gives a Dem:GOP:NoVote ratio (ie for Latinos: 41.4:20.4:36.3) (Note: these won’t add up to 100, because a small fraction of every group 𝑖 votes Third Party). The total votes is then found by multiplying these ratios by the total population of group 𝑖. For Latinos, their eligible voter population in 2020, per Pew above, was 32.3m. So the absolute value of Latino votes for each "candidate" would be:

• Dem: 0.414*32.3m = 13.25m
• GOP: 0.204*32.3m = 6.59m
• NoVote: 0.363*32.3m = 11.72m

Thus the 2020 Latino total vote breakdown Dem:GOP:NoVote is 13.25m:6.59m:11.72m.

The above is the method for the analysis of all race/income/gender groups, or two-way combinations of {race} × ({income} or {gender}) (the CNN exit poll data does not permit three-way combinations). Further, the US Census Office does not give gendered income distributions (only mean income values), so I did not do a gender-income analysis. Further, from the 2016 exit poll, I’m unable to look at race-income groups, so that analysis is left out for 2016.

For the age analysis, to find the percent of population of an age group, I took the population pyramid of the US from populationpyramid.net (th ODS file I used is here (clicking will download a file). Some method notes for that analysis are given in that section. For historical voter turnout by age bracket, I used this data from Statista.

Error: There are a few sources of error that could be included (ie to give something like a standard deviation, once errors are propagated through). I initially started writing this article quite flippantly, just to see what the rough picture is from the exit poll data - I was very annoyed seeing people directly comparing Dem:GOP pol proportions from 2020 and 2024, because changing turnout in general, and among component groups, is such a crucial part of the picture (if 50 𝑋 people vote blue and 50 𝑋 people vote red in 2020, it’ll look 50:50; then if 20 of the blue 𝑋 people don’t show up, and only 10 red 𝑋 people don’t show up, in 2024 - but no one changed their mind - the exit poll ratio will be 42.9:57.1. But that doesn’t mean Trump made gains - it means he just lost votes slower!). But I may include errors (the main quantifiable one being from the exit polls; the errors given a sample size are documented here).

(transcript, Oct 1 2024)

(1) Astead Herndon: So should she be saying that? Because she hasn’t really communicated more directly, hey, I’ve learned and I have changed from that thing. I mean, she’s tried to make a everything-for-everyone case. We talked to, actually, Bernie Sanders on this exact point, and he’s someone who’s making the argument that the Kamala Harris administration would be one that has a robust populism that progressives would have a voice in. And so I guess I’m asking, how can a Mark Cuban economy and a Bernie Sanders economy both exist under the same — It seems like both sides are pitching to me that the Kamala Harris administration would be one of them.

Mark Cuban: I hear you loud and clear.

Astead Herndon: Yeah.

Mark Cuban: But one was a month ago, and you haven’t heard about it since.

Astead Herndon: That’s fair. You’re saying she’s tacking towards you closer than she’s tacking towards them.

Mark Cuban: Right. I mean, she’s giving economy speech after economy speech. She’s talking about the things because if the number one issue is economy and the way to solve people and make people feel good about what she’s doing is to talk business and talk economy and talk opportunity, and that’s what she’s talked about. And I think you have to be in the middle to do that.

Astead Herndon: I hear that. You endorsed Nikki Haley in the Republican primary. And one question I wanted to ask, and it’s to get at the idea of what’s driving your support. But if she was the Republican nominee, would you still supporting Harris in this way? I asked to say like, is this about the affirmative advocacy for Harris and her policies or is this about rejecting Trump and his policies?

Mark Cuban: So if it was Nikki Haley versus Joe Biden, I probably would have voted for Nikki Haley.

Astead Herndon: OK.

Mark Cuban: Given that what I expect her policies would have been. And I’m socially liberal, but conservative fiscally. And so it depends on abortion and those things because that’s important to me. But with Vice President Harris, I think she’s truly on point with our mission. So this is an anti-Trump vote for me. I really like where she’s going. I mean, she’s very thorough. She’s very precise. She’s very well thought out. She takes an idea, she evaluates it. She may agree or may not agree with you, but you know it’s going to be completely and thoroughly thought through.

...

(2) Astead Herndon: Do you think that there’s been enough work done to make folks who are maybe fiscally conservative but open to voting for a Democrat feel comfortable with this candidate? And if not, what do you think the next steps or next message needs to be to get other people to the place that you are? Because theoretically, if the undecideds, if the center land where you are, she probably is in good shape for this election. So how does she get those folks there?

Mark Cuban: Repetition, just repeating her core principles, her policies, and letting people know that this is better business for the country of the United States of America. Donald Trump is the new Oprah of tax cuts.

You get a tax cut. You get a tax cut. You get a tax cut. And then on the other side of his mouth, he talks about reducing the deficit. Those two things don’t work together. Whereas with Kamala, she’s like, OK, we have to be more aware of the deficit, cutting interest rates will help. Hopefully the Fed keeps on along those lines.

But increasing taxes is going to reduce the deficit. I mean, her meeting with all these different businesses, from VCs to investors to Wall Street to bankers to little companies to big companies, that’s huge. That’s where she starts to communicate, because then as you get these individuals who are in the business community talking about those things, that conveys the confidence that she’s able to do this. You are not seeing that from the Donald Trump side.

Astead Herndon: The presidency you’re laying out, as for the hopes of Harris, is the same premise that Biden gave in 2020, that he would govern more from the center. I guess in retrospect, I guess I’d say it’s the inverse of Bernie Sanders because he was saying that Joe Biden ended up governing from more of a progressive perspective than he anticipated, and he liked that. I guess from your perspective, was Biden more progressive than expected and that was a bad thing?

Mark Cuban: That was a bad thing, yeah. I think he wasn’t a great communicator. And for all the reasons he’s not in the race anymore, there’s a lot of reasons potentially why. But I thought him not communicating well about business or to business people really hurt him.

I can’t tell you how many people I’ve talked to since she got in the race who are really successful business people who said, I tried to talk to Biden and nobody returned my call, whereas now I connect them to her immediate team. And they’re getting callbacks, they’re getting meetings, they’re getting phone calls. And I think that shift is enormous. And I think it really hurt Joe.

(3) Astead Herndon: One thing that’s sticking out to me, though, because you hear people say now that they don’t think they are getting a sense of where she’s ideologically landing from the last month or so. I hear you saying that you feel like you have gotten an increasing sense, and it’s one that’s more of a moderate candidate, and it’s one that’s actually owning that lane more than I think other people are giving her credit for. Is that a fair characterization of your view?

Mark Cuban: Yeah, no, for sure, because ideologues talk the party talk. Here’s who we are. This is what we believe in, and that’s it. She’s got her core values. But problem solvers talk to people, and they say, OK, we have people who are suffering in this country. What do we do? We want to see more businesses starting. What a great idea. Let’s increase the child tax credit for your first year, because not only can it help you with daycare, but if you’re in a business, you can bring somebody in to watch your infant that’s just been born. And that allows you to go back to work sooner.

(4) Astead Herndon: In my head, the billionaire class would break Republican, would break Trump supporters. Is that tax cuts? How should I understand the fact that you’re laying this out, and it feels very clear to you, you’re advocating folks around you, things like that. But it does seem more of an outlier than the rule. So what’s driving the rule?

Mark Cuban: So first, don’t judge me by my wallet. And I think that applies to everybody. When I was poor, you weren’t going to listen to me. But now, and I don’t — my friends aren’t billionaires per se. But when I do talk to wealthy people, I don’t think they think like that. I think they look to see what are the things that are important to them. And that falls into two categories. There are single issue. It may be Israel. It may be abortion. It may be the border. It could be any number of things. It may be tariffs.

Astead Herndon: Yeah.

Mark Cuban: And so if they’re single issue, that’s just who they are. And what I try to do is say, for all those things I mentioned, let me get you in touch with her team.

Astead Herndon: OK.

Mark Cuban: Maybe her perspective on the border is not what you think. She is not Joe Biden. This is her party. It’s not the Joe Biden party anymore. And then there’s the second category, which are, I think we can manipulate that other guy. If I can get close to him with donations, I can get him to listen to me and do what I want because I know he doesn’t get into the details.

And so the wealthy class, the billionaire class, if you will, either they feel they can influence him because he doesn’t really have strong positions or deep thought-through positions or they think there’s a significant benefit for the one issue that’s important to them. But I think the misunderstanding is it’s just taxes and money all the time.

Astead Herndon: Yeah. Yeah, that’s why I asked.

Mark Cuban: Yeah, but it’s not just that.

Astead Herndon: But the switch, to be clear, and this is just my last point, the switch has allowed you an opportunity to make a case to them to say, hey, go talk to them. And there’s a new — there’s a new level of openness.

Mark Cuban: She says it all the time. I want to talk to businesses, top to bottom.

Astead Herndon: OK.

Mark Cuban: I want to hear the best ideas. I think she said it in her speech in Pittsburgh. I want to hear the best ideas because that’s what makes this country better. I want there to be more entrepreneurs. All it takes is one to change the game and just make everything better for us. Whereas Donald Trump, when was the last time you heard him talk about entrepreneurs?

Astead Herndon: Yeah, yeah, yeah. I mean —

Mark Cuban: Startups.

(5) Astead Herndon: This will come out on Tuesday, so this is partly why I ask this question. Are you going to watch the vice presidential debate?

Mark Cuban: I don’t care. I really don’t care. I really don’t care.

Astead Herndon: Just because VPs aren’t people who matter most? Or why don’t you care?

Mark Cuban: Well, yeah. I mean, yeah, there’s the heartbeat away thing. But beyond that, they’re already picked. We can’t change it. There’s nothing we can do there.

And I mean, what do they do? I mean, what is it that Mike Pence said in his VP debate with the vice president that changed the course of history or really had an impact? And you go before that, what Biden said. Whatever it is, nobody remembers any of it.

Astead Herndon: I’m with you that it’s not usually an electoral point, but we are in a moment that should show how VPs end up mattering. Obviously, Harris ascended and all of that. There were so many questions last year about whether that was the right choice or not. I was just wondering whether any of the number twos impact some of your decision making, or is this for you a question of Harris and Trump?

Mark Cuban: This is Harris versus Trump. That’s the whole election. Who do you trust? Because the Republicans are going to say you had 3 and 1/2 years, which is not necessarily true when you’re second in command. And you can’t trust her. Why do you believe her now when you didn’t believe her then?

And it’s the same with Trump, right? You’ve been lying about everything you’re saying. You’re going to say you did this, and you didn’t do it. Why do I trust you? That’s what this election is about. Who can generate the most trust to the most number of people and get them to vote for you?

Whether Tim Walz — I think he’ll be — the reason to watch it, he’ll tell some great stories. And you want to see if JD Vance steps in something. But you also know JD Vance is going to be super intellectual. And it’ll be interesting to see if that works against him.

I think if I have a hope for Tim, it’s that he just comes across as a good, respectable guy and he doesn’t take the bait for anything. Because you know JD Vance is just going to try to just zing him, zing him, zing him, zing him, zing him like he’s in a school debate class. And I’m sure the governor has been through that before.

(6) Astead Herndon: So like I said to Mark Cuban, one of my big takeaways from our conversation is somewhat related to this open question of defining Kamala Harris and what she believes in. He thinks her ideology is clear, that Harris is no left winger, and that particularly when it comes to the economy, she’d be more open to business and billionaires than the Biden administration ever was, which is the exact opposite of what I heard from progressive leaders who cited things like Harris’s selection of running mate Tim Walz as a sign that the left would have a voice in a Harris-led Democratic Party.

Tonight, Walz and Republican vice presidential candidate JD Vance will face off in what could be the final debate of the 2024 cycle. We’ll be watching from Tim Walz’s former congressional district in Southern Minnesota. But we’ll also be listening to see which version of Kamala Harris does Tim Walz present, the progressive ally, as described by Bernie Sanders, or the pro-business moderate, as described by Mark Cuban.

And it was corporate Democrats—particularly lobbyists like Harris’s brother-in-law, former Uber executive Tony West, and David Plouffe—who held the most sway over Harris’s campaign. They advised her to cozy up to ultra-wealthy celebrities, Liz and Dick Cheney, and Mark Cuban, and avoid populist rhetoric that could have distanced her from the corporate elites who dominate the party. In 2024, the biggest spenders in Democratic Party politics weren’t progressives—it was AIPAC, cryptocurrency PACs, and corporate giants like Uber, all of whom poured millions into Democratic campaigns without regard for public opinion or the will of the people.

