Subscribe here: Apple Podcasts | Spotify | YouTube | Overcast | Pocket Casts
Every personal detail is a tell. From your choice of college major, to the industry you work in, to the company within that industry—each decision is part of a sorting phenomenon populating certain workplaces with Democrats and others with Republicans.
We’re socialized to not talk politics at the office to avoid polarizing issues that could break norms of professional behavior. But according to a new study from two Harvard researchers, that norm may have obscured a startling partisan divide at the workplace: Republicans and Democrats are sorting into different fields of study, industries, and companies.
Workers aren’t just pawns in this partisan sorting; they’re actively choosing it, although perhaps subconsciously. As the study authors Sahil Chinoy and Martin Koenen found, “The median Democrat or Republican would trade off 3% in annual wages for an ideologically congruent version of a similar job.”
“Is [3 percent] big or small?” Chinoy asks. “It’s less than something like health care. It’s sort of actually comparable in magnitude to some of these softer amenities, things like having a relaxed versus a fast pace of work, for example, or having training opportunities at work. People seem to care similarly about the ideological nature of the job.”
The following is a transcript of the episode:
Jerusalem Demsas: I have a bit of a weird job for several reasons, but for one: Many of my colleagues’ varied ideological commitments are pretty clear due to the nature of our work.
But I was curious about what workplaces look like in less overtly political places. Do people often know the political opinions of their colleagues and bosses? Could work be a place for the healthy mixing of people with different partisan identities?
Probably not. At least, that’s what I take away from a new paper called “Political Sorting in the U.S. Labor Market,” which argues that political segregation is extremely common in the workplace. According to the authors, “a Democrat or Republican’s co-worker is 10 percent more likely to share their party” than what you might expect based on where their workplace is located.
Why? Well, it’s largely because workers are opting into college majors, jobs, industries, and companies that correspond with their partisan identities. Republicans are more likely to have studied business, finance, engineering, and technology, while Democrats are more likely to have studied the arts, social sciences, and the humanities.
Industries themselves are therefore more likely to have employees of one party rather than the other, but even within industries, companies are attracting one party’s adherents over the other.
My name’s Jerusalem Demsas. I’m a staff writer at The Atlantic, and this is Good on Paper, a policy show that questions what we really know about popular narratives. My guest today is Sahil Chinoy, who co-authored this paper while finishing up his economics Ph.D. at Harvard. Sahil himself is on the job market, I’m sure headed to one of those ideologically diverse workplaces so common in academia.
Sahil, welcome to the show.
Sahil Chinoy: Thank you. Thanks so much for having me.
Demsas: So before you conducted your research project, did you think that workplaces were more or less segregated than other areas of life, like neighborhoods, schools, etcetera?
Chinoy: I think somewhat less. We’ve all heard this idea that there should be, quote, “no politics at work.” And I think I took that to heart and thought the workplace might be a uniquely important site where there’s less partisan or political segregation than some of the other environments that we inhabit, like schools and certainly neighborhoods. There’s a lot of attention paid to partisan segregation, particularly across space, across neighborhoods.
And I do think that I thought that the workplace would exhibit less sorting than that. How much less? I don’t know. I certainly thought that it can’t be the case that it’s perfectly even.
Demsas: Yeah, I mean, it’s like when you’re young, and you’re in, like, K-12 or something, it’s the time of your life when you’re repeatedly kind of interacting with people you didn’t choose to be interacting with. It’s your teachers, other kids at school.
And when you grow up, work is basically one of the only times when you’re being forced to go anywhere that you are not actively choosing who you’re going to be interacting with. Obviously, there’s a ton of segregation that comes from where people are in school.
But I think the fact that, like, this is arguably the only place that adults are interacting with people that they’re not opting into on a regular basis—that leads people to kind of feel like, Oh this is probably, like, the most generative space for reaching people across the partisan aisle.
But you do not find this. So walk me through your paper. What did you do to investigate whether workplaces were segregated by party?
Chinoy: Yeah. So just one point on that. I think that’s exactly right. I think the idea here is, like, people have less ability to choose who their co-workers are compared to other people that they interact with, and that’s what generates segregation.
I will note, though, that—it’s kind of funny. When I bring this point up to academics and to professors, there’s sort of the one group, I think, that doesn’t really see this. They’re like, What do you mean? We can perfectly choose who our colleagues are.
Demsas: (Laughs.) That’s, in fact, the whole point.
Chinoy: That’s what hiring is. That’s the whole point. But I think generally, yes. I think that’s the idea.
So we’re sort of starting with this basic premise that, you know, maybe the workplace is less politically segregated than some of these other environments—not such a hard question to pose or to ask. I think it’s just a hard question to answer.
And it’s a hard question to answer, I think, primarily because of a data constraint. We have very high-quality and large-labor-market surveys in the U.S., but they don’t ask questions about politics or partisan affiliation. And we have large and high-quality political surveys, but they usually don’t ask questions about where exactly people work. And even if they did, they’re large, but they’re not large enough to really capture who is working with whom.
And so the starting point of this paper is really to say we have this pretty, to some extent, obvious question: To what extent do Democrats work with Democrats and Republicans work with Republicans? And how are we going to answer it? And the way we answer it is by combining two sources of information: On the one hand, public LinkedIn profiles—so scraped LinkedIn profiles that list where everyone works and some other characteristics about them, often where they went to college or their educational background. And we combine that with administrative voter records.
