The fragile blue wall: analyzing geographies of the 2020 US presidential election

US presidential elections are peculiar contests based on mediation by an Electoral College in which votes are aggregated on a state-by-state basis. In 2020, as in 2016, the outcome was decided by a set of states where the two candidates were equally competitive: Michigan, Pennsylvania, and Wisconsin. Two geographical stories tend to dominate accounts of what happened in 2020. The first story is based on red (Republican) versus blue (Democratic) states, and the second story relies upon rural versus urban biases in support for the two parties. After showing how and where Donald Trump outperformed the expectations of pre-election polls, we consider these two geographical stories both generally, and more specifically, in relation to the crucial swing states. Through an examination of the successes of Joe Biden in Arizona and Georgia, two states long thought of as “red”, and the role of the suburbs and local particularities in producing this result, we conclude that the polarization of the United States into two hostile electorates is exaggerated. 

Only once in the past forty years has a US president been denied a second term in office. In 2020, President Donald Trump's 46.8 percent of the vote share was surpassed by the 51.3 percent for Joe Biden. Despite desperate social media efforts, public denials, legal actions, his refusal to concede, and the incitement of a violent and deadly insurrection, President Trump lost the election. Only the Electoral College's bias towards states with a knifeedge polarization between the two major parties, the Republicans and the Democrats, in which rural voters have a heavier weight in the outcome saved Trump from a crushing defeat. The US presidential election is an indirect election with votes aggregated individually by state to determine the outcome. It was arguably an existential election in the sense that Trump did not compete on policies as much as proposing himself alone as a representative of "true" Americans, and the 1950s America he was in the process of resurrecting, like he had done for himself after his own bout with Covid-19 and his miraculous "cure" in early October 2020 (O'Toole 2020).
Trump is a national-populist who portrays himself as an outsider, even though his entire business career in New York commercial real estate had been dependent on lobbying politicians and exploiting the income-tax code. He appeals to possible voters more as a Christian nationalist and scourge of the federal government than as a conventional politician, even as his main legislative accomplishments in office were very much in line with conventional Republican party positions -tax cuts for the wealthy and appointing ultraconservative federal judges -since the 1990s (e.g. Jones 2020; Lozada 2020).
The Democratic candidate Joe Biden represents a volatile coalition of groups held together by a loose ideology of inclusion, a commitment to active government, and a horror of Donald Trump. Biden was possibly the perfect candidate to both paper over the cracks in the Democratic coalition, given his moderate bona fides, and to bring back the voters in the swing states of Michigan, Pennsylvania, and Wisconsin, the so-called "blue wall", who had voted for Obama in 2008 but then had drifted away from the Democrats in 2016 (Peters 2020). After all, he had been Obama's vice-president and had been born in Scranton, Pennsylvania. In the face of a once in a generation pandemic and in the aftermath of his impeachment for inappropriate pressure placed on the president of Ukraine, Trump was seen as the underdog. A state-by-state predictive model using presidential approval ratings and the condition of state economies estimated a rather accurate outcome (with a reasonable allowance for error) in which the election would come down to the usual suspects: Michigan, Pennsylvania, and Wisconsin, with Biden as the likely winner (Enns and Lagodny 2020).
Long before the election Trump laid out a scenario in which if he lost he would claim the election was "rigged". He would then have his allies in crucial states, and in the courts, decide the election in his favor. He was particularly hostile to the use of mail-in ballots, used to a much greater extent in many states than typical given that a pandemic was raging, suggesting that they were both more subject to fraud and more beneficial to Democrats than had in fact ever been the case on either count in previous elections (Thompson et al. 2020). Trump seemed desperate from long before the election to prepare a fallback for his prospective defeat in which he would be a victim of malfeasance rather than the agent of his own defeat.
The heterogeneous state-by-state way in which federal elections are organized in the US leaves open the suspicion that any innovations, such as early or mail-in voting, could be compromised. Trump took advantage of this to avoid conceding defeat and to raise funds for his future either in or out of national politics. Including Michigan in the strategy proved especially reckless, however, given that Trump lost that state by more than 154,000 votes (Alberta 2020). Recounts only yield a few thousand votes at most and typically only a few hundred. Trying to have state election boards and courts make up for lost votes turned out to be more farcical than he could have anticipated, as his "personal lawyer", Rudy Giuliani, made a fool of himself and his client in multiple failed court filings and in disastrous press conferences in the aftermath of the election (Shubber 2020). The attempt at reversing the verdict of the electorate was based entirely around the notion that the election had been "fixed" by the Democrats in the largest cities in the swing states. Trump went so far as to claim that Biden had to prove that he had indeed won the election (Fox News 2020). America was "a place where there is no such thing as defeat, only broken scoreboards" (Schwartz 2020).