Waleed Shahid, The Nation, Nov 18 2024

This section is very rough, but gives a rough idea for now. The idea here is to give a timeline of events for reference, to show donation trends, to discuss rhetoric trends, to discuss Harris’s inner circle, and to highlight how Harris initially tried to balance populist messaging with the typical Democratic dependence on corporate donors. Quickly however, she decisively tipped towards big capital, and away from an economic message that could appeal to working Americans. The goal was to compensate with middle- and upper-class voters. As we now know, this failed catastrophically.

Figure t.1: Top: donations per day for Biden/Harris vs Trump, for the 2024 election. Bottom:: Cumulative contributions for Biden/Harris vs Trump for the 2024 election, and compared to the Biden vs Trump cumulative contributions for the 2020 election. The sharp rise in 2024 Democratic cumulative donations happens about 45.86% between "200 days" and "Election day", which is around 108 days before (July 20). However, there is some error in my estimate here - given the timeline above, this jump probably happened between July 21 4:31 PM and July 22 4:32 PM, as that was the time interval of Harris’s record-breaking fundraising of $81m in 24 hours, and this spot in question has the steepest slope. (source: NYT)

From Figure t.1 (bottom), we can clearly see the huge spike in donations starting at the beginning of Harris’s campaign (July 21).

Jacobin did an analysis, finding that Harris’s rhetoric focused on kitchen-table economic issues during and immediately after the DNC. Yet these topics became increasingly marginalized. Why? Because such rhetoric was was uncomfortable for the donor class - and they were well nestled in her inner circle.

The following is Harris’s inner circle, as reported by NYT (published Aug 17 2024, updated Sept 28 2024; this "inner circle" is the same since August publication). I’ve broken down the backgrounds of each here.

What can be observed is (A) a battery of attorneys (especially from Jenner and Block, and Paul, Weiss, Rifkind, Wharton and Garrison LLP), (B) along with close friends and family, many worked with Harris either during her Senate tenure, or for her 2020 presidential campaign (which makes sense), (C), a few strategy firms/NGOs with common membership here (ie Precision Strategies, ONE Campaign), (D) many old advisors to Bill Clinton, Harry Reid, Hillary Clinton, Rod Blagojevich (the one with the corruption scandal), and Barrack Obama, (E) several high-profile private sector leaders (ie Jalisa Washington-Price: VP of iHeartMedia, Tony West: high-level Uber executive), (F) several connections with the tech industry at large.

The latter is really interesting. The most obvious person here is Tony West - husband of Harris’s sister Maya - a top Uber executive. He’s credited with making Harris tone down populist anti-corporate messaging, to attract business support. But it goes much further than this. For example, one of the most interesting advisors here is Laphonza Butler. Her political career began as a union organizer in the SEIU across the country, including California (during which time she befriended Harris, and got the SEIU to endorse her 2010 AG run). In 2018, she joined SCRB Strategies, a political-consulting firm in CA. There, using her experience with the SEIU, she advised Uber on dealing with unions ata time when they were trying to stop pro-union legislation in California. After this, she left SCRB to join Airbnb. Another campaign advisor, Karen Dunn, defended Uber in 2017 in a case with Waymo over stealing tech secrets. She also represented Apple in court twice - the 2nd time, in a high-profile case against Epic Games. Apple’s victory in this case has strengthened tech monopolies’ control over their mobile app stores. Another advisor, David Plouffe, has been caught up in a variety of scandals, including taking paid speaking fees from Azerbaijan. After a career in the US legal system, as well as serving as a policy advisor to then president Obama, in August 2014, he was made Senior Vice President of Policy and Strategy at Uber. The company was then facing regulatory hurdles, and hoped Plouffe would help them get these lifted. In February 2017, he was fined by Chicago for not registering as a lobbyist; he had contacted mayor Rahm Emanuel about helping Uber with regulations. Speaking of Emanuel, Deanne Millison worked directly for him while Chicago mayor.

Another connection: the Chan Zuckerberg Initiative, with which both Plouffe and campaign manager Jen O’Malley Dillon have worked for or with. Dillon advised them while working at Precision Strategies (see (C) above), along with Gates Ventures (a VC firm founded by Bill Gates), GE, and Lyft.

The influence of big capital was reported by New York Times (Oct 14 2024):

Over two days, the pair held meetings across Wall Street, including at the offices of Lazard, an investment bank, and the elite law firm Paul, Weiss. Among the ideas the attendees pitched was to provide more lucrative tax breaks for companies that allowed their workers to become part owners, according to two people at the meetings. The campaign had already been discussing such an idea with an executive at KKR, the private equity firm.

A few days later, Ms. Harris endorsed the idea during her speech in Pittsburgh. “We will reform our tax laws to make it easier for businesses to let workers share in their company’s success,” she said.

The line, while just a piece of a much broader speech, was emblematic of Ms. Harris’s approach to economic policy since she took the helm of the Democratic Party in July. As part of a bid to cut into former President Donald J. Trump’s polling lead on the economy, her campaign has carefully courted business leaders, organizing a steady stream of meetings and calls in which corporate executives and donors offer their thoughts on tax policy, financial regulation and other issues.

The private feedback has, in sometimes subtle ways, shaped Ms. Harris’s economic agenda over the course of her accelerated campaign. At several points, she has sprinkled language into broader speeches that business executives say reflects their views. And, in at least one instance, Ms. Harris made a specific policy commitment — to pare back a tax increase on capital gains — after extended talks with her corporate allies.

They describe a Democratic campaign that is far more open to corporate input than the one President Biden had led for much of the election cycle. Ms. Harris’s team does not take all of the advice it is given. Some pleas for a more robust endorsement of big business have been ignored, for example, while some of the meetings have devolved into executives’ raising specific tax problems their companies face.

The friendlier corporate ties have nevertheless raised questions about whether Ms. Harris’s overtures are campaign-season coalition building — or a sign that she will take a more centrist tack if she wins the White House. On some issues, like a proposed tax on the ultrawealthy, her campaign has been studiously ambiguous, fueling uncertainty about what kind of an economic leader she would be.

...

Charles Kretchmer Lutvak, a spokesman for the Harris campaign, said that if she won, Ms. Harris would continue to listen to the business community as she pursued a “bold, capitalist vision” for the economy. “As president, Kamala Harris will continue to be open and pragmatic about how to deliver that vision for the American people,” he said.

In other words,

Here’s our golden boy, Mark Cuban:

[Mark Cuban:] “She says she’s open to inputs from independents and Republicans, she means it. She truly is open minded. I’ve put some wild things in her direction that they don’t laugh at,” he said. “People are trying to say, ‘Here are the progressive and liberal principles that have always been the principles of the Democratic Party.’ Those are gone. It’s Kamala Harris’s party now.”

This is no fluke. Here’s Mark Cuban’s remarks, in an interview with Astead Herndon on Oct 1 2024:

(1) Astead Herndon: So should she be saying that? Because she hasn’t really communicated more directly, hey, I’ve learned and I have changed from that thing. I mean, she’s tried to make a everything-for-everyone case. We talked to, actually, Bernie Sanders on this exact point, and he’s someone who’s making the argument that the Kamala Harris administration would be one that has a robust populism that progressives would have a voice in. And so I guess I’m asking, how can a Mark Cuban economy and a Bernie Sanders economy both exist under the same — It seems like both sides are pitching to me that the Kamala Harris administration would be one of them.

Mark Cuban: I hear you loud and clear.

Astead Herndon: Yeah.

Mark Cuban: But one was a month ago, and you haven’t heard about it since.

Astead Herndon: That’s fair. You’re saying she’s tacking towards you closer than she’s tacking towards them.

Mark Cuban: Right. I mean, she’s giving economy speech after economy speech. She’s talking about the things because if the number one issue is economy and the way to solve people and make people feel good about what she’s doing is to talk business and talk economy and talk opportunity, and that’s what she’s talked about. And I think you have to be in the middle to do that.

Astead Herndon: I hear that. You endorsed Nikki Haley in the Republican primary. And one question I wanted to ask, and it’s to get at the idea of what’s driving your support. But if she was the Republican nominee, would you still supporting Harris in this way? I asked to say like, is this about the affirmative advocacy for Harris and her policies or is this about rejecting Trump and his policies?

Mark Cuban: So if it was Nikki Haley versus Joe Biden, I probably would have voted for Nikki Haley.

Astead Herndon: OK.

Mark Cuban: Given that what I expect her policies would have been. And I’m socially liberal, but conservative fiscally. And so it depends on abortion and those things because that’s important to me. But with Vice President Harris, I think she’s truly on point with our mission. So this is an anti-Trump vote for me. I really like where she’s going. I mean, she’s very thorough. She’s very precise. She’s very well thought out. She takes an idea, she evaluates it. She may agree or may not agree with you, but you know it’s going to be completely and thoroughly thought through.

In short: (1) while progressives and big capital both claim that there’s room for their economic policy in a Harris campaign, but "one [the progressive argument] was a month ago, and you haven’t heard about it since". And (2) this ringing endorsement is coming from someone who said they would support Nikki Haley against Joe Biden! Throughout the interview, you can easily detect his malice towards Biden’s policies.

Perhaps the big bright spot - for working Americans and progressives at least - is the Tim Walz pick. How did this happen? It appears that, somehow, Martie Walsh and Harris became friends while Harris served as Senator, from 2017-2021. Recall that Walsh was favored by Sanders as Biden’s Labor Secretary pick. And he apparently helped vet running mate candidates, which perhaps explains how Tim Walz ever became the VP pick.

You’d think the big capital blob that gobbled up the Harris campaign would hate him. Yet in early August, Cuban praised the Tim Walz pick publicly. So what gives? In short, Cuban didn’t think he mattered - consider that Harris is young and healthy, after all. In this interview, you can tell how dismissive he is of Walz:

Some choice quotes: "Well, yeah. I mean, yeah, there’s the heartbeat away thing. But beyond that, they’re already picked. We can’t change it. There’s nothing we can do there". And "This is Harris versus Trump. That’s the whole election." And "I think if I have a hope for Tim, it’s that he just comes across as a good, respectable guy and he doesn’t take the bait for anything". In other words, while Cuban can’t do anything about the fact that Walz is the pick, he doesn’t think he’s important in the election. And if he does what’s right, he’ll just be a "good, respectable guy" - that’s what he means by "if I have a hope for Tim".

As Herndon summarizes:

(6) Astead Herndon: So like I said to Mark Cuban, one of my big takeaways from our conversation is somewhat related to this open question of defining Kamala Harris and what she believes in. He thinks her ideology is clear, that Harris is no left winger, and that particularly when it comes to the economy, she’d be more open to business and billionaires than the Biden administration ever was, which is the exact opposite of what I heard from progressive leaders who cited things like Harris’s selection of running mate Tim Walz as a sign that the left would have a voice in a Harris-led Democratic Party.

Tonight, Walz and Republican vice presidential candidate JD Vance will face off in what could be the final debate of the 2024 cycle. We’ll be watching from Tim Walz’s former congressional district in Southern Minnesota. But we’ll also be listening to see which version of Kamala Harris does Tim Walz present, the progressive ally, as described by Bernie Sanders, or the pro-business moderate, as described by Mark Cuban.

Perhaps Cuban was right to be dismissive - in that debate, Walz showed little signs of the progressive background which excited the left-wing of the Democratic Party. He had been completely subordinated.

Exit polls give two crucial pieces of data: (1) of course, is the Dem:GOP ratio reported by responding voters. But (2) is percent sample; circled in red below:

That means that, if the exit poll is any indication, that 23% of national 2024 voters were 30-44 years old. Now that might not actually be the case, but that’s what the exit poll would indicate, if it were a truly random sample of the 2024 voter population.

How can we know if this is accurate or not? Luckily, the Census Office studies this.

So if we look at 2020 - a year in which turnout was 65.3% - we can see that the NEP+ER poll indicates that 35% of voters were <$50k, 39% were $50k-$99.999k, and 26% were $100k+. What was the actual turnout?

The Census Office reports total citizens and total voters in 2020 for various income brackets for 180m people (and 165m citizens); see here, Table 7 - I’ll refer to this data as T7D20. If we assemble this data into our three icnome groups of interest (<$50k, $50k-$99.999k, $100k+), we find that the proportion of citizens in each group is 27.8%, 33.4%, and 38.7% respectively. In terms of raw data, it’s {35406e3, 42522e3, 49292e3}/127220e3. Further, the turnouts for the three income brackets would be 59.67%, 74.27%, and 82.80%. Overall, the Census survey here finds voter turnout at 68.6%. Since the best known 2020 turnout is proposed at 65.3%, we should rescale these turnouts by (65.3/68.6) for the three brackets: 56.80%, 70.69%, and 78.82%. Call this turnout set T20_S1.

