In my last post, I discussed a new PNAS paper by Vilbig and England showing that lower-income white Americans haven’t really changed how much they vote for Democrats over the past four decades. The raw vote shares for the bottom half of the income distribution in the Trump era look broadly similar to where they were in the 1980s. The big movement was mostly at the top, with economic and educational elites shifting dramatically toward Democrats.
Some of you pushed back and argued the paper (and my endorsement of it) is problematic. Some comments misfired. For instance, it’s simply not true that the results are driven by focusing solely on Whites. The paper itself has supplementary analyses using the whole sample, and the picture is virtually identical.
But other comments were more on point. They more or less asked the following: isn’t it true that the national context of each election is one crucial determinant of the vote that is changing over time and can thus skew the picture if we focus solely on raw shares? For example, Ronald Reagan won the popular vote by 10 and 18 points in 1980 and 1984. Recent elections, however, have been decided by low single digits. If lower-income whites are voting for Democrats at the same rate across those two very different environments, that actually signifies a relative (if not absolute) rightward shift to the Republicans.
So let’s make the adjustments and look at relative shifts.
Using the authors’ replication and data cleaning code, I first simply tried to replicate their main deciles figure (raw shares). I wasn’t perfectly successful, but after a few iterations I think I got close enough. Remember, this answers the simple question “What was this group’s Democratic vote share?”
Adjustment 1: Center on the same-year white mean
Now, instead of just asking “What was this group’s Democratic vote share?” (which, I should say, is a perfectly legitimate question), we can also ask “How far above or below the white average was this group in that specific election year?” (obviously important as well). This strips out year-specific noise and tells us whether a group’s relative position within the white electorate has changed.
And sure enough, you can see a clear relative shift when we look at it that way. The lowest income decile moves from about +18 points relative to the white mean in 1980 to about -4 points in 2020. The highest decile moves from about -12 to about +9, a rise of roughly 21 points. In relative terms, both groups have moved substantially over the whole 4-decade period. Interestingly, the shift at the bottom is almost insignificant if we focus solely on the last few elections (Obama-era to 2016/2020).
We can also aggregate into broader bins, like bottom 30% versus top 30%. It’s more or less the same here as above.
Adjustment 2: Center on the national popular vote
Okay, now let’s look at each group simply relative to the overall Democratic national two-party vote share in that election year. This directly addresses the fact that 1980 and 1984 were huge GOP years.
The bottom decile goes from about +16 points relative to the national vote in 1980 to about -10 in 2020, which is a huge relative decline. Though, in fairness, if we strip out the 1980 election, there’s virtually no movement on average all the way up to the 2020 election. In any case, the top decile has been on an upward tear consistently since 2000.
Let’s do bottom 30% vs. top 30% now. Not much movement among the lower classes here on average, except for the huge (and important) drop in 2020.
Adjustment 3: Add the 2024 election
Some of you objected to the fact that there’s no 2024 data in the PNAS paper. That didn’t bother me too much, because I was mostly interested in seeing if there’s any long-term decline, or at least medium-term decline from the Obama era through to Trump and Biden. I was sure that the 2024 election really was different and saw some huge movements across classes.
But let’s look at the data. Instead of relying on GSS, we can use ANES here.
First, the raw shares. The 2024 election was nothing special.
What about relative movements, so lower classes compared to white mean or the national popular vote? Yep, there is something to the idea. Here’s just one of the fairly representative comparisons.
So, are the Democrats losing working-class support (or however you want to call it: support of the worse-offs, the poor, lower half of society, whatever)? Not in absolute terms, but relatively speaking yes - at least in recent years. The relative drop is much less dramatic than I think many people imagine, especially if you look at the whole 4-decade period and focus on the averages. If you squint a bit, it almost goes away. Though it’s much clearer if we focus just on the very recent period (2020, 2024).
When I was working, now almost a decade ago, under the hypothesis of working-class support collapsing, I definitely had much larger and more consistent drops in mind. But perhaps I was just being naive. In any case, as long as people are clear what they mean with the Dems’ “collapsing” working-class support, we’re good.
There are many issues here (apologies for the length):
1. As often, the interpretation of time-series data depends on whether one focuses on levels or on changes. In the previous post, the claim was that “The poorest Americans (like the bottom or 2nd decile, but also the 4th and 5th) nowadays vote for the Dems just as much as they did in the 1980s”. By looking at the decile graph, though, one can see that in recent elections the changes were all negative. And, indeed, since 2012 the Dem support from the poorest five income deciles has only gone down. Nobody (at least nobody serious) claimed that poorer income deciles had voted en masse for Trump. What many noticed was that the Democratic Party was losing rapidly many “working class” voters, for example with many union members switching to Trump since 2016. This was important in close elections and did not foretell a good outcome for the future. So, if one had focused on the changes, the marked drop in 2024 was something that one could reasonably expect.
2. There were other indicators too. Voting the Democratic party is a binary decision but it does not capture how much one favorably one sees the party (the “latent index” of the Probit model for those who are interested). For example, since 2008, party identification with the Democratic Party has gone down almost each year (https://news.gallup.com/poll/700499/new-high-identify-political-independents.aspx). Favorability ratings have ebbed and flowed over the years but they recently cratered ( https://yougov.com/en-us/trackers/the-democratic-party-favorability).
3. Related to the point above, it would help to have a better model of "abandonment". Many couples do not divorce over a single minor episode. Rather, they grow apart over time. They may still go out together at social events but they are growing more and more disaffected with each other. It all may seem somewhat normal… until they call their lawyers. I have no idea how abandonment is modeled in politics (if it is at all) but I would not model it as a gradual constant descent, more like cracks developing over time until the cracks eventually precipitate a crisis.
4. The US is a low election turnout country. The PNAS paper is conditioned on individuals voting. It would help to tabulate % of voting by income deciles. If I recall correctly, in the 2016 primaries Trump did very well in the lower income districts. In my anecdotal evidence, the more run-down a house is, the higher the chance it has a Trump flag outside.
5. As an economist, I am embarrassed to have to point out to a sociologist that political identities are also socially determined! The graphs about the occupational and educational divide in the previous post are amazing (congratulations to the authors for doing them!). In the Democratic Party is now easier to find a Rhodes scholar (e.g., Buttigieg and Booker) than a plumber. So if you, as about 2 out of 3 Americans older than 25 years old, do not have a Bachelor degree (https://www.census.gov/newsroom/press-releases/2023/educational-attainment-data.html), it is not clear that you will feel a lot of identification with the Democratic Party. Sure, you may realize that the Republican Party may be even worse for you, but you will probably feel more like an hostage on Election Day than a committed voter. This has many implications. First, the % of those who identify as Independent is at all time high (see Gallup link above). Second, a party of highly educated may at times conduct pro-worker policies (Biden definitely did try) but they will fail to garner enough support for them because the “packaging” is suspicious. Third, a party of highly educated may feel disconnected from the lower rungs as it happened during the inflation in 2022 when many were complaining about their daily struggles and the administration, brandishing economic reports, replied that that cannot be true and it must be just “vibes” (“looked at the median earnings, they went up by x%!”).
6. I am less familiar with the European data but some qualitative data is hard to neglect. In Europe, I can only think of a major political figure who does not have a BA degree: the right-wing Italian Giorgia Meloni, who comes from a working class neighborhood of Rome. Her political opponent on the left, Elly Schlein, has a law degree, and three citizenships.
To sum up, I believe that those who deny the Brahmin left thesis are in denial.