To highlight the peculiarities of contemporary U.S. presidential elections, and to complement the often narrow and complex methods used to study electoral outcomes and political behavior, we offer an accessible approach that blends political inquiry with a few simple maps. Rather than provide incremental confirmations of accepted models of political behavior, our approach frames the 2020 election geographically to show how and where Trump lost, and why he lost. For instance, we show that the big cities were exactly not the places where the election was decided in terms of shifts since 2016, despite Trump's anti-urban rhetoric, and the rediscovery of the urban-rural divide by political scientists (Alberta 2019; Hohmann 2020a; Rodden 2019). We also discuss how the typical framing of voters and the American electorate, from the blind acceptance of census categories to the persistence of the red-state/blue-state dichotomy, in fact contribute to increasingly inaccurate polling, and propagate and perpetuate rather basic and limited understandings of American politics, voters and electoral outcomes. By identifying the limits to such approaches, a more complete understanding of the 2020 U.S. presidential outcome is achieved, as is the possibility of advancing electoral studies beyond ascribed individual voter characteristics.

THE 2020 US PRESIDENTIAL ELECTION
In an earlier paper we adopted a geographical approach to assessing the likelihood that Donald Trump could win again in 2020 by tapping into the places in the swing states of Michigan, Pennsylvania, and Wisconsin that he had switched from supporting Barack Obama in the majority in 2008 and 2012 to his side of the elec-toral ledger in 2016 (Agnew and Shin 2020). These were what we called "the counties that counted." In this paper we revisit the argument of that paper in the aftermath of the 2020 presidential election. First, we review briefly the fact that Trump did much better than the polls predicted both nationally and in the crucial so-called swing states of 2016. We then describe the two main geographical stories about the 2020 election and their respective limits to understanding what happened. This leads us to discuss where in the country Trump "overperformed" in 2020 compared to 2016. Even in defeat levels of support for Trump increased significantly in some places across the United States with respect to both red-and-blue state and rural-urban geographical dimensions discussed previously (Fessenden et al. 2020). As an incumbent president this is perhaps not that surprising, but Trump was a very polarizing figure and never really appealed much beyond the so-called base that he conjured up in 2016 (da Vinha and Ernst 2018).
Voter turnout was also up considerably compared to previous elections from which both presidential candidates benefited, but Biden more than Trump. Biden benefited in 2020 from an anti-Trump boost in voting particularly in the suburbs where Trump had done better than expected in 2016 (Burn-Murdoch and Zhang 2020; Bump 2020a). This was particularly important in Michigan and Pennsylvania where suburban voters were crucial in awarding those states' electoral votes to Biden along with some in places that had swung Trump's way in 2016 but left him in 2020 (Peters 2020;Witte 2020). But these were not so much former Trump voters as they were newly mobilized voters against Trump (e.g. Thomas et al. 2020).
This matters above all in terms of the final determination in the Electoral College. The Electoral College is undoubtedly biased, as is the US Senate, to favor the contemporary Republican Party with its reservoir of support in smaller, more homogeneously white and rural states (Millhiser 2021. This is partly why Trump could almost pull off the feat of winning again even in the face of a massive popular vote deficit. It is where the votes are, more than how many there are, that matters in a US presidential election. This is more so today than at any time since the late nineteenth century. More specifically, medium-size states that have comparable numbers of the polarized on both sides and a pool of non-polarized voters are crucial to the outcome (Smidt 2017). Inevitably, the red/blue state divide is institutionalized even as the processes that produce it are lodged at the more local levels like counties in which the urban-rural divide is not just predominant but in 2020 seemed even more so than previously (e.g. Kolko 2020; Economist 2020a; Thompson 2020a). We should not forget, however, that Trump could not win with rural votes alone. Rural voters only accounted for 14 percent of American voters in 2018. Trump's vote in Los Angeles County, for example, containing a city Trump had decried by saying it "looks like a third-world city," was 1.1 million. This number is equivalent to the same share of his popular vote as the 633 most rural counties combined (Van Dam 2020). Trump also still did well with higher-income voters, particularly men, in rich suburbs (Zhang and Burn-Murdoch 2020). Such voters are a segment of the traditional Republican Party's base that might be uncomfortable with Trump's rhetorical populism but supported his tax cuts and environmental deregulation.