Three things to notice. (1) T7D20 doesn’t cover everybody in the country, not even all 18+. But it’s still 180m people, so it’s a very good sample size, to say the least. (2) Notice in T7D20, these are proportions for individuals in households of such an income bracket (PIHIB), not proportion of households as such in each income bracket (PHIB). (3) Significantly, there’s also a large group for "Income not reported" (INR) (people who refused to answer, more or less) - 37664e3 citizens and 19527e3 voters.

What should the actual PIHIB be, for eligible voters?

In Census data for 2020, we get counts of individual males and females in households of a certain income bracket, by age - I’ll refer to this as PIHIB_D20. If we sum up for individuals 18+, we get that the PIHIB_D20 gives the PIHIB proportions as 29.8%, 29.5%, and 40.7% (for 2016: 34.84%, 30.62%, 32.4%; for 2018: 32.32%, 30.96%, 36.21%, 2022: 27.71%, 29.26%, 45.15%).

It’s safe to assume that the INR group in T7D20 have an income. So we can add these INR people into each T7D20 bracket in quantities to realize the PIHIB_D20 PIHIB distribution, and add a proportional amount of INR voters. Thus, although we’ll miss the income specific To get these proportions. This gives the total citizens for each bracket as 46906e3, 46322e3, and 63792e3, and total voters as 27090e3, 33550e3, and 48334e3, with resultant turnout at 57.75%, 72.43%, and 75.77%. And to adjust the turnout to estimated 2020 turnout (65.3%), we get 54.98%, 68.94%, and 72.12%. Call this turnout set T20_S2.

What would be the correct proportion samples (PS) to yield this turnout? Our equation is (PS/PP)*Tᵢ= Tⱼ, where PP is "percent of population", Tᵢ is turnout that year, and Tⱼ is the turnout for one of the three groups. If we use T20_S1, solving for PS, we get 25.95%, 31.96%, and 49.07%. This adds up to a bit more than 100 (106.98), so we normalize this to: 24.26%, 29.87%, and 45.87%. if we use T20_S2, solving for PS, we get 25.11%, 31.17%, and 44.90%. This adds up to a little bit more than 100 (101.18), so if we normalize this, we get 24.82%, 30.81%, and 44.38%.

What were the proportion samples for these income brackets from the exit polls NEP+ER (NEEP) and Fox+NORC+AP (FNAEP) in 2020? They were 35%, 39%, and 26% for NEEP, and 38%, 36%, and 26% for FNAEP. The sample proportions for the above are quite close (RMSD = 1.07 pp). The RMSD for NEEP is 13.02 pp, and for FNAEP is 13.40 pp. That is to say, the "typical" sample proportion deviation of both polls was about 13 pp from the actual value - far beyond the possible error given the sample size (which should be at most 3-5 pp).

Clearly then, although the exit polls appear to predict the election results well, there are deep flaws in the sampling. This would seem to imply the actual Dem:GOP ratios are off. However, if we plug in the income bracket Dem:GOP ratios in with the sample sizes to compute the total vote for Biden and Trump, using the equation:

TV{D,R}ⱼ = (PSⱼ*wT)*Tᵢ*PV{D,R}ⱼ

Where TV{D,R}ⱼ is the total votes from a bracket for Democrat (D) or Republican (R), wT is the total eligible population, and PV{D,R}ⱼ is the vote for either D or R specific to the income bracket, from the exit poll (see Data and Methods). If we apply this equation to both exit polls, for T20_S1, we get 78.66m/76.05m for NEEP, and 81.76m/74.66m for FNAEP. For T20_S2, we get 79.00m Biden votes and 75.79m Trump votes for NEEP, and for FNAEP we get 81.74m and 74.68m. Originally, the two exit polls suggest totals of 83.04m/72.61m and 81.90m/74.42m. The actual totals for 2020 were 81.28m and 74.22m. In addition, it appears that NEEP predicts from 2016 → 2020 the qualitative sample size changes for $50k+, but gets it wrong for <$50k (although overall, gets turnout increasing for only <$100k, and erroneously gets turnout falling for the $100k+).

What does this mean? It appears that the FNAEP gets very close to the real result for both sample sizes. But this doesn’t actually mean it’s reported values are correct: one reason for it’s robustness is that gains in the bottom group, at expense of upper group, are buffered, since both groups have positive ratios for Biden (54: 4 and 52: 47). Meanwhile, the middle group has a balanced ratio (49: 49), so losses or gains here cost/benefit either candidate equally. Yet a variety of PV{D,R}ⱼ values could yield these results. Meanwhile, the NEEP poll appears to fluctuate a lot with different sample sizes, but this is more because the Dem:GOP ratios are very asymmetric (55:44, 57:42, 42:54), so low/mid-end group gains/losses and high-end group gains/losses more sensitively affect vote totals.

Clearly these NEEP PV{D,R}ᵢ are not the actual vote ratios for the three income groups, but that doesn’t mean the FNAEP results are "correct" either. For example, if we put in the PS values of 0.26%, 09.87%, and 89.87% into the FNAEP PV{D,R}ⱼ, we get vote totals of 81.95m/74.79m. This appears close to correct, but clearly the PS values are highly unrealistic. In other words, these values - while they of course could be correct - are such that any sample size would give the right values.

How about their turnouts? While the Census doesn’t (to my knowledge) report how people voted, it does allow us to get a sense of turnout. This is one observable we can directly compare then, to measure how much the exit polls mismatch with reality. And further, if we characterize this mismatch, to see how likely (at least roughly so - a better likelihood estimation could be made than the one I will make, but again, I’m too lazy; just want to get a sense of things here) the predicted scenario in this article is (<$100k turnout falls, $100k+ turnout rises).

It appears that the likelihood that the <$50k turnout fell is 80.4% (NEEP: 100%; FNAEP: 60.8%); the likelihood that the $50k-$99.999k fell is 20.6% (NEEP: 34.8%, FNAEP: 6.5%); the likelihood that the $100k+ increased turnout is 7.1% (NEEP: 14.3%, FNAEP: 0%). Overall then, the likelihood of falling <$100k turnout, rising $100k+ turnout is 1.2% (NEEP: 5.0%, FNAEP: 0%). Not very likely! But also, something weird is going on - the amount of individuals in $100k+ households is growing rapidly, which is leading to these likelihoods. Is this an issue?

That has to be balanced in consideration that the share of the population in households that make $100k+ has increased steadily since 2016 - from 34% in 2016 to 48.7% in 2023. Part of this is inflation - $1.00 in 2016 is equivalent to $1.27 in 2023. Another factor is an increasing size of $100k+ households. According to the Census (file: here, Table A-2), 36.4% of households were $100k+ in 2016. Going off the raw numbers from the PIHIB data, we find in 2016, there were 1.91 people 18+ living in <$50k households, 2.16 people 18+ living in $50k-$99.999k households, and 1.82 18+ people 18+ living in $100k+ households. But in 2023, these values were 1.51, 1.92, and 2.34 respectively - a 28% increase in 18+ year olds in $100k+ households, while the adult size of $50k-$99.999k households fell 11.1%, and of <$50k households fell 21%. It seems that increasing adult members/$100k+ household, along with inflation, offer a meaningful explanation of the rise in adults in $100k+ households.

But if that’s the case, the meaning of "$100k+ household" starts to break down, as a category of upper/middle-class. For example, if a household makes $150k a year, but this should support 2.34 people, then the "average individual income" is $64.1k a year, which isn’t that high. For reference, non-family housheold median income (which would be closer to individual median income) is about $50k. So if the point of the "falling <$100k turnout, rising $100k+ turnout" is to gauge working class antipathy, and upper class engagement, then it would seem these household income brackets aren’t divided at useful lines (ie perhaps <$75k, $75k-$149.999k, $150k+ would be more meaningful).

Figure M: Plot of average people per household’s of a given income bracket (see table below).

Either something very interesting is going on, or there is a discrepancy between the two Census files.

What does a finer grain analysis reveal? If we look at overall PIHIB distributions (in 41 different household income brackets), they change as so:

Figurex 1

Previous Next

Figure xx.1: distribution of individuals, based on the income of the household they are in (ie PIHIB). The red vertical line indicates the median income. The black vertical lines indicates the H{X} bracket demarcation lines ($30k, $50k, $100k, $200k).

This seems to indicate that the number of people in $200k+ households has grown from to around 8% to 17.5% in 8 years! But looking at figure M, these households hold increasingly more persons each.

NEEP (but not FNAEP) reports results for subgroups of two brackets: for <$50k ({<$30k,$30k-$49.999k}) and for $100k+ ({$100k-$199.999k,$200k+}). Their distributions over these years (2016 - 2023) are reported in the table below, and the percent sampled by NEEP in the table below that. However, while the Census data allows (with some "smoothing") tracking of behavior for the two <$50k subgroups, it does not for the two $100k+ subgroups. So there isn’t much more to do here, except to notice sampling patterns within those subgroups (FNAEP give subgroups <$100k, but clearly the $100k+ group is of interest for this sampling problem and changing PIHIB distributions, so I have ignored those finer FNAEP samplings for now).

What if we use the PHIB distributions instead (ie distributions of household incomes, rather than apparent distribution of individuals within housecomes of particular incomes (PIHIB))? That is to say, if we set aside the interesting PIHIB trends (perhaps treating as a result of two different methodologies (since the finer-grained Census reporting doesn’t seem to indicate much increase of non-householder members in households; I can report this data, but for now, too lazy)), and assume the PHIB distributions reflect the actual income trends of the country.

Now the likelihood of lower <$50k turnout is 100%, lower $50k-$99.999k turnout is 53.9% (NEEP: 63.2%, FNAEP: 44.6%), and the likelihood of higher $100k+ turnout is 82.3% (NEEP: 100%, FNAEP: 64.6%) (scenario crude likelihood (not accounting for fact that sample sizes of each group must add to 100, just multiplying these probabilities): 44.4% overall, 63.2% for NEEP, and 36.7% for FNAEP). The likelihood of lower <$50k turnout and increased $100k+ turnout (and being agnostic to $50k-$99.999k turnout) is 82.3% (NEEP: 100%, FNAEP: 64.6%). With these likelihoods, then this article has a good shot of representing what happened, using the exit polls. I guess we won’t know til the Census report comes out for this election.

How about national polls? Are there issues there?

The closest poll in 2020 (and 2024) was from AtlasIntel (file here for 2020, Oct 26-28 2020). They give a total vote proportion of 50.9:46.2 (Biden:Trump); in reality, it was 51.3:46.8. So quite close, right? But the income brackets sampled were 38.8%, 31.2%, and 30.1% (RMSD with T20_S2: 11.54 pp). For the PV{D,R}ⱼ values, they give (60:36), (46:52), (44:54). Given the T20_S2 PS values, these would yield vote totals of 77.00m/77.53m. In the AtlasIntel 2020 poll, Trump and Biden represented 97.98% of voting respondents, indicating - with the above values - a total vote of 157.72m votes, and a Biden:Trump vote ratio of the total of 48.8: 49.2. Clearly not so accurate after all!

In other words, the AtlasIntel poll was correct on accident.

Age

By age (18-29, 30-44, 45-64, 65+), in 2020, the correct sample would have been 16.46%, 23.16%, 34.69%, and 25.69%. The NEEP poll gives 17%, 23%, 38%, and 22% (RMSD: 2.49). For FNAEP it was 16%, 25%, 32%, 27% (RMSD: 1.77). The AtlasIntel pre-election poll was 17.8%, 25.7%, 41.7%, 14.8% (RMSD: 6.63 pp).

Race and Gender

If we look at race, the proper samples based on the Census should be 74.04%, 12.93%, 9.12% and 3.9% for 2016 and 72.03% (-2.01 pp), 12.42% (-0.51), 10.74% (+1.62 pp), and 4.81% (+0.91 pp) in 2016.

The NEEP gives samples of 71, 12, 11, 4 (RMSD: 1.85) → 67, 13, 13, 4 (RMSD: 2.80). It get’s the white and Latino trend correct, but is flawed for the black and Asian trend. The FNAEP has (for white:black: Latino) 74: 11: 9 for 2020 (RMSD: 1.72), which rather than over-estimate non-whites, tends to under-estimate them (and doesn’t give specifics outside of this grouping). The AtlasIntel 2020 poll gives race composition of 75.9: 10.3: 9.9: 2.6 (RMSD: 3.83) (1.3 Other), which is quite far off as well.

In terms of gender, the expected samples - given the Census-reported turnout - would be male:female 47.08%: 52.92% (2016) and 47.57%: 52.43% (2020). The NEEP poll gives 47: 53 (2016) (RMSD: 0.08) and 49: 51 (2020) (RMSD: 1.43). The 2020 AtlasIntel poll gives a ratio of 48.2: 51.8 (RMSD: 0.63).

So What Does this Mean?

It’s not my opinion that the exit pollsters are a bunch of liars. Instead, I think the sorts of people who answer exit polls - call these PWAEE - aren’t representative of the electorate at large - call this EAL.