And so in the U.S., who you vote for is not public. But in 30 states and the District of Columbia, which party you register with is public. And so we combine those two sources of information, and that lets us see who is working with whom and everyone’s party affiliation. And that’s how we can quantify the magnitude of this partisan segregation at work.
Demsas: So you find that “a Democrat or [a] Republican’s co-worker is 10 percent more likely to share their party than expected based on local partisan shares.” Can you just unpack that finding? What does that mean?
Chinoy: Yeah. So the idea is: We want to benchmark the share of your co-workers who share your party affiliation against what we might expect. And so, you know, to give you a concrete example, say you are a Democrat, and you work at Google in Mountain View, then a high share of your co-workers are fellow Democrats. I think it’s something like 55 percent in our data. So that’s, you know, a relatively high share of Democrats. But what should we actually expect? Should we benchmark that against the share of Democrats in the U.S. as a whole?
And so our baseline measure takes that sort of realized share of co-partisans—so the 55 percent of Democrats that share your workplace—and divides it by the share of Democrats in your local labor market, which we operationalize as a commuting zone. Commuting zones are sort of aggregations of counties that are precisely designed to capture these kinds of commuting patterns, who could reasonably share your workplace.
And it turns out—I’m going to forget the exact number, but you know, it turns out—that share is quite high for Google in Mountain View. So you’re a Democrat at Google Mountain View—your co-workers are 6 percent more likely to share your party affiliation than you would expect based on the local shares of Democrats and Republicans. And then we generalize that for everyone. We do that for every person in our sample, tens of millions of people, and that’s how we arrive at that 10 percent number.
Demsas: This is, I think, a really great way of putting it, because, obviously, someone’s going to say, you know, Okay, Google—you’re in California. You’re in Silicon Valley. This is just a high-Democrat place, so are you just, like, looking at, basically, that there is geolocation sorting that’s already happening? And you’re saying it’s not just the fact that, like, there are locational differences in partisanship. It’s that even taking that into account, workplaces are even more segregated based on party than you would expect just by, like, walking around and taking a random sampling of people who live in the commuting zone that Google is in.
Chinoy: Yeah, that’s exactly right. And, you know, I can compare against the national shares, and if I do that, that number is 20 percent. It’s twice as high as the baseline statistic that I quoted for you earlier. And that’s a meaningful number, too, but what sort of intuitively accords with people’s sense of how to measure partisan segregation? It’s probably comparing against the local environment. And that’s why that 10 percent number is kind of our preferred estimate.
Demsas: How does this compare to the level of partisan segregation that we observe in other places? We know, for instance, that there’s partisan segregation happening in schools or in dating markets and churches and stuff like that. Is the workplace the most segregated based on party in America, or is this in line with other places?
Chinoy: Yeah, so it’s hard to answer this directly for every other social environment or every other group of people. I can tell you a couple things. So one is: I think a natural comparison is residential partisan segregation. This is something that people study a lot, right—the extent to which Democrats live on the same block as Democrats, and Republicans live next to other Republicans. And so we can sort of compare what I told you—that 10 percent number, that overexposure ratio—against partisan segregation across neighborhoods.
And you can define neighborhoods in different ways. One way to do it is a zip code. And when we do that, we find that partisan segregation at work is pretty similar. So, like, a little bit less than but overall pretty similar to partisan segregation across zip codes. We can go one step further and say, you know, maybe the zip code is a little bit bigger than what you have in mind when you think of neighborhood-level sorting. And so we have individual addresses in our data, and so we can say, you know, You have 15 co-workers. Let me figure out how many of them share your party affiliation, and let me look at our sample of the 15 people who live closest to you and figure out how many of those people share your party affiliation.
And when we do that, we find that workplace-level segregation, workplace-level overexposure ratio is a little bit less pronounced than that sort of nearest neighbor level of segregation, but still pretty similar, not so different. It’s not orders of magnitude different. So that’s kind of why we say that it’s a little less pronounced than residential segregation as a whole but still pretty sizable.
This isn’t in the paper, but we can also look at colleges. That’s the other thing that we can really observe well in our sample, and when we do that, we find that colleges are less segregated along party lines than our workplaces.
Demsas: Wow.
Chinoy: And sorry—that’s college cohorts.
Demsas: Wait. Sorry. Can you break that down? College cohort—you mean, the people who went to your college and then are in the class of 2017 as well?
Chinoy: Exactly. Yes.
Demsas: Okay, so that group, the people who went to my college and are in the same class as me, are less segregated than my workplace? Well, maybe not me, in particular, but on average.
Chinoy: Yeah. And I think a lot of that is just the size of these groups. College cohorts are quite big compared to workplaces, which tend to be relatively smaller. And so there’s a little bit more room for that kind of political diversity in college cohorts. And so the extent to which you think that’s an apples-to-apples comparison, I think it is up for debate because of that size issue.
Demsas: So you’re not going to take the hard position that colleges are more open than workplaces in America? (Laughs.)
Chinoy: I wouldn’t say they’re more open, but certainly you’re in this group of people that, for many people, is quite large and might include people from diverse geographic backgrounds. That’s also something that happens at colleges.
Demsas: So is this a function of income or racial or gender segregation? Like, how much of this can be explained by the fact that our workplaces are segregated by factors that are correlating with partisanship but are not partisanship, in particular?