After the review of where Trump over-performed we turn to what exactly happened in the three swing states that mattered most in 2016 and again in 2020 at the scale of counties. We call this the "fragile blue wall" because much of the media buzz about the election referred to the three states as the "blue wall" that Biden needed to rebuild. It turned out to be a fragile one. The purpose is to see which counties switched between the elections and the extent to which it was those counties or other ones that determined the electoral outcome. Younger voters seemed to show up in larger numbers than in 2016, particularly in counties with large universities, and this may well have been crucial for Biden because, if anything, Trump managed to increase his vote with older, white, and non-college educated voters in economically declining areas, with whom he had been most successful in 2016 (Siddiqui and Ngo 2020;Orr 2020).
Again, it was a fairly close run election in the Electoral College, particularly in Pennsylvania and Wisconsin, so Trump's over-performance, particularly in light of pre-election polling results, remains our focus. Perhaps Trump's very active campaigning by way of big rallies, despite a raging pandemic, particularly in these swing states, made a difference. In close races, such as those in the swing states, campaigns do seem to matter, if not more generally, because they tend to favor deepening of partisan polarization and stimulating turnout from targeted groups (Nickerson and Rogers 2020). Yet, in most counties where Trump campaigned immediately before the election turnout went up and his share of the vote went down suggesting strongly that his rallies mobilized opponents more than they did any "hidden" support on his behalf (Chinni 2020).
Finally, we use the cases of Arizona and Georgia, states Trump had won in 2016 but lost in 2020, to consider the relative weight to put on the red-blue state versus rural-urban narratives in assessing the way the elec-tion turned out. The suburbs of the largest cities and, in the case of Arizona, increased votes from people living on Native American reservations, proved crucial. How much did the results there, conventionally regarded as red states, conform geographically to those in the three swing states? Perhaps more states are becoming purple or magenta (e.g. Medina and Stephenson 2020 on Arizona)? What do the results across these five states tell us about likely trends in the future for the two political parties in presidential elections? Is the electorate truly as polarized as the two dominant narratives would have us believe?
Though Joe Biden won the presidency, Democrat losses in other races for Governors, Senators and US Representatives suggest that some voters split their votes and/or voted selectively across races (e.g. Hohmann 2020b; Penn 2020; Parti and Day 2020). The parties and their activists seem to be much more polarized than significant portions of the electorate (Hopkins 2017;Muirhead and Tulis 2020). Trump in particular, like he and Clinton both in 2016, tends to be viewed in terms of extreme character traits rather than in terms of partisan polarization per se (see Christenson and Weisberg 2019).
Local issues and candidates can then still surpass party or presidential affiliations, even as Trump himself gained support from new voters more radical than his party or relative to anything he had done in office. Still, some Republican-leaning voters showed up and voted for Biden, even as they also cast votes for downballot Senate and Congressional Representatives (Gerson 2020; Rauch 2020; Sargent 2020; Gabriel 2020). Trump was essentially wiped out in California, yet down ballot Republicans did better in 2020 than they had in any election year since 1998 (Siders 2020). This could be good news for liberal democracy in the face of the rise of populist politicians like Trump, and the sectarian politics that has become so evident in contemporary America (Graham and Svolik 2020;Finkel et al. 2020).
Whether or not the Republican Party moves beyond Trump and back towards a more pluralistic view of its role in American politics and away from the populist-authoritarian character it has increasingly displayed remains unknown (e.g. Boot 2020). To an extent, polarization is built into the American electoral system with its emphasis on winner-take-all in the Electoral College and the centrality of two political parties. The two parties become the singular focal points for all manner of social and ideological cleavages that in other countries with different electoral systems (e.g., proportional representation) are spread across multiple parties (e.g. Dimock and Wike 2020; Carothers and O'Donahue 2019).

TRUMP'S "OVER-PERFORMANCE" IN 2020
Arguably, Trump over-performed in 2020, particularly in the so-called swing states of Michigan, Pennsylvania, and Wisconsin but also more nationally (Wilkinson 2020). National opinion polls for months before the 2020 US presidential election showed Democrat Joe Biden well ahead of incumbent Republican Donald Trump. There was much talk about a "blue wave" that would sweep Biden into the White House and the Democrats into a majority in the Senate and increased representation in the House of Representatives. None of this turned out to be true (Tomasky 2020). US presidential elections are mediated by the aggregation of votes at the state level to produce a winner through the Electoral College: the national vote does not matter if a candidate can win by narrow margins in so-called swing states. This is what happened in 2016 when Trump lost the popular vote but won the presidency by narrow victories in Michigan, Pennsylvania, and Wisconsin, the swing states of that year. But even in these states and a number of other potential "battleground states", such as North Carolina and Florida, polls showed Biden with substantial leads in 2020 going into the election. In the end, the 2020 election came down to much the same scenario as 2016 but with Biden winning the crucial states plus a couple of others (Arizona and Georgia) that Trump won in 2016, but with huge nationwide increases in turnout (66.7% turnout in 2020 compared to 59.2 % of eligible voters in 2016). The increased turnout may suggest a collective sense of anxiety about the future of the country, plus the fact that so many people were at home during the pandemic, rather than a sudden explosion of democratic sentiment (Spence and Brady 2020).