Does this mean exit polls are garbage?

Well... kind of. I wouldn’t take the absolute Dem:GOP ratios in the exit polls at face value. However, assuming that trends in EAL are reflected in the PWAEE (ie if turnout of <$50k falls in the EAL, so will <$50k PWAEE turnout will fall; if Dems or GOP becomes more favorable among $100k+ EAL, they will also become more favorable among PWAEE), we can at least discern voting trends by comparing results in year X and X+1.

Looking at 2024 pre-election polling

The 2024 pre-election polling is full of this problem. Take this NBC poll: if you look at the detailed document on page 24, 35% of respondents described themselves as poor or working class. 42% call themselves "middle class" (and not "upper middle class"). Assuming this is largely, if not entirely, <$100k people, then that’s 77% of respondants - but only around 60% should be in this category. Clearly then, the $100k+ are under-represented. And thus the poll results, no matter how close or far they are, aren’t a good indicator of actual nationwide sentiment. Take this AtlasIntel poll from Nov 3-4, 2024 - on the eve of the election. It appears to call the election nearly right, giving a 50.0 vs 48.8 Trump:Harris vote share in a head-to-head race (actual: 50.0: 48.4). But only 27.9% of respondents (of 2703) were $100k+ - likely at least 17.1 pp too low. Below $50k was at 32.1% - not too far from a good sample - but $50k-$99.999k was 40.0% of the sample, about 10 pp too high. Given that the <$50k was close to the expected sample, perhaps its Dem:GOP ratio (48.9:48.5 Trump:Harris) is close-ish to reality. But who knows?

Here’s an interesting case. This Times/Sienna College poll (Oct 8 2024) - 2723 respondents reached by land-line and cell phone - got the overall Dem:GOP ratio wrong (49:46, vs 48.4:50.0), but the income sampling is actually pretty good: 24% <$50k, 28% $50k-$99.9k, and 40% $100k+ (7% refused to say). Interestingly, all income categories except $200k+ and "Refused" favored Trump, yet they still find Harris with the lead (they report income brackets in finer groups than the three here). I’m not sure how that works out. But also, the values for an income group don’t quite add up (ie for <$25k, Harris got 8%, Trump 8%, Other 15%; that adds up to 31%, not 100%). I’m not sure what’s going on here. What’s more odd: the sum for things like gender (<$25k: Men: 8%, Women: 9%; total=17%) don’t match with the sums for Dem:GOP:Other (ie <$25k: total=31%). I’m at a loss for what’s going on here, or why so many people reported "Other" for their vote. It seems that the reason the numbers don’t add up in an obvious way, is that the % reported (ie <$25k Harris voters = 8%) is supposed to sum to 100% across income groups. That is, to show where each candidate is getting the bulk of their votes. Therefore, each value should be multipled by 0.49 (if for Dem), by 0.46 (if for GOP), and by 0.05 (if for Other) (ie <$25k contributes 8%*0.49 = 3.92% to Harris’s total vote ratio, and 8%*0.46 = 3.68% to Trump’s total vote ratio, and 15%*0.05 = 0.75% to "Other’s" total vote ratio, with the sample size of <$25k already built in). This seems a weird approach, but I guess I see the logic. It seems then they are presenting the results backwards, to get the values across income groups to sum to 100%. Maybe this is normal though in cross-tabs, idk.

If I ignore that, and I normalize these values (ie <$25k’s 8:8:15 → 25.8:25.8:48.4), and multiply by 𝑛=2723 and each group’s "likely electorate" percentage (ie <$25k are 8% of the likely electorate here), I get 908 Harris voters, 922 Trump voters, and 866 Other voters. I’m not sure what to make of this enormous "Other" block (it’s significant across all income groups). The vote fractions for Dem:GOP:Other are 33.7%:34.2%:32.1%. If we take the ratio of Dem and GOP votes over Dem+GOP votes (adding up to 1830) (ie what seems a likely "total electorate"), we get Dem:GOP of 49.6%:50.4%. Okay, now we’re cooking.

Looking at the 2024 election results, we can see that about 1.7% of voters voted Third Party - in other words, 98.3% voted for one of the two big parties. Thus, a "proper electorate" would be 1830/0.983 = 1862 voters. Using this as our denominator for Dem and GOP votes (ie Dem vote fraction = 908/1862), we get a GOP:Dem:Third = 48.8%:49.5%:1.7%. This is actually pretty close to the final result (48.4%:50.0%), all things considered. This seems just a fluke though, since these vote ratios are backwards determined from the original 49%:46%:5%.

What are the historical trends for income group voter turnout? Below, I lay them out from 1992. Note that inflation over time means the <$50k, $50k-$99.999k, and $100k+ brackets have changing meanings. But I use them for a rough comparison over time.

A few things can be observed. First, it seems the argument of this article - a sharp drop in <$100k (or more conservatively, <$50k) turnout, along with a rise in $100k+ turnout (or more conservatively, a significantly less sharp fall) - has been observed only two times: the 1998 → 2002 midterm elections, and somewhat the 2006 → 2010 elections (though this seems more a result of $50k-$99.999k behavior).

I can’t seem to find appropriate exit polls for these elections though; however, we can look at the results of the election to get some clue. 1998 → 2002: Te 1998 elections were more favorable to Democrats in Congress, whereas the 2002 elections were more favorable to Republicans in Congress, although 2002 also saw a net Democrat win for governor races. At least in terms of Congress then, it was bad for Democrats. 2006 → 2010: the 2006 midterm saw large Democrat gains in Congress, and for governor races. The 2010 elections were the opposite, with huge Republican gains in Congress and for governor races. So, the trend appears to be when lower-income turnout falls, and higher-income turnout rises, its bad for Democrats and good for Republicans.

Notably, the opposite scenario (rising lower-income turnout, sinking upper-income turnout) has been observed apparently in the 2004 → 2008 presidential election (Obama’s big opening campaign) (<$50k vote trend Dem:GOP:other 2004 → 2008: 54.6:44.4:1 → 59.5:38.1:2.4, Δ = +4.9:-6.3:+1.4 (turnout adjusted: 28.4:23.1:0.5:47.9 → 31.5:20.2:1.3:47.1, Δ = +3.1:-2.9:+0.8:-0.8), and somewhat the 2012 → 2016 presidential elections (based on results here, it appears that the <$50k vote shifted in those years from Dem:GOP:other 60:38:2 → 53:41:6; Δ -7:+3:+4 (turnout adjusted: 29.3:18.6:1.0:51.1 → 26.1:20.2:3.0:50.7, Δ = -3.2:+1.6:+2.0:-0.4).

That is, when <$50k were relatively more mobilized, in 2008, they broke big for Democrats. In 2016, Democrats lost many voters here, some to Trump, but it appears equally, or likely more, to third party ("equally" if all the new voters from increased turnout just went to third party; this seems unlikely to me). This isn’t unanimously the complete opposite of the lower-income turnout fall, higher-income turnout rise scenario. However, in Obama’s case - with the promises of "hope and change" and healthcare amidst the financial crisis, clearly this is significant (and also significant that in 2016, the <$50k broke with Dems towards Trump as much, and likely more, as third parties.

Further, there’s three local maxima for <$50k turnout (a "local maxima" being a value higher than the values before or after it, but not necessarily the largest value in the set): 1992, 2008, and 2020. The values for 1992 are likely not quite for the same type of low-income group as <$50k represents today (for example, the Census used to report election-income data in much finer degrees for $0-$50k than today). If we just look at results from 2000, however, notice that (1) 2020 is the maxima of <$50k turnout, (2) 2008 - Obama’s campaign - is the second highest, and (3) 2016 is only the 4th highest of 6, barely recovering from the turnout fall from 2008 → 2012. In terms of the midterm <$50k slumps in 2002 and 2010, these were the 3rd and 2nd lowest <$50k midterm turnouts, with 2014 at #1 (which was also the lowest turnout point for other income groups).

With this in mind, 2008 appears to be the antithesis of the 2024 election (assuming the argument of this article is correct).

Notably, while many rightfully draw comparisons between Harris losing the popular vote, and Kerry losing the popular vote in 2004, these appear to be for very different reasons. 2004 saw rising turnout among both the <$50k and $50k+, although the $50k+ turnout rose slightly more. Whatever the reasons for his loss, it probably wasn’t former Democrat voters abandoning him to no-vote, very different from the apparent reason for Harris’s loss.

It’s notable, however, that 2024 turnout was the second highest in the time interval analyzed here (1992-2024) - compare this with 2004, at 60.9%. In fact, it appears that we have to go all the way back to 1968 (62.5%), 1964 (62.8%), 1960 (63.8% - with JFK!) to find comparable turnout - 62.5%, per Wikipedia (although given the trends from the University of Florida, it was likely a bit lower than this). In fact, looking back over time, this is the 4th highest turnout since (and including) 1908 - before that, turnout was much much higher (around 75%) than since (around 55%).

No, the Muslim/Arab/Third Party Vote Didn’t lose Harris a Single Swing State - More Garbage Data Interpretation

(Back to Table of Contents)

Along with Latinos, chauvinistic Harris supporters have also unleashed disgusting bile on Arab/Muslim voters (and the Third Parties they voted for). In fact, if we look at key swing states, we will find no example of a Green party spoiler, and two examples of a total third-party spoiler, which wouldn’t have changed the election anyways (following data from AP, states at 99%):

Total 2024 Trump vote lead in swing states: 763027

Biden 2020 lead: 236774 (NC: -74483, GA=+11779, AZ=+10457, WI=+20682, PA=+80555, NV=+33596, MI=+154188)

Only in Michigan and Wisconsin do we see total third party (and not just Greens) add up to the difference - which of course is to bulldoze over the vast ideological differences among the third parties. If we just look at leftist third parties (Green, de la Cruz (PSL), West (Justice for All), Kishore (Independent)) - typically considered, usually falsely, to be Democrat-voters-gone-rogue - we see no state changed by their vote. But even with these two states "spoiled" by the total third party, with a total of 25 EC votes, Harris would only be at 251 total EC votes. So, still not winning.

Still, despite Michigan not being the election pivot, people still like to get angry over something that didn’t lose them the election.

Nitpickers (ie people who militantly refuse to criticize the Harris campaign at all for losing the worst for Democrats since 2004, and delivering MAGA down-ballot wins across the country, such that they secure the House and Senate) have observed that total Arab/Muslim votes in Michigan for Trump and third parties surpass the difference there. For example, this Conversation article, the author analyzes voters in Dearborn, Dearborn Heights, and Hamtramck, observing Trump won 50332 votes. Notice with the Green Party votes, that seems to make the difference (50332+44697 = 95029 > 80618). What this misses (assuming that the Green Party vote total is largely from Muslim/Arabs, which I will for the sake of the argument) is that some in these areas were already Trump voters. If the argument is "angered by the Gaza Genocide, Arab/Muslim voters switched from the Democratic Party", then the already-Trump voters have to be left out of consideration, since they probably weren’t going to vote Democratic, regardless of anything in Gaza (except maybe switching to Democratic, had the Biden administration taken the humane and anti-genocidal course there).

For example, if we look at 2020 results in Dearborn MI, we see Biden won 30.7k votes, and Trump 13.2k votes. In 2024, Harris won 15k votes here, and Trump 18k votes. Thus, Trump "only" gained 4.8k votes (ie 27% of his votes were "new"). Since Dearborn’s 18k Trump votes are about 36% of those analyzed by the Conversation article, let’s take those as indicative of the overall pattern. That is, that only 27% of those 50332 votes were new, or 13590 new Trump votes. With the Green Party’s statewide 44697 votes (again, leaving aside if these are all Muslim/Arab, and ignoring that there were already Green voters before, and ignoring that some of those voters may have never voted before for anyone, given the US’s relatively low turnout rate), that’s at most 13590+44697=58287 votes of Arab/Muslim Michiganders who switched from Democratic to Trump, insufficient to overcome the 67028 threshold (since if Trump lost those 13590 voters, the difference would go down to 67028 votes). Although if we add in all of West’s votes (6721), and all of Kishore’s votes (2438) (totalling 9159), we just barely clear the threshold, 67446, by 418 votes.

What we’ve found then, is that if every single West, Kishore, and Stein voter were ex-Democrat Muslim/Arab voters, that along with the estimated 13.2k new Trump votes, they lost Harris Michigan by 418 votes.

So, did Muslim/Arabs lose Harris Michigan?

No.

Remember, we’re assuming - for the sake of the argument - that all the 3rd party votes come from Arabs/Muslims. This is obviously absurd. And only in this absurd scenario do we "conclude" that Muslim/Arab voters lost Harris Michigan, and just barely (418 voters would have been 0.007% of the total 5,638,701 MI votes).