Chinoy: Yeah, that’s a great question, I think a natural question. I think it is important to note, though, that overall, like, what is the phenomenon of interest? It’s partisan segregation without differencing out all of those other background characteristics, right?
And I think that there’s an analogy to gender segregation at work. You know, to what extent is gender segregation at work driven by different occupational choices? The different occupations that men and women sort into or are sorted into—that’s sort of an interesting question. At the end of the day, that’s part of the phenomenon of interest. That’s part of what creates segregation at work. And so I think a similar thing applies here.
That said, we can do, I think, the kinds of exercises you have in mind, where, instead of benchmarking just against the share of Democrats and Republicans in your local labor-market area, in your commuting zone, we can additionally incorporate information about those co-workers in predicting their partisanship.
And so what I mean is: If we knew not just where your co-workers lived, in terms of which commuting zone they live in, but also their exact year of birth and their gender and their race—things which are, as you know, very correlated with partisanship and politics in the U.S.—would we still be surprised that Democrats disproportionately work with Democrats, and Republicans disproportionately work with Republicans? So these are pretty fine predictors, right? We actually interact [based on] year of birth and gender and race. So for me, you know, an Asian male born in 1995 tends to be registered with the Democratic Party at a particular rate. We can incorporate that information and say, Is it still the case that we see this partisan sorting?
And we find that, indeed, that explains some segregation but certainly not all of it. And we can add even more predictors, so not just the education level of everyone, whether they went to high school or college or have a postgraduate degree—again, something that’s highly correlated with partisanship—but also the exact college that they went to. And we can show there are partisan differences across schools.
And we can incorporate that information as well. We can incorporate not just the exact college that you went to but what you studied when you were there. And there, again, we see large gaps in the college major choices of Democrats and Republicans, which I think actually is independently quite interesting also. So college and major and then industry and occupation—we can incorporate all of these predictors, and we still find that a Democrat or Republican’s co-worker is about 4.3 percent more likely to share their partisanship than we would expect. So we bring that ratio down from 10 percent to 4.3 percent. All of these things clearly matter, but they don’t explain all the segregation that we see.
Demsas: I think that’s really significant. I think when I first saw your abstract, I was like, Okay, well, is this just like, Black people are Democrats, and women are Democrats, and men are Republicans, when we’re looking at averages?
And seeing that significant difference even without that—and I mean, I take your point well that looking at partisanship is relevant, even if it is the case that race and gender are playing into that. Like, that overall partisanship still tells you something about workplaces in America.
But I also want to ask, because I think in your paper that you’re seeing more heterogeneity for different income bands and educational attainment, that there’s a different level of partisan segregation for people who make more money or for people who have, you know, graduate degrees or college degrees.
Can you tell me about that? What’s going on there?
Chinoy: Yeah, so that’s exactly right. You know, the 10 percent number that I was quoting for you before is an average across everyone in our sample. We can see what that looks like among subgroups and, in particular, we can see what that looks like among subgroups defined by education and by income.
When we do that for education, we see something pop there, which is that people with postgraduate degrees, people who are more highly educated, tend to be in workplaces that are more segregated. And there’s a little bit of something going on for high-school graduates versus people with bachelor’s degrees, but really where it tends to stand out the most is people with a postgraduate education.
Demsas: It’s you, Sahil. You’re causing all our problems.
Chinoy: (Laughs.) I mean, yeah, like, kind of, though, right? And you can think of stories why this might make sense. Maybe these people have more of an ability to choose an employer that really aligns with their ideological interests in a way that isn’t true for other groups of workers. I can’t say for sure why.
Demsas: Or they’re more motivated, right? Like, you might be more ideological or partisan.
Chinoy: Totally. Also true. And, actually, on that point, when we subset to people who have made campaign contributions and might be more politically motivated or politically interested kind of in the way you were describing, we also see that those people are experiencing more segregation at work, particularly the people who donate to very liberal or very conservative candidates.
And then you also asked about income. And here we see a little bit less of a clear pattern, actually. So the gradient seems more pronounced for education than for income. There’s a bit of a technical point here, which is that we don’t know someone’s exact income in the LinkedIn data. You don’t put them on your LinkedIn profile, mostly. So we infer it from where people live, based on the block group that they live on, and so you might worry that’s sort of an inexact measure of income, but the education measure we have is more specific. We try to do some things to alleviate that concern.
Overall, we stand by the idea that the gradient is stronger for education than for income.
Demsas: But for income, higher-income people are more likely to have a more-segregated workplace?
Chinoy: Barely, among our sample. And our sample is people who have LinkedIn profiles, and so that’s a higher-income slice of the population than the overall workforce. And among that sample, we don’t see too much in terms of the highest-income people among them experiencing more segregation than the lowest-income people.
Demsas: This is a bit afield from your specific paper, but I remember there being a lot of talk about how diverse workplaces were more creative, and I think that literature is actually kind of more mixed, so I don’t know how good that literature actually stands up. But there’s a lot of talk about how having kind of ideological diversity, background diversity, etcetera can make for more creative teams. Does your research look at whether these sorts of workplaces, you know, have any impact on productivity? Do you have thoughts on whether that would play out, given other research you’ve looked at?
Chinoy: Yeah, it’s a great question. I think that’s the next paper I want to write.
Certainly, there are countervailing forces here. It’s probably bad for, you know, workplaces to have 100 percent people who think a certain way. That seems not optimal. It also seems like you probably want to be able to get along with your co-workers to some extent. And, you know, if partisanship and political affiliation is a measure of that, then, probably, a little bit of homophily might actually be productive. So my speculation is that there would be a bit of a U shape there. I don’t know for sure. I think that would be super interesting to study.
