Abstract
White American voters have realigned among the two dominant parties by income and education levels. This article argues that the interaction of education and income provides a more insightful—and stark—display of this change than treating them individually. Each group of voters is associated with distinctive “first dimension” views of economic redistribution and “second dimension” preferences concerning salient sociopolitical issues of civic and cultural liberties, race, and immigration. Macro-level hypotheses are developed about the changing voting behavior of education-income voting groups along with micro-level hypotheses about the propensity of vote switching. The hypotheses are tested with data from the American National Election Studies 1952–2016. A profound realignment is revealed between (groups of) white voters and the two main US parties that is consistent with the theoretical expectations developed in the article.
Donald Trump’s victory in the 2016 US presidential election caught most observers by surprise. But should it have been surprising? To explore this question, we analyze the voting behavior of the American electorate during the last half-century. As in every election, there are particulars to the 2016 contest, but we find that it can be well described as continuing a long-term process of changing voting behavior of distinctive socioeconomic groups. The American experience is thus characterized by a theoretically expected realignment of voters and parties. The traditional New Deal alignment is gradually being replaced as we transition from an industrial to a knowledge society. In this article, we describe that process and characterize the emerging alignment.
Our theoretical framework describes the American electorate in terms of income (low vs. high), education (low vs. high), and—importantly—their interaction. Remarkable changes over the last four decades in the size and electoral importance of the resulting four groups—low-education/low-income, low-education/high-income, high-education/low-income, high-education/high-income—have changed the demands in the electoral arena. In a process described in this article, the New Deal core constituencies of the two main US parties—low-education/low-income voters for the Democrats and high-education/high-income voters for the Republicans—have become swing groups; the former swing groups are the parties’ new core constituencies (high-education/low-income voters for the Democrats and low-education/high-income voters for the Republicans).
Offering a theoretical account of this “polarity reversal” of the American electoral landscape, empirically tested at both the macro and micro levels, is the main contribution of this article. Our account relates to several prominent debates in the political science literature, but it has implications for public controversies about the strategy of center-left parties such as the Democrats, as well. On the theory side, first, the “what’s the matter with . . .” debate 1 disagrees about how the constituencies of the two main parties should be described—by income? education? class? geography? locality?—and how these constituencies may have changed over time. Has the “white working class” abandoned the Democratic Party? We will argue that this debate can be fruitfully advanced with our framework, which characterizes party constituencies by education-income groups. Second, the “democracy for realists” perspective paints a decidedly unflattering picture of citizens. But is it true that programmatic considerations do not motivate vote choice? Are the bonds between voters and parties based on policy preferences? Or is it all about identity politics? 2 We argue that the links between voters and parties are programmatic and present evidence consistent with such a perspective. Third, what happened in 2016? Did economic grievances or racial animus nudge more citizens than anticipated toward Trump? 3 Although we do not directly join this debate, we will evaluate the uniqueness of the 2016 contest from a long-term historical perspective. And we show that those with whom authoritarian and racist appeals by presidential candidates resonate most also tend to belong to socioeconomic groups affected—or prospectively threatened—by economic decline.
The developments we document and discuss in this article are characteristic of all modern democracies. Everywhere in advanced capitalist democracies, parties on the center-left and the center-right have undergone major changes. Everywhere, these changes have prompted public political debates about where and how these parties have to position themselves to win elections. One common dispute is over why the “working class”—typically understood as lower-income voters—has defected from center-left parties (or, alternatively, why center-left parties have abandoned the working class) and how it can be won back (or, alternatively, how it can be replaced).
Our analysis turns the tables on this debate in several ways. First, we show that the center-left party, in the United States at least, is being abandoned by lower-education/higher-income voters as much as by the working class (lower-education/lower-income). Second, we show that lower-income voters are divided sharply into two groups. One consists of highly educated people whose numbers and support for the Democratic Party are growing. Its members have become the core of center-left politics. The other group of low-income voters consists of those with lower education levels. Its members have drifted toward right-wing politics on the basis of appeals to authoritarian conceptions of social governance, racism, and xenophobia. But they have also become a “swing” group, up for grabs by either party, given their redistributive economic policy preferences. The working class, if anything, has not abandoned the center-left but become a more fickle, volatile electorate, easily swayed one way or another depending on the configuration of economic and noneconomic positions the candidates bundle together. The 2016 Trump election demonstrates this fact.
Our article focuses on American non-Hispanic white voters. Although racial sentiment as motivation of voting behavior among whites plays an important role in American politics—and in our analysis—we do not examine the racial identity of nonwhite voters. There are several reasons for this narrowing of our empirical focus. First, African Americans were the only quantitatively relevant nonwhite group of citizens until the turn of the millennium, and they switched to the Democratic Party almost entirely between the 1930s and 1960s. In other words, African Americans became almost all Democrats before the major changes occurring with the rise of the knowledge society since the 1970s and 1980s. The realignment of black voters predates available data, but it was one of several important precipitants of white realignment to the Republican Party, particularly in the South. By itself, however, it does not explain the full extent and general pattern of the realignment among distinctive groups of (white) voters to right-wing parties that can be found outside America as well.
Second, given the high propensity of African American voters to support Democratic candidates, and given their relatively small share in the electorate, there is simply not much variance to explore for subgroup analysis. 4 Third, the sample restriction increases the comparability of our results with the existing literature, which usually also focuses on white respondents when studying similar questions. Finally, we believe that our theoretical argument has analytical carrying capacity beyond the specific case of the United States to other advanced capitalist knowledge societies that do not have the exceptional experience of the United States with slavery and long-term domination of one ethnic group by another. 5
We first present a theoretical framework that lays out the policy space, demand- and supply-side developments, and the process of electoral realignment. From that framework we derive both macro-level and micro-level hypotheses, which are then empirically evaluated. In the conclusion, we relate our framework and findings to the three larger academic debates mentioned above and the strategic dilemmas articulated in the public debates not only of the US parties but also of center-left and center-right parties in other knowledge societies.
Theory
Partisan alignments are predictable and stable relationships between groups of voters with a set of political preferences and their choices for a particular political party. Partisan dealignment is the erosion of those patterns, while realignment is the emergence of new ones. Consequently, alignments are about the congruence between voters’ preferences and parties’ policy offerings, and spatial party competition is the tool by which to analyze them: parties offer bundles of issue positions in a low-dimensional space, and (groups of) citizens vote for parties that are proximate to them on issues that are important to them.
Any study of programmatic voter-party alignments and their change, therefore, must deal with four topics. (1) The policy space (or “electoral market”) in which party competition occurs. (2) The demand side (what voters want—or where they are located in the policy space). (3) The supply side (what parties offer—or where they are located in the policy space). And (4) how (groups of) voters decide to vote (or how the market clears). Realignment occurs when (groups of) voters adjust their partisan choices, based on the available partisan appeals, in a process of sorting and switching.
The process of realignment has thus a bottom-up and a top-down dynamic. Voters’ preferences result from economic and social experiences outside the realm of party competition. 6 But how they are activated as salient topics of partisan contention and crystallized around partisan alternatives depends on the strategic choices of politicians. We will first detail these four topics to develop our theoretical framework. We will then derive macro-level and micro-level hypotheses and test them.
Framework
Policy space
A programmatic perspective of links between voters and parties cannot do without a clearly defined policy space. What are the relevant issues on which parties stake out positions and that motivate voters’ choices? How many dimensions are there? What is their content? How salient are they? These are difficult questions, but fortunately we can build on a large and convincing literature.
There is broad agreement that American party competition is low-dimensional and that a principal axis of political competition evolves around economic issues. 7 There is less agreement about the number and content of possible political dimensions beyond that, although most accounts operate with a second, noneconomic dimension, often vaguely termed a “cultural” dimension, that is about “social issues” (guns, gays, and God; immigrants; non-Christian religions). 8
In US public opinion studies, this two-dimensional space is sometimes amended by a third dimension to map racial prejudice and ethnic-group boundary drawing. Although the noneconomic issues encompass a range of distinct topics in theory 9 —most important, questions of political governance (more libertarian or authoritarian views on civil liberties, participation, individual moral autonomy), racial prejudice, and citizenship (particularist or universalist-cosmopolitan positions on nation, trade, and immigration)—they are highly and increasingly correlated in practice. 10 Moreover, they are predicted by the same variables—with high education featuring most prominently as predictor of libertarian, racially unprejudiced, and cosmopolitan views. We therefore, theoretically and empirically, operate with a two-dimensional policy space.
The economic dimension ranges from “progressive” or “left” positions in favor of redistribution, economic regulation, and the like to a “conservative” or “right” position against redistribution, against taxes and government spending, and the like. Different scholars use different labels for the second dimension, and they label its extreme positions with different terminologies. For simplicity, we refer to it as the noneconomic dimension, which ranges from progressive or libertarian positions to conservative or authoritarian positions. It involves questions of civil liberties and law and order, gender roles and sexual orientation, tolerance for cultural diversity of beliefs and practices as well as differences of lifestyle, and acceptance of immigration from societies with different cultural practices. In the United States, it also includes racial animus.