(1) This ignores the fact that the Green’s have something of a voter base in the state already. The Green’s 2016 Stein-Baraka ticket got 51463 MI votes. From the city of Dearborn, the Green’s got 302 (straight party) + 817 (Stein-Baraka) = 1119, or 2.2% of MI Green votes. In 2020, the Green’s Hawkins-Walker ticket got 13718 votes (the Stein-Baraka votes were 2.06% of the presidential vote there, of a total 39614). From the city of Dearborn, the Green’s got 137 (straight party) + 208 (Hawkins-Walker) = 345 from city of Dearborn, or 2.5% of MI Green votes (the 208 Hawkins-Walker votes were 0.5% of the presidential vote there, of a total 44997). Already then, it’s apparent there were already many Green voters in the state besides Arabs/Muslims. (data for Dearborn Green vote in 2016 and 2020 from the city of Dearborn).

(2) We have an idea of the share of Green voter increase in Dearborn, and it’s dramatic, but not decisive for the state. Due to Harris’s support of the Gaza Genocide, Green’s share in the city of Dearborn went up to 22% of voters, according to Newsweek. Considering they already had about 2.2%-2.5% of voters, that’s a new 19.5%-19.8%. But let’s assume they’re all new voters, just for the sake of the argument.

If we take the range of total votes in Dearborn to be roughly in the range 39614-44997 (from 2016 and 2020, since it’s unclear so far, to me, what the turnout was here in 2024), that’s a possible 8715-9899 Green votes, or 19.5%-22.1% of the Green vote in the state. If we again assume Dearborn is about 36% of the Dearborn+Dearborn Heights+Hamtramck voters, that’s 24208-27497 votes.

So, with Trump’s new 13590 votes, the 27497 Green votes (taking the upper end), the West+Kishore 9159 votes (again, making the absurd assumption these are all from Muslim/Arab voters), we’re at 50246 voters - still below the 67028 threshold.

In other words, the only way we can properly "blame" Muslim/Arab voters for losing Michigan, is if we assume every single new left third party vote was from Muslim/Arabs. This is obviously not the case (ie we know that Stein got 22% of Dearborn votes, not the 36.0%-40.6% that would imply all of Stein’s votes were from Dearborn+Dearborn Heights+Hamtramck (since if we say Dearborn is about 36% of the voters here, then Stein would need to have had 44697*0.36 = 16091 votes in Dearborn; given our total vote range here of 39614-44997 (assuming comparable turnout to 2016 or 2020), that would give her 16091/{39614/44997} = 40.6%-36.0%).

So no, enraged Arab/Muslim voters didn’t lose Harris the election, unless we’re counting voters in Dearborn+Dearborn Heights+Hamtramck who were already voting Trump. But let’s take this extreme example further (And even if they did, that wouldn’t have won Harris the election, nor reversed her huge popular vote loss!

Finally, this isn’t to say Arab/Muslim voters should have voted Harris. She is/was vice president of an administration arming and supporting Israel’s genocide in Gaza. To boot, she continued to deny the genocide throughout the campaign, and blocked out Arab/Muslim speakers from events such as the Democratic National Convention. It is absurd to have expected them to vote "blue no matter who".

If you look at the raw exit poll Dem:GOP ratio, it looks as if the unmarried men group has shifted decisively for Trump, in terms of race, slightly for black voters, and much more for Latino voters. Why are they are all voting for him suddenly????

This is yet another mirage: it’s mostly former Democrat voters not coming to the polls, not switching to Trump (although some switching happened it appears) - but you need to consider turnout to get an idea of that. As it goes for most issues this election. And notably, every group besides married women seems to have swung to Trump. For married men though, this is a decline in turnout. For unmarried women, this is an actual gain. But that seems less interesting to yap about I guess.

For this, I used the Historical Marital Status Tables from the Census Office (file: Table MS-1), to compute {single/married} × {men/women}, nationally, and across {white,black,Latino}. Next, it seems I’d have to account for varying voter eligibility among gender (ie national: 49:51, black: 47:53). So first I checked what the actual male:female ratios were, and found something disturbing: those ratios are largely based on the actual voter population (I had assumed black eligiblity was a result of some diffferential incarceration rate among black men and women). That is, it seems there are actually about 47 men for every 53 black women (actually, the ratio appears slightly worse). The Census category for "Black" in this document includes mixed race people, so this doesn’t seem to be a clerical error of a man is "black only" or "black and mixed". This seems to imply a horrifying differential mortality rate among black men (unless there is a gigantic preferential abortion of black male babies, which seems unlikely to me, given the scale here).

That said, I won’t adjust the male:female distributions to account for gender differential voter eligiblity ratios, as it seems mostly a result of there being slightly more women than men in population groups.

The Census table gives total people in each category {un/married} × {gender}, for people 15 years and older. I compute a ratio from this for each group, and then apply this to the eligible voter population of the respective group. This assumes the ratio of the four groups {un/married} × {gender} are roughly similar in the whole population and the voter eligible population, which seems like a safe enough assumption to me (although ofc, there’s probably some considerations to make the analysis a little bit better).

National

White

Black

Latino

Feel free to peruse. Ofc, if we’re interested in the "incel" group, this would pop up among the unmarried men (UMM). Not every UMM is single, and not every single man is an incel - probably most aren’t. But that’s where the incels, by and large, will be. Ofc, there might be some married men (MM) incels ("incel" means a lot more than not getting laid, to be clear), but they probably are under-represented there).

Compared to 2016, we do see UMM gains for Trump, even more than for Harris - but these are in accord with gains across MM, unmarried women (UMW), and kinda also married women (MW). The latter is interesting as the only group that Harris sees a net loss in votes (mostly to no-vote it appears). Compared to 2020 among UMM, Trump makes gains, while Harris takes losses, but it appears no-vote is the bigger winner. More to the point, married men performed very similar. So this doesn’t appear as much an "incel" problem, although clearly there is a gendered difference.

Is there an "incel" problem here? Perhaps, but it seems more a man problem. Did women nearly save the day? Let’s be cautious.

One of the main observations in this article is a decline in Latino and black women vote, while white women vote increased. Interestingly, among white people, compared to 2020, it appears that married women flocked to Harris, while married men didn’t flock to either - both candidates lost voters MM voters. Both saw gains with UMM, and both saw gains with UMM, but Trump’s white woman gain (see later in the article) was mostly among UMW. So Harris won with white married women, and Trump with white unmarried women. Both made slight gains with UMM, but in terms of raw numbers, slightly less for Trump than with MW, and in terms of ratio, less with UMM than with any category.

For black voters, both candidates registered losses with UMM - no "incel" problem detected here. MM saw a decline in turnout, barely changing their actual votes, but among a growing black MM population, this translated to an apparent rise in support for Harris. Among UMW we barely see any changes. And among MW, we see a large decline in turnout, hurting both candidates, but Harris more. Although it’s worth noting that given the sample (and size, 𝑛 = 501 black MW in 2020), CNN didn’t feel comfortable reporting the Dem:GOP ratio. But the one used here is the one that yields the black Dem:GOP ratio as a whole, so it’s probably fine.

Latino voters are one of the "hottest" topics of discussion this election. Like with black MW, the Latino UMM vote was too small (𝑛 = 473) for CNN to feel comfortable reporting a Dem:GOP ratio. The one used here is the one required to make the group have the Latino Dem:GOP ratio in 2024. This will do for us. It turns out that Latino UMM saw a decline in turnout overall (assuming the sample is roughly representative), although Trump saw a gain of 70k votes, and Harris a loss of 590k votes. Whether or not there is an "incel" problem here, it seems extremely marginal. Trump’s gains amongst Latinos were virtually all from MM - gaining 1.47m votes, and Harris losing 0.62m. Among MW and UMW, both candidates saw losses, as turnout fell.

It’s possible that Latino UMM are under-represented here: the total vote gain from 2020 for Trump nationally among UMM was 1.01m, while the total from these race groups adds up to 0.5m. It’s possible UMM in other groups are under-represented, and/or some of the difference came from other race groups’ UMM. It is notable that Trump’s gains (excluding losses), in a national view, were 25.9% from UMM. I doubt this is all an "incel" problem though. And it’s also interesting that 73.3% of Trump’s gains were from UMM.

(Apparently, this question means Bachelor’s degree, since some respondnats had an Associate’s.)

Often, pundit’s will use "white no degree" (WND) as a proxy for "white working class". Yes, it’s true, on average someone with a college-degree makes more than someone without (maybe about a $10k difference, with big variation in what the "mean income" of "WND" is). But that doesn’t mean all WND are poor. In fact, many are quite wealthy. This is kind of like saying that, since on average women make $10k less than men, that women are a proxy for the working class. Yes, while true, women are more likely to be working class... they are also a very classed group themselves. This is also true of racial groups, which also see different patterns based on income, even if, for example, black people are disproportionately represented in the <$50k group. Using any of these as a proxy for class seems kind of silly, when you could just look at data by income. But alas.

Yet, for some reason, news sites don’t like to give breakdowns by income (neither for the NEP+ER poll, nor the NORC+Fox News+AP poll). Presumably, they could. See box why if you want.

In lieu of ABC or CNN or whoever (even Fox News and AP, with their alternative exit poll) presenting how voters by income answered questions, they instead use "how WND answered questions" to get at this (this is also a bit silly by its own logic: white people are disproportionately under-represented in the <$50k group; if the logic is we want a sense of the working class vote, shouldn’t we look at all non-degree holders? Not just the race under-represented in lower incomes? Alas. Not that that would even be representative of "the working class", but it would be more logically consistent). So, what’s the scoop here?

First, we should get some idea of the varying proportions of WND and white [Bachelor’s] degree+ holders (WDH) among white people (and other races, if the news sites cared to present that data). According the Census office, 62.2% of the population had some college, but no degree (or an associate’s). At the same time, in 2021, 46.4% of white’s had a bachelor’s or higher - or 53.6% were in the "white no degree" group (black no degree: 71.9%; Asian no degree: 39.0%; Latino: 79.4%).

The data released as a table from the Census Office (Table A-1) tells a slightly different story (giving rates for those 25 years an older), but it’s the same trend. I’ll use the percentages from the table below, as it at least has data for several years, and so we can at least consistently include the effect of changing degree-holding proportions among white people over elections.

Now with that figured out, it seems in 2016, WND:WDH was 62.7:37.3, in 2020 58.7:41.3, and 2024, we estimate 56.64:43.36. With that, let’s look at the exit poll results, by WND vs WDH.

Now let’s look at by income group behaviors for these two groups. I’m not going to attempt a full blown analysis at the moment (since presumably that would require some income distribution adjustment, and I’m feeling a bit lazy atm), but we can at least take a look.

H1 = <$30k; H2 = $30k-$49.999k; H3=$50k-$99.999k; H4=$100k-$199.999k; H5=$200k+

And for good measure, here’s whites by the same income breakdown for 2020 and 2024:

So what happened? Overall, we can see increasing turnout among WND over time, and a downward trend for WDH. Does that mean the white working class is making its voice heard? Not quite. You can see in Table OD.IIIa, among WND, those making <$50k saw declining sampling from 2020 → 2024, those with $50k-$99.999k saw somewhat stable sampling (up 2.9%), and those $100k+ saw increased sampling. And it’s almost the same trend for WDH, except that the $50k-$99.999k sees declining sampling, rather than stable. So, except for $50k-$99.999k, we see basically the same sampling trends.

In other words, WND isn’t actually a great proxy for the "white working class" - in 2020, WND $100k+ were 20% of that group’s sample, and in 2024, 29%. Whereas <$50k was 46% of the sample in 2020, and 37% in 2024.

So this returns us to the basic issue: we don’t have access to the subjective responses of the <$50k and <$100k as such. And we can’t really use this "white non-degree holder" as a suitable proxy. The other problem: to really see what the working class thinks, we have to also ask people who voted in 2020, who didn’t in 2024. Since it appears, overall, that <$100k lost voters for both candidates (or according to Fox News, $50k-$99.999k saw some gains for Trump, along with the expected increase in no-votes), this is the crucial demographic.

One interesting point: if we look at white $50k-$99.999k overall, we see an increase in votes for Harris, and a decline for Trump, with an increase in no-votes. Where does this come from? It appears from both WND and WDH, although more so from WND (sampling increases 2.9%, Harris gains in the exit polling 24→32 (+8), Trump falls 76→65 (-11)) (WDH sampling here falls 18.2%, but Harris gains 52→59 (+7), Trump falls 47→39 (-8); but since it’s a fall, more analysis is needed to see if there are any actual "gains" here, or if this is a mirage from falling Trump voters; we can, however, compute that in 2020, total votes went 8.78m:7.94m, and in 2024, 8.11m:5.36m, a Harris fall of 0.68m and Trump fall of 2.58m (thus an apparent increase in no-voters of 3.26m); and we can make a rough estimate that the size of this group fell by 1.14m (so actual non-voters increased around 2.12m); so it seems turnout did in fact fall, and both candidates lost voters with $50k-$99.999k WDH, just Trump lost them faster. We would need to know the actual size of this group, however, to be comfortable saying how many no-voters there were in either election, and thus what turnout was). "The white working class is turning more MAGA" seems a ludicrous idea.

