Are Americans considerably extra divided primarily based on the place they reside and their social identities? Or are tales of voters sorted into neat social and geographic enclaves overstated? Seo-young Silvia Kim finds that it’s not really easy to foretell how Americans will vote primarily based on their demographic teams—and it hasn’t gotten any simpler over time. David Darmofal finds that demographics are a bit extra predictive of geographic voting patterns, however spatial polarization has not elevated markedly over time. They each take the lengthy view, discovering that we aren’t as divided by social teams and geographies as we appear.
Guests: Seo-young Silvia Kim, American University; David Darmofal, University of South Carolina
Studies: “The Divided (But Not More Predictable) Electorate” and “Demography, Politics, and Partisan Polarization in the United States, 1828-2016”
Transcript
Matt Grossmann: Demographic and geographic polarization is overstated this week on the Science of Politics. For the Niskanen Center, I’m Matt Grossmann.
Are Americans considerably extra divided primarily based on the place they reside and their social identities? It actually appears that manner, with our city/rural divides and our growing divisions on race and training. It looks like it ought to now be straightforward to foretell how a person or a geographic space voted primarily based on a handful of variables, however taking an extended view makes the story extra difficult, with the tendencies a lot much less pronounced.
This week, I talked to Seo-young Silvia Kim of American University about her new working paper with [Yon Zelinsky 00:00:45], The Divided however Not More Predictable Electorate. She finds that it’s not really easy to foretell how Americans will vote primarily based on their demographic teams, and it hasn’t gotten any simpler over time. Instead, voters are more and more divided by partisanship.
I additionally talked to David Darmofal of the University of South Carolina about his [Springer 00:01:04] e book with Ryan Strickler, Demography Politics and Partisan Polarization within the United States. He finds that demographics are a bit extra predictive of geographic voting patterns, however spatial polarization has not elevated markedly over time. They each discover the standard tales of voters sorted into neat social and geographic enclaves to be overstated. Kim says demographics aren’t future relating to Americans voting and haven’t grow to be extra essential over time.
Seo-young Silvia Kim: Try as we’d, demographic labels don’t give a lot details about vote alternative all through the final 70 years. We quantify how a well-performing machine studying algorithm does with simply 5 variables: age, gender, race, training, and revenue, the massive 5 that the individuals suppose is demographics. And we discover that, on common, you may solely predict about 63.5% of the 2 social gathering vote decisions accurately on common, all through these years.
It additionally doesn’t improve over the interval of 1952 to 2020. So I believe this goes in opposition to lots of people’s instinct that the demographic group identities do actually decide political conduct resembling vote alternative. There’s quite a lot of punditry constructed round such notions. And additionally teachers, we consider that demographics is a robust and essential predictor that we should take note of. And on condition that we consider that demographic sorting has taken place, and that social gathering line voting has elevated, it will need to have been a pure conclusion to say that, primarily based on demographics, we will predict vote decisions higher. But it’s not likely that case.
Matt Grossmann: Darmofal finds that present geographic polarization doesn’t stand out. And even the newest choices might mirror outdated patterns.
David Darmofal: There’s actually been appreciable dialogue within the fashionable press about how we’re turning into a extra polarized nation. And a lot of this dialogue is concentrated, in fact, on geographic polarization, this notion of crimson states and blue states, and how we’re divided as a rustic between crimson states and blue states. What we wished to do on this e book was to put this present geographic polarization debate and dialogue in historic context, as a result of we will’t totally perceive the current with out understanding what got here earlier than it and how this up to date period suits into broader historic patterns.
For instance, is that this an anomalous interval, or is it reflective of previous patterns? And one in all our core findings is that we discover that, opposite to many fashionable accounts, we’re not dwelling in a very geographically polarized time. Americans aren’t any extra more likely to reside in landslide counties received by a presidential candidate by 20 factors or extra in latest many years than in earlier eras.
We additionally apply strategies of spatial evaluation to establish the political geography of county stage of voting in every presidential election, from 1828 by way of 2016. So that is mainly the whole interval of mass voter participation within the United States because the introduction of Jacksonian democracy.