The demand side
What determines citizens’ positions in the two-dimensional policy space? Once again, we can rely on a large literature but radically simplify its message: as the two main socioeconomic predictors of positions on the economic and noneconomic dimension, the literature has established that income—and to some degree, education—cleaves citizens on economic issues, whereas education divides them on noneconomic issues. We theoretically abstract from—but later empirically control for—a host of other socioeconomic endowments and experiences that, of course, also influence political preferences. Nevertheless, our simple model has striking empirical explanatory power.
In line with standard economic theory, we expect that citizens with higher income are less progressive on economic issues. 11 Furthermore, the social insurance literature has established that risk (future income streams) also shapes redistributive preferences. 12 Because higher education lowers risk exposure, higher education is associated with less support for progressive economic policies.
Education is the main socioeconomic correlate of issues related to the noneconomic dimension. It is well established that higher education is associated with more “libertarian” attitudes, whereas lower education is associated with attitudes that are more “authoritarian.” 13
Because income cleaves citizens on the economic dimension, and because education cleaves citizens on the noneconomic dimension, a natural way to divide the electorate into groups is by these two traits. The simplest way is to dichotomize and interact the two traits. 14 Doing so leads to four groups with distinct predicted preference profiles; 15 to some degree, these four education-income groups trace occupations. Data from the American Community Survey of the US Census (Census/ACS) allow us to explore the occupational composition of our education-income groups (see App. E).
Individuals in the high-education/high-income group tend to oppose progressive economic policies and tend to endorse libertarian noneconomic policies. Scientific-technical, financial-business, and general managerial occupations are prominent in this category. 16
Individuals in the high-education/low-income group tend to endorse libertarian noneconomic policies. They tend to be somewhat conflicted on economic issues: low income may tilt them toward redistribution, while high education pushes them in the opposite direction. In occupational terms, this group mainly consists of sociocultural professionals in social service, educational, cultural, and health care organizations.
Individuals in the low-education/high-income group tend to endorse authoritarian noneconomic policies and tend to oppose progressive economic policies. Small business owners and shopkeepers—particularly in construction, crafts, retail, and personal services—as well as some of their salaried associates populate this group.
Individuals in the low-education/low-income group tend to support progressive economic policies and tend to endorse authoritarian policies on the noneconomic dimension. In occupational terms, this group consists primarily of low-skill and intermediate routine blue-collar manufacturing or clerical-administrative jobs (the “working class”).
We will assume that the political preferences of the four education-income groups are relatively stable over time. Below, we will show that this assumption is empirically plausible. If so, demand-side changes are not driven by changing political preferences of individuals in the education-income groups. What has changed, rather, is the sizes of the four groups, a result of the transformation of the occupational structure in the transition from industrial to knowledge societies. These shifts, we submit, have altered the demand side on the economic and noneconomic dimension. Estimates based on Census/ACS data show that the sizes of the four groups in 1960 and 2015, respectively, were as follows:
Low-education/low-income: 64.4 percent (1960) → 52.4 percent (2015);
Low-education/high-income: 27.8 percent (1960) → 15.8 percent (2015);
High-education/low-income: 3.0 percent (1960) → 14.8 percent (2015);
High-education/high-income: 4.9 percent (1960) → 17.0 percent (2015).
The American National Election Studies (ANES) 17 data show a similar trend but quantitatively an even starker pattern (see App. E, Table E1; data refer to white voters who cast a two-party vote). For example, the low-education/low-income group drops from 50–60 percent in the 1950s to under 40 percent in 2016; over the same period the low-education/high-income group fell from the low 40s to the low 20s, whereas the high-education/low-income group ascends from almost nothing to one-sixth of all white respondents. Given these substantial changes in the sizes of the four education-income groups, we expect changes in aggregate political preferences in the two-dimensional space, the demand side in the voter-party realignment equation. With respect to the first dimension of economic issues, we expect continuity in terms of the salience of the dimension. But the composition of the pro- and anti-redistributive groups has changed: one unambiguously pro-redistribution group shrinks and is threatened (low-education/low-income), while another group leaning toward redistribution (high-education/low-income) grows in size. Likewise, in the anti-redistribution camp one group expands and seizes new opportunities (high-education/high-income), while another (low-education/high-income) is threatened by declining opportunities and consequent status anxieties. 18
With respect to the second dimension of noneconomic issues, however, the growth of higher education in knowledge societies produces an overall shift in demand in favor of libertarian positions as well as a leap in the salience of that dimension. It turns from a valence issue—with the vast majority of citizens flocking to a status quo supported by people with fixed, closed, authoritarian second-dimension dispositions—to a contested and salient dimension with wide spatial distribution of positions.
The two groups whose sizes change the most in the transition to the knowledge society are also farthest apart over second-dimension politics. In one corner is the rapidly growing high-education/low-income group. Individuals in that group support a destruction of the traditional authoritarian status quo on second-dimension issues, while also supporting economic redistribution, albeit with qualifications, as some may hope to rise into the group of high-income recipients. In the opposite corner are low-education/high-income citizens who face increasing risks to their social and economic status and an erosion in the size and power of their category. They oppose redistribution as a threat to current income; but they may have second thoughts, given their lack of education in labor markets that increasingly demand high qualifications. At the same time, their authoritarian predilections on second-dimension topics are challenged by a rising phalanx of educated low- and high-income professionals. Given that both groups have somewhat contradictory preferences on first-dimension economic issues but fully consistent preferences on second-dimension issues, appeals on noneconomic issues are likely to resonate with them more than appeals on first-dimension issues and to crystallize them around alternative political parties, provided politicians emphasize such second-dimension issues.
The other two groups are caught in even more complicated positions. Low-education/low-income voters are clearly wedded to a redistributive economic agenda and reject conservative market liberalism, but they are far from the libertarian zeal of the rising social category pushing for redistributive demands, (young) high-education/low-income social and cultural professionals. Conversely, high-education/high-income professionals may sympathize with a libertarian transformation of noneconomic issues up to a point, but they are repulsed by the strongly redistributive momentum of the low-income/high-education group that pushes the libertarian second-dimension agenda most.
It is this complex demand-side situation of political preferences into which politicians step on the supply side. In post–World War II industrial societies, parties essentially did not compete on second-dimension issues, once older conflicts around religion and urban-rural divisions led to settlements and converted from what Sani and Sartori called “divides of competition” into “divides of identification.” 19 The dominant parties were therefore ill prepared to cope with a differentiation of the preference space in salience and positional terms starting in the 1970s.
In sum, then, the demand-side premise of our analysis is as follows. (1) The income-by-education groups have distinct political preference orientations. (2) These preference orientations remain largely constant for each group over time. (3) However, the changing relative size of these groups has altered the demand side, mostly by creating a large libertarian second-dimension noneconomic issues camp, converting what was predominantly a valence dimension into a positional dimension and thereby increasing the salience of that second dimension. In the next subsection, we will argue that these developments have led political activists and politicians seeking electoral office to move the two major parties in the policy space.
The supply side
The demand-side developments described above are typical for any democracy undergoing the transition from an industrial to a knowledge society, and incremental realignment has occurred in most advanced democracies. 20 On the supply side, however, parties’ reactions to changing demand depend on the possibility of party entry. Where institutional constraints to party entry are weak—as in proportional electoral systems, especially with low electoral thresholds—party systems have experienced pronounced fragmentation. Electoral realignment in those cases leads to a proliferation of parties offering new policy bundles to satisfy unmet demand: “green” and other left-libertarian parties target high-education/low-income voters with progressive policies on both dimensions, radical right parties bundle pro-redistributive economic and authoritarian noneconomic issues, and so on.
Where institutional constraints to party entry are strong—as in “first past the post” electoral systems, particularly when combined with presidentialism, 21 as in the United States—the diversification of issue positions plays out within the existing two-party system. To represent the new demands, the old parties must bundle issues in a way that minimizes the defection of their traditional constituencies and maximizes the share of new supporters attracted by those new positions. 22 As a result, many voters are left with a choice between the lesser of two evils; they often have to make a “forced choice.” 23
This is the first of three features explaining why the American experience on the supply side is unique. The second US exceptionalism is that the party system inherited a deep racial divide over slavery and repression of African Americans in the South, 24 spreading with out-migration from the South to the entire country. 25 As a result, conflicts about ethnic group inclusion and group status have been salient since the 1950s, when the civil rights movement began to gain momentum. As another result, African American voters overwhelmingly support the Democratic Party. The third American exceptionalism is the role of evangelical “born again” Protestant Christianity—no equivalent exists in other knowledge societies. About one-quarter of the US population adheres to this belief. Members of this sizable group tend to hold decidedly authoritarian positions on the second dimension, even net of their income and education traits that predispose many against the libertarian wave. This reinforces the salience of the noneconomic issue dimension.