Here is some responses for that group though, if you’re interested.

2024 income data: G1 (36) 29:68, G2 (35): 32:65, G3(29): 34:65

From Tables OD.I and OD.II, we can estimate that the ND percent of white people in 2016, 2020, and 2024 was 62.7%, 58.7%, and 56.6%

2020 vote: Biden (28): 92:7, Trump (58): 3:96, Another (2): 33:47, didn’t vote (11): 39:56

In past year, inflation has caused your family: Severe hardship (27): 13:85, moderate hardship (53): 29:68, no hardship (19): 67:32

Most undocumented immigrants in the US should be: Offered chance at legal status (44): 65:34, deported (52): 6:92

Most important issue: foreign policy (3): NA:NA, abortion (12): 64:35, Economy (39): 11:88, Immigration (13): 2:96, Democracy (27): 69:30

2024 income data: G1 (46): 37:63, G2 (34): 24:76, G3 (20): 26:73

Union household? Yes (17): 40:57, No (83): 32:66

Is climate change a serious problem? Yes (59): 50:49, No (36): 4:96

View of Donald Trump: Favor (67): 2:98, Unfavor (31): 90:9

View of Joe Biden: Favor (32) 90:10, Unfavor (66): 1:99

2016 pres vote? Clinton (22): 91:7, Trump (57): 3:96, Someone else (5): NA:NA, did not vote (15): 51:46

Income: H1 (22): 45:55; H2 (24): 29:70; H3(34): 24:76, H4(17): 25:74

Most important issue to your vote? Racial inequality (9): 83:16; coronavirus (15): 77:23; economy (44): 5:94; crime and safety (13): 11:89; health care policy (14): NA:NA

Racism in the US is: most important problem (13): 83:12; an important (50): 38:60; minor problem (24): 15:84; not a problem at all (11): 5:94

View of BLM: Favor (42): 63:37; Unfavor (48): 4:96

View of Kamala Harris: Favor (35) 87:10; Unfavor (60): 2:96

In a November 2023 interview with Astead Herndon, she was clear and articulate, albeit with a heavy sprinkling of her "do the work" phrase. While I don’t agree with everything she stands for, she is very clear in expressing her political views - "smart on crime" (tackling "root causes" to reduce crime), a variety of social issues (ie women’s right to abortion, gun safety, etc), helping working families, and also helping business (this isn’t to say her record as California AG was "progressive", but my point is she is rhetorically comfortable and proficient here). It sounded like what you would expect from Biden’s VP - Biden policies, albeit a bit more pro-business, and viewing thinks more from a law enforcement perspective. Not something I wholly like, but she was clear in saying it.

In the interview, she sometimes comes off as defensive, repeatedly replying to Herndon’s questions with "could you be more specific" or "I don’t understand". Every once in awhile this is okay, but the frequency is a bit jarring. At the same time however, Herndon can be a forceful interviewer. This isn’t a bad thing either - he’s very good at formulating good questions in response to the interviewee, and trying to stick to the question if the interviewee doesn’t quite answer. He isn’t a "softball" interviewer, so to speak. For example, in his August 8th 2024 interview with Bernie Sanders, Sanders also engages combatively, but also productively (not gonna lie, I’ve really grown to like Herndon, listening to "The Run-Up").

In short, I think Harris is quite capable of clearly delivering a message she believes in (to some extent - perhaps the degree of her belief is up to debate), and is familiar with. My hunch is that her word salad speeches on topics like immigration and foreign policy reflect that these aren’t her forte (and it was frankly a poor decision for Biden to put Harris on the immigration beat at the start of his admnistration). But if the campaign message is "helping working families and business" - (not that I agree with the whole philosophy there) - I think she could perform well. And since I’m arguing she should have had a more Bernie-esque campaign, well, I think she could have done well. As for likeability? Being too combative? Maybe there is a racism-misogyny problem here, nationally. I’m willing to suppose that for the sake of the argument. Do I think that would have fatally wounded her, had she campaigned more for working Americans? Who knows, but my gut feeling is no. At very least, even if she lost, her defeat wouldn’t be as catastrophic downballot, giving the Republicans both the Senate and the House.

My view, reading and researching for this article, is that while Harris has these strengths (and to say otherwise is to mis-diagnose what her "inherent" flaws were as a candidate; although I think it’s fair to say she often seemed out of touch), her critical flaw isn’t simply that she bends easily with the donor winds, but that her close confidants (such as Tony West) - people that she listens to - are already primed for a more pro-big capital message, and less so a pro-working class message (certainly, this circle didn’t help her stay "in touch"). But the problem here isn’t purely Kamala Harris, the candidate on TV. And I haven’t said all of this just to damage control. She had strengths that could have been played - in fact, those strengths lie in messaging that includes policies for working Americans. Sure, she can sound combative, defensive (in an apparently negative way) - but this could also be endearing if she was defending and fighting for the right policies. Probably Bernie Sanders would sound annoying in his grumpy mode too, if he was defending big business, and leaving "the 99%" out to dry. Instead, Sanders argues on the other side of the line, and it makes him likeable - and from Mark Cuban’s interviews, it sounds like he finds him an annoying Lorax (not to say Harris has the same charisma as Bernie, but it’s also true that Bernie’s appeal goes far beyond his demeanor).

Let’s be clear - while a charismatic candidate is certainly an asset, it’s policy that wins elections (ie it’s hard to imagine someone less charismatic than Biden in 2020 - except, I guess, Biden in 2024; what Biden had in 2020 was both a COVID catastrophe for Trump, a mobilized base in the form of BLM (probably to BLM’s harm), and some progressive policies). This election (and 2016 as well) might appear to contradict this, but in both of these elections, the Democratic Party was very weak on policy for working Americans. What do we all remember about Obama’s 2008 campaign (after 1968, the 2nd highest turnout election after 2020)? Healthcare (and financial crisis relief promises). Sure, policy on various issues is necessary - for example, climate change is probably my #1 issue (and Hillary Clinton was far far better here) - but this has to be part of a broader platform that resonates with voters. And looking at her 2020 campaign - far better than the 2024 campaign - it seems she had the right instincts. But if nothing resonates in an election cycle, the reality show host carries the election, not thanks to his appeal as such, but more thanks to the Democratic Party bleeding millions of voters.

Perhaps I’m going too easy on Harris. Perhaps she never actually cared about working families. This seems credible, given how terrible, how pro-big capital her campaign was. And to be honest, I really don’t like her, and I’m not surprised by how she campaigned. My purpose here isn’t Harris apologism. It’s to discern where the 2024 election loss lies - is it in Harris herself (either as a black woman, or as, some argue, an irredeemably unlikeable, vapid person), or is it the campaign she ran, versus a campaign she could have run, and run well? If the key problem was Harris herself, there’s not much to learn, except that people don’t like politicians. Whoa, big surprise! Not much to learn if that’s the case, except to shit on the Democratic Party ad infinitum as "out of touch". Believe me, I enjoy it - and I do think it’s true, and a major flaw of the party. But we’ve known that since... forever, right? But if the loss lies in her terribly pro-capital campaign however, then there are lessons to be learned about America from this election, beyond ol’ reliable "The Democrats suck". Lessons not only for the Democratic Party, but for any political action, of any sort, in this country. Lessons about what people want, and what turns them off.

It should come as no great surprise that a Democratic Party which has abandoned working class people would find that the working class has abandoned them. First, it was the white working class, and now it is Latino and Black workers as well. While the Democratic leadership defends the status quo, the American people are angry and want change. And they’re right.

Today, while the very rich are doing phenomenally well, 60% of Americans live paycheck to paycheck and we have more income and wealth inequality than ever before. Unbelievably, real, inflation-accounted-for weekly wages for the average American worker are actually lower now than they were 50 years ago.

Today, despite an explosion in technology and worker productivity, many young people will have a worse standard of living than their parents. And many of them worry that Artificial Intelligence and robotics will make a bad situation even worse.

Today, despite spending far more per capita than other countries, we remain the only wealthy nation not to guarantee health care to all as a human right and we pay, by far, the highest prices in the world for prescription drugs. We, alone among major countries, cannot even guarantee paid family and medical leave.

Today, despite strong opposition from a majority of Americans, we continue to spend billions funding the extremist Netanyahu government’s all out war against the Palestinian people which has led to the horrific humanitarian disaster of mass malnutrition and the starvation of thousands of children.

Will the big money interests and well-paid consultants who control the Democratic Party learn any real lessons from this disastrous campaign? Will they understand the pain and political alienation that tens of millions of Americans are experiencing? Do they have any ideas as to how we can take on the increasingly powerful Oligarchy which has so much economic and political power? Probably not.

In the coming weeks and months those of us concerned about grassroots democracy and economic justice need to have some very serious political discussions.

Stay tuned.

Vermont Progressive Party, Nov 12 2024

Overall, turnout in 2024 was quite high, at 63.68%. This is up 4.48 pp (percentage points) from 2016 (turnout at 59.2%) and down 1.62 pp from 2020 (turnout at 65.3%) (notice these are different 2016 and 2020 turnout values than on Wikipedia: that’s because those values are not quite up to date, given new Census data and other calculations; see Data and Methods for further elaboration and links). You might notice - as I tried to spell out - 2024 turnout is up more from 2016 than it is down from 2020. Whatever the reason why (and I’ve nearly spelled them out already), it’s certainly a very different election.

That said, 2016 is just as valid (or invalid) a baseline as 2020 - in fact, perhaps less valid, if anything (given the turnout differences). And given the patterns just listed, the problems in this election are representative of a more fundamental trend than people relaxed from the COVID scare. To grapple with this election, we have to look at trends compared to 2016 and 2020, especially considering that Hilary at least talked about healthcare (as milquetoast as her proposals were), and Biden ran a relatively left-wing campaign, at least as far as the US in the past 60 years goes. By contrast, Harris ran the most right-wing Democratic campaign in at least 20 years. All of these factors play a part, and so it’s best to take a look at how things compare to prior elections to get a sense of what went wrong for the Democratic campaign.

So, despite high turnout (for the US at least, we’re in the same middling league as the Islamic Republic of Iran (note that our media frequently says that their turnout indicates their electoral system lacks legitimacy among Iranians; what does that say about us then?)), the Democrats still lose - and lose big. That seems to go against the most basic blue election heuristic: "higher turnout = better for Democrats, so get out the vote!" What actually happened then? Next time should Dems try to drive turnout down? (No)

(It’s also worth pointing out that among women, like nearly every other group, the <$100k turnout fell - only the $100k+ turnout increased. Thus, it appears this was an election for rich white women. But we don’t have the {income} × {race} × {gender} data to confirm this as such - but it seems, given we have {gender} × {income}, {race} × {income}, and {race} × {gender} data indicating this is the case, it would be pretty wild if it wasn’t actually the case)

Even if <$100k voters - axiomatically assumed to be reactionary dimwits "who don’t know what’s good for them" (and care less about kitchen-table policy than transphobic headlines) by the $100k+ talking heads - don’t like how she became presidential candidate (even if they think she just got their bc "DEI"), you can bet they’d turn out for her if she at least had the right policy. After all, they turned out for a demented has-been like Joe Biden in 2020 (which the $100k+ talking heads told us was all phony-baloney right up until the catastrophic 2024 debate, when his problems were rendered undeniable). Not cause voters liked seeing a nursing home patient run for president (nor liked being lied to constantly about how fit he was for office, despite our own lying eyes and ears), but because he at least had policies geared towards them (among COVID concerns).

I agree that him stepping down, and Harris taking the mantle, was the smart choice. While Dems should have had a primary, by July 2024, they had not. Biden deserves blame for much of this, being a stubbord genocidal jackass. But by that time, it was fait accompli, and Harris was at least VP, someone elected with Biden in 2020. I don’t think what happened in July was really that damaging - and we need to be specific about exactly what lost Dems this election, even if that process is indicative of the deep fundamental problems in the party. I think it was a smart move not only because voters can only tolerate so much dementia (no matter how good the policies are), but because she was an opportunity to start fresh: promise an end to the Gaza genocide, make policy adjustments from the Biden administration to help working class voters, etc. This seemed to be her direction, with the Tim Walz VP pick.