And one of many issues we discover then is that the spatial construction of partisan voting, the areas of Democratic and Republican regimes throughout the nation, has modified step by step over time fairly than in a haphazard, election-specific method.
Another attention-grabbing discovering is that you would be able to higher predict how counties voted within the 2016 presidential election between Donald Trump and Hillary Clinton from how they voted within the 1828 election between Andrew Jackson and John Quincy Adams, then from the 1976 contest between Jimmy Carter and Gerald Ford. Strong Jackson counties in 1828 tended to be robust Trump counties in 2016 and robust Adams counties, robust Clinton counties. In distinction, robust Carter counties in 1976 don’t are inclined to differ from different counties of their preferences for Trump or Clinton. And that will nicely mirror that Trump tapped into what Walter Russell Mead and others outline as a populist Jacksonian custom within the United States.
Finally, the concentrate on blue states and crimson states actually distorts our understanding of the political geography of voting within the US. They’re vital inside state variation and partisan voting. And satirically, we’re truly close to a excessive level within the contribution of inside state variation in partisan voting, in distinction to between state variation, exactly as our consideration as a rustic has more and more turned to the state stage debate of blue states versus crimson states.
In quick, there’s quite a lot of reality in former President Obama’s 2004 keynote speech. There’s quite a lot of Republicans in our so-called blue states and quite a lot of Democrats in our so-called crimson states. And if we actually need to perceive the political geography of elections, we’ve to have a look at the sub-state stage. We try this in our e book by specializing in counties, the bottom stage of aggregation for which we’ve full information on the contiguous United States for the total interval of mass voter participation within the US, and that’s the main focus of our e book.
Matt Grossmann: Kim and Zelinsky have been monitoring particular person voters because the Nineteen Fifties, whereas Darmofal and Strickler have been monitoring county stage patterns since 1828. Kim and Zelinsky anticipated to search out demographic sorority, however didn’t discover it.
Seo-young Silvia Kim: We have been questioning aloud concerning the diploma divide that was being repeatedly introduced up by the media. And we have been considering that yeah, yeah, the marginal distributions and the cross tabs in want appear to counsel a demographic sorting. Now, what occurs if we plug that right into a prediction algorithm? Does that additionally present up as we might predict? And we initially thought that this was going to be a very fast train, I believe. We anticipated to see outcomes in keeping with an growing demographic sorting. We anticipated this to finish shortly. Turns out that we have been actually puzzled by the preliminary outcomes.
Matt Grossmann: They discovered that utilizing demographics, a prediction algorithm, can’t get very far. It must know voter’s social gathering.
Seo-young Silvia Kim: What we do is we take surveys in presidential election years, and then take out the voters who say they’ve voted, and take out those that voted for both of the 2 main events. And then we prepare the machine studying algorithm to attempt to predict actually, rather well, given a coaching set about which social gathering this voter voted for.
And we take that actually well-trained machine and then plug an information set that the machine has by no means seen earlier than to the identical machine. And we discover that… I like to do that like a rating. And then the machine solely will get 63.5. That’s a D on my syllabus. Random guess offers you about 50% as a result of most presidential elections are fairly shut. So a low 60s accuracy is saying that the 5 variables of demographics shouldn’t be doing a complete lot. A celebration will get you as much as late 80s, and then on the rating e book, that implies that the machine studying algorithm on the rating e book, that implies that the machine studying algorithm has jumped from a rating of D to a B.
Matt Grossmann: Voter gaps by demographics don’t essentially imply simpler prediction.
Seo-young Silvia Kim: We are a joint distribution of votes versus the marginals, and I believe there isn’t any doubt that on the marginal stage it does appear to be the case that there’s a stronger affiliation for every two social gathering votes. But collectively thought-about collectively, you’re taking a voter and saying that as a substitute of simply this particular person as a Latino we’re now going to contemplate {that a} Latino particular person in his 40s earnings such and such revenue and educated with a school diploma. And then general, the predictive energy actually balances out and doesn’t give an overwhelming predictability to vote alternative. I believe that’s a serious factor. Cross tabs are actually concerning the marginals, however collectively, it doesn’t do very nicely.