In two-party systems with high entry thresholds, as long as the low-education/low-income category constituted the clear majority, why would both parties not have competed for this group through at least moderately redistributive appeals and authoritarian-nationalist-racist second-dimension politics? As we know from Przeworski and Sprague, 26 no social group votes entirely cohesively, and divisive second-dimension appeals always make a difference. Despite their historical numerical dominance, the working class never managed to assemble a “permanent majority.” Moreover, the lower turnout of less educated and less affluent voters whittled away the majority of the group before it lost its demographic majority status. Finally, on the political side, the activists that shaped party positions disproportionally came from the higher-education strata. As we shall see, the rise of high-education/low-income activists since the 1960s made a huge difference for the strategic positioning of the American Democrats—and, similarly, for social democratic center-left parties around the globe—just as mobilization of the numerically declining high-income/low-education did for parties of the right.
Generically, the process of parties’ repositioning in the policy space occurs in several typical steps that play out over long periods. To be sure, realignments may surface in sudden shifts of electoral patterns that inaugurate a new and irreversible pattern (“critical elections”). But more typically, they occur in a cumulative “secular” fashion—with the crystallization of voter-party links resulting from a process that is incremental and nonlinear. We concur with most investigations that the American realignment since the New Deal is better characterized by the “secular” realignment template than the “critical election” mold. 27
The first typical step in a realignment process is the emergence of a representation gap. In the case of the transition from industrial to knowledge societies, the unmet demand concerns libertarian second-dimension issues.
In the American case, two conditions silenced the potential salience of second-dimension issues in the New Deal alignment. To begin with, African Americans had not yet moved to or been politically mobilized in Northern cities in sufficient numbers to become critical tipping electorates over which Republicans and Democrats competed. Moreover, the share of highly educated individuals—who tend to hold libertarian, individualist, and cosmopolitan preferences—was too small to disturb the conservative post–World War II consensus. Consequently, the New Deal party system was one with (mild) partisan divisions over economic issues (social security, progressive income taxation, Keynesian demand management, regulation of corporations and utilities, etc.) but not over noneconomic issues. 28
Both conditions began to crumble in the 1950s. The political opportunity structure for black voters in the North improved dramatically, as they became critical to deciding electoral races between the parties. 29 Soon, the growing cohort of university students and African Americans discovered that they could link their demands for civil liberties and emancipation, respectively, in the civil rights movement. The size of the groups challenging the status quo on second-dimension politics—young college graduates and African Americans—grew, while the national parties tried to ignore the issues. Protest against the Vietnam War became a critical catalyst and conflict accelerator in the late 1960s. Among American whites, the first step in political realignment was the growing share of highly educated voters and the corresponding expansion of professionalized jobs, particularly in sociocultural professions, and particularly among those with relatively poor income prospects. College education did not automatically convert into high income, and a growing share of the (young) labor force began to experience that fact. 30
The second typical step in a realignment process is political activists’ attempts to close the representation gap. When party entry is difficult—as in two-party systems—activists have no option but to enter the existing parties. The typical result is protracted battles between different factions within a party. 31
In the American case, 32 party activists moved the Democratic Party toward libertarian positions—on questions of race, gender equality, sexual tolerance, and so on—starting in the late 1960s. The influx of libertarian progressives and radicals into the Democratic Party propelled this process and enabled them to challenge the party establishment in 1968 and to capture the presidential nomination with George McGovern in 1972. He was the first candidate to embrace both redistributive first-dimension and libertarian second-dimension appeals. A conservative backlash was triggered on the Republican side, among predominantly less educated and older electoral constituencies, primarily by white fundamentalist Christian evangelical activists, which moved the party firmly toward authoritarian positions on second-dimension issues.
The third typical step in a realignment process is the attempt of parties to gradually assimilate the new and existing issue positions into a single bundle. They take an increasingly clear position in the policy space. 33
In the American case, in both parties, the activists’ mobilization led to a process of “conflict extension” in which new libertarian and multicultural issue positions were added to the existing interparty divisions about economic distribution, compelling both new contenders for office and established incumbents to embrace such positions. 34 This process resulted in the increasingly clear combination of progressive and libertarian positions by the Democratic Party and conservative and authoritarian positions by the Republican Party. Citizens, then, still have a choice between the two parties, but their positions have changed over time. “Sorting,” described in the next subsection, completes the realignment process.
Sorting
After the parties have taken clear and salient positions in the two-dimensional policy space, the calculus of vote choice changes. Citizens now have to consider both economic and noneconomic issues when casting their vote. For some groups of voters, this involves no trade-offs. The progressive/libertarian Democratic positions replicate those of many high-education/low-income voters and make them the new core constituency of the Democratic Party. The conservative/authoritarian Republican positions replicate those of many low-education/high-income voters, making them the new core constituency of the Republican Party.
For other groups of voters, however, the new party positions do involve trade-offs. Low-education/low-income voters—the traditional core constituency of the Democratic Party—are close to the Democrats on economic issues but close to the Republicans on noneconomic issues. The reverse is true for high-education/high-income voters—the traditional core constituency of the Republican Party. They tend to share the libertarian positions of the Democratic Party but to prefer the Republican Party’s conservative stances on economic issues. These programmatic cross-pressures make the former core constituencies available. In the realignment process, low-education/low-income citizens and high-education/high-income citizens become the new swing groups. Therefore, the identity of swing and core voters flips (polarity reversal).
Many voters will not adapt their vote choice in light of the parties’ new issue bundling. Party affiliation often creates powerful affective bonds of identification between parties and voters, 35 and some voters prefer to adopt their party’s new issue positions as their own, rather than abandon their old party. In other words, rather than switch their party loyalty, some voters change their issue positions to match the (new) party line. This mechanism of motivated reasoning, 36 however, cannot account for realignment. That comes about through several mechanisms:
Generational replacement: Some voters need not update their attitudes, or change their vote choice, simply because they are born into the realigned party system. They need not revise their choices in the light of parties’ past stances and thus may not have to go through a process of vote switching. Younger voters, from the first time they vote, should have a greater propensity to align their partisan choice with the party that is only the destination of switching for older voters. Younger voters align their choices more with the parties’ current appeals. Therefore, sorting and realignment is also the result of generational replacement.
(De)mobilization: The parties’ new positions alienate and demobilize some voters but attract and turn out others. Defection from one party to another need not occur from one election to the next; it can play out over a series of elections. Realignment therefore also occurs through mobilization and demobilization of certain voting groups.
Direct vote switching: Faced with new issue bundles offered by the two parties, some voters will change their vote choice from one election to the next. Such direct vote switching should be motivated by programmatic considerations. This mechanism is the “least likely” scenario for critics of programmatic approaches to voting behavior—it is theoretically most dissonant with party identification theory—and we make it the focus of our micro-level empirical investigations below.
Hypotheses
The supply-side developments sketched above are well documented in the existing literature. Our interest is in the process of realignment at the level of (groups of) voters. In the empirical section, we therefore test two sets of hypotheses. One concerns the changing core constituencies of the two American parties at the macro level; the other concerns the process of vote switching at the micro level.
Changing core constituencies
In terms of voting behavior, the process of realignment we outline above should lead to the following empirical patterns:
Among low-education/high-income voters, left parties experience a steep, long-term erosion and right parties a strong, long-term growth in support. The group becomes the core of the new right-wing party electoral coalition.
The reverse applies to high-education/low-income voters. More than any other group, its members will flock to the left party alternative and abandon the right. This group constitutes the core of the left-wing party in knowledge societies.
Given the right-authoritarian appeal of the conservative alternative in the two-party system, members of the quantitatively growing former core group of the party, high-education/high-income voters, will increasingly be cross-pressured between the two parties and show some tendency to defect from the right. This group turns from a core partisan electorate into a swing group.
Conversely, low-education/low-income voters continue to tilt to the left because of their redistributive economic predilections; but they are drawn to the right by authoritarian governance appeals when they become salient. This group, too, will convert from a core partisan constituency (albeit on the left) to a swing group between the parties.
We summarize our theoretical account in Table 1. The table indicates the two main parties’ positions on first- and second-dimension issues in the industrial and knowledge society (top panel) and summarizes our claim of the polarity reversal of core and swing groups (bottom panel).
Political Realignment from Industrial to Knowledge Society.
Motivations for vote switching
Congruence between citizen preferences and party positions is a key ingredient in our account of electoral realignment. Our second set of hypotheses, therefore, evolves around the programmatic motivations of vote switching.
Electoral realignment tends to be a slow process brought about through different mechanisms, including some degree of vote switching. How does vote switching ever come about? It is most likely to occur among those without, or with only weak, party identification. Yet even voters with firm party identification are not entirely beyond vote switching. Panel studies show that on issues salient to party identifiers, they may well switch party support—rather than issue preferences—if their positions clash with those embraced by leaders of their party of identification and the issues are highly salient for them. 37 Moreover, the number of issues of disagreement may matter. In the language of John Zaller and Stanley Feldman’s theory of survey response, if voters sense a strong “balance of considerations” 38 about a range of issues that makes them diverge from positions of their party of identification, they may defect from their vote choice rather than revise their beliefs.