The polls of 2016 were wrong as well. They supposedly corrected their sampling methods to tap more potential "shy" Trump voters, and include more respondents from the demographic categories supposedly more Trump-friendly like non-college educated white men and women. In the aftermath of the election exit polls were even more unreliable than the opinion polls, with different ones producing totally different pictures of the electorate, both nationally and at the state level (Drezner 2020). Contemporary polling fails to do what it is supposed to do: predict the outcome of the election with a reliable narrative of why it turned out that way.
The problem is twofold. On the one hand, the demographic categories used in polling do not capture very well the different meanings they have for people living in different places. For example, Latinos in Florida helped win the state for Trump but in Arizona they helped Biden win, white Catholics in swing states swung to Biden but less so elsewhere, and Trump did even better with more affluent voters than in 2016 but less so in the crucial swing states of Michigan and Pennsylvania (Alcantara et al. 2020).
Consider, more specifically, the term Latino applied to a wide range of ethnic and national groups with very different histories across the United States. Why should we be surprised when Cubans in south Florida vote differently than people of Mexican and Salvadoran ancestry in California, or Mexican-Americans in south Texas than Mexican-Americans in New Mexico (Yglesias 2020; Rathbone 2020)? Similarly, Asian Americans in Georgia, though small in number, may have tipped the balance against Trump but elsewhere the net drift was probably in his favor (Wang 2020). Much of this difference may be attributed to the relatively high social status and dispersed residence patterns of this ethnically mixed group (Indian, Chinese, Korean, Vietnamese, etc.) in suburban Atlanta when compared to clusters of co-ethnics such as Vietnamese in southern California.
If in 2016 Trump had provided an obsessive focus on "illegal" immigrants and the southern US border, in the 2020 election campaign these issues were largely eclipsed by concerns about the economy and the pandemic. The gender and education categories also display considerable spatial variance in meaning that cannot simply be taken from census reports. A college-educated, professional woman in Arkansas, for example, is scarcer than a similar one in Massachusetts. This has consequences for self-image and possibly for political outlook. Even though there was a massive gender gap in 2020 with women voting net Democratic, as they had done increasingly since 1996, the reasons for this are not the same everywhere given gaps in education, race, ethnicity, and local histories (Zhang and Fox 2020; Maxwell and Shields 2019). White women, particularly in red states but also net nationally, still favored Trump, suggesting that the gender gap has all sorts of contingencies built into it (Lenz 2020).
On the other hand, polling itself has become increasingly unreliable because of the over-reliance on fading technologies like landlines, dishonest or disingenuous respondents, or a lack of willing respondents. Moreover, with the increasingly polarized population wherein the crucial swing voters are a smaller share of the total electorate, their relative turnout is probably what determines the final accuracy of the polls compared to actual votes cast for a given candidate (e.g. Hill 2020; Stephens-Davidowitz 2017). A plausible example of this in 2020 was the clear "evidence" from polling that states and local areas with higher levels of Covid-19 fatalities were going to be less likely to support Trump and other Republican candidates (Warshaw et al. 2020). This turned out to be problematic on two grounds. First, by the time of the election, the epicenter of fatalities moved to high Trump-supporting areas and away from areas where Trump was much less popular irrespective of the pandemic (like New York and New Jersey). Second, Trump supporters did not consider the pandemic to be a central issue in the election nor did they assign Trump any responsibility for its tragic outcomes (Florko 2020; Stacey 2020). The lack of areal sampling, and the assumption of a rational nexus between a pandemic and voting, were crucial errors. More generally, it seems that voters are unaware of the specific policy positions of candidates, and their responses cannot be trusted even when an issue is as visceral as a pandemic in which respondents may know someone who has tested positive or even died (Guntermann and Lenz 2020).