But then came in the donors (the same group who told Biden there would be no funding for the Democratic Party this election cycle if he didn’t step down - sure, that’s the right call, but how come these are the people with that deciding power?). And soon it became clear where her direction was: she decided to do the opposite of helping working Americans, continued to not use her VP office to exert any pressure to say Gazan lives, and lost the election for president, House, and Senate. It’s with the Harris campaign, the donors, and the Democratic Party machine involved, that the blame for this mega victory for MAGA lies.

Now that we know the election results, we can see one pollster that got it close: AtlasIntel (predicts: Harris:Trump 48.8%:50.0%; actual: 48.4%:50.0%). Here are the polling results nationally, and the sample sizes of the <$50k, $50k-$99.999k, and $100k+ (from here).

Keep in mind the deep issues with these polls (see here). Nonetheless, they give us some idea of trends at least, even if the actual values they report are probably wrong (even if they appear right: think, 30+70=100, but 50+50=100 as well. Just because we know the end result should be 100, if there’s only one true input set ({30,70} vs {50,50}, or any other variation that adds to 100), then simply getting the right ultimate result doesn’t mean you’re actually right; it means you’re lucky. But, trends are the key.

These results indicate that Harris and Trump barely budged in the polls starting July. Yet it’s worth noting how the sampling of <$50k steadily falls from 34.3% to 32.1% (-2.2 pp), and the sampling of $100k+ rises from 25.9 to 27.9 (+ 2.0 pp) ($50k-$99.999k stays near constant, rising 0.2 pp). This indicates - if sampling is any indication of a likely voter (given that at most 1.4% of respondants said they wouldn’t vote, but turnout was only 63.38%) - from the July 2024 baseline, <$50k engagement declined, whereas $100k+ engagement rose. (NEP+ER exit poll sample: 27:32:40; Fox+NORC+AP exit poll sample: 32:37:30)

Polling from Data for Progress backs this up: in PA, campaigning on Cheney’s supports increased enthusiasm for 29% of voters (skewing Democrat - not great GOP outreach, eh) and made 25% less enthusiastic (net 4% voters made enthusiastic). Yet campaigning on economic issues increased enthusiasm among 42% of voters, and made only 20% of voters less enthusiastic (net 22% of voters made enthusiastic). Similar trends were registered in Michigan.

Among both PA and MI voters concerned about the economy, they pointed to inflation as the #1 issue (77% and 73%). At the same time, populist economic policies (ie lowering childcare costs, expanding Medicare, strengthening anti-monopoly laws, cracking down on price gouging, increasing taxes on billionaires and corporations) were strongly favored by Trump voters here. And majorities/pluralities said voting for Trump was driven by economic motivations.

On the campaign trail, voters tended to hear Harris talk about the economy less in both states; instead, they saw Trump as the "change" candidate.

It seems like there was a lot of room for improvement for Harris, even if it was a tough race.

I’ve found a big pet peave since I read Taylor’s remarks: political pundits ignoring turnout. For example, if in 2020, 50 people of the 𝑋 group vote blue, and 50 vote red (and a little thing up top: % of all respondents 15%). Then in 2024, 15 of the blue voters decide not to vote, and 5 red decide not to vote. Now the exit polls show 𝑋 voters at 44:56. Oh no, a rightward shift! Why are all the 𝑋 people so right-wing now? But wait, you missed something: now the % of all respondents is 12%. No, 𝑋 didn’t convert to the GOP. They became less blue faster than they became less red. Say 𝑋 people are 14% of the population. So their turnout in 2020 was (15/14)*0.653=70.0%, and in 2024, it was (12/14)*0.6338=54.3%. For some reason, no one is interested in that. So since we’re ignoring that, then yea, the only explanation is people converting to red. garbage in, garbage out.

I will accumulate articles as they come.

Here’s another beauty (from Reuters).

Here’s a piece of work from the AP. It does the same old thing: just looks at the exit poll two numbers.

"Thus, Trump gains around 2.9-3.0m votes, and Harris lost 6.4-6.5m votes. Overall, our results seem to short Harris an extra 2.4m votes, and short Trump 0.7m votes..."

The latter seems a tolerable error, but 2.4m missing votes raises my eyebrows. There are a few issues here: in 2020, it predicts Trump with 72.62m votes, and Biden 83.05m votes - Biden overperforming and Trump underperforming both by about 1.7m, while getting the Third Party vote about right (2.82m vs an actual 2.92m). In 2024, it predicts Harris with 74.14m votes (within ~0.66m expected), and like 2020, shorts Trump, giving him 74.77m (short 2.3m votes). Unlike 2024, it over-predicts third parties, giving them 4.6m votes, when in reality, they got a total of 2.5m. So in 2024, probably third-party voters were a bit over-represented, and Trump voters over-represented. What exactly happened in 2020? The "missing vote" problem lies, from what I can tell, in some combination of sampling error, and mismatch between how income data is recorded in the exit polls, and the distributions used here. To the question "what’s your income", household distributions match up well with the responses. But the result isn’t always perfect. In 2020, more than half of black respondents reported an income of $50k-$99.999k - while such a household income bracket was then only 27.8% of black households. The result was a predicted turnout of 120% for this group (see Figure 1, bottom). This suggests some subtle distortions are present in the analysis here (similarly, Latino $50k-$99.999k in 2020 ends up with a turnout of 100.5%). But if we take a crude approach, and lop off 20% of the black $50k-$99.999k Dem vote in 2020 (votes which, it appears, are impossible), we lose 1.93m Dem votes from 2020, shrinking the 2020, 2024 gap that much - 80% of Harris’s "missing" votes (and likewise, the data predicts Biden with 83.05m votes in 2020, about 1.77m more than he actually got). Probably this group’s turnout wasn’t exactly 100%, but it gives us an idea where the main data issues lie. Why does the issue happen at all? This may be a sampling error, or mismatch in how "how much your income is" and "household income" are understood/reported. Either way, it seems probable that Harris’s drop-off in the $50k-$99.999k group is ~2m less than it appears.

Probably this group (which was the most favorable income group to Biden) was over-represented in exit polls, at the expense of groups that favored Trump - which appears to be the $100k+ (which registered a fairly low turnout of 42.4%, based on the exit poll + Census data). So, as a rough estimate, it’s likely the bulk of Trump’s missing 2020 vote is in the $100k+ group - thus its no-Votes are 1.7m less that year, and Trump "gained" 1.7m less in this group in 2024. Meanwhile, Harris "lost" 2m less in the $50k-$99.999k group, and its no-Vote fell in 2024 by 2m "less". Meanwhile, it’s unclear where the over-represented Third Party votes lie in 2024. According to the data here, among $50k-$99.999k they had 2.2% of eligible voters (1.49m), among $100k+ 1.9% eligible voters (1.86m), and among <$50k 1.6% eligible voters (1.26m). So... idk where they are. Somewhere in there. But, other than possibly netting Trump a marginal net gain in the $50k-$99.999k group of at most 0.84m votes (rather than a marginal loss of 0.65m votes), none of the qualitative patterns in gross votes would change if, for any income group, all of the 3rd Party votes were allocated to him (in that group).

It could also be that the exit polls only report in whole number percentages. The problem is apparent if you compute the predicted popular vote percentage from any of the questions - you’ll notice you’ll get different percentages for different questions (when they should all add up to the same, since its the same 15590 people answering the questions). This could explain, from what I can tell (at the most extreme possibility; $50k-$99.999k sampled 38.6, and goes 56.6:42.4, and $100k+ sampled 26.4, and goes 41.6:54.4; <$50k sampled still at 35, but goes 54.6:44.4), gives Trump 1.63m votes (largely making up his shortage), but only takes 0.70m votes from Biden. That that can’t do the whole thing suggests some sampling distortion in the poll itself. Nothing horribly bad - nothing is perfect - but probably isn’t simply the use of whole numbers.

Demonstrating some mismatch between how income is reported, for 2020 $50k-$99.999k black voters, 52% of 1929 black respondents surveyed said they were in this group. According to the error chart from Edison (the pollster), the ± sampling error should be 4%. But if we even put in 52-4 = 48%, we get ((48)/27.8)*0.640 = 110.5% turnout. Perhaps the error lies in how turnout was computed - ie error in the black population sampled from the greater whole? Looking at the same chart, for our whole sample of 15590 respondents, and 13% saying they were black, our ± sampling error should be 1%. So at lowest, black turnout in 2020 was ((13-1)/13.3)*0.653 = 58.9%. Plugging back in to the income group above, we get $50k-$99.999k black voter turnout at (48/27.8)*0.589 = 101.7%. Even at the extremes of our confidence intervals, we are getting an impossible value - this indicates either a sampling systemic error, and/or mismatch between how "income" is understood by respondents and the Census Office.

How to Interpret Even THESE Results Like Garbage

Even with these results here, it’s still possible to make a garbage interpretation of the data. For example, looking at Figure 7, you can observe among women a net decrease in Harris votes, and a net increase in Trump votes (compared to 2020). You might conclude "women are switching to Trump!" This would be wrong. First, the race perspective.

Looking at Figure 3, you’ll notice that Harris’s losses are from Latino and black women - both groups which also saw Trump losses (albeit smaller than Harris’s losses). On the other hand, Harris and Trump both gained among white women. Harris’s white women gains couldn’t overcome her Latino+black women losses however, while Trump’s gains could. So, overall, women voters didn’t "switch" to Trump, but Harris lost POC women voters faster than she could gain them among white women, while Trump gained white women voters faster than he lost POC women voters. On the other hand, it looks like male Democrat voters just stayed home, and Trump voters behaved per usual. In fact, Trump gained Latino men voters, but lost white men voters. At the same time, Harris lost among both.

Or by income, notice that Harris lost with <$100k women voters, and made big gains with $100k+. While Trump made some gains in the <$100k group, notice from the ratio plot that some Democrat <$50k women voters switched either to no-vote or to Trump, while in the $50k-$99.999k, Democrat women voters simply stayed home. Again, the problem isn’t "women switching" (save for 3.3% of <$50k women voters, which is 1% of all women), but Harris losing votes to no-vote among the <$100k (while making lesser gains among the $100k+), and Trump gaining from no-vote. For males, it’s a very similar story: Trump didn’t have a stable base, his gains just ended up equaling his losses. Both Harris and Trump saw losses in the <$100k male group, and gains in the $100k+ group (and again, Harris’s <$100k losses were greater than her $100k+ gains, while for Trump, the losses were more modest than his gains, giving him a net increase).

It’s the same story with race: the only group that saw net gains across income groups was Latinos. No race saw gains for either candidate across gender, but among Latinos Harris did see losses across Latino male and female (and in terms of ratio, losses with both black men and women).

The point is, we have to be careful in saying "group 𝑋 behaved in such and such way". We have to look who among that group did what. Even across {race} × {income}, there is much heterogeneity. My reason for focusing on income here is simple: across race groups, <$100k either sees an increase in no-votes, or only a marginal increase, indicating this "wasn’t an election for them".

What was the trend in the seven swing states of 2020 and 2024 (MI, PA, WI, NC, GA, AZ)? I’ll also include Ohio (since we have income data from exit polls for Ohio, along with Michigan. The classic duo!)

† turnout values for 2020 University of Florida (these turnout rates were updated in 2024, given Census data); for 2024 University of Florida

MI: Michigan; PA: Pennsylvania; WI: Wisconsin; NC: North Carolina; GA: Georgia; AZ: Arizona; NV: Nevada; OH: Ohio

How does the turnout compare with the nation? First, we need to take averages, ± the standard error of the mean (SEM). While standard deviations (SD) tell us about the spread of data, the SEM tells us the error on the mean - that is, how "precise" is the mean value calculated? The SEM is calculated by dividing the SD of a sample by the square root of the sample size. Let’s "learn by doing".

We see turnout increases in five (MI, PA, WI, GA, NV) swing states - these give a mean value of +1.35 pp, and a standard deviation of 0.1247... pp. To find the SEM here, we divide the standard deviation (0.1247 pp) by the square root of the sample size (√5): SEM = 0.1247/√5 = 0.06. Thus we report the mean turnout of this group as +1.35 pp ± 0.06 pp. This is the procedure used for all such values to follow in this section. Two swing states (NC, AZ) saw declining turnout - the two on average at -1.85 ± 0.83 pp. Altogether, the swing state + Ohio turnout change was +0.20 ± 0.57 pp, and without Ohio, +0.44 ± 0.61 pp.

Now how to compare with national turnout?

Of interest are values for which the difference from the national value is statistically significant (SS). This is when mean ± 2*SEM (standard error of the mean) doesn’t overlap with the national value. For example, swing state + Ohio turnout pp change is reported as +0.20 ± 0.57 pp, thus the lower bound of interest is 0.20 - (2*0.57) = -0.94 pp, and the upper bound of interest is 0.20 + (2*0.57) = 1.34 pp. Thus it’s 95% confidence interval is [-0.94 pp, 1.34 pp]. If the national value lies outside this, we can say with 95% confidence that the difference is SS (ie not due to variations in the sample (ie these 8 states) which are similar to those in the national sample (ie all 50 states)), but to some other factor.