If our account of realignment from an industrial to a knowledge society is correct, vote switchers should be motivated by programmatic considerations. Moreover, these policy motivations should differ between defection from the left and defection from the right. 39 Further, the considerations of switchers should set members of the emerging “core” electorates of left and right parties in a knowledge society—high-education/low-income voters on the left; low-education/high-income voters on the right—apart from members of the new “swing” groups receiving mixed cues from the parties (low-education/low-income; high-education/high-income). Finally, swing groups switch parties strongly motivated by the new dimension if there is supply-side convergence between parties on the old dimension.
Our first hypothesis builds on the classic insight that positions favoring redistribution should have a structural majority in capitalist economies because the poor outnumber the rich. 40 It is therefore difficult for parties opposed to redistribution to attract majorities purely on economic appeals. The transition to knowledge societies increases the salience of second-dimension issues for all parties and should therefore become a major consideration for switchers in addition to concerns about economic redistribution. But it is particularly plausible for parties of the economic market-liberal right to emphasize second-dimension issue positions, as such parties are less able to attract particularly lower-income voters on liberal economic-distributive policy positions. And the greater the material inequality of income and wealth, the more conservatives rely on noneconomic issue appeals to be electorally competitive. 41 For the American context, this logic yields the following hypothesis:
H1: Vote switchers from Democrats to Republicans tend to be relatively more motivated by second-dimension positions of their target party. Switchers from Republicans to Democrats, by contrast, may be motivated by either economic or noneconomic issue positions.
Our second hypothesis concerns the motivations of core- and swing-group switchers in the two-party system in transition to knowledge societies. Core groups align with their parties of destination on both dimensions. Members of the core categories who switch to the congruent party may therefore do so for reasons of either or both of the critical dimensions. Low-education/high-income voters, for example, flock to Republicans based on their anti-redistributive first-dimension economic positions as well as their authoritarian second-dimension positions. Conversely, high-education/low-income voters prefer left parties (Democrats) because they typically embrace progressive and libertarian positions. By contrast, swing voters will be drawn to parties with postindustrial configurations of appeals only by their positions on one issue dimension. Low-income/low-education voters may vote Democrat based on economic issues, but they may vote Republican because of noneconomic issues. The reverse tends to be the case for high-education/high-income voters. They may abandon the Republican Party because of noneconomic issue positions, but they are drawn to it because of economic issue positions.
H2: Members of the new swing groups—belonging to the low-education/low-income and high-education/high-income categories—will change party choice based on second-dimension issues, as they are still aligned on first-dimension positions of their party of departure. In contrast, either policy dimension motivates vote switching by members in the new core groups—high-education/low-income voters on the left and low-education/high-income voters on the right—as they are congruent with both.
Third, if (swing) voters perceive parties as converging on the old (economic) dimension, their vote choice should be more strongly motivated by the new (noneconomic) dimension. For example, in the face of perceived economic policy convergence, white low-education/low-income voters will be more inclined to vote for Republicans on authoritarian considerations. In the same example, high-education/high-income voters will be more attracted to the Democrats on libertarian second-dimension positions.
H3: When parties’ positions converge on one issue dimension that is fundamental for a partisan alignment, swing voters will primarily vote on a different dimension, provided that party positions are still divided on it.
Empirics
We will first briefly offer empirical probes of our assumptions regarding the relevance of the two issue dimensions for vote choice as well as the relative positions of the education-income groups in the policy space. We then test our macro- and micro-level hypotheses.
Our main data source is the American National Election Study (ANES) Time Series Study. For a subset of years (1992, 2000–2016), it contains a set of items on economic and noneconomic issues that we use to construct scales for the first and second dimension. We are then able to measure the position of respondents (and groups of respondents) in the two-dimensional policy space. Appendix A contains details.
Figure 1 shows that issue positions on the two dimensions are powerful predictors of vote choice. It shows the predicted probability of vote choice in presidential elections, as a function of economic and noneconomic positions, based on a logit estimation that includes a large set of control variables 42 (see App. B for the estimation results). The figure shows that—all else equal—a one-standard-deviation increase in “progressive” preferences on either dimension increases the probability of voting for the Democratic presidential candidate by about twenty percentage points.
Vote Choice as a Function of Policy Preferences (Two-Dimensional Policy Space).
We have argued that the four education-income groups have particular, distinct, and stable positions in the two-dimensional space. Figure 2 shows that this is plausible. The figure shows the four education-income groups in a two-dimensional policy space, based on a regression with a range of controls (we demean for each year to abstract from the ebb and flow of aggregate positions in the electorate). 43 The four groups hold distinct preferences—they are about one standard deviation apart on each dimension—and they are as theoretically expected. High-education/high-income voters, for example, combine a mildly anti-redistributive position that sets them apart from high-education/low-income voters with a libertarian position on noneconomic issues that separates them from low-education/high-income voters. When measured as the Euclidean distance from the origin of the two-dimensional space, the Democrats’ and Republicans’ core electoral constituencies in the knowledge society—high-education/low-income voters on the left, low-education/high-income voters on the right—tend to be more extreme than the two swing groups and, of course, not conflicted by being pulled toward opposite parties, depending on the dimension one investigates. And the most extreme positions on both dimensions are taken by Democratic (D) and Republican (R) Party voters.
Positioning of Education-Income Groups and Parties in the Two-Dimensional Policy Space.
In the figure, each marker symbol represents the estimate for one ANES survey with available data (1992, 2000–2016). The figure reveals a pattern of stability of the political preference configurations characteristic of the four education-by-income groups, at least for this period. Comparable data further back in time are not available. But note, for example, that Seymour Lipset’s conception of “working-class authoritarianism,” 44 postulated with the limited data available at the time, is quite closely related to the preference profile we find for the low-education/low-income category. Low social status, captured by education and income, has consistently been associated with authoritarianism. 45
Macro-Level Evidence of Incremental Realignment
In order to show how the education/income-based constituency support for Democrats and Republicans has changed over time, we again use data from the ANES presidential election year surveys. For this analysis, we have data from 1952 to 2016. As discussed above, we dichotomize the income and education variables (high income: sixty-eighth percentile or higher; high education: college degree or higher) and categorize respondents in one of four education-income groups. 46
The voting tendencies of these four groups chronicle the seismic shifts that characterize American electoral realignment over the last sixty years. To quantify these shifts, we estimate logit models on a dependent variable that equals 1 for respondents who voted for the Democratic presidential nominee and 0 for respondents who voted for the Republican nominee. The Democratic Party’s electoral coalition heavily relies on ethnic minorities. Among non-Hispanic white voters, Democrats find it increasingly difficult to win a majority. It is this development that we want to study and one reason for restricting the analyses to non-Hispanic whites.
We calculate the predicted probabilities of supporting the Democratic nominee for each of the four education-income groups (non-Hispanic whites only), based on models that control for a range of factors: gender, age, region (South), and church attendance, all interacted with election years. 47 To facilitate pattern recognition, we demean each group’s predicted probability by the overall vote for the Democratic Party in a given election, thus showing disproportionate support for the Democratic nominee in a given election-year (see Fig. 3; for the estimation results, see App. C). We also add trend lines that can show whether there is a long-term change of a group’s partisan voting propensity.
Support for Democratic Presidential Candidate by Education-Income Group.
The figure reveals stark trends: sixty years ago, low-education/low-income and high-education/high-income citizens were the fundamental core support groups of Democrats and Republicans, respectively. But by the 2010s, their trend lines place them in the mid-range between the two parties. By contrast, the initially tiny high-education/low-income category and the very large low-education/high-income categories started out in the mid-range, as swing groups up for grabs by both parties, yet slightly skewed toward Democrats, confirming the party’s hegemony during the New Deal era. But by the 2010s, the trend lines show they have moved in opposite directions and become the core support groups of the two parties. 48
Education appears to be the major story in the vote choice realignment, with the growing share of highly educated white respondents lining up more behind the Democrats and the declining share of less educated white voters rallying to the Republicans. So why do we emphasize the interaction of education and income? We do so because the intercepts and slopes—the starting points and trends—vary across the education-income groups. It seems unlikely that all low-education and all high-education voters will converge onto similar partisan voting probabilities. On the low-education side, low-income voters start out with a higher baseline (intercept) favoring the Democrats and show a flatter slope of decline, including the 2016 election, than the high-income voters who start out lower and move to the Republicans at a higher rate. On the high-education side, the slope of high-income voters moving toward the Democrats is steeper than that of low-income professionals, and they have substantially different intercepts. Net of the 2016 election, it may be fanciful to expect low- and high-income professionals to converge on the same voting probabilities. 49
How does the 2016 Trump-Clinton presidential election stack up against previous elections and voting trends? Was it a break with the past, a pattern shift that might announce the presence of a “critical election,” or a continuation of incremental secular realignment anticipated by elections prior to 2016? We opt for the second interpretation, although there is a qualification. On the one hand, for each of the education-income groups, the partisan-choice shifts from 2012 to 2016 are in the direction of the predicted trend lines. Moreover, excluding the 2016 election from calculation would still give similar intercepts and trend line gradients, albeit somewhat flatter slopes for low-education/low-income and high-education/high-income voters: the New Deal core groups of the two parties loosen ties to their primary parties only hesitantly. On the other hand, at the margin there is something unique about the 2016 election: the realignment process seems to be accelerated by large movements of the new swing groups (low-education/low-income and high-education/high-income voters). We offer our explanation for this pattern in the next subsection.