Subsequently, many commentators turned by necessity to considering more aggregate or ecological explanations for what had happened. Of course, survey aficionados warn that this is probably to engage in the ecological fallacy: making inferences about individual persons from aggregates or groups. What they fail to recognize, beyond the problems with polling and surveys noted previously, is the danger of the individualist fallacy which lies in regarding people as isolated individuals, independent from each other, who identify perfectly with ascribed census categories. People, and more specifically voters, are social beings embedded in relationships with each other, and the places in which they live. It is well known, for example, that human health challenges tend to cluster geographically, and lifetime prospects for social mobility tend to depend heavily on where you live and work more than just on adding up individual or household traits (Agnew 2018). Thus, there is a strong argument for grounding survey data and/or engaging in analysis of aggregate conditions in themselves not just because of the dilemmas of polling, but because the causal pathways to explaining voting behavior cannot be adequately examined solely in terms of the putatively national demographic traits of individual persons (Agnew and Shin 2020; Davis 2020).

TWO GEOGRAPHICAL STORIES OF THE 2020 US PRESIDENTIAL ELECTION
Two dominant geographical stories of the 2020 presidential election prevailed as syntheses or nationwide framings of various sorts of empirical data. The first story is the continuing allusion to a fundamental division between "blue" (i.e., Democratic) and "red" (i.e., Republican) states (see Hopkins 2017). Hence, the election is determined by what happens in a few "purple" states. This narrative privileges the historical analysis of the two political parties, their relative embedding in different regions over time, the correlation between their incidence and the distribution of different ethnic groups, and the cultural-economic history of different states in terms of industrialization, labor organization, and social attitudes. Accordingly, southern states tend to be more conservative ideologically than others on a range of issues because of the history of slavery, evangelical religiosity, and hostility to the federal government as a result of being on the losing side in the Civil War.
If at one time these states were heavily Democratic, since the Nixon presidency they have moved inexorably in presidential elections towards the Republican Party as that party became increasingly based around positions on issues attractive to southern whites. As the ethnic complexion of America has shifted, the nexus of attitudes long associated solely with the South has nationalized through the Republican Party to whites across the entire United States (Maxwell and Shields 2019). Of course, this diffusion has met with differential reception across the country because of local and regional contingencies such as degrees of urbanization, patterns of recent economic growth, and religious affiliations of various sorts. Nationalizing the white electorate to the benefit of Trump thus proved more difficult than this story might suggest.
The second story, has become more popular than the first, particularly in the aftermath of this election, but is still hardly novel (see Rodden 2019). This is the notion that the polarization of the electorate into two opposing camps is essentially a rural-urban divide. Trump-supporters are overwhelmingly located in the more ruralsmall town areas of the country, and Biden-supporters reside largely in those areas of the country that are more urbanized, dynamic economically, and outward looking. Given the majoritarian character of US elections, large numbers of Democrats pooled up in cities are outvoted by the smaller margins needed for Republicans to win in elections in rural areas. Thus, the Republican Party benefits in squeezing more Representatives and better presidential outcomes from the relative bias against cities with their large "wasted" majorities compared to tighter elections where Republicans have the edge in rural areas.
Drawing electoral boundaries to put Democrats in fewer districts makes this even more the case in congressional elections. Of course, the urban bias of the Democratic vote also makes it harder to develop positions on economic and cultural issues that travel well outside their strongholds (Thompson 2020b). This poses a major challenge to a party that is already more a congeries of groups with different interests and identities, unlike the Republican Party with its ideological consistency and overwhelmingly white base (Grossmann and Hopkins 2016). An analysis of counties at the scale of the United States leaves the impression that the long-running trend of denser places with respect to population becoming more Democratic, and less dense ones becoming more Republican, has in fact increased. In 2020 voters in the least dense counties (i.e., bottom 20% by population density) favored Donald Trump by 33 percent, up from 32 percent in 2016, whereas voters in the most urban counties favored Joe Biden by 25 percent compared to 25 percent for Hillary Clinton in 2016 (Economist 2020b).