The national turnout fell from 2020 to 2024 by 63.38% - 65.30% = -1.62 pp. Since this is outside of the [-0.94 pp, 1.34 pp] 95% confidence interval, we can say - with 95% confidence - that the difference is SS, and not due to a shared underlying variability in the sample (although to be really good, we should figure out the state-by-state change-in-turnout ± SEM; but I’ll keep it simple for now, as I’m being lazy).

In fact, we can see that not just the seven swing states + Ohio, but the seven swing states and the five positive-turnout-change swing states have a SS difference with the national turnout change. Only the two negative-turnout-change swing states are in line with the nation. Overall then, it seems there’s something different about swing states. I guess the obvious hypothesis here is that they’re different, because they’re swing states, which mobilizes their voters.

In fact, looking at the data for state turnout shifts from 2020 to 2024, the only non-seven-swing-states states to see increased turnout were Nebraska (+0.07 pp), New Hampshire (+1.60 pp), New Mexico (+0.17 pp), North Dakota (+1.06 pp), Rhode Island (+0.81 pp), Vermont (+0.80 pp), and Virginia (+0.44 pp) (altogether: +0.71 ± 0.19 pp). 6/14 states with increased turnout were "swing". The seven largest turnout falls were Alaska (-10.18 pp), California (-5.23 pp), Hawaii (-4.18 pp), Illinois (-5.85 pp), New Jersey (-7.09 pp), Oregon (-4.10 pp), and Washington (-4.05 pp) (altogether: -5.81 ± 0.78 pp). Notice, other than Alaska, these are all Democratic strongholds. Compare this with now-considered GOP strongholds: Texas (-2.25 pp), Ohio (-0.33 pp), West Virginia (-0.77 pp), Tennessee (-0.67 pp), Indiana (-1.41 pp), Missouri (-0.08 pp), and Florida (-2.93 pp) (altogether: -1.21 ± 0.37 pp).

These seem to reflect trends seen elsewhere: falling turnout for Harris and Trump, but falling faster for Harris. Although we would need more details about these states to be sure.

With turnout analyzed, how about gross voter changes?

The results are interesting. Only MI, PA, and AZ follow the national gross vote trend (Harris losing votes compared to 2020, Trump gaining) (interestingly, Ohio sees a net loss for both), yet differ from the nation in terms of turnout. All other swing states defy the nation, and show a gain for both candidates. That NC saw increased votes for both candidates, while seeing declining turnout, indicates either a substantial growth in its eligible voter population, or a significant decline in third party voters there. Only AZ then follows the national trend of declining turnout, while Harris sees net vote losses, and Trump sees net gains.

Now let’s look swing state by swing state, by income.

A big issue is that I don’t have an estimate for household income distributions for each. But since we’re interested in turnout trends, I’ll use the national bracket percentage population (PP) for each state. To compute a group "turnout" then, I’ll use the standard formula in this article: turnout = (PS/PP)*Tᵢ (see Data and Methods section). This might get the actual value wrong group-to-group in 2020 and 2024. But the goal is more to discern trends. So, for example, if 35% of Wisconsin’s population made <$50k a year in 2020, and 34% in 2024, those values would differ from the national values (32.4% → 31.3%), but the downward trend would be shared.

And in the formula (PS/PP)*Tᵢ (where PP is that percentage, ie our hypothetical WI 35% and 34%), this downward trend would be picked up in the resulting turnout changes, even if the actual values are different. This can still lead to issues, but at least it gives us a rough idea, with a reasonable enough input value.

For this analysis, NEP+ER only have income data for the states of Michigan and Ohio in 2024 (hence why Ohio was included here). Fox+NORC+AP has data for all the seven swing states and Ohio. The data will be included as available below.

Looking at these results, notice that in every single swing state + Ohio, the <$50k vote turnout falls (-7.59 ± 1.00 pp), while the $100k+ vote increases (7.56 ± 1.06 pp). And in all but Michigan and North Carolina, the $50k-$99.999k voter turnout falls as well (-1.06 ± 1.12 pp; among the six falling states, -2.57 ± 0.55 pp). All of this aligns with the national trend of falling turnout <$100k, and rising turnout for the $100k+. But it does differ in the magnitude: nationally (according to the Fox+NORC+AP exit poll, to be consistent), the <$50k voter turnout fell by 11.50 pp, for the $50k-$99.999k by 0.70 pp, and for the $100k+ turnout rose by 4.30 pp (according to the NEP+ER poll, the trend was -15.60 pp, -19.30 pp, and +19.90 pp).

In other words, comparing national vs swing results from the same exit poll (Fox+NORC+AP), we see that while the same general trend exists (falling <$100k turnout, rising $100k+ turnout), the swing state <$50k turnout fall was less sharp in the swing states than nationally, the rise in $100k+ turnout was sharper (the $50k-$99.999k group appears to be largely consistent with the national trend). Thus, the swing states + Ohio (or without Ohio) follow the national income group pattern, but see a net rising turnout of 0.2 ± 0.57 pp, compared to the national fall in turnout of 1.9 pp.

In terms of apparent changing vote ratios (again, assuming that trends in income bracket composition nationally are representative of these states), we see the following:

Now let’s compare the results for the Fox+NORC+AP exit poll between swing states and the nation (table SSA.IV.a vs SSA.IV.c). Of interest of swing state + Ohio values for which the difference is statistically significant (SS). To review from above (you can skip if you remember), which we can say is when mean ± 2*SEM (standard error of the mean) doesn’t overlap with the national value. For example, swing state + Ohio <$50k pp change in no-vote is reported as 7.59 ± 1.00, thus the lower bound of interest is 7.59 - 2*1.00 = 5.59, and the upper bound of interest is 7.59 + 2*1.00 = 9.59. Thus it’s 95% confidence interval is [5.59,9.59] - so if the national value lies outside this (which it does), we can say with 95% confidence that the difference is SS (ie not due to variations in the sample (ie these 8 states) which are similar to those in the national sample (ie all 50 states)), but to some other factor.

The effects of a less-sharp <$50k voter turnout decline (no votes less by 3.9 pp) are less bad losses for Harris (by 2.9 pp), with no SS change for Trump. Among the $50k-$99.999k, the trend isn’t SS different from the nation. Among the $100k+ is a larger decline in no-votes compared to the nation, translating into a larger gain for Harris. Overall then, while the swing states follow the same general trend as the nation (ie lower <$100k turnout, hurting Harris more than Trump, higher $100k+ turnout, helping Harris more than Trump, but not enough to compensate), the trend was more favorable for Harris in the swing states than for the nation as a whole.

Looking at just Ohio and Michigan (Table SSA.IV.b) - since those are the only two states with income data from the 2024 NEP+ER exit poll - there are some differences between the Fox+NORC+AP and the NEP+ER exit polls. But they share some features. First, the <$50k bracket was more favorable to Trump than the nation, and with lower no-votes - it seems Trump mobilized more voters here than compared to the nation. On the other hand, the $50k-$99.999k bracket was better for Harris here than across the nation. Interestingly, they conflict on how the $100k+ went for Harris. The Fox+NORC+AP poll indicates they were more favorable to Harris, and the NEP+ER poll indicates they were less favorable to her.

Comparing the Fox+NORC+AP results in Tables SSA.IV.a and SSA.IV.b, the confidence intervals overlap for nearly every category, except the trend for $100k+ Democrat voters, indicating $100k+ voters in Michigan and Ohio were harder on Harris than the swing states as a whole.

Overall, it seems both polls agree that in Michigan and Ohio, <$50k voters were more mobilized for Trump, and $50k-$99.999k voters went more for Harris. The swing states as a whole, however, were generally more favorable to her than the nation, with higher turnout overall.

So What’s this Mean?

It may appear that voter engagement increased in swing states compared to the nation, but it’s important to keep in mind that we still see the same trend: falling turnout for the <$100k, and rising turnout for the $100k+ (with the trend difference, overall, helping Harris, with some exceptions, such as Michigan). Although it shouldn’t be forgotten that these are swing states after all, where turnout is expected to be higher (at very least because campaigns generally put more focus on them).

*Note that these are changes in votes from neighborhoods that are majority 𝑋. So if a neighborhood has 55% black people, then they are part of the group in that row, even though the vote pattern in such a community isn’t 100% representative of black behavior.

†Thus these fractions are more to get a sense of scale, than a hard precise/accurate measurement.

Gay-Dagnogo said she’s among a group of Black political leaders whose concerns were ignored by the campaign.

“I wish I wasn’t right,” she said Wednesday. “It was too little too late and that’s always the Democratic way. I don’t know if they’ll learn. They’ve become immune to the stench of dysfunction.”

The Harris campaign engaged Black voters with near-daily events at churches, union halls, house parties, Divine Nine Greek organizations, barber shops and beauty salons. They brought in actors, musicians, political leaders and influencers to persuade voters.

Branden Snyder, director of Michigan Election Defense Coalition, said he encountered voters who didn’t think Harris was talking enough about improving the economy and working class issues. Harris had a plan to provide down-payment assistance to first-time homebuyers and tax credits for new parents, but voters weren’t familiar with it.

...

Benji Ballin, 28, was a paid canvasser for the Harris campaign who said he encountered enough Detroiters on the fence to make him nervous. He said some voters planned to vote in down-ballot races but not the presidential race.

Clips circulating on social media ahead of the election showed Harris saying she wouldn’t do anything differently than Biden and supported economic policies that would benefit all of society rather than Black people specifically. Social media users sharing the interviews said it showed Harris wasn’t focused on helping Black voters.

Scott Holiday, director of nonprofit advocacy group Detroit Action, said Harris couldn’t articulate a winning message on improving the economy. Instead, Holiday said voters were told to fear Trump.

“We saw the best turnout in 2008 and that happened through hope and change – people wanted something and were willing to (vote) to get it,” Holiday said. “That simply wasn’t the case here. The majority of Democratic messaging was ‘be afraid of this man.’ I don’t like Trump’s messaging but he was giving his base something to fight for.”

Justin Onwenu, a Detroit activist and entrepreneurship director for the city, said rage against Trump didn’t translate into a Democratic victory in 2016. Onwenu, a delegate for Bernie Sanders in 2020, was knocking doors for the Harris campaign in Detroit.

...

Snyder worked for Clinton’s 2016 campaign. He said the Harris team made a similar mistake by trying to pull Republicans to their side. Allying with former Vice President Dick Cheney and his daughter may have done more harm than good, he said.

Snyder said former President Bill Clinton also “didn’t do any favors” by saying Israel was forced to kill civilians during a campaign stop in Benton Harbor. Arab American and Muslim communities were deeply angered by Harris’ support for Israel’s wars in Lebanon and Gaza.

...

Trump overtly pledged to end the wars during visits with Muslim leaders in Michigan.

Harris pledged to secure a ceasefire but ignored calls to end military aid for Israel. During her final campaign rally in Michigan, Harris said she would “will do everything in my power to end the war in Gaza.”

Holiday said the Harris campaign may have focused too much on the prospect of a national abortion ban. Though reproductive freedom is vitally important, he said, “people wanted to vote with their wallet.”

Gay-Dagnogo agreed. She argues the campaign put too much focus on protecting abortion rights in a state that already mobilized voters around that issue in 2023.

“Black elected women (told the campaign) the community is tired of hearing about abortion,” Gay-Dagnogo said. “It’s an important issue but we want to get bread and butter issues in the way the inner city needs to hear. The housing crisis, the cost of food.”

Democratic vice presidential nominee Tim Walz made abortion the centerpiece of his closing argument to Detroiters during a Monday campaign visit.

Snyder said the emphasis on the potential threat Trump poses to democratic norms, while important, was a nebulous issue for working class voters.

“Don’t get me wrong, I believe it and understand the concerns of Project 2025, but for a lot of Black voters democracy doesn’t work in a lot of places already,” Snyder said. “I don’t know if the hand wringing around fascism was enough to move people.”

† W: White, non-Hispanic (unless otherwise noted); B: Black; L: Latino ("Latino or Hispanic of any race"); A: Asian; AI: American Indian and Alaska Native alone

‡ here, five values are reported; the first are W:B:L:A:AI; the fifth is "Native Hawaiian annd Other Pacific Islander alone"

† Note that these turnout values may be different than in table t.I, because the values here are pulled from Wikipedia (for the convenient table formatting), and may be slightly out-of-date with AP reporting. And it seems, some odd discrepancy between county and statewide vote (not sure if there is a methodological reason why, or...)