Micro-Level Evidence of Vote Switching
This section documents how individual voting behavior cumulatively realizes the secular partisan realignment of education-by-income groups over time. We pursue this objective by analyzing the vote switching of individuals between two presidential elections. Our sample consists of white respondents who cast a vote either for the Democratic (D) or Republican (R) presidential candidate in two consecutive elections. This gives us four patterns of vote choice in two consecutive presidential elections: DD and RR (voting for the Democratic or Republican candidate in both elections), as well as DR (voting for the Democratic candidate in the last election and for the Republican candidate in the current election) and RD (the reverse pattern). The latter two cases—DR and RD—are patterns of vote switching that are our dependent variable(s) in the analysis below. 50
We note that our measure of vote switching is based not on panel data but on a recall item available in most ANES surveys. 51 Respondents are likely to make errors in their recall of past choices, and they tend to underreport such changes. 52 However, existing investigations do not reveal an obvious bias in voting recalls that would undermine our testing of the micro-level realignment hypotheses. 53 Many vote switchers are likely to have low political sophistication and will not process partisan policy cues, and highly sophisticated voters make smaller moves in the policy space in multiparty systems and under conditions of party polarization. 54 All this, however, is likely to bias our test against our own hypotheses, as we are looking for systematic tendencies in the vote choices of switchers that do respond to partisan appeals in a highly polarized two-party system.
Aggregate vote switching—the share of DR plus RD voters—has trended downward since the 1950s, from about 20 percent to about 10 percent. Party loyalty—DD and RR—has increased, accordingly, from about 80 percent to about 90 percent of respondents (Fig. 4). The figure also shows that party loyalty is typically higher among Republican voters, that is, RR is typically more common than DD.
Voting Patterns across Two Presidential Elections.
According to our first hypothesis, vote switchers from the Democrats to the Republicans (DR) are primarily motivated by second-dimension issues. Switchers in the other direction (RD), by contrast, could be motivated by issues on either policy dimension. To test this, we estimate logit models with a binary “direct” vote switching indicator as the dependent variable and economic and noneconomic preferences as the explanatory variables, along with the usual set of controls. We construct two different vote switching indicator variables, which serve as our dependent variables: the first codes defection from the Republican to the Democratic presidential candidate (RD vs. RR), and the second codes defection in the opposite direction (DR vs. DD). We estimate each of the two models (DR switchers; RD switchers) for three different time-periods: 1992–2016 (all available years), 1992–2012 (all available years without the 2016 contest), and 2016 alone. 55 Figure 5 summarizes the results as coefficient plots.
Vote Switching as a Function of Economic and Noneconomic Issues (Coefficient Plot for Different Time-Periods).
In line with our hypothesis concerning cross-party differences in vote switching, close analysis of Figure 5 reveals that different motivations separate those who switch to the Democrats (RD) from those who switch to the Republicans (DR). As hypothesized, it is DR switchers who are more strongly motivated by second-dimension issue positions. This finding applies to all elections, with or without the 2016 presidential election in which, however, the role of second-dimension positions was most overwhelming. By contrast, RD switchers are more evenly motivated by economic and noneconomic considerations, except in the 2016 election, when the latter prevail. In most elections, however, RD switchers find disagreements with the Republicans—as their party of origin—just as important on noneconomic issues. We return below to the reasons why 2016 stands out.
For concreteness, Figure 6 plots predicted probabilities to defect from R to D (left panel) and D to R (right panel) as a function of (individual-level) distance from RR and DD positions, respectively, based on the 1992–2016 estimates. Voters closest to their party of origin have a very small probability of defecting to the opposite side (to the Democrats in the left panel, with positive standard deviations indicating more redistributive first-dimension and more libertarian second-dimension positions; to the Republicans in the right panel, with positive standard deviations meaning more market-liberal first-dimension positions and more authoritarian governance positions on the second dimension). The probability of defection for voters with average distance (0 in the figure) is about 10 percent. That probability rises with increasing distance, but it rises furthest for Democratic defectors to the Republicans based on noneconomic authoritarian issue positions in the right panel (DR defections). A voter with a two-standard-deviation distance from DD positions on the economic dimension has a 23 percent probability of defecting to the Republican candidate and a 33 percent probability with the same distance on noneconomic issues.
Probability of Vote Switching as a Function of Economic and Noneconomic Issues.
Our second hypothesis makes predictions about vote switchers’ policy considerations, depending on their education-income position. Specifically, the theory expects the new swing categories in the knowledge society—low-education/low-income and high-education/high-income voters—to display more switching on second-dimension issues away from the parties whose core support category they were in the industrial society. Because of more authoritarian second-dimension preferences, low-education/low-income voters abandon the Democrats, while former Republican high-education/high-income voters defect motivated by more libertarian second-dimension views than embraced by their party. With regard, however, to high-education/low-income and low-education/high-income voters—the emerging partisan “core” electorates of Democrats and Republicans in the knowledge society—switchers in both categories may be motivated by either economic or noneconomic considerations: low-education/high-income voters prefer the Republicans’ more market oriented and authoritarian governance appeals, while high-education/low-income voters opt for Democrats on the basis of both economic-redistributive and libertarian governance policies.
To assess this proposition, we estimate the same models as in Figure 5 but for each of the four education-income groups in the white electorate separately. 56 Figure 7 displays the relevant coefficients. Since we are interested in the relative size of the coefficients for the four education-income groups, the figure also reports whether the noneconomic coefficient is larger than the economic coefficient (see the symbols next to DR or RD, respectively).
Vote Switching by Education-Income Groups as a Function of Economic and Noneconomic Issues (Coefficient Plot).
We need to disaggregate by party switching direction to get to our critical hypotheses. Among Republicans switching to the Democrats (RD vs. RR), we expect the swing group of high-education/high-income (HE-HI) voters to be particularly motivated by second-dimension considerations, and that indeed turns out to be the case: members of that category abandon the Republicans primarily based on their personal libertarian second-dimension views that are incompatible with an increasingly authoritarian party (bottom panel). All other categories of RD switchers show no difference between the relative influences of first- and second-dimension considerations in motivating moves from the Republicans to the Democrats (top three panels).
For Democrats switching to the Republicans (DR vs. DD), we expect the swing group of low-education/low-income (LE-LI) voters to be motivated primarily by noneconomic issues but less so the other categories. Results reported in the top panel reveal that, indeed, for this swing group the noneconomic considerations are much more important to switch to the Republicans than economic issues. The coefficient of noneconomic issues is 2.6 times larger than the coefficient of economic issues, and the relative difference in issue impact is bigger than that for any of the other education-income categories of DR switchers. Nevertheless, noneconomic issues also outweigh economic issues for DR switchers in the Democrats’ core constituency in the knowledge society—high-education/low-income voters (HE-LI). By contrast, the white low-education/high-income (LE-HI) category whose members abandon the Democrats in droves (DR) appear to do so more on economic rather than noneconomic considerations (although the relevant coefficients are statistically not significantly different). 57
Our third hypothesis argues that vote switching motivated by noneconomic considerations is more likely if parties take—or are perceived to voice—similar positions on economic issues, encouraging voters to choose based on second-dimension differences between party positions. We have already observed that in 2016 the probability that switchers voted on noneconomic issue positions—and especially, but not exclusively, on racial bias 58 —was particularly large (Fig. 5). This unique aspect of the 2016 election plausibly accounts for the particularly pronounced swing of low-education/low-income voters to Trump and the particularly pronounced swing of high-education/high-income voters in the other direction (Fig. 3).
No data exist to test this proposition directly. However, we can probe its plausibility indirectly, using, once again, the ANES data. The ANES survey asks respondents to place the Democratic and Republican candidates as well as their parties on a seven-point liberal-conservative scale (which correlates strongly with both the economic and noneconomic scale). Figure 8 displays average placements for each available year, disaggregated by switcher/standpatter partisan voting.
Perceptions of Candidates and Parties by Voting Behavior.