I think that where people do study this question, specifically, is among corporate leaders. It’s been, like, historically, a little bit easier to get information about the political affiliation and donation behavior of executives and board members, and so people have focused on that. There is one paper that basically claims that increasing political polarization of corporate America is not in the financial interest of shareholders. I’d have to remember exactly what they study in that paper, but I think they’re looking at the alignment between corporate executives and their boards, the political alignment, and looking at what happens when they leave.
Demsas: Yeah, we can put that paper in show notes for people to take a look at if they want.
There’s another paper—also, we’ll put that in the show notes—from Christopher Rosen last January [2024], and they found that employees experience negative affect after overhearing political conversations at work. Essentially, the effects are amplified when employees think their co-workers are less similar to them. So you’re more negative when you overhear someone who’s a Democrat, and you’re a Republican having kind of a conversation. And they’re attenuated when you overhear someone who has your viewpoints or you feel like it’s aligned with your ideological or partisan goals.
And so that seems pretty straightforward there, and I agree, these countervailing forces here feel difficult to sort out, particularly because an individual firm’s goal might be to increase the amount of good feelings that people within their company feel, but an industry or our goal as a society to try to create the most productive companies might be to have a lot more frictions happening in the workplace for the societal benefits that might bring.
It’s also like, the incentives are also countervailing here. There’s not really an incentive from workplace leadership, maybe, to try and make their workplaces more diverse in ideology, which I feel like is why there was such a push to try to find productivity benefits from ideological diversity—to try to incentivize this kind of corporate shift. But it seems rough.
Chinoy: I think that’s a good point. And I think the other thing I would add is that the conversation we just had is kind of focused on the interests and the efficiency and productivity of companies and workers. We also might have a social interest in partisan mixing and people who don’t think the same way politically interacting with each other.
Again, this is the kind of thing that, on its face, seems right. Like, we probably want people to interact with people who don’t think like them. Actually saying why that is good, politically, is not terribly obvious. Is it going to reduce support for, like, political violence or things like that? Probably not.
I think people have shown that people don’t generally believe in that kind of thing anyways. But you know, the extent to which I think mixing between Democrats and Republicans is good for our politics, I think that’s sort of another reason to be interested in this issue.
Demsas: Yeah, I want to emphasize for listeners, we’re not saying that, like, all these workplaces—and you’re not finding that all these workplaces—are 100 percent Democrat or 100 percent Republican. It’s just more likely to be. And so people might look around and say, I know the conservative at my job, and it’s like, Yeah, you know the conservative at your job.
So I think what you just said kind of segues into another finding in your paper, which is that there’s a persuasive aspect to this too. Just to tell you my prior, I feel like if I was constantly surrounded by very right-wing people at my job, I would probably only become more left-wing. But I do think that maybe your paper indicates that that effect is not the same for everyone—or maybe I’m just wrong about myself. Maybe I would be persuaded. But tell me about that finding. What did you see?
Chinoy: Yeah, so the story there is mixed. I don’t think what you’re saying is wrong about yourself. So the idea here is: What could explain political segregation? Well, one channel that could explain political segregation is sort of this conformity effect. People become like their co-workers, like their workplace over time. If I end up in a workplace with a lot of Democrats, I might be more likely to affiliate with the Democratic Party, and sort of vice versa with Republicans.
This has been shown, actually, with neighborhoods, and so people tend to adopt, to some extent, the partisanship of their neighbors or the people that they live around. And so we were interested in whether something similar could be happening at work. I’ll spare you the details of exactly how we estimate this, unless you want me to get into it, but the basic finding is that we find that the workplace can causally affect people’s partisan affiliation but only for people who don’t start out as committed Democrats or Republicans.
And so you’re not really getting people to change their mind. I think that’s kind of consistent with, perhaps, the story that you were telling about yourself. But people who start out as either independents or who start out as not registered with a particular political party, we find that moving to a workplace where the co-workers are more Democratic or more Republican tends to make those individuals more Democratic or more Republican, on average.
So there is some evidence of a little bit of an effect of the workplace on an individuals’ partisan politics or party affiliation. It’s not as big as in the case of neighborhoods, and so it seems like this channel has less power to explain the segregation that we see.
And in particular, the timing of this is kind of interesting. It looks less like the case that people switch to a new workplace and then adopt the politics of that workplace, and rather the case that people update their own party registration and then move to a compatible workplace, a co-partisan workplace. And that’s what kind of leads us to think, like, maybe it’s less the case that people are picking up the politics of their workplace and more the case that people are selecting jobs or workplaces, in part, based on politics or things correlated with their own party affiliation. And that’s the direction that we go in the paper.
[Music]
Demsas: After the break: how firms use partisan language to appeal to Democratic versus Republican job seekers.
[Break]
Demsas: What is driving this, I think, is a very useful thing to spend some time on now. I could theorize a bunch of different streams by which partisan sorting shows up, like word of mouth and recommendations, or it might be driven through partisan networks, or Democrats are more likely to be academics, and Republicans are more likely to be petroleum engineers or business owners or whatever. And when you’re able to drill down into how people are sorting, what part of the employment timeline is this actually coming up in?