Matt Grossmann: And further demographics don’t assist.
Seo-young Silvia Kim: We did attempt including different variables resembling faith. Because our information set is proscribed within the sense that we’ve to make the variables constant from the Nineteen Fifties to 2020, there’s some limitation. But what we added was whether or not the voter was Christian, Catholic, Jewish, or in any other case. We additionally tried including preliminary geographics, which was whether or not the voter lived within the Southern States or not. Neither did actually a lot of something. The accuracy remains to be in its 60s. It doesn’t improve over time.
Matt Grossmann: [inaudible 00:10:58] at spatial polarization, additionally discovering no elevated predictability or an apparent shift to new partisan regimes.
David Darmofal: We discover little proof that latest many years have seen a rise in landslide counties. Instead, the sample is according to earlier eras. Now, there was an uptick on this in 2016. We, in fact, don’t embody analyses for 2020 in our e book. Our e book covers from 1828 by way of 2016. And it’ll be attention-grabbing to see whether or not that continues to extend. But when positioned in a historic perspective, latest many years don’t look like a interval of elevated spatial polarization. And clearly one other facet of this query of spatial polarization is that this notion of Democratic and Republican regimes of partisan voting. Looking at these Democratic and Republican regimes over time they alter step by step over time and in comprehensible manner. And I simply need to briefly talk about the strategies right here that we used to establish the spatial construction of voting conduct from 1828 by way of 2016.
What we’re doing is utilizing a set of diagnostics, the Global Moran’s I and the native Moran’s I, to establish whether or not counties as a complete exhibited comparable partisan voting as their neighboring counties, and we’re utilizing contiguous County as our definition of neighbors right here. And then if we establish that on the international stage or we don’t establish on the international stage, though we do in every election, we then look to the native stage to establish which counties particularly are auto correlated with their neighbors. And so we will then use these native Moran’s I’s with a Moran scatter plot to establish which counties exhibit comparable voting patterns as their neighboring contiguous counties at charges which can be extra Democratic than the nationwide common, that are correlated with their neighbors at charges much less Democratic than the nationwide common, which have increased assist for the Democrats than their neighboring counties, which have decrease assist for the Democratic candidate than the neighboring counties, and which counties are uncorrelated with their neighbors.
And so what we do then use, we use these spatial assessments to establish the place contiguous counties with comparable ranges of Democratic or Republican assist are situated within the United States. And mainly, what we’re right here is areas of robust Democratic assist and areas of robust Republican assist. And what we discover is that these regimes actually change step by step over time in comprehensible methods. There have a tendency to not be robust election to election fluctuations. That differs for some outlier Presidential elections resembling 1964 and 1972, however usually extra Democratic elements of the nation are usually extra Democratic for a number of elections and the identical for extra Republican areas within the nation. And, curiously, the relative areas of Democratic and Republican power stay such whether or not the nation that had a detailed election or a landslide election.
So, for instance, the 1984 and 2000 maps that present spatially auto correlated areas of Democratic and Republican power are fairly comparable regardless of 1984 being a landslide election and 2000 being a fairly shut election. The imply ranges of Democratic and Republican assist shifted between these elections, however the relative areas of Democratic and Republican power stay fairly comparable.
Matt Grossmann: Density does matter however the city rural divide began within the Twenties.
David Darmofal: One of the issues we do in our e book is we do take a look at density as a predictor of partisan voting in every presidential election from 1828 by way of the current. And we do discover that inhabitants density begins predicting partisan voting on the county stage starting within the Twenties or so. So there’s little or no constant sample in that previous to the Twenties or so. But since then, it’s been a reasonably constant phenomenon the place extra densely populated counties are inclined to have increased ranges of assist for the Democratic candidate than do extra much less densely populated counties.