What is noteworthy in Figure 8 is that all voting categories—except Republican standpatters (RR)—perceived the Republicans’ 2016 presidential candidate Donald Trump as substantially more moderate than his party, and as more moderate than most Republican presidential candidates since 1980. For those Democrats who switched to the Republican candidate in 2016 (DR), this unique placement of the Republican candidate is particularly stark. DR vote switchers at no time since 1972 perceived a Republican candidate in a more moderate position (Fig. 8, lower right panel).
At first sight, the fact that DR switchers in 2016 perceived Trump as a centrist, however, may be interpreted as a matter of motivated reasoning and dissonance reduction: since most DR switchers were less educated whites with lower incomes, and thereby are likely to prefer some measure of economic redistribution, they might have wanted to make themselves believe that Trump was moderate on economic policy and therefore placed him in an overall centrist position. This perception would have removed cognitive dissonance and inhibitions to vote for the Republican based on sociopolitically authoritarian, and often racist, positions that were served by Trump’s rhetoric. But DR switchers were not the only ones who perceived the Republican presidential candidate, as opposed to the Republican Party, as moderate in 2016. Even voter categories that had no motivational incentives interpreted Trump as moderate: Democratic standpatters (DD) and even former Republicans switching to the Democratic candidate (RD). Only Republican standpatters (RR) saw no difference between the Republican Party and its presidential candidate, in line with what motivated reasoning theory would expect for this group.
The perception of Trump’s centrist stances on economic issues was particularly consequential for low-education/low-income voters, the “swing” category in the knowledge society. These voters are tempted to defect from the center-left party when cued by a right-wing alternative’s more authoritarian second-dimension appeal. In 2016, a perceived convergence of the Republican presidential candidate on centrist economic positions—underlined by promises to protect the public pension system in the United States and to replace the Affordable Care Act (“Obamacare”) by medical care provisions that are cheaper and more accessible to hitherto uncovered groups—made it easier for low-education/low-income voters to choose the Republican based on second-dimension considerations.
The empirics of vote switching in 2016 bear out this expectation. Table 2 shows the average size (as a percentage of the white ANES voter sample) and positions on the two policy dimensions (standardized between 0 and 1) of respondents belonging to the four major vote patterns (whites only): RR (Romney/Trump: 47.6 percent); DD (Obama/Clinton: 40.8 percent); RD (Romney/Clinton: 3.1 percent); and DR (Obama/Trump: 8.5 percent). The table disaggregates the largest category, DR Obama-Trump voters, by education-income categories. On average, DR Obama-Trump voters have second-dimension positions that are similar to those of RR voters, albeit somewhat less extreme. On the first dimension, however, white low-education/low-income DR voters are close to the Democratic Party and candidate appeal, when compared to the other socioeconomic DR switcher categories, but also to RR or even RD voters more broadly. The white working-class DR voters hence did not abandon the Democrats based on their economic preferences. They switched in light of noneconomic considerations, and this is likely to have been encouraged by the Republican presidential candidate’s moderate posturing on economics and social policy in the 2016 campaign.
Policy Position of Certain Vote Switcher Groups in 2016.
Overall, considerations of one’s own well-understood economic and noneconomic interests appear to affect the conduct of voters. Over a sequence of elections, parties attract—through programmatic appeals on both policy dimensions—new partisan “core” groups whose socioeconomic endowments align. Figure 3 documents the resulting realignment between the two parties and the education-income groups. We believe it is the result of at least three mechanisms, and we have empirically documented the third—and most challenging—in this article. First, generational change, a process in which younger voters with more aligned preferences replace older voters, can result in realignment. In particular, younger low-education/high-income and low-education/low-income voters are more likely to support the Republican presidential candidates than their older peers. Equivalently, younger high-education/high-income and high-education/low-income voters are more likely to support the Democratic presidential candidate than their older peers. 59 Second, the parties’ new positions alienate and demobilize some voters but attract and turn out others. Defection from one party to another does not have to be instantaneous but can play out over a series of elections with the same logic as direct vote switching. 60
Third, realignment stems from direct vote switching when voters change their partisan support from one election to the next, motivated by issue positions. We have focused on direct vote switching, because it is the least likely scenario for critics of programmatic approaches to voting behavior, and we have found much evidence that is consistent with our theoretical framework. Over a number of elections, members of the new partisan core categories whose socioeconomic endowments align with one of the parties’ programmatic appeals on both policy dimensions switch toward the supplier that satisfies their demands. By contrast, the new swing categories are more hesitant to realign because none of the parties fully satisfies their preference schedules. 61 If low-education/low-income voters switch party allegiance away from their preferred choice on economic-distributive grounds (the Democrats), it is because of second-dimension considerations that have gained salience with the erosion of the New Deal alignment in industrial society. A perceived convergence of the Republican candidate on centrist, moderate economic-distributive policies and a polarization of the two candidates’ positions on second-dimension issues accelerates the realignment, as evidenced by the 2016 Trump-Clinton contest.
Conclusion
We have proposed a simple characterization of the two US parties’ electorates—four groups defined by the interaction of education and income—that (we believe) goes a long way in properly characterizing the long-term realignment process. We also have empirically evaluated various predictions that stem from our theoretical framework—both at the macro and micro level—and have found much evidence consistent with them. Therefore, we are carefully optimistic that our account is useful for making sense of the realignment process in the United States over the last few decades. We also believe that it advances the three vibrant larger debates on American voting behavior, as well as the policy debate, that we mentioned at the beginning of this article.
First, with respect to the “what’s the matter with . . .” debates—and, more generally, the question to what degree does income structure voting behavior 62 —our account takes a clear position: a focus on income or education alone misses critical aspects of the white American realignment process. Neither the high-education/low-income nor the low-education/high-income group features in the literature. But it is these groups that become the white “core” electoral constituencies of the two American parties.
More specifically, Larry Bartels appears to be more correct than Thomas Frank but for the wrong reasons. It is certainly the case that white low-income voters still have a greater propensity to support the Democrats than high-income voters, but not because low-skill blue-collar and clerical workers stick to the party in exceptional proportions. It is rather because high-education/low-income service professionals become the left-wing spearhead of the Democratic Party. Hence, Frank is at least partially right in that a critical segment of low-income whites gradually abandons their traditional partisan homes, while both miss the Democratic leanings of the high-education/low-income category.
Conversely, both Frank and Bartels—and others 63 —insufficiently appreciate the deep divide among affluent white voters. True, the majority of the affluent still supports the Republicans, but it is a subgroup never clearly distinguished in the Frank-Bartels debate—the low-education/high-income whites—that expresses this partisan preference, whereas the high-education/high-income whites started trending the other way toward the Democrats some time ago. The Frank-Bartels debate thus fails to describe properly the two “core” electorates of the Democrats and Republicans in the knowledge society. Instead, they are preoccupied with the New Deal core groups, gradually becoming electoral swing groups in the knowledge society, exhibiting more divided and volatile partisan affiliations with Democrats and Republicans.
Second, with respect to the debate whether ties between voters and parties are programmatic or based on identity politics, we strongly side with the former view. In our theoretical framework, electoral competition is programmatic by assumption. But we generate specific predictions about what kind of voters are motivated by what kind of issues under what kind of circumstances—and these hypotheses seem to have strong empirical support. We therefore conclude that there is plenty of evidence consistent with what Christopher Achen and Larry Bartels dismiss as a “folk theory”: 64 issue-position-based partisan accountability (“responsible partisan governance”).
In our opinion, Achen and Bartels are more successful in offering examples that cast doubt on the responsible partisan governance perspective than in offering a crisp identity theory that could be contrasted with it. For example, do politicians’ appeals to certain groups—defined by class, geography (urban/rural), religion, ethnicity, race, and so forth—resonate with voters because they trigger a sense of identity or because they signal policy positions close to those groups? In other words, identity appeals and policy objectives are closely intertwined, and identities may serve as a useful heuristic for navigating a complex political landscape to pursue one’s interests. 65 Most of the macro-level and micro-level white voter realignment in the United States, in transition from an industrial to a knowledge society, therefore can be told both as a story of realigning group identities with parties—of the white working class (low-education/low-income), the petty bourgeoisie (low-education/high-income), the urban bohemian bourgeois (high-education/low-income or high-income), and a global finance professional caste (high-education/high income)—as well as an account of realigning economic and noneconomic interests around competing partisan packages. And for both accounts, at the micro level, only a minority of group members need to go through the trouble of decoding the politicians’ appeals in order to create macro-level patterns of group alignments consistent with an interest-based account. 66
Regarding the identity debate, perhaps it makes sense to distinguish a comparative-static and a dynamic perspective. In a comparative-static perspective, at every point, fixed and often affective group and party identifications explain the bulk of the variance in voter choices. In a dynamic perspective, however, the conduct of vote switchers demonstrates that, at the margin, partisan policy messages and voter information processing are relevant for incremental realignment. In our account, generational replacement, (de)mobilization, and the small number of vote switchers bring about realignment over a long period of time. Our framework suggests that vote switchers’ motivations and orientations are programmatic, and our empirical analysis yields evidence consistent with that claim. Over a sequence of elections, the actions of small, moderately informed critical minorities in each education-by-income category generate cumulative effects that manifest themselves in partisan realignments among socioeconomic and ethnic groups and that are consistent with a responsible partisan government process of accountability.