Chinoy: Yeah, that’s right. So it’s not a straightforward answer, in the sense that certainly it’s the case that we can see in our data that Democrats and Republicans are choosing different schools and majors and occupations and industries. And so, clearly, all of that matters. It sort of limits the available workplaces and the kinds of people that you could even possibly interact with at work. That is something we can account for statistically when we measure segregation. I kind of described how we did that earlier, and we still find that there’s residual segregation.
So what is explaining that residual? There are a couple of different ideas. And I think the two probably main ones are workers selecting partisan or compatible workplaces, or it’s some kind of employer-discrimination channel, where employers are hiring people of a particular party or want things correlated with a particular partisanship, partisan identity.
And we focus on worker selection in this paper, kind of for the reason that I mentioned, that suggestive timing of when people are moving to co-partisan workplaces after sort of updating their party registration. And we focus on the worker-selection channel. And we try to say, you know, Is it the case that workers are actually selecting jobs based on something related to their political identity and something related to how they perceive the politics of the company?
We do this in a survey, and sort of the key question, I think, to begin studying this is, you know, What do workers actually know about a company when they choose whether or not to apply for a job there? It could be the case that segregation is driven because Democrats, you know, want to work at companies with more Democrats, but, like, is that something you really know when you apply for a job? You don’t know that it’s 60 percent Democrats versus 40 percent or something like that.
And so to study this, we look at our big data set that we’ve assembled of all these companies and shares of Democrats and Republicans who actually work there. And we look at how these companies are signaling. We look at the language that they used to describe themselves, and we see how that correlates with partisanship, with the shares of Democrats and Republicans who are actually at that company. And we actually find that there’s quite a bit of signal here, that the Democratic companies are advertising themselves in a way that’s quite different from the Republican companies, even within the same industry.
And a lot of the actual signaling language probably won’t surprise you. It’s words related to the environment and diversity and community for the Democratic companies, and sort of the absence of those for the Republican companies. But the fact that these signals kind of come through so clearly in our data kind of leads us to study the extent to which this can drive sorting. And I’ll pause there for a second, but I can tell you more about that.
Demsas: I would like to talk more about that because I saw the ideological signals and company descriptions, and this is on LinkedIn, right? So how are Democratic firms versus Republican firms describing themselves?
Chinoy: Yes, exactly. So, again, I want to emphasize this is all within industries. So we’re not just comparing, you know, nonprofits, which tend to have more Democratic employees, to, say, energy companies or oil-and-gas firms, which tend to have more Republican employees. This is saying, Take two firms within the same industry. Look at the text that they use, the words that they use on LinkedIn to describe themselves. And we find empirically that there are these words that are quite correlated with the partisanship of the employees.
And again, it’s a lot of the kind of bundle of things related to ESG practices and things related to diversity initiatives and things that are related to more subtle, perhaps, things, like, We’re a company that really emphasizes community and teamwork among our co-workers—that tends to be empirically more Democratic—versus, We’re a company that really emphasizes customer service and efficiency and excellence. That tends to actually be more correlated with companies that have more Republicans, which maybe wasn’t necessarily obvious to me.
Demsas: Yeah. I saw this other paper come out recently by Erika Kirgios and her co-authors that looks at whether communicating measurable diversity goals attracts or repels historically marginalized job applicants. It’s a bit orthogonal to the broader conversation, but I think it plays into this part of your paper quite well.
They look at whether “adding a measurable goal to a public diversity commitment,” like, instead of, quote, “We care about diversity.” You might say, like, We care about diversity and plan to hire at least one woman or racial minority for every white man we hire. And they look at whether that impacts application rates from women and racial minorities. They find that it increases application likelihood among those groups by 6.5 percent, without sacrificing candidate quality. Interestingly, it’s mostly driven by white women. I’m not even sure that the racial-minority finding is statistically significant, though they do find that it’s positive.
I think about this in relation here to whether there are different subgroups that are more motivated to find a job with more of their co-partisans at work and how that changes with racial and gender—different subgroups. I don’t know if you have thoughts on that.
Chinoy: Yeah, that’s interesting. There’s a lot of interesting work studying how these kinds of job ads affect who applies. The other big one I alluded to before is about ESG practices. And there’s another case, I think, where you can think of ways to make that, like, verifiable. You could say that this company actually has some particular ESG designation.
And I think what is maybe interesting about our survey and our paper is that we actually aren’t signaling anything that is explicitly verifiable. There’s no stamp associated with, you know, a company that is more pro-Democratic or more pro-Republican. And yet, this sort of matters. And yet, job seekers clearly seem to pick up on these things and care about them.
Demsas: I think someone listening to this might feel like, Okay, is this just a function of this current moment right now? Like, overt partisan politics in the workplace is much more commonplace in recent years. And this prevailing narrative about, like, Politics shouldn’t be discussed at work. You know, Politics, religion—you know, leave that outside the workplace. That’s kind of an older view of the workplace.
When exactly were you conducting this? And do you have any sense of whether or not this is just, like, a 2017–2022 moment?
Chinoy: Yeah, it’s a great question. So to answer the question, our LinkedIn data is a snapshot from 2022, and our voter-file data, which we’re kind of using to track how people’s party affiliation changes as they move from workplace to workplace, starts in 2012. And so it’s covering, you know, a more recent time period, for sure.
The question of, Can we measure how this is changing over time? I think it is super interesting and a little bit hard, in the sense that you can see in our LinkedIn data, where everyone was working in 2010, but then you worry, Who are the kinds of people who have gone through and listed where they worked in 2010? And so you worry about selection, and so I find that kind of a hard question to answer.