Matt Grossmann: But he does discover proof according to a shift towards nationalized elections.
David Darmofal: One attention-grabbing discovering that will probably converse to the nationalization of the citizens is that this. We make use of a set of spacial fashions for every Presidential election from 1828 by way of 2016, and so the political geography that we establish, the spatial construction of partisan voting that we establish could possibly be produced broadly talking by any of both of two sorts of processes. On the one hand, neighboring counties may exhibit comparable ranges of Democratic or Republican assist, as a result of the residents of those native electorates straight work together with one another and form one another’s partisan voting conduct. So that’s a behavioral diffusion story and that’s according to a spatial lag mannequin the place there are spatial dependents pertaining to the dependent variable.
Alternatively, these counties might not, the individuals might not work together with one another throughout these county traces. They could also be extra atomistic models, however they could be responding to nationwide stimuli. And so their correlation could be then produced not by direct behavioral diffusion throughout these native electorates, however as a substitute on account of admitted co-variates from our fashions. And so any residual spatial auto correlation you discover then is according to a spatial error mannequin. So we run diagnostics, Lagrange Multiplier Test, to find out whether or not a spatial lag mannequin or a spatial error mannequin is extra applicable. And what we discover is that the spatial lag mannequin, the behavioral diffusion on the native stage, appears to be driving the political geography of voting conduct that we establish conditional on our co-variates for many of American historical past till latest many years. Now, what we’ve executed is we’ve seen that in latest elections, the spatial error mannequin is rather more relevant, and that could possibly be according to the thought according to Robert Putnam’s argument that individuals are interacting much less.
… according to Robert Putnam’s argument that individuals are interacting much less, they’re responding as a substitute to nationwide stimuli. Obviously Dan Hopkins’s latest glorious e book on the nationalization of political conduct within the United States speaks to this as nicely. It could possibly be that these native electorates are behaving equally now as a result of they’re all responding to nationwide stimuli.
Matt Grossmann: One factor that could possibly be driving that’s polarization primarily based on social gathering identification. [inaudible 00:18:29] Zalensky tried to determine which components matter past demographics, deciding on half.
Seo-young Silvia Kim: Once that we set up the very fact about predictability of demographics, we have been interested by what variables add power to prediction, proper? And so what we generally take into account are three separate issues, express events and affiliation, partisanship labels, whether or not you establish as a robust Republican, weak Republican, leaning Republican and such. You even have symbolic ideology, whether or not you suppose your self as a liberal or a conservative and the spectrum on that. You even have difficulty positions, whether or not you assist or oppose abortion and such. So we tried three fashions during which we added these three individually to demographics. And we discover that every one of them improve predictability, however partisanship simply outperforms the opposite two.
Matt Grossmann: Party is turning into extra predictive and it encapsulates identities and ideology.
Seo-young Silvia Kim: You add a celebration ID to demographics, and then the scoreboard abruptly jumps to an accuracy of 87.3 ranges. And over time, on common, yearly the accuracy will increase by .18 share factors. So in a decade, meaning virtually two share level improve. But now on condition that it was already fairly excessive to start with, I believe that’s a exceptional factor. I believe the outcomes mirror a partisan, an ideological sorting that we’ve been discussing within the self-discipline.
That as soon as partisan ship is included, the opposite variables don’t appear to do nicely, I believe that speaks to the thought of hardness in ship as a brilliant id, proper? When we take into account individually operational or symbolic ideology as a substitute of partisanship, it did enhance predictability, nevertheless it didn’t do in addition to merely together with the partisanship label. I believe that’s a exceptional proof saying that partisanship is known as a tremendous id that encompasses all the things and sends a a lot, a lot stronger sign than the ideology variables that we often take into account.
Matt Grossmann: Partisanship’s predictive energy may point out that social gathering is an efficient presidential alternative, however that’s much less possible.