Two strands of the rural-urban argument are apparent (Agnew and Shin 2019). One emphasizes the largely economic forces in play with the so-called rural areas seen as lagging or "left behind" in terms of economic growth and employment prospects. These areas also suffer from aging populations and outmigration, as younger people move to more dynamic metropolitan centers. In 2020, counties that gave Biden a majority of their votes together account for fully 70 percent of US GDP (Muro et al. 2020). On the other side of the ledger, so to speak, Mansfield et al. (2019), for example, show that attitudes to trade in places with import challenged industries, became more negative towards trade and international trade agreements in the aftermath of the 2008-9 financial recession. But they also suggest that increased "ethnocentrism" was also involved. A second focuses on the status anxieties of people who are overwhelmingly white, and who live in rural and declining industrial areas. Their declining economic condition and limited fortunes seem to correspond closely with the real or perceived increase in the number of immigrants, and unwanted cultural changes that are being forced upon them. This is what explains the popular anger, resentment, and racism that Donald Trump has tapped into since he descended the escalator in his office tower in New York City in 2015 to declare his candidacy for the US presidency. It is President Trump who would rescue the country from the "carnage" (largely associated with urban areas and their populations) he himself identified as undermining the imaginary country represented by his people in the "heartland" (e.g. Wuthnow 2018; Agnew and Shin 2019; Bartels 2020; Hohmann 2020a).

MAPPING TRUMP IN 2020: RED VERSUS BLUE AND RURAL VERSUS URBAN
The Electoral College places a premium on so-called swing states in which the likely voters of the two major parties are evenly split, and there is a significant number of inactive and indifferent voters who move one way or another between the parties in subsequent elections. In 2016 the presidential election came down to narrow vote advantages for Donald Trump in the three states of Michigan, Pennsylvania, and Wisconsin, but he lost the national popular vote to Hillary Clinton. Though Joe Biden squeezed out narrow wins in the states of Arizona and Georgia, and also enjoyed a massive victory in the national popular vote, 2020 came down to the same three states again. Nevertheless, what we emphasize is that irrespective of the popular vote victory for Biden, largely attributable to massive majorities in California and New York, Trump actually over-performed relative to expectations nationally and in these three crucial states.
Examining Trump's electoral performance in 2020 compared to 2016 across all US counties is a good way to begin examination of the geography of the 2020 presidential contest (Fessenden et al. 2020). In this way we can identify the places with the greatest shifts and speculate on the processes that produced this map. We then examine one of the most important forces behind Trump's over-performance in 2020: the surprising increase in votes as a result of an average 7.5% increase in turnout nationwide between the two elections. Of course, this increase did not accrue to Trump alone: far from it. He lost the election nationwide and in the crucial swing states. But along with some critical shifts in groups typically more supportive of Democrats than Republicans (some Latino voters particularly in Texas and Florida), this is what made the election closer than pre-election polling and punditry suggested.
Three features stand out on the map of swings between 2016 and 2020 ( Figure 1). The first is the relative stability of the counties. Very few show shifts of more than a few percentage points in either direction. This fits the conventional wisdom that the country as a whole is fairly fixed in its partisan orientations, and using the county as the basic unit of analysis makes it clear (Sances 2019). Second, though there are swaths of the country in light blue or light pink, there is also considerable geographic variation beyond the simple red versus blue at the scale of the states. Third, there are also some significant regional and local effects in swings between 2016 and 2020. Several of these reflect peculiarities of the 2016 election. For example, the relatively high swings to Trump in Utah and southern Idaho are down to the fact that in 2016 a third-party candidate from Utah did well in precisely these areas. But others reflect real swings to Trump as opposed to Biden. This is the case in the Rio Grande Valley counties of Texas, and in Miami-Dade County in south Florida. These swings were crucial in keeping the states in question in Trump's column in the Electoral College even as in Texas, for example, the Dallas-Fort Worth urban area swung to Biden and Jacksonville and some rural parts of Florida went in that direction too.
The main story of the map, however, given that it is not weighted by population, is that Biden did much better in most of the major suburban areas of the country where most of the national population now lives, particularly around Atlanta in Georgia, Philadelphia in Pennsylvania, Denver in Colorado, and Houston, Austin, San Antonio, and Dallas-Fort Worth in Texas. Added to Biden's large majorities in most of the cities in question (although the city of Philadelphia shows a swing to Trump) and elsewhere (particularly all along the west coast) this accounted for his large national popular majority of votes. The impression given by the map undoubtedly overemphasizes the rural areas of the Midwest, Prairies and the Mississippi Valley where swings to Trump were large in percentage terms but small in overall numbers given the low population densities of these regions (Le Monde 2020). This leaves an impression of a larger red/blue gap then may actually be the case when population distribution is taken into account.