I think that there’s suggestive stuff. So, you know, if you ask people in surveys if they’re willing to leave a job over political differences, you’ll find that it’s the case that young people are much more likely to say that. Now, was that true in 1962? I’m not quite sure. But that sort of points in that direction. You find that in other countries—in Brazil, for example—that this kind of political assortative matching has been increasing over time. You find that among corporate boards, again, this kind of thing has been increasing over time.
And so I think there are a lot of sort of suggestive indications that this might be something that is more pronounced today than it was a couple of decades ago. It’s really hard to say for sure, though.
Demsas: Yeah. It’s funny, too, because part of what’s happened over this time period, at least in the United States, is that our parties have become much more sorted on ideology. And so in the 1950s or whatever, there were a lot more conservative Democrats and liberal Republicans, and that has changed significantly over time. And now most people who are ideologically liberal have sorted into the Democratic Party, so it becomes kind of difficult to measure this if you’re looking at just partisan measurements. Like, maybe there was more partisan diversity at workplaces in 1962, but ideological diversity was still really low, and people were still sorting. So I think there’s quite a difficulty with measurement there.
Chinoy: That’s a good point. It makes me sort of think of one other point, which is that for a smaller set of our sample, about 10 percent of our sample, we can measure their donation behavior. And so there, we can actually look at sort of a within-party measure of sorting. We can say, Is it the case that the more liberal Democrats, as measured by who they’re actually donating to, are sorting into workplaces with more Democrats, and vice versa for Republicans? And there, we also find pretty strong patterns. So there is an extent to which this exists even within party.
Demsas: Another question I have is about, like, obviously 2022—pretty tight labor market. So workers had tons of choice and ability to sort based on a bunch of different amenities. And I wonder if you think that this kind of sorting happens less in a high-unemployment environment. So, you know, God forbid, when there’s a recession at some point, or there’s a period of high unemployment, do you expect this kind of ideological sorting to go down?
Chinoy: Yeah, so I think the answer is yes. And I think that, again, that would be a pretty cool follow-up paper. I think that’s something that we can probably study directly in the data that we have.
I think that one piece of evidence here is, you know, how important are these kinds of what we call “ideological amenities” or “partisan amenities”? How much are these characteristics about a job—how much do workers value them, relative to other things that they might care about in a job? Our survey is designed to precisely measure the quantitative trade-offs people would make for these kinds of ideological amenities, you know, trading off against wages. It turns out to be about 3 percent of their salary.
Demsas: (Laughs.) That’s wild.
Chinoy: Is that big or small? I think it depends on your priors. But, you know, it’s certainly way smaller than what people would pay for health care or, you know, some of these sort of more—
Demsas: How do you measure that? How do you know someone’s willing to trade 3 percent of their salary?
Chinoy: Yeah. So this is kind of what I was getting at before. We pick up these ideological signals in the way these companies describe themselves. We use them to generate synthetic job ads. We want to ask workers in a wide range of occupations and industries about these different kinds of jobs. How do we do that without me sitting down and writing, you know, 10,000 job ads?
This turns out to be exactly the kind of thing that ChatGPT is good at, a large language model is good at. We can give it these kinds of ideological signals that we find, in our data, are correlated with companies with more Democrats and Republicans. We can give it a particular occupation and industry, and it’ll come back with a job ad that does emphasize these signals or doesn’t emphasize these signals, and then we ask workers about them in an online survey.
We ask them to explicitly make choices between these companies that are framed in different ways, and we vary the wages associated with these job ads, and that’s how we can sort of capture the strength of this trade-off. And so that’s where we get this 3 percent number. Is it big or small? It’s less than something like health care. It’s sort of actually comparable in magnitude to some of these softer amenities, things like having a relaxed versus a fast pace of work, for example, or having training opportunities at work.
People seem to care similarly about the ideological nature of the job. Of course, the difference is that our ideological amenities are precisely designed to split Democrats and Republicans. You know, Democrats care about the liberal one, and Republicans care about the other one. Whereas, Democrats and Republicans care similarly about, say, a relaxed pace of work. And so those other amenities can’t generate segregation, but the stuff that we study and design actually can generate segregation.
Demsas: I’m not surprised that people would trade off a little bit on wages in order to feel more comfortable at work with their ideological co-partisans. But I wonder if you were to tell people, Hey—your revealed preference is that you would sacrifice, like, X thousand dollars a year. Do you actually want to take that trade? With remote work, for instance, they’re doing these experiments now where people are like, Yeah, I will take a pay cut in order to be able to be fully remote. I wonder if people would explicitly say, Yes, in order to be at a more Republican firm, I will give you $3,000. I wonder if that’s a self-conception problem that we might run into if you made that explicit.
Chinoy: Yeah. So certainly we’re asking people in the survey to kind of make these trade-offs explicitly. You know, it’s job A or job B, and it’s $3,000 or not. I will say, also, that in the observational data, we don’t know individuals’ wages directly.
Again, that’s the problem I mentioned. People don’t list their salary in their LinkedIn profile, but we do know something about where they live and their occupation and their industry. We know what college they went to. And so we can take similar Democrats and Republicans—similar in terms of their demographics and where they live, and in terms of what exact college they went to, which is a pretty good measure of education or perhaps labor-market skills—and find that the Democrats are consistently choosing occupations and industries that pay less than the Republicans. And so there’s certainly some evidence, or some suggestive evidence, that there’s some trade-off that people are making between fit with the workplace, or their job more generally, and the actual salary associated with that job.