Seo-young Silvia Kim: Well, it’s totally doable that it’s not that vote alternative is set by partisanship, however you want a sure presidential candidate and then you definately change your partisan affiliation. In information resembling voter recordsdata, you rarely see individuals altering their artists and affiliation, particularly to not one other social gathering. So I believe the possibilities of which can be uncommon.
Matt Grossmann: Voters could possibly be sorting on identifications and views fairly than group membership.
Seo-young Silvia Kim: First of all, events and ideological sorting and stuff [inaudible 00:21:49] happening. It’s very, very robust. There is little doubt to that. As to social sorting, I believe our outcomes can go hand in hand with the truth that there’s social sorting as a result of we’re utilizing labels as a substitute of id. So we’re saying that you simply belong to this group, we mechanically assume that you simply carry this group id. But the low stage of power that we discover in projectability would possibly point out that it’s not likely concerning the goal labels which can be placed on you. It’s actually about how strongly you establish with these teams. So I believe our outcomes may go hand in hand with that and say that these goal labels that we actually construct our discussions round, they don’t appear to be as robust as we consider them to be. Instead, it’s actually about how strongly you consider this group [inaudible 00:22:51] and a gaggle voting ought to be or the way you establish strongly with these [inaudible 00:23:00].
Matt Grossmann: Kim says partisan alignment is actual, however not realignment primarily based on demographics.
Seo-young Silvia Kim: The two tales of realignment all collectively. So there’s a partisan realignment, the power of events. There was an period of independence and then we’re abruptly figuring out extra and extra social gathering, we’re extra concerned in politics. And that’s the alignment story that I see in a single finish. And I believe that’s undoubtedly true there. But the second realignment could be about social teams. And I believe whereas social teams can have many nuances and layers, as a result of these are the simplest labels we will muster, we are inclined to concentrate on these teams resembling age and race and gender and so on. And I believe in that sense, we’re saying that these labels, they may not be doing a lot. The power of realignment won’t be so pronounced as to be proven within the information.
Matt Grossmann: And she says that geographic areas may nonetheless be extra predictable primarily based on demographics.
Seo-young Silvia Kim: If you are taking that to a neighborhood stage, I believe that’s a very completely different story. Let’s say that you simply’re a Brooklyn resident that has robust opinions about housing costs, costs in inexpensive housing. And then between that, I believe inside these small neighborhoods, I believe demographics can do loads to find out vote decisions come what may. So I believe our outcomes can undoubtedly go hand in hand with the outcomes on spatial polarization, as a result of that has quite a lot of details about the place the voters reside and we don’t.
Matt Grossmann: [inaudible 00:24:56] agrees that demographics would possibly predict mixture voting with out predicting particular person stage voting.
David Darmofal: Clearly you would discover that demographics wouldn’t essentially more and more predict voting conduct on the particular person stage. But you would possibly see if there’s migration or differential generational substitute that depends upon demographics, we may see demographics more and more predicting mixture voting in a manner that it doesn’t, once more, more and more predict particular person stage voting. In different phrases, if counties or different models have gotten extra demographically homogeneous, you’d probably see stronger democratic or Republican assist in these models, even when it’s not rising in its results on the particular person stage. So for instance, let’s say there’s Latino or Latina voting. Let’s say that that doesn’t predict voting any extra.. Maybe it was rising in its power for a number of years, however that’s leveled off now. But in the event you had growing in migration of Latino populations, that are nonetheless strongly Democratic, you’ll see probably a robust relationship between the native demographic make-up of counties and their partisan voting.
Matt Grossmann: But mixture stage relationships can mislead. Black inhabitants was related to Democratic voting when blacks have been prevented from voting within the South primarily based on assist from Southern Democratic segregationists. After the Voting Rights Act, it was finally related primarily based on enfranchised black voters. [inaudible 00:26:28] and Strickler take a look at demographic predictability over time, discovering robust and rising relationships, together with black and immigrant populations.