Third, what, if anything, was special about the 2016 election? From a historical perspective, and based on our theoretical framework, the observed voting behavior in 2016 continues a long-term trend, although it does so at an accelerated pace. For example, the strong support of the “white working class” (low-education/low-income, in our framework) for Trump is surprising only if one fails to realize that this group stopped being the core constituency of the Democratic Party quite some time ago. On the other hand, that the Democrats lost low-education/low-income voters by double digits is unprecedented.
Nevertheless, there are specific features of the 2016 election. The accelerated realignment trends in 2016 are due to the exceptional importance of noneconomic issues 67 in combination with the unique perceived convergence of the candidates on economic issues (as discussed in Hypothesis 3). Trump, indeed, was not a run-of-the-mill Republican candidate and, at the margin, could broaden the party’s appeal in ways that won him the primary election season and the Electoral College majority, albeit not the popular vote. Under these conditions, we have argued, the polarity reversal—core groups become swing groups; swing groups become core groups—advances faster.
Let us close on a note concerning party strategy. Inside established center-left and center-right parties that were hegemonic in the era of industrial society and New Deal politics, heated debates are ensuing on how to confront the imperatives of political realignment, particularly where these parties have lost electoral support. A “nostalgic” camp advocates a return to tried-and-true politics on the economic-distributive dimension emphasizing contrasting leftist and rightist positions concerning universalistic social policies (pensions, health care, education, family, and unemployment) and taxation. An opposed “progressivist” camp argues for abandoning the economic-distributive dimension entirely and positioning the partisan alternatives based on cultural identities embedded in a broader set of divides between libertarians and authoritarians on noneconomic issues, while occupying a technocratic middle ground on economic issues.
If our analysis of divisions in the electorate is correct, politicians in a knowledge society will not enjoy the luxury of reducing the political-issue space to one dimension. This prediction applies particularly in polities with low entry thresholds for new competitors. European social democrats may win back working-class voters with first-dimension economic-redistributive politics, but only at the expense of seeing high-education/lower-income sociocultural professionals flock to the Green Party and others with more libertarian appeals. Conversely, Christian Democrats and Conservatives may invoke low taxation and private property, but they will lose many voters if they do not offer answers to questions of immigration, multiculturalism, and law and order.
Under American institutional supply-side and historical demand-side conditions, the conflicts of two-dimensional politics surface in intraparty factionalism rather than fragmentation. And there is no easy way out of the agony of intraparty conflict on both sides of the aisle. A victory of those who push for identity politics only, as much as a victory of those who demand pure strategies of economic policymaking, is likely to alienate critical voter groups that are decisive for each party’s victory or loss in the zero-sum game over control of legislative seats and the presidential office.
Footnotes
Appendix A
Appendix B
Logit Estimation for Predicted Probabilities in Figure 1.
| Vote for Democratic Candidate | |
|---|---|
| First dimension (progressive to conservative) | −1.125
**
(0.051) |
| Second dimension (conservative to progressive) | 1.152
**
(0.052) |
| LE-HI | −0.178 (0.110) |
| HE-LI | −0.175 (0.108) |
| HE-HI | −0.339
**
(0.104) |
| Female | 0.045 (0.075) |
| South | −0.316
**
(0.083) |
| Age | 0.005
*
(0.002) |
| Frequent church attendance | −0.793
**
(0.099) |
| Born again | −0.548
**
(0.0−91) |
| Year fixed effects | Yes |
| Constant | 0.208 (0.159) |
| Pseudo R2 | 0.369 |
| Observations | 7,310 |
#p < .1, *p < 0.05, **p <0.01.
Appendix C
Logit Estimation on Which Predicted Probabilities in Figure 3 Are Based.
| LE-HI | −0.166(0.133) |
| HE-LI | −0.443(0.527) |
| HE-HI | −0.863 ** (0.280) |
| LE-HI × year=1956 | 0.080(0.192) |
| LE-HI × year=1960 | 0.140(0.225) |
| LE-HI × year=1964 | −0.457 * (0.206) |
| LE-HI × year=1968 | 0.259(0.216) |
| LE-HI × year=1972 | −0.197(0.194) |
| LE-HI × year=1976 | −0.331 # (0.198) |
| LE-HI × year=1980 | −0.087(0.246) |
| LE-HI × year=1984 | −0.453 * (0.213) |
| LE-HI × year=1988 | −0.230(0.225) |
| LE-HI × year=1992 | −0.373 # (0.224) |
| LE-HI × year=1996 | −0.142(0.264) |
| LE-HI × year=2000 | −0.133(0.285) |
| LE-HI × year=2004 | −0.552 # (0.325) |
| LE-HI × year=2008 | −0.670 * (0.297) |
| LE-HI × year=2012 | 0.070(0.204) |
| LE-HI × year=2016 | 0.112(0.222) |
| HE-LI × year=1956 | 0.208(0.647) |
| HE-LI × year=1960 | 0.742(0.770) |
| HE-LI × year=1964 | −0.483(0.629) |
| HE-LI × year=1968 | 0.124(0.607) |
| HE-LI × year=1972 | 0.601(0.575) |
| HE-LI × year=1976 | 0.206(0.585) |
| HE-LI × year=1980 | 0.348(0.593) |
| HE-LI × year=1984 | 0.930(0.573) |
| HE-LI × year=1988 | 0.483(0.576) |
| HE-LI × year=1992 | 0.360(0.572) |
| HE-LI × year=1996 | 0.553(0.579) |
| HE-LI × year=2000 | 0.757(0.574) |
| HE-LI × year=2004 | 0.583(0.593) |
| HE-LI × year=2008 | 0.621(0.580) |
| HE-LI × year=2012 | 0.842(0.547) |
| HE-LI × year=2016 | 1.313 * (0.552) |
| HE-HI × year=1956 | 0.253(0.375) |
| HE-HI × year=1960 | 0.114(0.432) |
| HE-HI × year=1964 | −0.509(0.371) |
| HE-HI × year=1968 | 0.465(0.371) |
| HE-HI × year=1972 | 0.849 * (0.339) |
| HE-HI × year=1976 | −0.102(0.347) |
| HE-HI × year=1980 | 0.285(0.402) |
| HE-HI × year=1984 | 0.521(0.347) |
| HE-HI × year=1988 | 0.546(0.345) |
| HE-HI × year=1992 | 0.214(0.334) |
| HE-HI × year=1996 | 0.512(0.357) |
| HE-HI × year=2000 | 1.123 ** (0.352) |
| HE-HI × year=2004 | 0.954 * (0.372) |
| HE-HI × year=2008 | 0.307(0.372) |
| HE-HI × year=2012 | 1.088 ** (0.313) |
| HE-HI × year=2016 | 1.783 ** (0.315) |
| Constant | 0.196(0.238) |
| Pseudo R2 | 0.058 |
| Observations | 17,998 |
Estimation includes variables for age, gender, South, and frequent church attendance, each interacted with election year.
Standard errors in parentheses. Whites only.
p < .1, *p < 0.05, **p < 0.01.
Appendix D
The ANES information does not allow us to construct an occupational variable that covers a long time series. The General Social Survey (GSS) 68 cumulative file 1972–2018, in contrast, does, but it is not a voting survey. However, it contains survey items about vote choice (and hypothetical vote choice) in the last presidential election (if68, pres72, pres76, pres80, pres84, pres88, pres92, pres96, pres00, pres04, pres08, pres12, pres16; if68who, if72who, if76who, if80who, if84who, if88who, if92who, if96who, if00who, if04who, if08who, if12who, if16who). They let us reproduce Figure 3 using occupational instead of education-income groups (Fig. D1).
Appendix E
In Table E1, the two low-income groups (LE-LI + HE-LI) do not necessarily add up to 66 percent because the sample is restricted to white voters who cast a two-party vote, while low- and high-income are defined relative to the entire income distribution. Table E2 shows the size of low- and high-income groups for different sample restrictions.