Demsas: There’s Gallup polling from February of last year that asked about U.S. employees’ experience with political conversations at work. And conservatives were much more likely to say that they had a discussion with co-workers about politics: 60 percent of conservatives versus 48 percent of liberals. That is contrary to, at least, my expectations. I’d expect parity, or maybe liberals would be doing it more. I don’t know why I had that expectation, but I was surprised.
It makes me think that maybe there’s some sort of mobilization aspect happening here, if conservatives are saying, in 2024, that they’re having conversations with co-workers about politics, and then, all of a sudden, conservatives win a trifecta. I don’t know if that’s playing into it.
Chinoy: That’s interesting. That probably goes against my priors a little bit too. I think I would have expected liberals or Democrats to be having more of these conversations at work. That’s interesting.
I think that, certainly, studying mobilization—it’s actually not clear to me, right? If you’re part of the majority group at your workplace, and then everyone’s like, Hey, let’s go vote for our guy, for our candidate. Is that actually going to make you more likely to turn out? Or is there some sort of backlash effect if you’re a minority and you say, you know, I really hate all these conversations I’m having with my co-workers. I’m going to go try to vote them out of office, or something like that? It’s not super clear to me what direction that goes in. I think that it is a great question.
Demsas: I’m revealing a lot about my psychology in this episode, going, like, Well, if I had 60 percent people who disagreed with me, then I’m definitely gonna go vote, you know? So maybe that’s not the average experience for people. (Laughs.)
Chinoy: Well, to tie this back, again, I think we find these, like, pretty heterogeneous effects on partisanship for people who start out as committed, versus not. And so I think there’s some sense in which maybe people who are younger and who are susceptible to political influence might adopt the politics of their workplace and perhaps turn out, and then people who already have a particular ideological stance or particular partisan attachment might be motivated to turn out as a backlash against the prevailing politics of the workplace. I’m not sure. I think that’s an interesting question.
Demsas: So we mentioned a couple times—I mean, you’re on the job market right now. You’ve mentioned academia a bit. I mean, have you seen any of this playing out in your own field? Like, this kind of sorting?
Chinoy: Yes. And I think that one fascinating thing, I think, is the sorting across college majors, which is something that we can see—sorting across colleges but also across college majors. And we can see this explicitly in our data. It turns out that economics is pretty much in the middle, which, when I tell economists, makes them very happy that it’s a discipline that doesn’t necessarily lean so far one way or another.
But certainly, higher education leans to the left in our data. Certainly, elite colleges lean to the left in our data. Certainly, many academic disciplines lean to the left in our data. And so I guess for lack of a more sophisticated answer, if you’re looking for a place with a good deal of partisan segregation, looking at universities is not a bad place to look.
Demsas: The mechanisms are interesting because you have this self-sorting. You lean a lot on people choosing these sorts of majors that are kind of correlated with their partisan identities or may help shape their partisan identities. And then they choose workplaces and things like that, and colleges, and down the line, etcetera.
But is there any impact that you can find on the employer side of selection? Like, I don’t know if you’re experiencing this at all, too, but there’s some level to which, when you’re in a job interview, they’re trying to suss out if you’re a good fit for the company. And part of that fit, I assume, might be ideological or partisan.
Chinoy: I think that’s absolutely right. And so I think that just isn’t the main mechanism or the main channel that we study in this paper, honestly, because we have to choose something. And so we focus on the worker side. Again, there’s evidence—from Brazil, in particular, there’s a paper looking at employer, you could call it, discrimination. There are audit studies in the U.S. looking at callback rates for résumés that signal whether you’re a Democrat or Republican, and there’s some evidence for that as well. So yes, I do think that kind of employer-selection story is certainly part of what’s going on.
Demsas: I think your focus on the employee side is actually really interesting, because I think it raises questions about allowing this kind of free choice, how that can lead to, maybe, societally suboptimal ends. There’s a new paper in the American Political Science Review from Jon Green and several of his co-authors that looks at demand for echo chambers.
And we think about echo chambers, and we’re like, Oh man, social media. It’s such an echo chamber. We’re talking about it as if it’s kind of imposed on us. And their intro of that paper has something interesting. They argue that “networked curation processes lead information consumption on social media in particular to be more politically homogeneous than [this] empirical literature has thus far suggested. However, this is more a reflection of democracy than a threat to democracy—a product of individuals engaging with information, and each other, on their own terms.”
Essentially, people are choosing to follow certain people. They follow creators. They follow influencers. They follow their friends. They follow people that make them feel good about themselves. There was this big outrage recently. I don’t know if you noticed—people were following the VP account on Instagram and then were shocked to realize that they were now following J. D. Vance, and now a mass unfollowing happened. And it’s like, you had curated your Instagram feed to be people you wanted to follow, and all of a sudden you see, like, J. D. Vance being inaugurated on your feed, and you’re like, How did I get on here? What’s going on?
And it’s an interesting question about—in previous generations, people went and they bought a newspaper, and you couldn’t just choose to take the parts of the newspaper that you wanted. You had to take it all. And you don’t do that anymore. So I don’t know if you have thoughts on that. Or should you want to change this, are there even ways to change this?