Donald Trump received around 10 million more votes in 2020 than he did in 2016. He lost because Joe Biden received even more votes, and already had an edge in terms of Hillary Clinton's total in 2016. What is most noticeable, however, is that although Trump did pick up more votes in predominantly rural America relative to Biden, he also picked up most of his new votes in urban areas (Figure 2). With the exception of Florida where Trump was the net winner of new voters across the state, Trump was the narrow loser across much of the rest of urban and suburban America. So, even as he lost, Trump picked up significant votes in and around Los Angeles, San Diego, Phoenix, Houston, and Salt Lake City. His problems with respect to turnout were, above all, in Michigan and, to a lesser extent, in Pennsylvania, and Wisconsin where suburban voters in Philadelphia, Pittsburgh, Detroit, and Milwaukee proved critical in turning out in greater numbers for his opponent. So, it was not in the cities or the countryside that the 2020 US presidential election was decided but in the suburbs of the largest cities in the three swing states. Neither the red/blue state nor the rural/urban narratives as outlined adequately accounts for this reality; being "in-between" was determining (Badger and Bui 2020; Burn-Murdoch and Zhang 2020). In 2016 the story in the three swing states was somewhat different. As we argued previously (Agnew and Shin 2020), the 2016 story was one of Trump taking over rural and declining industrial areas in the three states that had historically been volatile electorally but had voted for Obama in 2008 and 2016. According to Ballotpedia (2017), 206 counties nationwide voted for Trump in 2016 that had voted for Obama in 2008 and 2012. The 206 counties were spread over 34 states. It was where their numbers were concentrated in key states, however, that was crucial. Michigan had 12 "pivot" counties, Pennsylvania had 3, and Wisconsin had 23 (Figure 3). This is where the voters who allowed Trump to eke out his victory in the Electoral College were located as he was losing the nationwide popular vote to Hillary Clinton.
In 2016 Trump won the three states of Michigan, Pennsylvania, and Wisconsin by net 77,744 votes, mostly concentrated in a number of largely rural and exurban counties in the three states. These voters seem to be mainly white working-class voters who never obtained college degrees. Like their peers across the country, having supported Obama's campaigns, they turned away from Hillary Clinton and voted for Trump. Thus, a key element in the outcome of the 2016 presidential election were those voters largely in the Upper Midwest and Pennsylvania who had backed President Obama in 2008 (and 2012) but then reversed course to support Donald Trump in 2016. Nationally about 9 percent of Obama voters went for Trump in 2016, about 5 percent of the total electorate (Sides et al. 2018).
So, what happened to the role of the swing counties in the swing states that we identified as crucial to Trump's victory in 2016? By and large they stayed with him in 2020. Economic stasis, plus the federal mismanagement of the pandemic, might have led to some questioning of the prior move to Trump (e.g. Casselman and Russell 2019;Warshaw et al. 2020;Dawsey 2021). In fact, the pandemic did not seem to be a major factor in undermining Trump's support (Bump 2021). Trump's trade disputes with China and other countries certainly did not help these beleaguered places, but neither did they seem to harm him politically (e.g. Langevin 2020; Brown 2020; Tita and Mauldin 2020). As we argued previously, much of Trump's appeal in 2016 was emotional rather than cognitive, and based on anxiety about the future (Agnew and Shin 2020).
Perhaps the street protests and urban violence of summer 2020 magnified on social media and on right- wing radio and television played some role in keeping fearful residents of many of the rural and exurban counties in Trump's column. This did not seem to be the case in advance of the election in the Detroit suburbs, however (Jamerson 2020). Unemployment as a result of measures taken to deal with the pandemic seems to have had longer lasting effects in areas that voted more for Biden than for Trump except in the case of Michigan where the counties with the highest rates voted net for Trump (Koeze 2020). In 2020 only two counties in Pennsylvania and one in Michigan flipped back to Biden after having gone to Trump after Obama in 2016 (Bump 2020b). Erie County was one of those in Pennsylvania. The outcome there in 2020 was remarkably close (Maher and Zitner 2020).
In 2020 the counties of 2016 no longer counted. What happened was that in a race to tap increased turnout Biden beat Trump in the suburbs of the main urban areas, particularly around Detroit and Philadelphia, while holding on to majorities, although diminished in Philadelphia, in the big cities across the three states. Biden built up large majorities of votes in the same counties that Clinton won in 2016, but enjoyed larger majorities that gave him the narrow margins he needed at the state-level. There had been some signs of this trend in the 2018 Midterm elections, particularly in southeastern Michigan (Sarbaugh- Thompson and Thompson 2019). What was more surprising was Biden's relative success in the Milwaukee suburbs (see, e.g. Weichelt 2021). At the same time, and suggesting a powerful suburban anti-Trump vote more than a repudiation of the Republican Party tout court, some of the suburban districts that gave Biden his victory across the three states also elected Republicans to Congress and to state legislatures (Gabriel 2020). This blue wall is indeed fragile, and in future presidential elections it could crumble.