David Darmofal: Other covariates that we do study as nicely are, for instance, proportion African-American, the African-American proportion of the inhabitants within the county and the immigrant populations in counties as nicely. And so mainly what we discover is that the proportion African-American in a county, the scale of the native county stage black inhabitants, is a reasonably robust and consistent-
Well, a black inhabitants is a reasonably robust and constant predictor of partisan voting on the county stage, for fairly a very long time now. Going again, for instance, to 1900, we’re discovering this constant sample. And mainly what we’re discovering is that counties which have bigger African American populations, are usually extra more likely to vote Democratic.
And we truly see rising and stronger impact of that within the Sixth American Party system, within the present American social gathering system, because the late Sixties. Interestingly, once more, the scale of the native African American inhabitants is positively related to assist for the Democratic Party on the county stage in all of those elections, nevertheless it was not, in significantly anomalous elections. So, the 1964 presidential election is the one election that we discover, going again fairly a methods, during which bigger black populations have been related to decrease assist for the Democratic candidate.
And in fact, 1964 is previous to the Voting Rights Act of 1965. And you would nicely see this as probably an instance of white voters in counties that had giant black populations voting for Goldwater. Of course, we don’t have the person stage information, so we don’t know what’s producing that, nevertheless it stands proud fairly clearly because the one election during which bigger black populations have been related to stronger assist for the Republican candidate.
And then the opposite factor we do study is immigrant populations. The dimension of the proportion of overseas born in a inhabitants, in a county, fairly. And there we discover fairly constant outcomes because the Clinton years, mainly, after many elections the place there wasn’t an impact, whereby counties which have giant immigrant populations have stronger ranges of Democratic assist. And this might converse probably to Democratic beneficial properties because the nation has diversified, demographically. And these areas with giant overseas born populations are having a very robust, pro-Democratic impact of getting native immigrant populations.
Matt Grossmann: Kim additionally says demographics may nonetheless be crucial to prove, which could additionally present up in these geographic divisions.
Seo-young Silvia Kim: Because you’ve already conditioned the turnout, you would possibly discover just a little relationship with demographics and vote alternative, but when demographics has a bigger impact on figuring out turnout, we already know that we see a big turnout differential by demographics. And I believe that implies that it’s not that demographics shouldn’t be helpful in anyway, it’s that after these individuals have determined to prove, it’s much less to be informative about vote alternative.
Matt Grossmann: But she says we’ve gone too far in emphasizing demographic divides.
Seo-young Silvia Kim: Journalists, practitioners, they have to actually take into account rethinking the practices of generally studying an excessive amount of into marginal modifications in every polling, every post-election cross tabs.
I believe generally the soundness of voting conduct actually outweighs what we see to be very marginal modifications in every election. And I believe this actually is compounded by the horse race protection in primaries and normal elections. In my opinion, I believe there’s additionally a possible hurt by such reporting, that by emphasizing the divides by demographic teams, you create an affective emotions of distance between demographic teams. And I don’t suppose there was an ample evaluation of potential adverse results of such media and [inaudible 00:31:24].
Matt Grossmann: She’s now whether or not our overemphasis has led to polarization.
Seo-young Silvia Kim: I’d have an interest to see whether or not the elite communications have actually influenced public perceptions of the demographic divide, out of proportion. If that’s the case, I believe that could possibly be one other supply of affective polarization and emotional distances between teams. And I believe that’s actually dangerous for the democracy. And I need to see if there are methods during which to cut back, and bash these myths if we will.
Matt Grossmann: There’s much more to study. The Science of Politics is out there biweekly from the Niskanen Center and a part of the Democracy Group Network. I’m your host, Matt Grossman. If you favored this dialogue, you could need to try the podcast, Our Body Politic in our community, together with episodes on talking with Hispanic Republicans and ladies of coloration within the Republican Party. Or our prior episodes on explaining the city rural divide, or decoding the early outcomes of the 2020 election. Thanks to Seo-young Silvia Kim, and David Darmofal for becoming a member of me. Please try The Divided (But Not More Predictable) Electorate, and Demography, Politics, and Partisan Polarization within the United States, 1828–2016, and then pay attention in subsequent time.