Chinoy: Yeah. I think that’s a really good point and interesting question: Is partisan segregation bad? Should we be worried about it? I think it is a very important, interesting question.
I will note that when labor economists study racial and gender segregation at work, they have a different set of motivations in mind. They have equity motivations in mind. These are protected categories. We really are worried, in particular, about differences in pay between groups doing the same work, whereas I don’t think we have the same equity motivations for being worried about Democrats and Republicans at different workplaces.
Demsas: DEI for conservatives in the academy is a very controversial conversation. (Laughs.)
Chinoy: Exactly. Yeah. So without going there, I think that we have different motivations to study partisan segregation at work versus these other forms of segregation.
I think that—and this is kind of getting to a conversation that we had a little bit earlier—it’s probably good that people get to do what they want and put themselves in workplaces where they feel happy and where the organization kind of aligns with their goals. It’s probably good that people work with co-workers who they get along with. It probably makes them, to some extent, more productive. Does it make the actual firms more or less productive? I think that is an open question, and certainly one that factors into this calculus about whether we think this is ultimately helpful or harmful.
And then, of course, I think the other really hard question to answer, which we talked about before, is: What does this do for democracy generally? Should we think that it’s bad that this place where we thought that partisans might be mixing more than other environments actually isn’t going to provide that kind of kind of mixing? What are, exactly, the consequences of that lack of contact between people who don’t think the same way? I think it is a hard question to nail down. I think we have this intuitive sense that probably it’s not so good if we really segregate ourselves politically. But actually quantifying or measuring and thinking about, What are the effects of that? I think is still an open question.
Demsas: One note of hope I might relay here is that—I thought about this in the context of my job, which is, obviously, not the average job in the United States, but I come into contact, in the context of my work, with people who don’t work here, all the time.
So for instance, I might come into contact with people who I’m interviewing, who are different from me, when I’m walking down the street, doing man-on-the-street interviews. But if I also conceptualize other jobs—if you’re in the service industry and you’re a restaurant worker, maybe all your co-workers behind the bar are on the same team as you, but you’re serving customers and talking to them and interacting with them, and that may also lead to a lot of that cross-partisan contact.
I think it’s both difficult—impossible, maybe even—and undesirable from a business perspective to be able to even do that sorting. Obviously, at some level, businesses do this, right? Like, if you have a rainbow flag in the window or something, you’re signaling to people. But, you know, in general, most jobs force you to interact with people outside of your workplace. And that sorting may happen much less in that context. So you could think of, like, your workplace as your home, versus, you know, when you go outside, and then you’re like, Okay, well, I’m interacting with people who are different than me, but I have a place to go back and, you know, dish with my co-workers about how rude they were.
Chinoy: Yeah, so I think that studying the extent to which different occupations interact with customers or with the public and whether that sort of has some bearing on these effects on political views, I think would also be interesting. When they say the workplace is a context for this kind of cross-cutting discourse, I think what they usually have in mind is, like, with fellow co-workers. But certainly, you’re right that those aren’t the only people that you interact with in the context of work. And so that would be super cool to study.
Demsas: Well, that’s a great place for, I think, our final question, which is: What is an idea that you once thought was a great one but ended up only being good on paper?
Chinoy: So I thought about this a bit, and if you’ll forgive me, I think I’m going to mention another academic paper, which is related but not exactly the same. So one thing that I’m super interested in, as someone who’s interested in politics and demographics and data, is the extent to which these demographic characteristics, some of which we’ve been talking about, are really predictive of politics and party affiliation and things like race and gender and age.
And I think what is tempting, then, is to say, There are these strong correlations that exist between politics and demographics. We know something usually about demographic trends in the U.S., whether a particular racial group is growing, whether people are becoming more educated, on average, or not. And so using that to make predictions about what’s going to happen to politics and to elections, I think, is really tempting.
This is the idea that, you know, people are becoming more educated, on average, and more-educated people tend to vote for Democrats, and so the Democrats are going to do better in the future. There are various versions of this argument, and it’s quite tempting to make, but it turns out it doesn’t really work.
There’s a paper by one of my advisors, Vincent Pons, as well as Jesse Shapiro and Richard Calvo where they test this. So they look at the correlation between politics and demographics in a particular election. They say, If these correlations were the same in the next election and we sort of just tracked the evolution of demographics from election to election, how good would that prediction be? And it turns out that it’s quite bad. It’s sort of worse than just guessing that it’s gonna be 50–50, Democrat or Republican.
And I think that sort of goes to show that—it’s kind of interesting that these demographic correlations are so strong in the moment. But also, these trends are kind of slow moving, and politics responds kind of quickly, and parties respond to where they see their electoral advantages.
Demsas: Demographics are not destiny.
Chinoy: Yeah, I could have just said that, and I think I probably would have gotten the same point across. But this is a longer way of saying that.
Demsas: No, it’s great. We’ll put the papers in the show notes as well. But thank you so much, Sahil. This was fantastic.
Chinoy: Thanks for your time.
[Music]
Demsas: Good on Paper is produced by Rosie Hughes. It was edited by Dave Shaw, fact-checked by Ena Alvarado, and engineered by Erica Huang. Our theme music is composed by Rob Smierciak. Claudine Ebeid is the executive producer of Atlantic audio. Andrea Valdez is our managing editor.
And hey, if you like what you’re hearing, please leave us a rating and review on Apple Podcasts.
I’m Jerusalem Demsas, and we’ll see you next week.