EXAGGERATING GEOGRAPHICAL POLARIZATION?
One of the takeaways from the three swing states is that the hold of candidates of both parties across all of them is tenuous at best. These states remain purple. At the same time, two other states, Arizona and Georgia, came over to Biden in 2020 after being consistently red states for many years. They too are at least now light purple. In both of them Trump's vote held up, particularly in white rural areas, but was outweighed by swings to Biden in the more urbanized-suburban counties and minority-majority counties (Native American in Arizona and African American in Georgia). In terms of mobilizing likely voters in 2020, Biden managed to outdo Trump in these two states even as he lost in other states, like Florida and North Carolina, that many commentators thought were more likely to turn blue this time around.
The stories of the two states seem different beyond the elemental role of the suburbs in producing the outcome in each: the suburbs of Phoenix in the first and those of Atlanta in the second. There is a definite distance decay effect in Trump votes with the largest percentages at the county level in the places most distant from city centers, ceteris paribus. Only in heavily minority counties do we see the exceptions to the rule. Of course, most Trump voters are still in the major urban and suburban areas. They were just outnumbered there in 2020 (e.g. Medina and Stephenson 2020). This explains the continuing success of down-ballot Republicans in those areas even as Trump was losing.
In Arizona, Biden's relative success in 2020 can be explained by three factors. One is the massive in-migration into the main urban-suburban areas of people from west coast states like California and Washington, who bring with them different political sensibilities from those of long-term residents in the state (Balk 2020). The second is the reaction against Trump's toxic personality, particularly because of his personal attacks in 2016 on the popular, late-Senator from Arizona, John McCain. McCain's widow, a lifelong Republican, endorsed Joe Biden in 2020. Finally, although perhaps exaggerated around the time of the election, turnout among Native Americans in Arizona was much higher in 2020 than previously, and this undermined the simple rural-urban dimension as fundamental in this particular state (Caldera 2020). Much of this increase could be put down to negative appraisals of the role of the federal government under Trump in addressing the pandemic as it affected life profoundly on the state's native reservations, specifically in the northeastern and central southern regions of the state.
In Georgia a massive registration campaign, directed primarily at African American voters, but also more generally in the aftermath of a controversial governor's election in 2018 when many African Americans were disenfranchised, was undoubtedly one factor in producing a much larger turnout in the Atlanta metropolitan area than in recent presidential elections (Kim 2020;McWhirter 2020). As with Arizona, recent immigration from the Northeast and California (including Asian Americans), as well as from Latin America has created a more complex electorate than the historic black-white bifurcation, and its tendency to reproduce the blue-red polarization. Thus, in Georgia suburban districts are now much more likely to produce wins for Democrats across all offices up for election than previously was the case (Badger 2020). Only either voter suppression of groups likely to vote Democratic, legitimized perhaps by Trump's unsubstantiated but rallying charge of "massive fraud" denying him victory in 2020, or a shift in the character of the Republican Party from a white nationalist to a more inclusive if still conservative party could undermine this trend (Corasaniti and Rutenberg 2020).
The overriding geographical message from the 2020 presidential election is twofold. The first is that the red state-blue state distinction oversimplifies and minimizes the complexity of the American electorate. Shades of purple are the new colors of, and within, many states. Yet it seems clear that many Trump voters do live in places and in relation to media that limit their access to people who do not think like them compared to many Biden supporters (Bump 2020c;Andrews and McGill 2020). But the evidence from 2020 is also that the vote was less polarized geographically at the state level than in 2016 (Kolko and Monkovic 2020). The accession of Arizona and Georgia into the ranks of the swing states is clear evidence for this conclusion. The fact that Michigan, Pennsylvania, and Wisconsin were once again on the front lines of the 2020 election reinforces this view. Vote mixing up-and-down the ballot in many suburban areas, as well as vote switching from one party to the other everywhere, is further evidence against seeing an absolute red versus blue story at the state level as having permanent value. The second message is even clearer. The 2020 US presidential election was won and lost in suburban areas; not as a result of a fixed ruralurban opposition. Whether this is down to the relative unpopularity of Trump among voters who might otherwise vote Republican in presidential elections remains to be seen. The presence of so many non-polarized voters, however, even in the presence of Trump as a presidential candidate, suggests that not all is as bleak or as dire as the stories of a terminally polarized America suggest.