Appendix F
| 1992–2016 |
1992–2016 |
1992–2016 |
1992–2012(4) | 1992–2012(5) | 1992–2012(6) | 2016 |
2016 |
2016 |
|
|---|---|---|---|---|---|---|---|---|---|
| Dependent variable | DR = 1, RD = 1; DD = 0, RR = 0 | RD = 1, RR = 0 | DR = 1, DD = 0 | DR = 1, RD = 1; DD = 0, RR = 0 | RD = 1, RR = 0 | DR = 1, DD = 0 | DR = 1, RD = 1; DD = 0, RR = 0 | RD = 1, RR = 0 | DR = 1, DD = 0 |
| Economic dimension | 0.674
**
(0.044) |
0.712
**
(0.064) |
0.567
**
(0.067) |
0.723
**
(0.050) |
0.746
**
(0.069) |
0.589
**
(0.078) |
0.431
**
(0.105) |
0.543
*
(0.192) * |
0.523
**
(0.141) |
| Noneconomic dimension | 0.864
**
(0.050) |
0.825
**
(0.078) |
0.942
**
(0.079) |
0.702
**
(0.055) |
0.709
**
(0.083) |
0.845
**
(0.093) |
1.527
**
(0.125) |
1.672
**
(0.260) |
1.239
**
(0.163) |
| Female | 0.052 (0.087) |
0.036 (0.128) |
0.038 (0.124) |
0.045 (0.097) |
-0.013 (0.137) |
0.079 (0.146) |
0.153 (0.203) |
0.537 (0.366) |
−0.034 (0.253) |
| South | 0.171
#
(0.094) |
0.203 (0.133) |
0.187 (0.139) |
0.120 (0.105) |
0.172 (0.144) |
0.077 (0.162) |
0.508
*
(0.221) |
0.385 (0.373) |
0.649
*
(0.288) |
| Age | −0.009
**
(0.003) |
−0.010
*
(0.004) |
−0.007
#
(0.004) |
−0.009
**
(0.003) |
−0.010
*
(0.004) |
−0.006 (0.005) |
−0.009 (0.006) |
−0.010 (0.011) |
−0.013 (0.008) |
| Frequent church attendance | −0.489
**
(0.110) |
−0.852
**
(0.153) |
0.076 (0.170) |
−0.510
**
(0.121) |
−0.787
**
(0.161) |
−0.043 (0.196) |
−0.339 (0.276) |
−1.243
*
(0.539) |
0.540 (0.370) |
| Born again | 0.046 (0.100) |
−0.238
#
(0.141) |
0.457
**
(0.151) |
0.101 (0.110) |
−0.237 (0.150) |
0.633
**
(0.175) |
−0.215 (0.246) |
−0.252 (0.430) |
0.121 (0.321) |
| LE-HI | −0.174 (0.118) |
−0.268 (0.171) |
−0.103 (0.174) |
−0.167 (0.133) |
−0.410
*
(0.184) |
0.088 (0.207) |
−0.054 (0.271) |
1.196
*
(0.544) |
−0.561 (0.343) |
| HE-LI | −0.210 (0.132) |
−0.424
*
(0.198) |
0.134 (0.192) |
−0.181 (0.147) |
−0.474
*
(0.213) |
0.332 (0.224) |
−0.238 (0.310) |
0.232 (0.623) |
−0.358 (0.386) |
| HE-HI | −0.363
**
(0.119) |
−0.647
**
(0.178) |
−0.002 (0.183) |
−0.313
*
(0.135) |
−0.891
**
(0.199) |
0.438
*
(0.211) |
−0.483
#
(0.263) |
0.635 (0.536) |
−1.116
**
(0.378) |
| Constant | −1.259
**
(0.193) |
−0.359 (0.261) |
−3.179
**
(0.361) |
−1.210
**
(0.210) |
−0.236 (0.276) |
−3.346
**
(0.398) |
−2.683
**
(0.432) |
−3.813
**
(0.840) |
−1.964
**
(0.526) |
| Year fixed effects | Yes | Yes | Yes | Yes | Yes | Yes | No | No | No |
| AIC | 3,762.130 | 1,859.969 | 1,793.613 | 3,034.227 | 1,591.581 | 1,339.965 | 700.120 | 249.829 | 442.279 |
| Pseudo R2 | 0.197 | 0.237 | 0.213 | 0.175 | 0.237 | 0.178 | 0.326 | 0.308 | 0.337 |
| Log likelihood | −2.3e+03 | −1.2e+03 | −1.1e+03 | −1.8e+03 | −1.0e+03 | −797.056 | −502.687 | −164.674 | −316.833 |
| N | 6,197 | 3,364 | 2,833 | 4,620 | 2,568 | 2,052 | 1,577 | 796 | 781 |
p < .1, *p < 0.05, **p < 0.01.
Appendix G
Logit Models Underlying Figure 7.
| LE-LI |
LE-HI |
HE-LI |
HE-HI |
LE-LI |
LE-HI |
HE-LI |
HE-HI |
LE-LI |
LE-HI |
HE-LI |
HE-HI |
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Dependent variable | DR = 1, RD = 1, DD = 0, RR = 0 | DR = 1, RD = 1, DD = 0, RR = 0 | DR = 1, RD = 1, DD = 0, RR = 0 | DR = 1, RD = 1, DD = 0, RR = 0 | RD = 1, RR = 0 | RD = 1, RR = 0 | RD = 1, RR = 0 | RD = 1, RR = 0 | DR = 1, DD = 0 | DR = 1, DD = 0 | DR = 1, DD = 0 | DR = 1, DD = 0 |
| Distance on economic dimension | 0.521
**
(0.060) |
0.805
**
(0.108) |
0.945
**
(0.145) |
0.898
**
(0.115) |
0.623
**
(0.090) |
0.839
**
(0.157) |
1.291
**
(0.238) |
0.645
**
(0.170) |
0.354
**
(0.091) |
0.828
**
(0.170) |
0.868
**
(0.237) |
1.202
**
(0.203) |
| Distance on noneconomic dimension | 0.807
**
(0.070) |
0.552
**
(0.110) |
1.334
**
(0.169) |
1.062
**
(0.126) |
0.679
**
(0.117) |
0.735
**
(0.175) |
1.286
**
(0.249) |
1.057
**
(0.192) |
0.929
**
(0.118) |
0.299
#
(0.168) |
1.485
**
(0.267) |
1.218
**
(0.200) |
| Female | 0.285
*
(0.125) |
−0.215 (0.204) |
−0.224 (0.256) |
−0.178 (0.209) |
0.213 (0.192) |
−0.367 (0.296) |
−0.133 (0.399) |
0.111 (0.289) |
0.384
*
(0.180) |
−0.199 (0.299) |
−0.579 (0.374) |
−0.500 (0.326) |
| South | 0.089 (0.133) |
−0.006 (0.244) |
0.708
**
(0.264) |
0.386
#
(0.220) |
−0.202 (0.198) |
0.410 (0.335) |
1.080
**
(0.412) |
0.755
*
(0.297) |
0.328
#
(0.193) |
−0.256 (0.386) |
0.631 (0.400) |
−0.181 (0.390) |
| Age | −0.008
*
(0.004) |
−0.006 (0.008) |
−0.004 (0.008) |
−0.016
*
(0.008) |
−0.004 (0.005) |
−0.009 (0.011) |
−0.024
*
(0.012) |
−0.029
**
(0.010) |
−0.010
#
(0.005) |
−0.003 (0.011) |
0.012 (0.011) |
0.006 (0.013) |
| Frequent church attendance | −0.663
**
(0.159) |
−0.124 (0.247) |
−1.118
**
(0.347) |
−0.035 (0.249) |
−1.288
**
(0.245) |
−0.305 (0.331) |
−1.761
**
(0.520) |
−0.120 (0.323) |
0.032 (0.235) |
0.318 (0.411) |
−0.270 (0.554) |
0.450 (0.437) |
| Born again | 0.098 (0.136) |
−0.250 (0.238) |
0.571
*
(0.288) |
−0.029 (0.276) |
−0.207 (0.201) |
−0.455 (0.335) |
0.132 (0.414) |
−0.116 (0.352) |
0.594
**
(0.202) |
0.169 (0.376) |
1.103
*
(0.471) |
0.187 (0.517) |
| Year fixed effects | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| Constant | −1.210
**
(0.257) |
−1.173
**
(0.429) |
−2.201
**
(0.539) |
−1.767
**
(0.461) |
−0.190 (0.358) |
−0.572 (0.578) |
−1.390
#
(0.799) |
−1.142
#
(0.589) |
−3.435
**
(0.504) |
−2.981
**
(0.849) |
−3.317
**
(0.905) |
−4.428
**
(1.093) |
| Pseudo R2 | 0.160 | 0.183 | 0.340 | 0.269 | 0.270 | 0.235 | 0.375 | 0.234 | 0.175 | 0.176 | 0.389 | 0.391 |
| Observations | 2,477 | 1,021 | 1,096 | 1,603 | 1,369 | 658 | 520 | 817 | 1,108 | 363 | 576 | 786 |
p < .1, *p < 0.05, **p < 0.01.
Acknowledgements
We would like to thank the editors for helpful comments and suggestions. We would like to thank participants at the workshop “From Preferences to Politics in the New Economy” at Nuffield College, Oxford University, July 2017. We would also like to thank discussants and audiences at the APSA Annual Meetings in 2013 and 2018. Philipp Rehm would like to acknowledge the hospitality of both Princeton University’s Department of Politics and the Hanse-Wissenschaftskolleg.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Philipp Rehm would like to acknowledge the support of the EURIAS program (co-funded by Marie Skłodowska-Curie Actions, under the 7th Framework Programme).